Microsoft Programmer's Library 1.3/Microsoft_Programmers_Library.7z/MPL/masm/masmpgmr.txt

Text File | 2013-11-08 | 1.8 MB | 46,684 lines

 Microsoft  Macro Assembler - Programmer's Guide








 ────────────────────────────────────────────────────────────────────────────
              Microsoft (R) Macro Assembler - Programmer's Guide

                                 Version 6.0
 ────────────────────────────────────────────────────────────────────────────


               For MS (R) OS/2 and MS-DOS (R) Operating Systems








 Microsoft Corporation

 Information in this document is subject to change without notice and does
 not represent a commitment on the part of Microsoft Corporation. The
 software described in this document is furnished under a license agreement
 or nondisclosure agreement. The software may be used or copied only in
 accordance with the terms of the agreement. It is against the law to copy
 the software on any medium except as specifically allowed in the license or
 nondisclosure agreement. No part of this manual may be reproduced or
 transmitted in any form or by any means, electronic or mechanical, including
 photocopying and recording, for any purpose without the express written
 permission of Microsoft.
 RESTRICTED RIGHTS: Use, duplication, or disclosure by the U.S. Government is
 subject to restrictions as set forth in subparagraph (c)(1)(ii) of the
 Rights in Technical Data and Computer Software clause at DFARS 252.227-7013
 or subparagraphs (c)(1) and (2) of Commercial Computer Software
 ─Restricted Rights at 48 CFR 52.227-19, as applicable.
 Contractor/Manufacturer is Microsoft Corporation, One Microsoft Way,
 Redmond, WA  98052-6399.


 (C) Copyright Microsoft Corporation, 1991. All rights reserved.

 Printed in the United States of America.


 Microsoft, MS, MS-DOS, CodeView, QuickC,
 and XENIX are registered trademarks and Making it all make sense,
 Microsoft QuickBasic, QuickPascal, and Windows are trademarks of Microsoft
 Corporation.

 U.S. Patent No. 4,955,066

 Hercules is a registered trademark of Hercules Computer
 Technology.

 IBM is a registered trademark of International Business
 Machines Corporation.

 Intel is a registered trademark of Intel Corporation.

 NEC and V25 are registered trademarks and V35 is a trademark
 of NEC Corporation.

 Document No. LN06556-0291

 10 9 8 7 6 5 4 3 2 1







 Introduction
      New and Extended Features in MASM 6.0
             New MASM Language Features
             ML and MASM Command Lines
             Compatibility with Earlier Versions of MASM
      Scope and Organization of this Book
      Books for Further Reading
      Document Conventions
      Getting Assistance and Reporting Problems

 Chapter 1  Understanding Global Concepts

      1.1   The Processing Environment
             1.1.1    8086-Based Processors
             1.1.2    Operating Systems
             1.1.3    Segmented Architecture
             1.1.4    Segment Protection
             1.1.5    Segmented Addressing
             1.1.6    Segment Arithmetic
      1.2   Language Components of MASM
             1.2.1    Reserved Words
             1.2.2    Identifiers
             1.2.3    Predefined Symbols
             1.2.4    Integer Constants and Constant Expressions
             1.2.5    Operators
             1.2.6    Data Types
             1.2.7    Registers
             1.2.8    Statements
      1.3   The Assembly Process
             1.3.1    Generating and Running Executable Programs
             1.3.2    Using the OPTION Directive
             1.3.3    Conditional Directives
      1.4   Related Topics in Online Help

 Chapter 2  Organizing MASM Segments

      2.1   Overview of Memory Segments
      2.2   Using Simplified Segment Directives
             2.2.1    Defining Basic Attributes with .MODEL
             2.2.2    Specifying a Processor and Coprocessor
             2.2.3    Creating a Stack
             2.2.4    Creating Data Segments
             2.2.5    Creating Code Segments
             2.2.6    Starting and Ending Code with .STARTUP and .EXIT
      2.3   Using Full Segment Definitions
             2.3.1    Defining Segments with the SEGMENT Directive
             2.3.2    Controlling the Segment Order
             2.3.3    Setting the ASSUME Directive for Segment Registers
             2.3.4    Defining Segment Groups
      2.4   Related Topics in Online Help

 Chapter 3  Using Addresses and Pointers

      3.1   Programming Segmented Addresses
             3.1.1    Initializing Default Segment Registers
             3.1.2    Near and Far Addresses
      3.2   Specifying Addressing Modes
             3.2.1    Register Operands
             3.2.2    Immediate Operands
             3.2.3    Direct Memory Operands
             3.2.4    Indirect Memory Operands
      3.3   Accessing Data with Pointers and Addresses
             3.3.1    Defining Pointer Types with TYPEDEF
             3.3.2    Defining Register Types with ASSUME
             3.3.3    Basic Pointer and Address Operations
      3.4   Related Topics in Online Help

 Chapter 4  Defining and Using Integers

      4.1   Declaring Integer Variables
             4.1.1    Allocating Memory for Integer Variables
             4.1.2    Data Initialization
      4.2   Integer Operations
             4.2.1    Moving and Loading Integers
             4.2.2    Pushing and Popping Stack Integers
             4.2.3    Adding and Subtracting Integers
             4.2.4    Multiplying and Dividing Integers
      4.3   Manipulating Integers at the Bit Level
             4.3.1    Logical Operations
             4.3.2    Shifting and Rotating Bits
             4.3.3    Multiplying and Dividing with Shift Instructions
      4.4   Related Topics in Online Help

 Chapter 5  Defining and Using Complex Data Types

      5.1   Arrays and Strings
             5.1.1    Declaring and Referencing Arrays
             5.1.2    Declaring and Initializing Strings
             5.1.3    Processing Arrays and Strings
      5.2   Structures and Unions
             5.2.1    Declaring Structure and Union Types
             5.2.2    Defining Structure and Union Variables
             5.2.3    Referencing Structures, Unions, and Fields
             5.2.4    Nested Structures and Unions
      5.3   Records
             5.3.1    Declaring Record Types
             5.3.2    Defining Record Variables
             5.3.3    Record Operators
      5.4   Related Topics in Online Help

 Chapter 6  Using Floating-Point and Binary Coded Decimal Numbers

      6.1   Using Floating-Point Numbers
             6.1.1    Declaring Floating-Point Variables and Constants
             6.1.2    Storing Numbers in Floating-Point Format
      6.2   Using a Math Coprocessor
             6.2.1    Coprocessor Architecture
             6.2.2    Instruction and Operand Formats
             6.2.3    Coordinating Memory Access
             6.2.4    Using Coprocessor Instructions
      6.3   Using Emulator Libraries
      6.4   Using Binary Coded Decimal Numbers
             6.4.1    Defining BCD Constants and Variables
             6.4.2    Calculating with BCDs
      6.5   Related Topics in Online Help

 Chapter 7  Controlling Program Flow

      7.1   Jumps
             7.1.1    Unconditional Jumps
             7.1.2    Conditional Jumps
      7.2   Loops
             7.2.1    Loop-Generating Directives
             7.2.2    Writing Loop Conditions
      7.3   Procedures
             7.3.1    Defining Procedures
             7.3.2    Passing Arguments on the Stack
             7.3.3    Declaring Parameters with the PROC Directive
             7.3.4    Using Local Variables
             7.3.5    Creating Local Variables Automatically
             7.3.6    Declaring Procedure Prototypes
             7.3.7    Calling Procedures with INVOKE
             7.3.8    Generating Prologue and Epilogue Code
      7.4   DOS Interrupts
             7.4.1    Calling DOS and ROM-BIOS Interrupts
             7.4.2    Replacing or Redefining Interrupt Routines
      7.5   Related Topics in Online Help

 Chapter 8  Sharing Data and Procedures among Modules and Libraries

      8.1   Selecting Data-Sharing Methods
      8.2   Sharing Symbols with Include Files
             8.2.1    Organizing Modules
             8.2.2    Declaring Symbols Public and External
             8.2.3    Positioning External Declarations
      8.3   Using Alternatives to Include Files
             8.3.1    PUBLIC and EXTERN
             8.3.2    Other Alternatives
      8.4   Developing Libraries
             8.4.1    Associating Libraries with Modules
             8.4.2    Using EXTERN with Library Routines
      8.5   Related Topics in Online Help

 Chapter 9  Using Macros

      9.1   Text Macros
      9.2   Macro Procedures
             9.2.1    Creating Macro Procedures
             9.2.2    Passing Arguments to Macros
             9.2.3    Specifying Required and Default Parameters
             9.2.4    Defining Local Symbols in Macros
      9.3   Assembly Time Variables and Macro Operators
             9.3.1    Text Delimiters (< >) and the Literal-Character
                         Operator (!)
             9.3.2    Expansion Operator (%)
             9.3.3    Substitution Operator (&)
      9.4   Defining Repeat Blocks with Loop Directives
             9.4.1    REPEAT Loops
             9.4.2    WHILE Loops
             9.4.3    FOR Loops and Variable-Length Parameters
             9.4.4    FORC Loops
      9.5   String Directives and Predefined Functions
      9.6   Returning Values with Macro Functions
      9.7   Advanced Macro Techniques
             9.7.1    Nesting Macro Definitions
             9.7.2    Testing for Argument Type and Environment
             9.7.3    Using Recursive Macros
      9.8   Related Topics in Online Help

 Chapter 10  Managing Projects with NMAKE

      10.1  Overview of NMAKE
      10.2  Running NMAKE
      10.3  NMAKE Description Files
             10.3.1    Description Blocks
             10.3.2    Pseudotargets
             10.3.3    Comments
             10.3.4    Macros
             10.3.5    Inference Rules
             10.3.6    Directives
             10.3.7    Preprocessing Directives
             10.3.8    Extracting Filename Components
      10.4  Command-Line Options
      10.5  NMAKE Command File
      10.6  The TOOLS.INI File
      10.7  Inline Files
      10.8  Sequence of NMAKE Operations
      10.9  A Sample NMAKE Description File
      10.10 Differences between NMAKE and MAKE
      10.11 Using NMK
      10.12 Using Exit Codes with NMAKE
      10.13 Related Topics in Online Help

 Chapter 11  Creating Help Files with HELPMAKE

      11.1  Structure and Contents of a Help Database
             11.1.1    Contents of a Help File
             11.1.2    Help File Formats
      11.2  Invoking HELPMAKE
      11.3  HELPMAKE Options
             11.3.1    Options for Encoding
             11.3.2    Options for Decoding
             11.3.3    Options for Help
      11.4  Creating a Help Database
      11.5  Help Text Conventions
             11.5.1    Structure of the Help Text File
             11.5.2    Local Contexts
             11.5.3    Context Prefixes
             11.5.4    Hyperlinks
      11.6  Using Help Database Formats
             11.6.1    QuickHelp Format
             11.6.2    Rich Text Format
             11.6.3    Minimally Formatted ASCII Format
      11.7  Related Topics in Online Help

 Chapter 12  Linking Object Files with LINK

      12.1  Overview
      12.2  LINK Output Files
      12.3  LINK Syntax and Input
             12.3.1    The objfiles Field
             12.3.2    The exefile Field
             12.3.3    The mapfile Field
             12.3.4    The libraries Field
             12.3.5    The deffile Field
             12.3.6    Examples
      12.4  Running LINK
             12.4.1    Specifying Input with LINK Prompts
             12.4.2    Specifying Input in a Response File
      12.5  LINK Options
             12.5.1    Specifying Options
             12.5.2    The /ALIGN Option
             12.5.3    The /BATCH Option
             12.5.4    The /CO Option
             12.5.5    The /CPARM Option
             12.5.6    The /DOSSEG Option
             12.5.7    The /DSALLOC Option
             12.5.8    The /EXEPACK Option
             12.5.9    The /FARCALL Option
             12.5.10   The /HELP Option
             12.5.11   The /HIGH Option
             12.5.12   The /INCR Option
             12.5.13   The /INFO Option
             12.5.14   The /LINE Option
             12.5.15   The /MAP Option
             12.5.16   The /NOD Option
             12.5.17   The /NOE Option
             12.5.18   The /NOFARCALL Option
             12.5.19   The /NOGROUP Option
             12.5.20   The /NOI Option
             12.5.21   The /NOLOGO Option
             12.5.22   The /NONULLS Option
             12.5.23   The /NOPACKC Option
             12.5.24   The /OV Option
             12.5.25   The /PACKC Option
             12.5.26   The /PACKD Option
             12.5.27   The /PADC Option
             12.5.28   The /PADD Option
             12.5.29   The /PAUSE Option
             12.5.30   The /PM Option
             12.5.31   The /Q Option
             12.5.32   The /SEG Option
             12.5.33   The /STACK Option
             12.5.34   The /TINY Option
             12.5.35   The /W Option
             12.5.36   The /? Option
      12.6  Setting Options with the LINK Environment Variable
             12.6.1    Setting the LINK Environment Variable
             12.6.2    Behavior of the LINK Environment Variable
             12.6.3    Clearing the LINK Environment Variable
      12.7  Using Overlays under DOS
             12.7.1    Restrictions on Overlays
             12.7.2    Specifying Overlays
             12.7.3    How Overlays Work
             12.7.4    Overlay Interrupts
      12.8  Linker Operation under DOS
             12.8.1    Segment Alignment
             12.8.2    Frame Number
             12.8.3    Segment Order
             12.8.4    Combined Segments
             12.8.5    Groups
             12.8.6    Fixups
      12.9  LINK Temporary Files
      12.10 LINK Exit Codes
      12.11 Related Topics in Online Help

 Chapter 13  Module-Definition Files

      13.1  Overview
      13.2  Module Statements
             13.2.1    Syntax Rules
             13.2.2    Reserved Words
      13.3  The NAME Statement
      13.4  The LIBRARY Statement
      13.5  The DESCRIPTION Statement
      13.6  The STUB Statement
      13.7  The EXETYPE Statement
      13.8  The PROTMODE Statement
      13.9  The REALMODE Statement
      13.10 The STACKSIZE Statement
      13.11 The HEAPSIZE Statement
      13.12 The CODE Statement
      13.13 The DATA Statement
      13.14 The SEGMENTS Statement
      13.15 CODE, DATA, and SEGMENTS Attributes
      13.16 The OLD Statement
      13.17 The EXPORTS Statement
      13.18 The IMPORTS Statement
      13.19 Related Topics in Online Help

 Chapter 14  Customizing the Microsoft Programmer's WorkBench

      14.1  Setting Switches
             14.1.1    Changing Current Assignments and Switch Settings
             14.1.2    Editing the TOOLS.INI Initialization File
      14.2  Assigning Functions to Keystrokes
      14.3  Writing Macros
             14.3.1    Macro Syntax
             14.3.2    Macro Responses
             14.3.3    Macro Arguments
             14.3.4    Macro Conditionals
             14.3.5    Recording Macros
             14.3.6    Temporary Macros
      14.4  Related Topics in Online Help

 Chapter 15  Debugging Assembly-Language Programs with CodeView

      15.1  Understanding Windows in CodeView
      15.2  Overview of Debugging Techniques
      15.3  Viewing and Modifying Program Data
             15.3.1    Displaying Variables in the Watch Window
             15.3.2    Displaying Expressions in the Watch Window
             15.3.3    Displaying Local Variables
             15.3.4    Using Pointers to Display Arrays and Strings
             15.3.5    Displaying Structures
             15.3.6    Using Quick Watch
             15.3.7    Displaying Memory
             15.3.8    Displaying the Processor Registers
             15.3.9    Modifying the Values of Variables, Memory,
                          and Registers
      15.4  Controlling Execution
             15.4.1    Continuous Execution
             15.4.2    Single-Stepping
             15.4.3    Changing the Program Display Mode
      15.5  Replaying a Debug Session
      15.6  Advanced CodeView Techniques
      15.7  CodeView Command-Line Options
      15.8  Customizing CodeView with the TOOLS.INI File
      15.9  Related Topics in Online Help

 Chapter 16  Converting C Header Files to MASM Include Files

      16.1  Basic H2INC Operation
      16.2  H2INC Syntax and Options
      16.3  Converting Data and Data Structures
             16.3.1    User-Defined and Predefined Constants
             16.3.2    Variables
             16.3.3    Pointers
             16.3.4    Structures and Unions
             16.3.5    Bit Fields
             16.3.6    Enumerations
             16.3.7    Type Definitions
      16.4  Converting Function Prototypes
      16.5  Related Topics in Online Help

 Chapter 17  Writing OS/2 Applications

      17.1  OS/2 Overview
      17.2  Differences between DOS and OS/2
      17.3  A Sample Program
      17.4  Building an OS/2 Application
      17.5  Binding OS/2 MASM Programs
      17.6  Register and Memory Initialization
      17.7  Other OS/2 Utilities
      17.8  Module-Definition Files
      17.9  Related Topics in Online Help

 Chapter 18  Creating Dynamic-Link Libraries

      18.1  DLL Overview
      18.2  DLL Programming Requirements
             18.2.1    Separate Stack and Data Requirement
             18.2.2    Floating-Point Math Requirement
             18.2.3    Re-entrance Requirement
             18.2.4    Segment Strategy in a DLL
      18.3  Writing the DLL Code
             18.3.1    Choosing Module Attributes
             18.3.2    Defining Procedures and Data
             18.3.3    Creating Initialization and Termination Code
      18.4  Building the DLL
             18.4.1    Writing the Module-Definition File
             18.4.2    Generating an Import Library with IMPLIB
             18.4.3    Creating and Using the DLL
      18.5  Related Topics in Online Help

 Chapter 19  Writing Memory-Resident Software

      19.1  Terminate-and-Stay-Resident Programs
             19.1.1    Structure of a TSR
             19.1.2    Passive TSRs
             19.1.3    Active TSRs
      19.2  Interrupt Handlers in Active TSRs
             19.2.1    Auditing Hardware Events for TSR Requests
             19.2.2    Monitoring System Status
             19.2.3    Determining Whether to Invoke the TSR
      19.3  Example of a Simple TSR: ALARM
      19.4  Using DOS in Active TSRs
             19.4.1    Understanding DOS Stacks
             19.4.2    Determining DOS Activity
             19.4.3    Interrupting DOS Functions
             19.4.4    Monitoring the Critical Error Flag
      19.5  Preventing Interference
             19.5.1    Trapping Errors
             19.5.2    Preserving an Existing Condition
             19.5.3    Preserving Existing Data
      19.6  Communicating through the Multiplex Interrupt
             19.6.1    The Multiplex Handler
             19.6.2    Using the Multiplex Interrupt Under DOS Version 2.x
      19.7  Deinstalling TSRs
      19.8  Example of an Advanced TSR: SNAP
             19.8.1    Building SNAP.EXE
             19.8.2    Outline of SNAP
      19.9  Related Topics in Online Help

 Chapter 20  Mixed-Language Programming

      20.1  Naming and Calling Conventions
             20.1.1    Naming Conventions
             20.1.2    The C Calling Convention
             20.1.3    The Pascal Calling Convention
             20.1.4    The Standard Calling Convention
      20.2  Writing the Assembly-Language Procedure
      20.3  The MASM/High-Level-Language Interface
             20.3.1    The C/MASM Interface
             20.3.2    The FORTRAN/MASM Interface
             20.3.3    The Basic/MASM Interface
             20.3.4    The Pascal/MASM Interface
             20.3.5    The QuickPascal/MASM Interface
      20.4  Related Topics in Online Help

 Appendix A  Differences between MASM 6.0 and 5.1

      A.1   New Features of Version 6.0
             A.1.1    The Assembler, Environment, and Utilities
             A.1.2    Segment Management
             A.1.3    Data Types
             A.1.4    Procedures, Loops, and Jumps
             A.1.5    Simplifying Multiple-Module Projects
             A.1.6    Expanded State Control
             A.1.7    New Processor Instructions
             A.1.8    Renamed Directives
             A.1.9    Macro Enhancements
             A.1.10   MASM 6.0 Programming Practices
      A.2   Compatibility between MASM 5.1 and 6.0
             A.2.1    Rewriting Code for Compatibility
             A.2.2    Using the OPTION Directive
             A.2.3    Changes to Instruction Encodings

 Appendix B  BNF Grammar


 Appendix C  Generating and Reading Assembly Listings

      C.1   Generating Listing Files
             C.1.1    Generating a First Pass Listing
             C.1.2    Controlling the Contents of the Listing File
             C.1.3    Controlling Listing Information on Macros
             C.1.4    Controlling the Page Format
             C.1.5    Precedence of Command-Line Options and Listing
               Directives
      C.2   Reading the Listing File
             C.2.1    Code Generated
             C.2.2    Error Messages
             C.2.3    Symbols and Abbreviations
             C.2.4    Reading Tables in a Listing File

 Appendix D  MASM Reserved Words

      D.1   Operands and Symbols
             D.1.1    Special Operands for the 80386/486
             D.1.2    Predefined Symbols
      D.2   Registers
      D.3   Operators and Directives
      D.4   Processor Instructions
             D.4.1    8086/8088 Processor Instructions
             D.4.2    80186 Processor Instructions
             D.4.3    80286 Processor Instructions
             D.4.4    80286 and 80386 Privileged-Mode Instructions
             D.4.5    80386 Processor Instructions
             D.4.6    80486 Processor Instructions
             D.4.7    Instruction Prefixes
      D.5   Coprocessor Instructions
             D.5.1    8087 Coprocessor Instructions
             D.5.2    80287 Privileged-Mode Instruction
             D.5.3    80387 Instructions

 Appendix E  Default Segment Names


 Appendix F  Error Messages

      F.1   BIND Error Messages
      F.2   CodeView Error Messages
      F.3   EXEHDR Error Messages
      F.4   HELPMAKE Error Messages
             F.4.1    HELPMAKE Fatal Errors
             F.4.2    HELPMAKE Errors
             F.4.3    HELPMAKE Warnings
      F.5   H2INC Error Messages
             F.5.1    H2INC Fatal Errors
             F.5.2    H2INC Compilation Errors
             F.5.3    H2INC Warnings
      F.6   IMPLIB Error Messages
             F.6.1    IMPLIB Fatal Errors
             F.6.2    IMPLIB Errors
      F.7   LIB Error Messages
             F.7.1    LIB Fatal Errors
             F.7.2    LIB Errors
             F.7.3    LIB Warnings
      F.8   LINK Error Messages
             F.8.1    LINK Fatal Errors
             F.8.2    LINK Errors
             F.8.3    LINK Warnings
      F.9   ML Error Messages
             F.9.1    ML Fatal Errors
             F.9.2    ML Errors
             F.9.3    ML Warnings
      F.10  NMAKE Error Messages
             F.10.1   NMAKE Fatal Errors
             F.10.2   NMAKE Errors
             F.10.3   NMAKE Warnings
      F.11  PWB.COM Error Messages
      F.12  PWBRMAKE Error Messages
             F.12.1   PWBRMAKE Fatal Errors
             F.12.2   PWBRMAKE Warnings

 Glossary


 Index




 Introduction
 ────────────────────────────────────────────────────────────────────────────

 The Microsoft (R) Macro Assembler Programmer's Guide provides the
 information you need to write and debug assembly-language programs with the
 Microsoft Macro Assembler (MASM), version 6.0. This book documents enhanced
 features of the language and the programming environment for MASM 6.0. It
 also describes new features that take advantage of the capabilities of the
 80386/486 processors.

 The Programmer's Guide is written for experienced programmers who know
 assembly language and are familiar with an assembler. The book does not
 teach the basics of assembly language; it does explain Microsoft-specific
 features. If you want to learn or review the basics of assembly language,
 refer to "Books for Further Reading" later in this introduction.

 The documentation for MASM 6.0 is an integrated set, comprehensive and
 cohesive. This book emphasizes writing efficient code with the new and
 advanced features of MASM. Installing and Using the Professional Development
 System explains not only how to set up MASM 6.0 but also how to use the
 extensive online reference system, the Microsoft Advisor.

 Installing and Using also introduces the integrated environment called the
 Programmer's WorkBench (PWB) and shows how to manage development projects
 with it. The Microsoft Macro Assembler Reference provides a full listing of
 all MASM instructions, directives, statements, and operators, and it serves
 as a quick reference to utility commands.

 For more information on these same topics, see the online Microsoft Advisor,
 which is a complete reference to Macro Assembler language topics, to the
 utilities, and to PWB. You should be able to find most of the information
 you need in the Microsoft Advisor. The printed documents give more in-depth
 and background information.


 New and Extended Features in MASM 6.0

 Version 6.0 of MASM differs from version 5.1 in many ways, from optional
 extensions to features that replace or modify previous assembler behavior.

 MASM 6.0 includes the Programmer's WorkBench, an integrated software
 development environment, and the CodeView (R) source-level debugger. From
 within PWB you can edit, build, debug, or run a program, and you can perform
 most of these operations with either menu selections or keyboard commands.
 You can also customize PWB to suit your individual programming and editing
 requirements and preferences.


 New MASM Language Features

 MASM 6.0 includes a number of new features, described in the list below,
 designed to make programming more efficient and intuitive and to increase
 your productivity. For example, MASM's new high-level-language features mean
 that you can get the speed of assembly language with the ease of high-level
 languages. You can also maintain your programs more easily.


   ■   MASM 6.0 has many enhancements related to types. You can now use the
       same type specifiers in initializations as in other contexts (BYTE
       instead of DB). You can also define your own types, including pointer
       types, with the new TYPEDEF directive. See Chapter 3, "Using Addresses
       and Pointers," and Chapter 4, "Defining and Using Integers."

   ■   The syntax for defining and using structures and records has been
       enhanced. You can also define unions with the new UNION directive. See
       Chapter 5, "Defining and Using Complex Data Types."

   ■   MASM now generates complete CodeView information for all types. See
       Chapter 3, "Using Addresses and Pointers," and Chapter 4, "Defining
       and Using Integers."

   ■   New control-flow directives let you use high-level-language constructs
       such as loops and if-then-else blocks defined with .REPEAT and .UNTIL
       (or .UNTILCXZ); .WHILE and .ENDW; and .IF, .ELSE, and .ELSEIF. The
       assembler generates the appropriate code to implement the control
       structure. See Chapter 7, "Controlling Program Flow."

   ■   MASM now has more powerful features for defining and calling
       procedures. The extended PROC syntax for generating stack frames has
       been enhanced in version 6.0. You can also use the PROTO directive to
       prototype a procedure, which you can then call with the INVOKE
       directive. INVOKE automatically generates code to pass arguments
       (converting them to a related type, if appropriate) and make the call
       according to the specified calling convention. See Chapter 7,
       "Controlling Program Flow."

   ■   MASM optimizes jumps by automatically determining the most efficient
       coding for a jump and then generating the appropriate code. See
       Chapter 7, "Controlling Program Flow."

   ■   Maintaining multiple-module programs is easier in MASM 6.0. The
       EXTERNDEF and PROTO directives make it easy to maintain all global
       definitions in include files shared by all the source modules of a
       project. See Chapter 8, "Sharing Data and Procedures among Modules and
       Libraries."


 The assembler has many new macro features that make complex macros clearer
 and easier to write:


   ■   You can specify default values for macro arguments or mark arguments
       as required. And with the VARARG keyword, one parameter can accept a
       variable number of arguments.

   ■   You can implement loops inside of macros in various ways. For example,
       the new WHILE directive expands the statements in a macro body while
       an expression is not zero.

   ■   You can define macro functions, which return text macros. Several
       predefined text macros are also provided for processing strings. Macro
       operators and other features related to processing text macros and
       macro arguments have been enhanced. For more information on all these
       macro features, see Chapter 9, "Using Macros."


 Finally, MASM 6.0 has improved customizable capabilities:


   ■   With the new .STARTUP and .EXIT directives you can automatically
       generate appropriate start-up and exit code for DOS or OS/2 modules.
       See Chapter 2, "Organizing MASM Segments."

   ■   MASM 6.0 supports flat memory model, available with OS/2 version 2.0.
       In flat model, segments can be as large as 4 gigabytes instead of 64K
       (kilobytes). Offsets are 32 bits instead of 16 bits. See Chapter 2,
       "Organizing MASM Segments."

   ■   The program H2INC.EXE converts C include files to MASM include files
       and translates data structures and declarations. See Chapter 16,
       "Converting C Header Files to MASM Include Files."


 MASM 6.0 includes many other minor new features as well as extended support
 for features of earlier versions of MASM. These features are listed in
 Appendix A, "Differences between MASM 6.0 and 5.1," with cross-references to
 the chapters where they are discussed in detail.


 ML and MASM Command Lines

 MASM 6.0 provides a new command-line driver, ML, which is more powerful and
 flexible than the previous driver (MASM). ML assembles and links with one
 command. The old MASM driver command syntax is still supported, however, to
 support existing batch files and makefiles that use MASM command lines.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 The name MASM has traditionally been used to refer to the Microsoft Macro
 Assembler. It is used in that context throughout this book. But MASM also
 refers to MASM.EXE, which has been replaced by ML.EXE. In MASM 6.0, the
 MASM.EXE file is a small utility that translates command-line options to
 those accepted by ML.EXE, and then calls ML.EXE. The distinction between
 ML.EXE and MASM.EXE is made whenever necessary. Otherwise, MASM refers to
 the assembler and its features.
 ────────────────────────────────────────────────────────────────────────────


 Compatibility with Earlier Versions of MASM

 In many cases, MASM 5.1 code will assemble without modification under MASM
 6.0. However, MASM 6.0 provides a new OPTION directive that lets you
 selectively modify the assembly process. In particular, you can use the M510
 argument with OPTION or the /Zm command-line option to set most features to
 be compatible with version 5.1 code.

 See Appendix A, "Differences between MASM 6.0 and 5.1," for information
 about obsolete features that will not assemble correctly under MASM 6.0. The
 appendix also discusses how to update code to use the new features.


 Scope and Organization of this Book

 The Programmer's Guide describes how to get the most out of the Microsoft
 Macro Assembler 6.0 and the Programmer's WorkBench. The book is arranged by
 topic, with each topic answering a question or solving a problem. The last
 section in each chapter lists topics in the online reference system that
 provide additional information.

 The Programmer's Guide is divided into three parts:

 Part 1, "Programming in Assembly Language," explains how to program
 efficiently using both the new and old features of MASM. It reviews the
 basic components of assembly language and also describes the new and
 enhanced features.

 Part 2, "Improving Programmer Productivity," introduces the utility programs
 included with MASM 6.0. These programs can help you program more quickly and
 efficiently. For example, the chapters in Part 2 show you how to
 automatically update your project (Chapter 10), use program lists as input
 (Chapter 11), use the Microsoft linker (LINK) (Chapter 12), write
 module-definition files (Chapter 13), customize PWB to suit your programming
 style (Chapter 14), use the CodeView debugger to record and play back a
 debugging session (Chapter 15), and easily port data structures from C
 programs to MASM programs (Chapter 16).

 Part 3, "Advanced Topics," covers specialized areas. It describes how to
 write programs to run under OS/2 (Chapter 17) and how to build dynamic-link
 libraries (Chapter 18). Chapter 19 shows how to write a
 terminate-and-stay-resident (TSR) program. Chapter 20, on mixed-language
 programming, defines the calling conventions and equivalent data types that
 allow MASM to call and be called by C, FORTRAN, Basic, and Pascal.

 In addition, six appendixes and a glossary detail the features of MASM 6.0.
 Of particular interest are Appendix A, "Differences between MASM 6.0 and
 5.1," and Appendix B, "BNF Grammar." Appendix A lists the new features of
 MASM 6.0 and also explains how to update MASM 5.1 code. The BNF grammar, or
 Backus-Naur Form for grammar notation, lets you determine the exact syntax
 for any MASM language component. It clearly defines recursive definitions
 and shows all the available options for any placeholder. Other appendixes
 cover assembly listings, reserved words, default segment names, and error
 messages.


 Books for Further Reading

 The following books may help you learn to program in assembly language or
 write specialized programs. These books are listed only for your
 convenience. Microsoft makes no specific recommendations concerning any of
 these books.


 Books about Programming in Assembly Language

 Abrash, Michael, Zen of Assembly Language.
     Glenview, IL: Scott, Foresman and Co., 1990.

 Duntemann, Jeff, Assembly Language from Square One: For the PC AT and
 Compatibles.
     Glenview, IL: Scott, Foresman and Co., 1990.

 Fernandez, Judi N., and Ashley, Ruth, Assembly Language Programming for the
 80386.
     New York: McGraw-Hill, 1990.

 Miller, Alan R., DOS Assembly Language Programming.
     San Francisco: SYBEX, 1988.

 Scanlon, Leo J., 80286 Assembly Language Programming on MS-DOS Computers.
     New York: Brady Communications, 1986.

 Turley, James L., Advanced 80386 Programming Techniques.
     Berkeley, CA: Osborne McGraw-Hill, 1988.


 Books about DOS and BIOS

 "Article 11." MS-DOS Encyclopedia.
     Redmond, WA: Microsoft Press, 1988. Contains information about
 terminate-and-stay-resident programs.

 Duncan, Ray, Advanced MS-DOS.
     2nd ed. Redmond, WA: Microsoft Press, 1988.

 Jourdain, Robert, Programmer's Problem Solver for the IBM PC, XT and AT.
     New York: Brady Communications, 1986.

 Microsoft MS-DOS Programmer's Reference.
     Redmond, WA: Microsoft Press, 1986-87.

 Norton, Peter and Wilton, Richard, The New Peter Norton Programmer's Guide
 to the IBM PC and PS/2.
     Redmond, WA: Microsoft Press, 1988.

 Wilton, Richard, Programmer's Guide to PC & PS/2 Video Systems.
     Redmond, WA: Microsoft Press, 1987.


 Books about OS/2

 Duncan, Ray, Advanced OS/2 Programming.
     Redmond, WA: Microsoft Press, 1989.

 ───, Essential OS/2 Functions.
     Redmond, WA: Microsoft Press, 1989.

 Letwin, Gordon, Inside OS/2.
     Redmond, WA: Microsoft Press, 1989.

 OS/2 Programmer's Reference.
     4 vols. Redmond, WA: Microsoft Press, 1989.


 Books about Other Topics

 Nelson, Ross P., The 80386 Book.
     Redmond, WA: Microsoft Press, 1988.

 Startz, Richard, 8087/80287/80387 for the IBM PC and Compatibles.
     Bowie, MD: Robert J. Brady Co., 1988.

 Writing ROMable Code in Microsoft C.
     Costa Mesa, CA: SSI Corporation.


 Document Conventions

 The following document conventions are used throughout this manual:

 Example of                        Description
 Convention
 ────────────────────────────────────────────────────────────────────────────
 SAMPLE2.ASM                       Uppercase letters indicate file names,
                                   segment names, registers, and terms used
                                   at the command level.

 .MODEL                            Boldface type indicates
                                   assembly-language directives,
                                   instructions, type specifiers, and
                                   predefined macros, as well as keywords
                                   in other programming languages.

 placeholders                      Italic letters indicate placeholders for
                                   information you must supply, such as a
                                   file name. Italics are also occasionally
                                   used for emphasis in the text.

 target                            This font is used to indicate example
                                   programs, user input, and screen output.

 ;                                 A semicolon in the first column of an
                                   example signals illegal code. A
                                   semicolon also marks a comment.

 SHIFT                             Small capital letters signify names of
                                   keys on the keyboard. Notice that a plus
                                   (+) indicates a combination of keys. For
                                   example, CTRL+E means to hold down the
                                   CTRL key while pressing the E key.

 «argument»                        Items inside double square brackets are
                                   optional.

 {register|memory}                 Braces and a vertical bar indicate a
                                   choice between two or more items. You
                                   must choose one of the items unless
                                   double square brackets surround the
                                   braces.

 Repeating elements...             A horizontal ellipsis (...) following an
                                   item indicates that more items having
                                   the same form may appear.

 Program                           A vertical ellipsis tells you that part
 .                                 of a program has been intentionally
 .                                 omitted.
 .
 Fragment


 Getting Assistance and Reporting Problems

 If you need help or think you have discovered a problem in the software,
 please provide the following information to help us locate the problem:


   ■   The version of DOS or OS/2 that you are running

   ■   Your system configuration: the type of machine you are using, its
       total memory, and its total free memory at assembler execution time,
       as well as any other information you think might be useful

   ■   The assembly command line used, or the link command line if the
       problem occurred during linking

   ■   Any object files or libraries you linked with if the problem occurred
       at link time


 If your program is very large, please try to reduce its size to the smallest
 possible program that still produces the problem.

 Use the Product Assistance Request form at the back of this book to send
 this information to Microsoft. If you have comments or suggestions regarding
 any of the books accompanying this product, please indicate them on the
 Document Feedback Card at the back of this book.

 If you are not a registered Macro Assembler owner, you should fill out and
 return the Registration Card. This enables Microsoft to keep you informed of
 updates and other information about the assembler.






 Chapter 1  Understanding Global Concepts
 ────────────────────────────────────────────────────────────────────────────

 With the development of the Microsoft Macro Assembler (MASM) version 6.0,
 you now have more options available to you for approaching a programming
 task. This chapter explains the general concepts of programming in assembly
 language, beginning with the environment and reviewing the components you
 need to work in the assembler environment. Even if you are familiar with
 previous versions of MASM, you should examine this chapter for information
 on new terms and features.

 The first section of the chapter takes a look at the available processors
 and operating systems and how they work together. It also discusses the
 relationship of segmented architecture to assembly programming and the
 differences it makes for programming in OS/2 rather than in DOS.

 The second section describes some of the language components of MASM that
 are common to most programs, such as reserved words, constant expressions,
 operators, and registers. The rest of this book assumes that you understand
 the information presented in this section.

 The last section summarizes the assembly process, from assembling a program
 through running it. You can affect this process by the way you develop your
 code. Finally, this section explores how you can change the assembly process
 with the OPTION directive and conditional assembly.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 This manual does not cover information specific to programming for Microsoft
 Windows(tm). For information on this, see the Microsoft Windows Software
 Development Kit.
 ────────────────────────────────────────────────────────────────────────────


 1.1  The Processing Environment

 The processing environment for MASM 6.0 includes the processor on which your
 programs run, the operating system your programs will use, and the aspects
 of the segmented architecture that influence the choice of programming
 models. This section summarizes these elements of the environment and how
 they affect your programming choices.


 1.1.1  8086-Based Processors

 The 8086 "family" of processors uses segments to control data and code. The
 later 8086-based processors have larger instruction sets and more memory
 capacity, but they still use the same segmented architecture. Knowing the
 differences between the various 8086-based processors can help you select
 the target processor for your programs.

 The instruction set of the 8086 processor is upwardly compatible with its
 successors. To write code that runs on the widest number of machines, select
 the 8086 instruction set. By choosing to use the instruction set of a more
 advanced processor, you increase the capabilities and efficiency of your
 program, but you also reduce the number of systems on which the program can
 run.

 Table 1.1 lists modes, memory, and segment size of processors on which your
 application may need to run. Each processor is discussed in more detail
 below.

 Table 1.1  8086 Family of Processors

╓┌────────────┌───────────────────┌──────────────────┌───────────────────────╖
              Available           Addressable        Segment
 Processor    Modes               Memory             Size
 ────────────────────────────────────────────────────────────────────────────
 8086/8088    Real                1 megabyte         16 bit

 80186/80188  Real                1 megabyte         16 bit

              Available           Addressable        Segment
 Processor    Modes               Memory             Size
 ────────────────────────────────────────────────────────────────────────────

 80286        Real and Protected  16 megabytes       16 bit

 80386        Real and Protected  4 gigabytes        16 or 32 bit

 80486        Real and Protected  4 gigabytes        16 or 32 bit

 ────────────────────────────────────────────────────────────────────────────



 Processor Modes - Real mode allows only one process to run at a time. The
 DOS operating system runs in real mode. The OS/2 operating system can
 execute programs written for DOS, but is designed to provide capabilities
 available only in protected mode. In protected mode, more than one process
 can be active at any one time. Memory accessed by these different processes
 is protected from access by another process.

 Protected-mode addresses do not correspond directly to physical memory.
 Under protected-mode operating systems, the processor allocates and manages
 memory dynamically. Additional privileged instructions initialize protected
 mode and control multiple processes. Section 1.1.2 provides more information
 on operating systems.

 8086 and 8088 - The 8086 is faster than the 8088 because of its 16-bit data
 bus; the 8088 has only an 8-bit data bus. The 16-bit data bus allows you to
 use EVEN and ALIGN on an 8086 processor to word-align data and thus improve
 data-handling efficiency. Memory addresses on the 8086 and 8088 refer to
 actual physical addresses.

 80186 and 80188 - These two processors are identical to the 8086 and 8088
 except that new instructions have been added and several old instructions
 have been optimized. These processors run significantly faster than the
 8086.

 80286 - The 80286 processor adds some instructions to control protected
 mode, and it runs faster. It also provides the optional protected mode that
 can be used by the operating system to allow multiple processes to run at
 the same time. The 80286 is the minimum for running 16-bit versions of OS/2.

 80386 - Unlike its predecessors, the 80386 processor can handle both 16-bit
 and 32-bit data. It is fully software-compatible with the 80286. It
 implements many new hardware-level features, including virtual paged memory,
 multiple virtual 8086 processes, addressing of up to four gigabytes of
 memory, and specialized debugging registers.

 Under DOS, the 80836 supports all the instructions of the 80286 as well as
 several additional ones. It also allows limited use of 32-bit registers and
 addressing modes. The 80386 operates at faster processor speeds than the
 80286 and is the minimum for running 32-bit versions of OS/2 and other
 32-bit operating systems.

 80486 - The 80486 processor is an enhanced version of the 80386, with
 instruction "pipelining" that executes many instructions two to three times
 faster. It incorporates an enhanced version of the 80387 coprocessor, as
 well as an 8K (kilobyte) memory cache. The 80486 includes several new
 instructions and is fully compatible with 80386 software.

 8087, 80287, and 80387 - These math coprocessors work concurrently with the
 8086 family of processors. Performing floating-point calculations with math
 coprocessors is up to 100 times faster than emulating the calculations with
 integer instructions. Although there are technical and performance
 differences among the three coprocessors, the main difference to the
 applications programmer is that the 80287 and 80387 can operate in protected
 mode. The 80387 also has several new instructions. The 80486 does not use
 any of these coprocessors; its floating-point processor is built in and is
 functionally equivalent to the 80387.


 1.1.2  Operating Systems

 With MASM, you can create programs that run under DOS, Windows, or OS/2─or
 all three, in some cases. For example, ML.EXE can produce executable files
 that run in any of the target environments, regardless of the programmer's
 environment. For information on building programs for different
 environments, see "Building and Running Programs" in PWB's online help.

 DOS and OS/2 provide different processing modes. DOS uses the single-process
 real mode. OS/2 uses the multiple-process protected mode. While OS/2 can
 also run in real mode, this book assumes it is being used in protected mode.


 DOS and OS/2 differ primarily in system access methods, size of addressable
 memory, and segment selection. Table 1.2 summarizes these differences.

 Table 1.2  The DOS and OS/2 Operating Systems

                         Available               Contents
 Operating   System      Active      Addressabl  of Segment  Word Length
 System      Access      Processes   e Memory
                                                 Register
 ─────────────────────────────────────────────────────────────────────────────
 DOS (and    Direct to   One         1 megabyte  Actual      16 bit
 OS/2 1.x    hardware                            address
 real mode)

 OS/2 1.x    Operating   Multiple    16          Segment     16 bit
 protected   system                  megabytes   selectors
 mode        call

 OS/2 2.x    Operating   Multiple    4           Segment     32 bit
             system                  gigabytes   selectors
             call

 ─────────────────────────────────────────────────────────────────────────────


 DOS - In real-mode programming, you can access system functions by calling
 DOS, calling the basic input/output system (BIOS), or directly addressing
 hardware. Access is through DOS interrupt 21h.

 Protected-mode programs cannot directly access hardware ports.

 OS/2 1.x - As you can see in Table 1.2, protected mode allows for much
 larger data structures than real mode, since the addressable memory is
 extended to 16 megabytes. In protected mode, segment registers contain
 segment selectors rather than actual segment values. These selectors cannot
 be calculated by the program; they must be obtained by calling the operating
 system. Programs that attempt to calculate segment values or to address
 memory directly do not work.

 Note that protected-mode operating systems such as XENIX (R) and OS/2
 provide system functions for memory and hardware accesses that would be
 prohibited with direct processor commands. This software interface permits
 access without the possibility of corrupting memory or crashing the system.


 Protected mode uses privilege levels to maintain system integrity and
 security. Programs cannot access data or code that is in a higher privilege
 level. Some instructions that directly access ports or clear interrupts
 (such as CLI, STI, IN,  and OUT) are available at privilege levels normally
 used only by systems programmers.

 OS/2 protected mode enforces the separation of segment values. The segments
 have selectors that have no relationship to the offset. The operating system
 combines the segment and offset so that your programs can address up to 16
 megabytes of virtual memory in a 16-bit system.

 OS/2 2.x and flat model eliminate segments.

 OS/2 2.x - OS/2 2.x uses an unsegmented architecture. (See Section 1.1.3.)
 It creates a "flat model" in which the entire address space is within one
 32-bit segment. Section 2.2.1, "Defining Basic Attributes with .MODEL,"
 explains how to use the flat model. In a 32-bit system, you can access up to
 four gigabytes of virtual memory. (The term "virtual memory" means that if
 the programs running under OS/2 request more memory than is physically
 available, part of the memory is temporarily swapped out to disk.) Since
 code, data, and stack are in the same segment, the value of segment
 registers never needs to change. Internal mechanisms of OS/2 2.x implement
 protection at a lower level.


 1.1.3  Segmented Architecture

 The 8086 processors differ from many other microprocessors in that they use
 a segmented architecture: that is, each address is represented in two
 parts─a segment and an offset. Segmented addresses affect many aspects of
 assemblylanguage programming, especially addresses and pointers.

 Only 64K of data can be addressed by a 16-bit segment address.

 Segmented architecture was originally designed to enable a 16-bit processor
 to access an address space larger than 64K. (Section 1.1.5, "Segmented
 Addressing," explains how the processor uses both the segment and offset to
 create addresses larger than 64K.) DOS is an example of an operating system
 that uses segmented architecture on a 16-bit processor.

 With the advent of protected-mode processors such as the 80286, segmented
 architecture gained a second purpose. Segments can separate different blocks
 of code and data to protect them from undesirable interactions. OS/2 1.x is
 an operating system that takes advantage of the protection features of the
 16-bit segments on the 80286.

 Segmented architecture went through another significant change with the
 release of 32-bit processors, starting with the 80386. These processors are
 backward compatible with the older 16-bit processors, but they also offer a
 32-bit mode that minimizes the memory limitations of a 16-bit segmented
 architecture. Both offer paging to maintain segment protection. XENIX 386 is
 an example of a 32-bit segmented operating system using segment protection.


 OS/2 2.x takes advantage of the 32-bit processors to allow a nonsegmented
 memory configuration. The processor still uses 32-bit segments, but from the
 user's viewpoint, there is only one segment. The flat memory model used by
 OS/2 2.x places code and data in a single segment. See Section 2.2.1,
 "Defining Basic Attributes with .MODEL," for more information about the flat
 memory model.


 1.1.4  Segment Protection

 Segmented architecture is an important part of the OS/2 memory-protection
 scheme. In a "multitasking" operating system where numerous programs can run
 simultaneously, programs must not access the code and data of another
 process without permission.

 In DOS, the data and code segments are usually allocated adjacent to each
 other, as shown in Figure 1.1. In OS/2, the data and code segments may be
 anywhere in memory. The programmer knows nothing about their location and
 has no control over it. The segments may even be moved to a new memory
 location or swapped to disk while the program is running.

 (This figure may be found in the printed book.)

 Segment protection prevents a bug in one program from corrupting another
 program.

 Segment protection makes software development easier and more reliable in
 OS/2 than in DOS because, in OS/2, any illegal access is detected
 immediately. The operating system intercepts illegal memory accesses,
 terminates the program, and displays a message. This makes the bug easier to
 track down and fix.

 In DOS, an illegal access is not detected and may not cause an error until
 later, when another part of the program attempts to use the corrupted
 memory.


 1.1.5  Segmented Addressing

 Segmented addressing is the internal mechanism that combines a segment value
 and an offset value to create an address. The two parts of an address are
 represented as

   segment:offset

 The segment portion is always a 16-bit value. The offset portion is a 16-bit
 value in 16-bit mode or a 32-bit value in 32-bit mode.

 In real mode, the segment value is a physical address that has an arithmetic
 relationship to the offset value. The segment and offset together create a
 20-bit physical address (explained in the next section). Although 20-bit
 addresses can access up to one megabyte of memory, the operating system on
 IBM (R) PCs and compatibles uses part of this memory, leaving 640K of memory
 for programs.


 1.1.6  Segment Arithmetic

 Manipulating segment and offset addresses directly in real-mode programming
 is called "segment arithmetic." Programs that perform segment arithmetic are
 not portable to protected-mode operating systems, where addresses do not
 correspond to a known segment and offset.

 The segment selects a region of memory; the offset selects the byte within
 that region.

 To perform segment arithmetic successfully, it helps to understand how the
 processor combines a 16-bit segment and a 16-bit offset to form a 20-bit
 linear address. In effect, the segment selects a 64K region of memory, and
 the offset selects the byte within that region. Here's how it works:


   1.  The processor shifts the segment address to the left by four binary
       places, producing a 20-bit address ending in four zeros. This
       operation has the effect of multiplying the segment address by 16.

   2.  The processor adds this 20-bit segment address to the 16-bit offset
       address. The offset address is not shifted.

   3.  The processor uses the resulting 20-bit address, often called the
       "physical address," to access an actual location in the one-megabyte
       address space.


 Figure 1.2 illustrates this process.

 (This figure may be found in the printed book.)

 A 20-bit physical address may actually be specified by 4,096 equivalent
 segment:offset addresses. For example, the 20-bit physical address 0F800 is
 equivalent to 0000:F800, 0F00:0800, or 0F80:0000.

 You may need to convert two segmented addresses with different segments to
 segmented addresses with the same segment to write TSRs (see Chapter 19), to
 write code to handle huge arrays, or to determine the size of an area of
 memory.


 1.2  Language Components of MASM

 Programming with MASM requires that you understand the MASM concepts of
 reserved words, identifiers, predefined symbols, constants, expressions,
 operators, data types, registers, and statements. This section defines
 important terms and provides lists that summarize these topics. See online
 help or the MASM Reference for detailed information.


 1.2.1  Reserved Words

 A reserved word has a special meaning fixed by the language. You can use it
 only under certain conditions. MASM's reserved words include:


   ■   Instructions, which correspond to operations the processor can execute

   ■   Directives, which give commands to the assembler

   ■   Attributes, which provide a value for a field, such as segment
       alignment

   ■   Operators, which are used in expressions

   ■   Predefined symbols, which return information to your program


 MASM reserved words are not case sensitive except for predefined symbols
 (see Section 1.2.3).

 Use OPTION NOKEYWORD if you want to use a reserved word in another context.


 The assembler generates an error if you use a reserved word as a variable,
 code label, or other identifier within your source code. However, if you
 need to use a reserved word for another purpose, the OPTION NOKEYWORD
 directive can selectively disable a word's status as a reserved word.

 For example, to remove the STR instruction, the MASK operator, and the NAME
 directive from the set of words MASM recognizes as reserved, use this
 statement in the code segment of your program prior to the first reference
 to STR, MASK, or NAME:

   OPTION NOKEYWORD:<STR MASK NAME>

 The OPTION directive is discussed in Section 1.3.2. Appendix D provides a
 complete list of MASM reserved words.


 1.2.2  Identifiers

 Identifiers are names of variables of a given type.

 An identifier is a name that you invent and attach to a definition.
 Identifiers can be symbols representing variables, constants, procedure
 names, code labels, segment names, and user-defined data types such as
 structures, unions, records, and types defined with TYPEDEF. Identifiers
 longer than 247 characters generate an error.

 Certain restrictions limit the names you can use for identifiers. Follow
 these rules to define a name for an identifier:


   ■   The first character of the identifier can be an alphabetic character
       (A-Z) or any of these four characters: @  _   $  ?

   ■   The other characters in the identifier can be any of the characters
       listed above or a decimal digit (0-9)


 Avoid starting an identifier with the at sign (@), because MASM 6.0
 predefines some special symbols starting with @ (see Section 1.2.3).
 Beginning an identifier with @ may also cause conflicts with future versions
 of the Macro Assembler.

 The symbol--and thus the identifier--is visible as long as it remains within
 scope. (See Section 8.2, "Sharing Symbols with Include Files," for
 additional information about visibility and scope.)


 1.2.3  Predefined Symbols

 Macros and conditionalassembly blocks often use predefined symbols.

 The assembler includes a number of predefined symbols (also called
 predefined equates). You can use these symbol names at any point in your
 code to represent the equate value. For example, the predefined equate
 @FileName represents the base name of the current file. If the current
 source file is TASK.ASM, the value of @FileName is TASK. The MASM predefined
 symbols are listed below according to the kinds of information they provide.
 Case is important only if the /Cp option is used. (See online help on ML
 command-line options for additional details.)


 Predefined Symbols for Segment Information

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Symbol                            Description
 ────────────────────────────────────────────────────────────────────────────
 @code                             Provides the name of the code segment,
                                   except in tiny model when it returns
                                   DGROUP.

 @CodeSize                         Returns an integer representing the
                                   default code distance.

 @CurSeg                           Returns the name of the current segment.

 @data                             Expands to DGROUP except in flat model.

 @DataSize                         Returns an integer representing the
                                   default data distance.

 @fardata                          Represents the name of the segment
                                   defined by the .FARDATA directive.

 @fardata?                         Represents the name of the segment
 Symbol                            Description
 ────────────────────────────────────────────────────────────────────────────
@fardata?                         Represents the name of the segment
                                   defined by the .FARDATA? directive.

 @Model                            Returns the selected memory model.

 @stack                            Expands to DGROUP for near stacks or
                                   STACK for far stacks. (See Section 2.2.3,
                                   "Creating a Stack.")

 @WordSize                         Provides the size attribute of the
                                   current segment.




 Predefined Symbols for Environment Information

 Symbol                            Description
 ────────────────────────────────────────────────────────────────────────────
 @Cpu                              Contains a bit mask specifying the
                                   processor mode.

 @Environ                          Returns values of environment variables.

 @Interface                        Contains information about the language
                                   parameters.

 @Version                          Represents the text equivalent of the
                                   MASM version number. In MASM 6.0, this
                                   expands to 600.



 Predefined Symbols for Date and Time Information

 Symbol            Description
 ────────────────────────────────────────────────────────────────────────────
 @Date             Supplies the current system date.
 @Time             Supplies the current system time.


 Predefined Symbols for File Information

 Symbol                            Description
 ────────────────────────────────────────────────────────────────────────────
 @FileCur                          Names the current file (base and suffix).

 @FileName                         Names the base name of the main file
                                   being assembled as it appears on the
                                   command line.

 @Line                             Gives the source line number in the
                                   current file.



 Predefined Functions for Macro String Manipulation

 Symbol    Description
 ────────────────────────────────────────────────────────────────────────────
 @CatStr   Returns concatenation of two strings.
 @InStr    Returns the starting position of a string within another string.
 @SizeStr  Returns the length of a given string.
 @SubStr   Returns substring from a given string.


 1.2.4  Integer Constants and Constant Expressions

 An integer constant is a series of one or more numerals followed by an
 optional radix specifier. For example, in these statements

   mov     ax, 25
           mov     ax, 0B3h

 the numbers  25  and  0B3h  are integer constants. The  h  appended to  0B3
 is a radix specifier. The specifiers are


   ■   y  for binary (or  b  if radix is less than or equal to 10)

   ■   o  or  q  for octal

   ■   t  for decimal (or  d  if radix is less than or equal to 10)

   ■   h  for hexadecimal


 The default radix is decimal.

 Radix specifiers can be either uppercase or lowercase letters; sample code
 in this book uses lowercase. If no radix is specified, the assembler
 interprets the integer according to the current radix. The default radix is
 decimal, but it can be changed with the .RADIX directive.

 Hexadecimal numbers must always start with a decimal digit (0-9). If
 necessary, add a leading zero to distinguish between symbols and hexadecimal
 numbers that start with a letter. For example,  ABCh  is interpreted as an
 identifier. The hexadecimal digits A through F can be either uppercase or
 lowercase letters. Sample code in this book uses uppercase letters.

 Values of integer constants and expressions are known at assembly time.

 Constant expressions contain integer constants and (optionally) operators
 such as shift, logical, and arithmetic operators, and can be evaluated. The
 assembler evaluates them at assembly time. (In addition to constants,
 expressions can contain labels, types, registers, and their attributes.)
 Constant expressions do not change value during program execution.

 Symbolic Integer Constants - You can define symbolic integer constants with
 either of the data assignment directives, EQU or the equal sign (=). These
 directives assign values to symbols during assembly, not during program
 execution. Symbols defined as integer constants can then be used in
 subsequent statements as immediate operands having the assigned value.
 Symbolic constants are often used to assign mnemonic names to constant
 values, which makes your code more readable and easier to maintain.

 The assembler does not allocate data storage when you use either EQU or =.
 Instead, it replaces each occurrence of the symbol with the value of the
 expression.

 Symbols defined with EQU cannot be redefined.

 The difference between EQU and = is that integers defined with the =
 directive can be changed in your source code, but those defined with EQU
 cannot. Once a symbolic integer constant has been defined with the EQU
 directive, attempting to redefine it generates an error. The syntax is

   symbol EQU expression

 The symbol must be a unique name. The expression can be an integer, a
 constant expression, a one- or two-character string constant (four-character
 on the 80386/486), or an expression that evaluates to an address. If a
 constant value used in numerous places in the source code needs to be
 changed, you modify the expression in one place rather than throughout the
 source code.

 The following example shows the correct use of EQU to define symbolic
 integers.

   column  EQU    80                ; Constant -    80
   row     EQU    25                ; Constant -    25
   screen  EQU    column * row      ; Constant  - 2000
   line    EQU    row               ; Constant -    25

          .DATA

          .CODE
          .
          .
          .
          mov     cx, column
          mov     bx, line

 The value of a symbol defined with the = directive can be different at
 different places in the source code. However, a constant value is assigned
 during assembly for each use, and that value does not change at run time.

 The syntax for the = directive is

 symbol = expression

 Size of Constants - The default word size for MASM 6.0 expressions is 32
 bits. This behavior can be modified using OPTION EXPR16 or OPTION M510. Both
 of these options set the expression word size to 16 bits, but OPTION M510
 affects other assembler behavior as well (see Appendix A).

 It is illegal to change the expression word size once it has been set with
 OPTION M510, OPTION EXPR16, or OPTION EXPR32, but you can repeat the same
 directive in a file. This can be useful for putting an OPTION EXPR16 in
 every include file, for example.


 1.2.5  Operators

 Operators are used in expressions. The value of the expression is determined
 at assembly time and does not change when the program runs.

 Operators should not be confused with processor instructions. The reserved
 word ADD is an instruction. The plus sign (+) is an operator. For example,
 Amount+2  is a valid use of the plus operator (+); it tells the assembler to
 add  2 to  Amount, which might be a value or an address. This operation,
 which occurs at assembly time, is different from the ADD instruction, which
 tells the processor to perform addition at run time.

 The assembler evaluates expressions that contain more than one operator
 according to the following rules:


   ■   Operations in parentheses are always performed before any adjacent
       operations.

   ■   Binary operations of highest precedence are performed first.

   ■   Operations of equal precedence are performed from left to right.

   ■   Unary operations of equal precedence are performed right to left.


 The order of precedence for all operators is listed in Table 1.3. Operators
 on the same line have equal precedence.

 Table 1.3  Operator Precedence

╓┌───────────────────┌───────────────────────────────────────────────────────╖
 Precedence          Operators
 ────────────────────────────────────────────────────────────────────────────
 1                   ( ), [ ]
 2                   LENGTH, SIZE, WIDTH, MASK
 Precedence          Operators
 ────────────────────────────────────────────────────────────────────────────
2                   LENGTH, SIZE, WIDTH, MASK
 3                   . (structure-field-name operator)
 4                   : (segment-override operator), PTR
 5                   LROFFSET, OFFSET, SEG, THIS, TYPE
 6                   HIGH, HIGHWORD, LOW, LOWWORD
 7                   + ,- (unary)
 8                   *, /, MOD, SHL, SHR
 9                   +, - (binary)
 10                  EQ, NE, LT, LE, GT, GE
 11                  NOT
 12                  AND
 13                  OR, XOR
 14                  OPATTR, SHORT, .TYPE
 ────────────────────────────────────────────────────────────────────────────



 1.2.6  Data Types

 A "data type" describes a set of values. A variable of a given type can have
 any of a set of values within the range specified for that type.

 The intrinsic types for MASM 6.0 are BYTE, SBYTE, WORD, SWORD, DWORD,
 SDWORD, FWORD, QWORD, and TBYTE. These types define integers and binary
 coded decimals (BCDs); they are discussed in Chapter 6. The signed data
 types SBYTE, SWORD, and SDWORD are new to MASM 6.0. They are useful in
 conjunction with directives such as INVOKE (for calling procedures) and .IF
 (introduced in Chapter 7). The REAL4, REAL8, and REAL10 directives can be
 used to define floating-point types. See Chapter 6.

 Previous versions of MASM have separate directives for types and
 initializers. For example, BYTE is a type and DB is the corresponding
 initializer. The distinction has been eliminated for MASM 6.0. Any type
 (intrinsic or user-defined) can be used as an initializer.

 MASM does not have specific types for arrays and strings. However, it allows
 a sequence of data units to be treated as arrays, and character (byte)
 sequences to be treated as strings. (See Section 5.1, "Arrays and Strings.")


 Types can also have attributes such as langtype and distance (NEAR and FAR).
 See Section 7.3.3, "Declaring Parameters with the PROC Directive," for
 information on these attributes.

 You can also define your own types with STRUCT, UNION, and RECORD. The types
 have fields that contain string or numeric data, or records that contain
 bits. These data types are similar to the user-defined data types in
 high-level languages such as C, Pascal, and FORTRAN. (See Chapter 5,
 "Defining and Using Complex Data Types.")

 The TYPEDEF directive defines aliases and pointer types.

 You can define new types, including pointer types, with the TYPEDEF
 directive, which is also new to MASM 6.0. TYPEDEF assigns a qualifiedtype
 (explained below) to a typename.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 The concept of the qualifiedtype is essential to understanding many of the
 new features in MASM 6.0, including prototypes and the .IF and INVOKE
 directives. Descriptions of these topics in later chapters refer to this
 section.
 ────────────────────────────────────────────────────────────────────────────

 Once assigned, the typename can be used as a data type in your program. Use
 of the qualifiedtype also allows the CodeView debugger to display
 information on the type. You cannot use a qualifiedtype as an initializer,
 but you can use a type defined with TYPEDEF.

 The qualifiedtype is any MASM type (such as structure types, union types,
 record types, or an intrinsic type) or can be a pointer to a type with the
 form

   «distance» PTR «qualifiedtype»

 where distance is NEAR, FAR, or any distance modifier. See Section 7.3.3,
 "Declaring Parameters with the PROC Directive," for more information on
 distance.

 The qualifiedtype can also be any type previously defined with TYPEDEF. For
 example, if you use TYPEDEF to create an alias for BYTE, as shown below,
 then you can use that  CHAR  type as a qualifiedtype when defining the
 pointer type  PCHAR.

   CHAR    TYPEDEF BYTE
   PCHAR   TYPEDEF PTR CHAR

 Section 3.3, "Accessing Data with Pointers and Addresses," shows how to use
 the TYPEDEF directive to define pointers.

 Since distance and qualifiedtype are optional syntax elements, you can use
 variables of type PTR or FAR PTR. You can also define procedure prototypes
 with qualifiedtype. See Section 7.3.6, "Declaring Procedure Prototypes," for
 more information about procedure prototypes.

 Several rules govern the use of qualifiedtype:


   ■   The only component of a qualifiedtype definition that can be
       forwardreferenced is a structure or union type identifier.

   ■   If distance is not specified, the right operand and current memory
       model determine the type of the pointer. If the operand following PTR
       is not a distance or a function prototype, the operand is a pointer of
       the default data pointer type in the current mode. Otherwise, the type
       of the pointer is the distance of the right operand.

   ■   If .MODEL is not specified, SMALL model (and therefore NEAR pointers)
       is the default.


 A qualifiedtype can be used in seven places:

╓┌─────────────────────────────────────┌─────────────────────────────────────╖
 Use                                   Example
 ────────────────────────────────────────────────────────────────────────────
 In procedure arguments                proc1 PROC pMsg:PTR BYTE

 In prototype arguments                proc2 PROTO pMsg:FAR PTR WORD

 With local variables declared inside  LOCAL pMsg:PTR
 procedures

 Use                                   Example
 ────────────────────────────────────────────────────────────────────────────

 With the LABEL directive              TempMsg LABEL PTR WORD

 With the EXTERN and EXTERNDEF         EXTERN pMsg:FAR PTR BYTE
 directives                            EXTERN MyProc:PROTO

 With the COMM directive               COMM var1:WORD:3

 With the TYPEDEF directive            PPBYTE TYPEDEF PTR PBYTE PFUNC
                                       TYPEDEF PROTO MyProc



 Section 3.3.1 shows ways to write a TYPEDEF type for a qualifiedtype.
 Attributes such as NEAR and FAR can also be applied to a qualifiedtype.

 You can also determine an accurate definition for TYPEDEF and qualifiedtype
 from the BNF grammar definitions given in Appendix B. The BNF grammar
 defines each component of the syntax for any directive, showing the
 recursive properties of components such as qualifiedtype.


 1.2.7  Registers

 All the 8086 processors have the same base set of 16-bit registers. Some
 registers can be accessed as two separate 8-bit registers. In the 80386/486,
 most registers can also be accessed as extended 32-bit registers.

 Figure 1.3 shows the registers common to all the 8086-based processors. Each
 register has its own special uses and limitations.

 (This figure may be found in the printed book.)

 80386/486 Only - The 80386/486 processors use the same 8-bit and 16-bit
 registers that the rest of the 8086 family uses. All of these registers can
 be further extended to 32 bits, except segment registers, which always
 occupy 16 bits. The extended register names begin with the letter "E." For
 example, the 32-bit extension of AX is EAX. The 80386/486 processors have
 two additional segment registers, FS and GS. Figure 1.4 shows the extended
 registers of the 80386/486.

 (This figure may be found in the printed book.)


 1.2.7.1  Segment Registers

 At run time, all addresses are relative to one of four segment registers:
 CS, DS, SS, or ES. (The 80386/486 processors add two more, FS and GS.) These
 registers, their segments, and their purpose are listed below:

 Register and Segment              Purpose
 ────────────────────────────────────────────────────────────────────────────
 CS (Code Segment)                 Contains processor instructions and
                                   their immediate operands.

 DS (Data Segment)                 Normally contains data allocated by the
                                   program.

 SS (Stack Segment)                Creates stacks for use by PUSH, POP,
                                   CALLS,
                                   and RET.

 ES (Extra Segment)                References secondary data segment. Used
                                   by string instructions.

 FS, GS                            Provides extra segments on the
                                   80386/486.


 1.2.7.2  General-Purpose Registers

 Operations on registers are usually faster than operations on memory
 locations.

 The AX, DX, CX, BX, BP, DI, and SI registers are 16-bit general-purpose
 registers. They can be used for temporary data storage. Since the processor
 accesses registers more quickly than it can access memory, you can speed up
 execution by keeping the most frequently used data in registers.

 The 8086 family of processors does not perform memory-to-memory operations.
 Thus, operations on more than one variable often require the data to be
 moved into registers.

 Four of the general registers, AX, DX, CX, and BX, can be accessed either as
 two 8-bit registers or as a single 16-bit register. The AH, DH, CH, and BH
 registers represent the high-order 8 bits of the corresponding registers.
 Similarly, AL, DL, CL, and BL represent the low-order 8 bits of the
 registers. All the general registers can be extended to 32 bits on the
 80386/486.


 1.2.7.3  Special-Purpose Registers

 The 8086 family of processors has two additional registers whose values are
 changed automatically by the processor.

 SP (Stack Pointer) - The SP register points to the current location within
 the stack segment. Pushing a value onto the stack decreases the value of SP
 by 2; popping from the stack increases the value of SP by 2. With 32-bit
 operands on 80386/486 processors, SP is increased or decreased by 4 instead
 of 2. Call instructions store the calling address on the stack and decrease
 SP accordingly; return instructions get the stored address and increase SP.
 SP can also be manipulated as a general-purpose register with instructions
 such as ADD.

 Only the processor can change IP.

 IP (Instruction Pointer)  - The IP register always contains the address of
 the next instruction to be executed. You cannot directly access or change
 the instruction pointer. However, instructions that control program flow
 (such as calls, jumps, loops, and interrupts) automatically change the
 instruction pointer.


 1.2.7.4  Flags Register

 Flags reveal the status of the processor.

 The 16 bits in the flags register control the execution of certain
 instructions and reflect the current status of the processor. In 80386/486
 processors, the flags register is extended to 32 bits. Some bits are
 undefined, so there are actually 9 flags for real mode, 11 flags (including
 a 2-bit flag) for 80286 protected mode, 13 for the 80386, and 14 for the
 80486. The extended flags register of the 80386/486 is sometimes called
 "Eflags."

 Figure 1.5 shows the bits of the 32-bit flags register for the 80386/486.
 Only the lower word is used for the other 8086-family processors. The
 unmarked bits are reserved for processor use; do not modify them.

 (This figure may be found in the printed book.)

 The nine flags common to all 8086-family processors are summarized below,
 starting with the low-order flags. In these descriptions, "set" means the
 bit value is 1, and "cleared" means the bit value is 0.

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Flag                              Description
 ────────────────────────────────────────────────────────────────────────────
 Carry                             Set if an operation generates a carry to
                                   or a borrow from a destination operand.

 Parity                            Set if the low-order bits of the result
                                   of an operation contain an even number
                                   of set bits.
 Flag                              Description
 ────────────────────────────────────────────────────────────────────────────
                                  of set bits.

 Auxiliary Carry                   Set if an operation generates a carry to
                                   or a borrow from the low-order four bits
                                   of an operand. This flag is used for
                                   binary coded decimal (BCD) arithmetic.

 Zero                              Set if the result of an operation is 0.

 Sign                              Equal to the high-order bit of the
                                   result of an operation (0 is positive, 1
                                   is negative).

 Trap                              If set, the processor generates a
                                   single-step interrupt after each
                                   instruction. A debugging program can use
                                   this feature to execute a program one
                                   instruction at a time.

 Flag                              Description
 ────────────────────────────────────────────────────────────────────────────

 Interrupt Enable                  If set, interrupts are recognized and
                                   acted on as they are received. The bit
                                   can be cleared to turn off interrupt
                                   processing temporarily.

 Direction                         Set to make string operations process
                                   down from high addresses to low
                                   addresses; can be cleared to make string
                                   operations process up from low addresses
                                   to high addresses.

 Overflow                          Set if the result of an operation is too
                                   large or small to fit in the destination
                                   operand.




 1.2.8  Statements

 Statements are the line-by-line components of source files. Each MASM
 statement specifies an instruction or directive for the assembler.
 Statements have up to four fields. The syntax is shown below:

   «name» «operation» «operands»
   «;comment»

 The fields are explained below:

 Field                             Purpose
 ────────────────────────────────────────────────────────────────────────────
 name                              Defines a label that can be accessed
                                   from elsewhere in the program. For
                                   example, it can name a variable, type,
                                   segment, or code location.

 operation                         States the action of the statement. This
                                   field contains either an instruction or
                                   an assembler directive.

 operands                          Lists one or more items on which the
                                   instruction or directive operates.

 comment                           Provides a comment for the programmer.
                                   Comments
                                   are for documentation only; they are
                                   ignored by the
                                   assembler.


 The following line contains all four fields:

   mainlp: mov     ax, 7   ; Comments follow the semicolon

 Here,  mainlp  is the label,  mov  is the operation, and  ax  and  7  are
 the operands, separated by a comma. The comment follows the semicolon.

 All fields are optional, although certain directives and instructions
 require an entry in the name or operand field. Some instructions and
 directives place restrictions on the choice of operands. By default, MASM is
 not case sensitive.

 Each field (except the comment field) must be separated from other fields by
 white-space characters (spaces or tabs). MASM also requires code labels to
 be followed by a colon, operands to be separated by commas, and comments to
 be preceded by a semicolon.

 The backslash character joins physical lines into one logical line.

 A logical line can contain up to 512 characters and occupy one or more
 physical lines. To extend a logical line into two or more physical lines,
 put the backslash character (\) as the last non-whitespace character before
 the comment or end of the line. You can place a comment after the backslash
 as shown in this example:

   .IF     (x > 0)    \  ; X must be positive
            &&     (ax > x)   \  ; Result from function must be > x
            &&     (cx == 0)     ; Check loop counter too
            mov    dx, 20h
           .ENDIF

 Multiline comments can also be specified with the COMMENT directive. The
 assembler ignores all code between the delimiter character following the
 directive and the line containing the next instance of the delimiter
 character. This example illustrates the use of COMMENT.

   COMMENT ^               The assembler
                           ignores this text
   ^       mov     ax, 1   and this code


 1.3  The Assembly Process

 Creating and running an executable file involves several processes:


   ■   Assembling the source code into an object file

   ■   Linking the object file with other modules or libraries into an
       executable program

   ■   Loading that program into memory

   ■   Running the program


 Once you have written your assembly-language program, MASM provides several
 options for assembling it. The OPTION directive, new to MASM 6.0, has
 several different arguments that let you control the way MASM assembles your
 programs.

 You can control assembly behavior with conditional assembly.

 Conditional assembly allows you to create one source file that can generate
 a variety of programs, depending on the status of various
 conditional-assembly statements.


 1.3.1  Generating and Running Executable Programs

 This section briefly lists all the actions that take place during each of
 the assembly steps. You can change the behavior of some of these actions in
 various ways, for example, by using macros instead of procedures, or by
 using the OPTION directive or conditional assembly. The other chapters in
 this book discuss specific programming methods; this list simply gives you
 an overview.


 1.3.1.1  Assembling

 The ML.EXE program does two things to create an executable program. First,
 it assembles the source code into an intermediate object file. Second, it
 calls the linker, LINK.EXE, which links the object files and libraries into
 an executable program (usually with the .EXE extension).

 At assembly time, the assembler


   ■   Evaluates conditional-assembly directives, assembling if the
       conditions are true.

   ■   Expands macros and macro functions.

   ■   Evaluates constant expressions such as  MYFLAG AND 80H, substituting
       the calculated value for the expression.

   ■   Encodes instructions and nonaddress operands. For example,  mov cx, 13
        can be encoded at assembly time because the instruction does not
       access memory.

   ■   Saves memory offsets as offsets from their segment.

   ■   Passes segments and segment attributes to the object file.

   ■   Saves placeholders for offsets and segments (relocatable addresses).

   ■   Outputs a listing if requested.

   ■   Passes messages (such as INCLUDELIB and .DOSSEG) directly to the
       linker.


 See Section 1.3.3 for information about conditional assembly; see Chapter 9
 for macros. Chapters 2 and 3 give further details about segments and
 offsets, and Appendix C explains listing files.


 1.3.1.2  Linking

 Once your source code is assembled, the resulting object file is passed to
 the linker. At this point, the linker may combine several object files into
 an executable program.

 At link time, the linker


   ■   Combines segments according to the instructions in the object files,
       rearranging the positions of segments that share the same class or
       group.

   ■   Fills in placeholders for offsets (relocatable addresses).

   ■   Writes relocations for segments into the header of .EXE files (but not
       .COM files).

   ■   Writes an executable image.


 Section 2.3.4, "Defining Segment Groups," defines classes and groups.
 Chapter 3, "Using Addresses and Pointers," explains segments and offsets.


 1.3.1.3  Loading

 The operating system loads the file generated by the linker into memory.
 When the executable file is loaded into memory, DOS


   ■   Reads the program segment prefix (PSP) header into memory.

   ■   Allocates memory for the program, based on the values in the PSP.

   ■   Loads the program.

   ■   Calculates the correct values for absolute addresses from the
       relocation table.

   ■   Loads the segment registers SS, CS, DS, and ES with values that point
       to the proper areas of memory.

   ■   Loads the instruction pointer (IP) to point to the start address in
       the code segment and the stack pointer (SP) to point to the stack.

   ■   Begins execution of the program.


 The process is similar for OS/2.

 See Section 1.2.7, "Registers," for information about segment registers, the
 instruction pointer (IP), and the stack pointer (SP). See MASM online help
 or a DOS reference for more information on the PSP.


 1.3.1.4  Running

 Your program is now ready to run. Some program operations cannot be handled
 until the program runs, such as resolving indirect memory operands. See
 Section 7.1.1.2, "Indirect Operands."


 1.3.2  Using the OPTION Directive

 The OPTION directive lets you modify global aspects of the assembly process.
 With OPTION, you can change command-line options and default arguments.
 These changes affect only statements that follow the use of OPTION.

 For example, you may have MASM code in which the first character of a
 variable, macro, structure, or field name is a dot (.). Since a leading dot
 causes MASM 6.0 to generate an error, you can use this statement in your
 program:

   OPTION DOTNAME

 This enables the use of the dot for the first character.

 Changes made with OPTION override any corresponding command-line option. For
 example, suppose you compile a module with this command line (which enables
 M510 compatibility):

   ML /Zm TEST.ASM

 but this statement is in the module:

   OPTION NOM510

 From this point on in the module, the M510 compatibility options are
 disabled.

 The lists below explain each of the arguments for the OPTION directive. You
 can put more than one OPTION statement on one line if you separate them by
 commas.


 Options for M510 Compatibility

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Argument                          Description
 ────────────────────────────────────────────────────────────────────────────
 CASEMAP: maptype                  CASEMAP:NONE (or /Cx) causes internal
                                   symbol recognition to be case sensitive
 Argument                          Description
 ────────────────────────────────────────────────────────────────────────────
                                  symbol recognition to be case sensitive
                                   and causes the case of identifiers in
                                   the .OBJ file to be the same as
                                   specified in the
                                   EXTERNDEF, PUBLIC, or COMM statement.
                                   The default is CASEMAP:NOTPUBLIC (or
                                   /Cp). It specifies case insensitivity
                                   for internal symbol recognition and the
                                   same behavior as CASEMAP:NONE for case
                                   of identifiers in .OBJ files.
                                   CASEMAP:ALL (/Cu) specifies case
                                   insensitivity for identifiers and
                                   converts all identifier names to
                                   uppercase.

 DOTNAME | NODOTNAME               Enables the use of the dot (.) as the
                                   leading character in variable, macro,
                                   structure, union, and member names.
                                   NODOTNAME is the default.
 Argument                          Description
 ────────────────────────────────────────────────────────────────────────────
                                  NODOTNAME is the default.

 M510 | NOM510                     Sets all features to be compatible with
                                   MASM version 5.1, disabling the SCOPED
                                   argument and enabling OLDMACROS,
                                   DOTNAME, and, OLDSTRUCTS. OPTION M510
                                   conditionally sets other arguments for
                                   the OPTION directive. The default is
                                   NOM510. See Appendix A for more
                                   information on using OPTION M510.

 OLDMACROS | NOOLDMACROS           Enables the version 5.1 treatment of
                                   macros. MASM 6.0 treats macros
                                   differently. The default is NOOLDMACROS.



 OLDSTRUCTS | NOOLDSTRUCTS         Enables compatibility with MASM 5.1 for
                                   treatment of structure members. See
 Argument                          Description
 ────────────────────────────────────────────────────────────────────────────
                                  treatment of structure members. See
                                   Section 5.2 for information on
                                   structures.

 SCOPED | NOSCOPED                 Guarantees that all labels inside
                                   procedures are local to the procedure
                                   when SCOPED (the default) is enabled.




 Options for Procedure Use

 Argument                          Description
 ────────────────────────────────────────────────────────────────────────────
 LANGUAGE : langtype               Specifies the default language type (C,
                                   PASCAL, FORTRAN, BASIC, SYSCALL, or
                                   STDCALL) to be used with PROC, EXTERN,
                                   and PUBLIC. This use of the OPTION
                                   directive overrides the .MODEL directive
                                   but is normally used when .MODEL is not
                                   given.

 EPILOGUE: macroname               Instructs the assembler to call the
                                   macroname to generate a user-
                                   defined epilogue instead of the standard
                                   epilogue code when a RET instruction is
                                   encountered. See Section 7.3.8.

 PROLOGUE: macroname               Instructs the assembler to call
                                   macroname to generate a user-
                                   defined prologue instead of generating
                                   the standard prologue code. See Section
                                   7.3.8.

 PROC: visibility                  Allows the default visibility to be set
                                   explicitly. The default visibility is
                                   PUBLIC. The visibility can also be
                                   either EXPORT or PRIVATE.



 Other Options

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Argument                          Description
 ────────────────────────────────────────────────────────────────────────────
 EXPR16 | EXPR32                   Sets the expression word size to 16 or
                                   32 bits. The default is 32 bits. The
                                   M510 argument to the OPTION directive
                                   sets the word size to 16 bits. Once set
                                   with the OPTION directive, the
                                   expression word size cannot be changed.

 EMULATOR | NOEMULATOR             Controls the generation of
                                   floating-point instructions. The
                                   NOEMULATOR option generates the
                                   coprocessor instructions directly. The
                                   EMULATOR option generates instructions
                                   with special fixup records for the
                                   linker so that the Microsoft
 Argument                          Description
 ────────────────────────────────────────────────────────────────────────────
                                  linker so that the Microsoft
                                   floating-point emulator, supplied with
                                   other Microsoft languages, can be used.
                                   It produces the same result as setting
                                   the /Fpi command-line option. You can
                                   set this option only once per module.

 LJMP | NOLJMP                     Enables automatic conditional-jump
                                   lengthening. The default is LJMP. See
                                   Section 7.1.2 for information about
                                   conditional-jump lengthening.

 NOKEYWORD:<keywordlist>           Disables the specified reserved words.
                                   See Section 1.2.1, "Reserved Words," for
                                   an example of the syntax for this
                                   argument.

 NOSIGNEXTEND                      Overrides the default sign-extended
                                   opcodes for the AND, OR, and XOR
 Argument                          Description
 ────────────────────────────────────────────────────────────────────────────
                                  opcodes for the AND, OR, and XOR
                                   instructions and generates the larger
                                   non-sign-extended forms of these
                                   instructions. Provided for compatibility
                                   with NEC V25 (R) and NEC V35(tm)
                                   controllers.

 OFFSET: offsettype                Determines the result of OFFSET operator
                                   fixups. SEGMENT sets the defaults for
                                   fixups to be segment-
                                   relative (compatible with MASM 5.1).
                                   GROUP, the default, generates fixups
                                   relative to the group (if the label is
                                   in a group). FLAT causes fixups to be
                                   relative to a flat frame. (The .386 mode
                                   must be enabled to use FLAT.) See
                                   Appendix A for more information.

 READONLY | NOREADONLY             Enables checking for instructions that
 Argument                          Description
 ────────────────────────────────────────────────────────────────────────────
READONLY | NOREADONLY             Enables checking for instructions that
                                   modify code segments, thereby
                                   guaranteeing that read-only code
                                   segments are not modified. Replaces the
                                   /p command-line option of MASM 5.1. It
                                   is useful for OS/2, where code segments
                                   are normally read-only.

 SEGMENT: segSize                  Allows global default segment size to be
                                   set. Also determines the default address
                                   size for external symbols defined
                                   outside any segment. The segSize can be
                                   USE16, USE32, or FLAT.




 1.3.3  Conditional Directives

 MASM 6.0 provides conditional-assembly directives and conditional-error
 directives. You can also use conditional-assembly directives when you want
 to test for a specified condition and assemble a block of statements if the
 condition is true. You can use conditional-error directives when you want to
 test for a specified condition and generate an assembly error if the
 condition is true.

 Both kinds of conditional directives test assembly-time conditions, not
 run-time conditions. Only expressions that evaluate to constants during
 assembly can be compared or tested. Predefined symbols are often used in
 conditional assembly. See Section 1.2.3.


 Conditional-Assembly Directives

 The IF and ENDIF directives enclose the statements to be considered for
 conditional assembly. The optional ELSEIF and ELSE blocks follow the IF
 directive. There are many forms of the IF and ELSE directives. Online help
 provides a complete list.

 The syntax used for the IF directives is shown below. The syntax for other
 condition-assembly directives follow the same form.

   IF expression1
   ifstatements
   [[ELSEIF expression2
   elseifstatements]]
   [[ELSE
   elsestatements]]
   ENDIF



 The statements following the IF directive can be any valid statements,
 including other conditional blocks, which in turn can contain any number of
 ELSEIF blocks. ENDIF ends the block.

 The statements following the IF directive are assembled only if the
 corresponding condition is true. If the condition is not true and an ELSEIF
 directive is used, the assembler checks to see if the corresponding
 condition is true. If so, it assembles the statements following the ELSEIF
 directive. If no IF or ELSEIF conditions are satisfied, the statements
 following the ELSE directive are assembled.

 For example, you may want to assemble a line of code only if a particular
 variable has been defined. In this example,

   IFDEF   buffer
   buff    BYTE    buffer DUP(?)
           ENDIF

 buff  is allocated only if  buffer  has been previously defined.

 The following list summarizes the conditional-assembly directives:

 Directive                         Use
 ────────────────────────────────────────────────────────────────────────────
 IF and IFE                        Tests the value of an expression and
                                   allows
                                   assembly based on the result.

 IFDEF and IFNDEF                  Tests whether a symbol has been defined
                                   and allows assembly based on the result.

 IFB and IFNB                      Tests to see if a specified argument was
                                   passed to a macro and allows assembly
                                   based on the result.

 IFIDN and IFDIF                   Compares two macro arguments and allows
                                   assembly based on the result. (IFDIFI
                                   and IFIDNI perform the same action but
                                   are case insensitive.)





 Conditional-Error Directives

 You can use conditional-error directives to debug programs and check for
 assembly-time errors. By inserting a conditional-error directive at a key
 point in your code, you can test assembly-time conditions at that point. You
 can also use conditional-error directives to test for boundary conditions in
 macros.

 Like other severe errors, those generated by conditional-error directives
 cause the assembler to return a nonzero exit code. If a severe error is
 encountered during assembly, MASM does not generate the object module.

 For example, the .ERRNDEF directive produces an error if some label has not
 been defined. In this example, .ERRNDEF at the beginning of the conditional
 block makes sure that a  publevel  actually exists.

   .ERRNDEF    publevel
           IF          publevel LE 2
           PUBLIC      var1, var2
           ELSE
           PUBLIC      var1, var2, var3
           ENDIF

 These directives use the syntax given in the previous section. The following
 list summarizes the conditional-error directives.

 Directive                 Use
 ────────────────────────────────────────────────────────────────────────────
 .ERR                      Forces an error where the directives occur in
                           the source file. The error is generated
                           unconditionally when the directive is
                           encountered, but the directives can be placed
                           within conditional-assembly blocks to limit the
                           errors to certain situations.

 .ERRE and .ERRNZ          Tests the value of an expression and
                           conditionally generates an error based on the
                           result.

 .ERRDEF and               Tests whether a symbol is defined and
 .ERRNDEF                  conditionally generates an error based on the
                           result.

 .ERRB and .ERRNB          Tests whether a specified argument was passed to
                           a macro and conditionally generates an error
                           based on the result.

 .ERRIDN and               Compares two macro arguments and conditionally
 .ERRDIF                   generates an error based on the result. (
                           .ERRIDNI and .ERRDIFI perform the same action
                           but are case sensitive.)



 1.4  Related Topics in Online Help

 In addition to information covered in this chapter, information on the
 following topics can be found in online help.

╓┌─────────────────────────────────────┌─────────────────────────────────────╖
 Topic                                 Access
 ────────────────────────────────────────────────────────────────────────────
 Predefined symbols                    From the "MASM 6.0 Contents" screen,
                                       choose "Predefined Symbols"

 Operator precedence                   From the list of tables on the "MASM
                                       6.0 Contents" screen, choose
                                       "Operator Precedence"

 Data types                            Choose "Directives" from the "MASM
 Topic                                 Access
 ────────────────────────────────────────────────────────────────────────────
Data types                            Choose "Directives" from the "MASM
                                       6.0 Contents" screen; then choose
                                       "Data Allocation" or "Complex Data
                                       Types" from the resulting screen

 Registers                             From the "MASM 6.0 Contents" screen,
                                       choose "Language Overview"; then
                                       choose "Processor Register Summary"

 Processor directives                  To see a table of directives, choose
                                       "Processor Selection" from the "MASM
                                       6.0 Contents" screen

 Conditional assembly and conditional  Choose "Directives" from the "MASM
 errors                                6.0 Contents" screen

 EVEN, ALIGN,                          From the "MASM 6.0 Contents" screen,
 OPTION                                choose "Directives," then
                                       "Miscellaneous"
 Topic                                 Access
 ────────────────────────────────────────────────────────────────────────────
                                      "Miscellaneous"

 Radix specifiers                      From the "MASM 6.0 Contents" screen,
                                       choose "Language Overview"

 ML command-line options               From the "Microsoft Advisor Contents"
                                       screen, choose "Macro Assembler"
                                       from the " Command Line" list








 Chapter 2  Organizing MASM Segments
 ────────────────────────────────────────────────────────────────────────────

 A segment is a collection of instructions or data whose addresses are all
 relative to the same segment register. The code in your assembly-language
 program defines and organizes them.

 Segments can be defined by using simplified segment directives or full
 segment definitions. Section 2.2, "Using Simplified Segment Directives,"
 covers the directives you can use to begin, end, and organize segment
 program modules. It also discusses how to access far data and code with
 simplified segment directives.

 Section 2.3, "Using Full Segment Definitions," describes how to order,
 combine, and divide segments, as well as how to use the SEGMENT directive to
 define full segments. It also tells you how to create a segment group so
 that you can use just one segment address to access all the data.

 Most of the information in this chapter also applies to writing modules to
 be called from other programs. Exceptions are noted when they apply. See
 Chapter 8, "Sharing Data and Procedures among Modules and Libraries," for
 more information about multiple-module programming.


 2.1  Overview of Memory Segments

 A physical segment is an area of memory in which all locations are
 contiguous and share the same segment address. A segment always begins on a
 16-byte (paragraph) boundary (unless an alignment attribute is specified
 with ALIGN). While 16-bit segments can occupy up to 64K (kilobytes), 32-bit
 segments can be as large as 4 gigabytes.

 Segments reflect the architecture of the original 8086 processor. Prior to
 the 80386 processors and OS/2 2.x, assembly-language programming meant using
 segmented memory. A flat address space is now available on 80386/486
 processors in 32-bit mode. This space is still segmented at the hardware
 level, but it allows you to ignore most segmentation concerns.

 Segments provide a means for associating similar kinds of data. Most
 programs have segments for code, data, constant data, and the stack. These
 logical segments are allocated by the assembler at assembly time.

 You can define segments in two ways: with simplified segment directives and
 with full segment definitions. You can also use both kinds of segment
 definitions in the same program.

 Simplified segment directives are easier to use than full segment
 definitions.

 Simplified segment directives hide many of the details of segment definition
 and assume the same conventions used by Microsoft high-level languages. (See
 Section 2.2.) The simplified segment directives generate necessary code,
 specify segment attributes, and arrange segment order.

 Full segment definitions require more complex syntax but provide more
 complete control over how the assembler generates segments. (See Section
 2.3.) If you use full segment definitions, you must write code to handle all
 the tasks performed automatically by the simplified segment directives.


 2.2  Using Simplified Segment Directives

 Structuring a MASM program using simplified segments requires use of several
 directives to assign standard names, alignment, and attributes to the
 segments in your program. These directives define the segments in such a way
 that linking with Microsoft high-level languages is easy.

 The simplified segment directives are .MODEL, .CODE, .CONST, .DATA, .DATA?,
 .FARDATA, .FARDATA?, .STACK, .STARTUP, and .EXIT. These directives and the
 arguments they take are discussed in the following sections.

 The main module is where execution begins.

 MASM programs consist of modules made up of segments. Every program written
 only in MASM has one main module, where program execution begins. This main
 module can contain code, data, or stack segments defined with all of the
 simplified segment directives. Any additional modules should contain only
 code and data segments. Every module that uses simplified segments must,
 however, begin with the .MODEL directive.


 The following example shows the structure of a main module using simplified
 segment directives. It uses the default processor (8086), the default
 operating system (OS_DOS), and the default stack distance (NEARSTACK).
 Additional modules linked to this main program would use only the .MODEL,
 .CODE, and .DATA directives and the END statement.

   ; This is the structure of a main module
   ;  using simplified segment directives

           .MODEL small, c ; This statement is required before you
                           ;  can use other simplified segment
                           ;  directives

           .STACK          ; Use default 1-kilobyte stack

           .DATA           ; Begin data segment

                           ; Place data declarations here

           .CODE           ; Begin code segment
           .STARTUP        ; Generate start-up code

                           ; Place instructions here

           .EXIT           ; Generate exit code
           END

 A module must always finish with the END directive.

 The .DATA and .CODE statements do not require any separate statements to
 define the end of a segment. They close the preceding segment and then open
 a new segment. The .STACK directive opens and closes the stack segment but
 does not close the current segment. The END statement closes the last
 segment and marks the end of the source code. It must be at the end of every
 module, whether or not it is the main module.


 2.2.1  Defining Basic Attributes with .MODEL

 The .MODEL directive defines the attributes that affect the entire module:
 memory model, default calling and naming conventions, operating system, and
 stack type. This directive enables use of simplified segments and controls
 the name of the code segment and the default distance for procedures.

 You must place .MODEL in your source file before any other simplified
 segment directive. The syntax is

   .MODEL memorymodel «, modeloptions »

 The memorymodel field is required and must appear immediately after the
 .MODEL directive. The use of modeloptions, which define the other
 attributes, is optional. The modeloptions must be separated by commas. You
 can also use equates passed from the ML command line to define the
 modeloptions.

 The list below summarizes the memorymodel field and the modeloptions fields
 (language, operating system, and stack distance):

 Field                             Description
 ────────────────────────────────────────────────────────────────────────────
 Memory model                      TINY, SMALL, COMPACT, MEDIUM, LARGE,
                                   HUGE, or FLAT. Determines size of code
                                   and data pointers. This field is
                                   required.

 Language                          C, BASIC, FORTRAN, PASCAL, SYSCALL, or
                                   STDCALL. Sets calling and naming
                                   conventions for procedures and public
                                   symbols.

 Operating system                  OS_OS2 or OS_DOS. Determines behavior of
                                   .STARTUP and .EXIT.

 Stack distance                    NEARSTACK or FARSTACK. Specifying
                                   NEARSTACK groups the stack segment into
                                   a single physical segment (DGROUP) along
                                   with data. SS is assumed to equal DS.
                                   FARSTACK does not group the stack with
                                   DGROUP; thus SS does not equal DS.


 You can use no more than one reserved word from each field. The following
 examples show how you can combine various fields:

   .MODEL   small                  ; Small memory model
           .MODEL   large, c, farstack     ; Large memory model,
                                           ;  C conventions,
                                           ;  separate stack
           .MODEL   medium, pascal, os_os2 ; Medium memory model,
                                           ;  Pascal conventions,
                                           ;  OS/2 start-up/exit

 The next four sections give more detail on each field.


 Defining the Memory Model

 MASM supports the standard memory models used by Microsoft high-level
 languages─tiny, small, medium, compact, large, huge, and flat. You specify
 the memory model with attributes of the same name placed after the .MODEL
 directive. Your choice of a memory model does not limit the kind of
 instructions you can write. It does, however, control segment defaults and
 determine whether data and code are near or far by default (see Table 2.1).


 Table 2.1  Attributes of Memory Models

╓┌─────────────┌─────────────┌─────────────┌────────────────┌────────────────╖
 Memory Model  Default Code  Default Data  Operating        Data and Code
                                           System           Combined
 ────────────────────────────────────────────────────────────────────────────
 Memory Model  Default Code  Default Data  Operating        Data and Code
                                           System           Combined
 ────────────────────────────────────────────────────────────────────────────
 Tiny          Near          Near          DOS              Yes

 Small         Near          Near          DOS, OS/2 1.x    No

 Medium        Far           Near          DOS, OS/2 1.x    No

 Compact       Near          Far           DOS, OS/2 1.x    No

 Large         Far           Far           DOS, OS/2 1.x    No

 Huge          Far           Far           DOS, OS/2 1.x    No

 Flat          Near          Near          OS/2 2.x         Yes

 ────────────────────────────────────────────────────────────────────────────



 When writing assembler modules for a high-level language, you should use the
 same memory model as the calling language. Generally, choose the smallest
 memory model available that can contain your data and code, since near
 references are more efficient than far references.

 The predefined symbol @Model returns the memory model. It encodes memory
 models as integers 1 through 7. See Section 1.2.3 for more information on
 predefined symbols, and see online help for an example of how to use them.

 The seven memory models supported by MASM 6.0 divide into three groups.

 Small, Medium, Compact, Large, and Huge Models - The traditional memory
 models recognized by many DOS and OS/2 1.x languages are small, medium,
 compact, large, and huge. Small model supports one data segment and one code
 segment. All data and code are near by default. Large model supports
 multiple code and multiple data segments. All data and code are far by
 default. Medium and compact models are in between. Medium model supports
 multiple code and single data segments; compact model supports multiple data
 segments and a single code segment.

 Huge model implies individual data items larger than a single segment, but
 the implementation of huge data items must be coded by the programmer. Since
 the assembler provides no direct support for this feature, huge model is
 essentially the same as large model.

 In each of these models, you can override the default. For example, you can
 make large data items far in small model, or internal procedures near in
 large model.

 Tiny Model - OS/2 does not support tiny model, but DOS does under MASM 6.0.
 This model places all data and code in a single segment. Therefore, the
 total program size can be no more than 64K. The default is near for code and
 static data items; you cannot override this default. However, you can
 allocate far data dynamically at run time using DOS memory allocation
 services.

 Tiny model produces DOS .COM files. Specifying  .MODEL tiny  automatically
 sends a /TINY to the linker. Therefore, /AT is not necessary with  .MODEL
 tiny. However, /AT does not insert a .MODEL directive. It only verifies that
 there are no base or pointer fixups, and sends /TINY to the linker.

 Flat Model - The flat memory model is a nonsegmented configuration available
 for 32-bit operating systems. It is similar to tiny model in that all code
 and data go in a single 32-bit segment.

 OS/2 2.x uses flat model when you specify the .386 or .486 directive before
 .MODEL FLAT. All data and code (including system resources) are in a single
 32-bit segment. Segment registers are initialized automatically at load
 time; the programmer needs to modify them only when mixing 16-bit and 32-bit
 segments in a single application. CS, DS, ES, and SS are all assumed to the
 supergroup FLAT. FS and GS are assumed to ERROR, since 32-bit versions of
 OS/2 reserve the use of these registers. Addresses and pointers passed to
 system services are always 32-bit near addresses and pointers. Although the
 theoretical size of the single flat segment is four gigabytes, OS/2 2.0
 actually limits it to 512 megabytes in flat model.


 Choosing the Language Convention

 The language type is most important when you write a mixed-language program.


 The language option facilitates compatibility with high-level languages by
 determining the internal encoding for external and public symbol names, the
 code generated for procedure initialization and cleanup, and the order that
 arguments are passed to a procedure with INVOKE. It also facilitates
 compatibility with high-level-language modules. The PASCAL, BASIC, and
 FORTRAN conventions are identical. C and SYSCALL have the same calling
 convention but different naming conventions. OS/2 system calls require the
 PASCAL calling convention for OS/2 1.x, but require the SYSCALL convention
 for OS/2 2.x. Specifying STDCALL for the calling convention enables a
 different calling convention and the same naming convention (see Section
 20.1).

 Procedure definitions (PROC) and high-level procedure calls (INVOKE)
 automatically generate code consistent with the calling convention of the
 specified language. The PROC, INVOKE, PUBLIC, and EXTERN directives all use
 the naming convention of the language. These directives follow the default
 language conventions from the .MODEL directive unless you specifically
 override the default. Chapter 7, "Controlling Program Flow," tells how to
 use these directives. You can also use the OPTION directive to set the
 language type. (See Section 1.3.2.) Not specifying a language type in either
 the .MODEL, OPTION, EXTERN, PROC, INVOKE, or PROTO statement causes the
 assembler to generate an error.

 The predefined symbol @Interface provides information about the language
 parameters. See online help for a description of the bit flags.

 See Chapter 20, "Mixed-Language Programming," for more information on
 calling and naming conventions. See Chapter 7, "Controlling Program Flow,"
 for information about writing procedures and prototypes. See Chapter 8,
 "Sharing Data and Procedures among Modules and Libraries," for information
 on multiple-module programming.


 Specifying the Operating System

 The operating-system options (OS_DOS or OS_OS2) are arguments of .MODEL.
 They specify the start-up and exit code generated by the .STARTUP and .EXIT
 directives. (See Section 2.2.6.) If you do not use .STARTUP and .EXIT, you
 can omit this option. The default is OS_DOS.


 Setting the Stack Distance

 The NEARSTACK setting places the stack segment in a group, DGROUP, shared
 with data. The .STARTUP directive then generates code to adjust SS:SP so
 that SS (Stack Segment register) holds the same address as DS (Data Segment
 register). If you do not use .STARTUP, you must make this adjustment
 yourself or your program may fail to run. (See Section 2.2.6 for information
 about start-up code.) In this case, you can use DS to access stack items
 (including parameters and local variables) and SS to access near data.
 Furthermore, since stack items share the same segment address as near data,
 you can reliably pass near pointers to stack items.

 Having SS equal to DS gives some programming advantages.

 The FARSTACK setting gives the stack a segment of its own. That is, SS does
 not equal DS. The default stack type, NEARSTACK, is a convenient setting for
 most programs. Use FARSTACK for special cases such as memory-resident
 programs and dynamic-link libraries (DLLs) when you cannot assume that the
 caller's stack is near.

 The stack specification also affects the ASSUME statement generated by
 .MODEL and .STACK. You can use the predefined symbol @Stack to determine if
 the stack location is DGROUP (for near stacks) or STACK (for far stacks).


 2.2.2  Specifying a Processor and Coprocessor

 MASM supports a set of directives for selecting processors and coprocessors.
 Once you select a processor, you must use only the instruction set available
 for that processor. The default is the 8086 processor. If you always want
 your code to run on this processor, you do not need to add any processor
 directives.

 To enable a different processor mode and the additional instructions
 available on that processor, use the directives .186, .286, .386, and .486.


 The .286P, .386P, and .486P directives enable the instructions available
 only at higher privilege levels in addition to the normal instruction set
 for the given processor. Privileged instructions are not necessary for
 writing applications, even for OS/2. Generally, you don't need privileged
 instructions unless you are writing operating-systems code or device
 drivers.

 Processor directives affect availability of various MASM language features.


 In addition to enabling different instruction sets, the processor directives
 also affect the behavior of extended language features. For example, the
 INVOKE directive pushes arguments onto the stack. If the .286 directive is
 in effect, INVOKE takes advantage of operations possible only on 80286 and
 later processors.

 Use the directives .8087 (the default), .287, .387, and .NO87 to select a
 math coprocessor instruction set. The .NO87 directive turns off assembly of
 all coprocessor instructions. Note that .486 also enables assembly of all
 coprocessor instructions because the 80486 processor has a complete set of
 coprocessor registers and instructions built into the chip. The processor
 instructions imply the corresponding coprocessor directive. The coprocessor
 directives are provided to override the defaults.


 2.2.3  Creating a Stack

 The stack is the section of memory used for pushing or popping registers and
 storing the return address when a subroutine is called. The stack often
 holds temporary and local variables.

 If your main module is written in a high-level language, that language
 handles the details of creating a stack. Use the .STACK directive only when
 you write a main module in assembly language.

 The .STACK directive creates a stack segment. By default, the assembler
 allocates 1K of memory for the stack. This size is sufficient for most small
 programs.

 To create a stack of a size other than the default size, give .STACK a
 single numeric argument indicating stack size in bytes:

   .STACK  2048            ; Use 2K stack

 For a description of how stack memory is used with procedure calls and local
 variables, see Chapter 7, "Controlling Program Flow."


 2.2.4  Creating Data Segments

 Programs can contain both near and far data. In general, you should place
 important and frequently used data in the near data area, where data access
 is faster. This area can get crowded, however, because (in 16-bit operating
 systems) the total amount of all near data in all modules cannot exceed 64K.
 Therefore, you may want to place infrequently used or particularly large
 data items in a far data segment.

 The .DATA, .DATA?, .CONST, .FARDATA, and .FARDATA? directives create data
 segments. You can access the various segments within DGROUP without
 reloading segment registers (see Section 2.3.4, "Defining Segment Groups").
 These four directives also prevent instructions from appearing in data
 segments by assuming CS to ERROR. (See Section 2.3.3 for information about
 ASSUME.)


 Near Data Segments

 The .DATA directive creates a near data segment. This segment contains the
 frequently used data for your program. It can occupy up to 64K in DOS or 512
 megabytes under flat model in OS/2 2.0. It is placed in a special group
 identified as DGROUP, which is also limited to 64K.

 Near data pointers always point to DGROUP.

 When you use .MODEL, the assembler automatically defines DGROUP for your
 near data segment. The segments in DGROUP form near data, which can normally
 be accessed directly through DS or SS.

 You can also define the .DATA? and .CONST segments that go into DGROUP
 unless you are using flat model. Although all of these segments (along with
 the stack) are eventually grouped together and handled as data segments,
 .DATA? and .CONST enhance compatibility with Microsoft high-level languages.
 In Microsoft languages, .CONST is used for defining constant data such as
 strings and floating-point numbers that must be stored in memory. The .DATA?
 segment is used for storing uninitialized variables. You can follow this
 convention if you wish. If you use C start-up code, .DATA? is initialized to
 0.

 You can use @data to determine the group of the data segment and @DataSize
 to determine the size of the memory model set by the .MODEL directive. The
 predefined symbols @WordSize and @CurSeg return the size attribute and name
 of the current segment, respectively. See Section 1.2.3, "Predefined
 Symbols."


 Far Data Segments

 The compact, large, and huge memory models use far data addresses by
 default. With these memory models, however, you can still use .DATA, .DATA?,
 and .CONST to create data segments. The effect of these directives does not
 change from one memory model to the next. They always contribute segments to
 the default data area, DGROUP, which has a total limit of 64K.


 When you use .FARDATA or .FARDATA? in the small and medium memory models,
 the assembler creates far data segments FAR_DATA and FAR_BSS, respectively.
 You can access variables with:

   mov     ax, SEG farvar2
           mov     ds, ax

 See Section 3.1.2 for more information on far data.


 2.2.5  Creating Code Segments

 Whether you are writing a main module or a module to be called from another
 module, you can have both near and far code segments. This section explains
 how to use near and far code segments and how to use the directives and
 predefined equates that relate to code segments.


 Near Code Segments

 The small memory model is often the best choice for assembly programs that
 are not linked to modules in other languages, especially if you do not need
 more than 64K of code. This memory model defaults to near (two-byte)
 addresses for code and data, which makes the program run faster and use less
 memory.

 When you use .MODEL and simplified segment directives, the .CODE directive
 in your program instructs the assembler to start a code segment. The next
 segment directive closes the previous segment; the END directive at the end
 of your program closes remaining segments. The example at the beginning of
 Section 2.2, "Using Simplified Segment Directives," shows how to do this.

 You can use the predefined symbol @CodeSize to determine whether code
 pointers default to NEAR or FAR.


 Far Code Segments

 When you need more than 64K of code, use the medium, large, or huge memory
 model to create far segments.

 The medium, large, and huge memory models use far code addresses by default.
 In the larger memory models, the assembler creates a different code segment
 for each module. If you use multiple code segments in the small, compact, or
 tiny model, the linker combines the .CODE segments for all modules into one
 segment.

 The assembler assigns names to code segments.

 For far code segments, the assembler names each code segment MODNAME_TEXT,
 in which MODNAME is the name of the module. With near code, the assembler
 names every code segment _TEXT, causing the linker to concatenate these
 segments into one. You can override the default name by providing an
 argument after .CODE. (See Appendix E, "Default Segment Names," for a
 complete list of segment names generated by MASM.)

 With far code, a single module can contain multiple code segments. The .CODE
 directive takes an optional text argument that names the segment. For
 instance, the example below creates two distinct code segments, FIRST_TEXT
 and SECOND_TEXT.

   .CODE   FIRST
           .
           .       ; First set of instructions here
           .
           .CODE   SECOND
           .
           .       ; Second set of instructions here
           .

 Whenever the processor executes a far call or jump, it loads CS with the new
 segment address. No special action is necessary other than making sure that
 you use far calls and jumps. See Section 3.1.2, "Near and Far Addresses."

 ────────────────────────────────────────────────────────────────────────────
 NOTE
 The ASSUME directive is never necessary when you change code segments. In
 MASM 6.0, the assembler always assumes that the CS register contains the
 address of the current code segment or group. See Section 2.3.3 for more
 information about ASSUME used with segment registers.
 ────────────────────────────────────────────────────────────────────────────


 2.2.6  Starting and Ending Code with .STARTUP and .EXIT

 The easiest way to begin and end a program is to use the .STARTUP and .EXIT
 directives in the main module. The main module contains the starting point
 and usually the termination point. You do not need these directives in a
 module called by another module.

 .STARTUP generates the start-up code required by either DOS or OS/2.

 These directives make programs easy to maintain. They automatically generate
 code appropriate to the operating system and stack types specified with
 .MODEL. Thus, you can specify the program is for a different operating
 system or stack type by altering keywords in the .MODEL directive.

 To start a program, place the .STARTUP directive where you want execution to
 begin. Usually, this location immediately follows the .CODE directive:

   .CODE
           .STARTUP
           .
           .       ; Place executable code here
           .
           .EXIT
           END

 Note that .EXIT generates executable code, while END does not. The END
 directive informs the assembler that it has reached the end of the module.
 All modules must end with the END directive whether you use simplified or
 full segments.

 If you do not use .STARTUP, you must give the starting address as an
 argument to the END directive. When .STARTUP is present, the assembler
 ignores any argument to END.

 The code generated by .STARTUP depends on the operating system specified
 after .MODEL.

 If your program uses DOS for its operating system (the default), the
 initialization code sets DS to DGROUP, and adjusts SS:SP so that it is
 relative to the group for near data, DGROUP. To initialize a DOS program
 with the default NEARSTACK attribute, .STARTUP generates the following code:


   @Startup:
           mov     dx, DGROUP
           mov     ds, dx
           mov     bx, ss
           sub     bx, dx
           shl     bx, 1   ; If .286 or higher, this is
           shl     bx, 1   ;  shortened to shl bx, 4
           shl     bx, 1
           shl     bx, 1
           cli             ; Not necessary in .286 or higher
           mov     ss, dx
           add     sp, bx
           sti             ; Not necessary in .286 or higher
           .
           .
           .
           END     @Startup

 A DOS program with the FARSTACK attribute does not need to adjust SS:SP, so
 it just initializes DS:

   @Startup:
          mov     dx, DGROUP
          mov     ds, dx
          .
          .
          .
          END     @Startup

 OS/2 initializes DS so that it points to DGROUP and sets SS:SP as desired.
 Thus, when the OS_OS2 attribute is given, .STARTUP generates only a starting
 address. This does not show up in the listing file, however, since the /Sg
 option for listing files shows only the generated instructions.

 When the program terminates, you can return an exit code to the operating
 system. Applications that check exit codes usually assume that an exit code
 of 0 means no problem occurred and that 1 means an error terminated the
 program. The .EXIT directive accepts the exit code as its one optional
 argument:

   .EXIT   1       ; Return exit code 1

 This directive generates a DOS interrupt or OS/2 system call, depending on
 the operating system specified in .MODEL. The code generated under DOS
 depends on the argument provided to .EXIT. One example is

   mov     al, value
           mov     ah, 04Ch
           int     21h

 if a return value is specified. The return value can be a constant, a memory
 reference, or a register that can be moved into the AL register. If no
 return value is specified, the first line in the example code above is not
 generated.

 For OS/2, .EXIT invokes DosExit if you provide a prototype for DosExit and
 if you include OS2.LIB. The listing file shows the statements generated by
 INVOKE if the /Sg command-line option is specified. If you specify a return
 value as an expression, the code generated passes the expression instead of
 the register contents to the DosExit function. See Chapter 17 for
 information on writing programs for OS/2.


 2.3  Using Full Segment Definitions

 If you need complete control over segments, you can fully define the
 segments in your program. This section explains segment definitions,
 including how to order segments and how to define the segment types.

 If you write a program under DOS without .MODEL and .STARTUP, you must
 initialize registers yourself and use the END directive to indicate the
 starting address. Under OS/2 you do not have to initialize registers.
 Section 2.3.2, "Controlling the Segment Order," describes typical start-up
 code.


 2.3.1  Defining Segments with the SEGMENT Directive

 The SEGMENT directive begins a segment, and the ENDS directive ends a
 segment:

   name SEGMENT «align» «READONLY»
   «combine» «use» «'class'»
   statements
   name ENDS

 The name defines the name of the segment. Within a module, all segment
 definitions with the same name are treated as though they reference the same
 segment. The linker also combines identically named segments from different
 modules unless the combine type is PRIVATE. In addition, segments can be
 nested.

 Options used with the SEGMENT directive can be in any order.

 The optional types that follow the SEGMENT directive give the linker and the
 assembler instructions on how to set up and combine segments. The list below
 summarizes these types; the following sections explain them in more detail.


 Type                              Description
 ────────────────────────────────────────────────────────────────────────────
 align                             Defines the memory boundary on which a
                                   new segment begins.

 READONLY                          Tells the assembler to report an error
                                   if it detects an instruction modifying
                                   any item in a
                                   READONLY segment.

 combine                           Determines how the linker combines
                                   segments from different modules when
                                   building executable files.

 use (80386/486 only)              Determines the size of a segment. USE16
                                   indicates that offsets in the segment
                                   are 16 bits wide. USE32 indicates 32-bit
                                   offsets.

 class                             Provides a class name for the segment.
                                   The linker automatically groups segments
                                   of the same class in memory.


 Types can be specified in any order. You can specify only one attribute from
 each of these fields; for example, you cannot have two different align
 types.

 Once you define a segment, you can reopen it later with another SEGMENT
 directive. When you reopen a segment, you need only give the segment name.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 The PAGE align type and the PUBLIC combine type are distinct from the PAGE
 and PUBLIC directives. The assembler distinguishes them by means of context.
 ────────────────────────────────────────────────────────────────────────────


 Aligning Segments

 The optional align type in the SEGMENT directive defines the range of memory
 addresses from which a starting address for the segment can be selected. The
 align type can be any one of these:

 Align Type                        Starting Address
 ────────────────────────────────────────────────────────────────────────────
 BYTE                              Next available byte address.

 WORD                              Next available word address.

 DWORD                             Next available doubleword address.

 PARA                              Next available paragraph address (16
                                   bytes per paragraph). Default.

 PAGE                              Next available page address (256 bytes
                                   per page).


 The linker uses the alignment information to determine the relative starting
 address for each segment. The operating system calculates the actual
 starting address when the program is loaded.


 Making Segments Read-Only

 The optional READONLY attribute is helpful when creating read-only code
 segments for protected mode or when writing code to be placed in read-only
 memory (ROM). It protects against illegal self-modifying code.

 The READONLY attribute causes the assembler to check for instructions that
 modify the segment and to generate an error if it finds any. The assembler
 generates an error if you attempt to write directly to a read-only segment.



 Combining Segments

 The optional combine type in the SEGMENT directive defines how the linker
 combines segments having the same name but appearing in different modules.
 The combine type controls linker behavior, not assembler behavior. The
 combine types are described in full detail in online help and are summarized
 below.

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Combine Type                      Linker Action
 ────────────────────────────────────────────────────────────────────────────
 PRIVATE                           Does not combine the segment with
                                   segments from other modules, even if
                                   they have the same name.
                                   Default.

 PUBLIC                            Concatenates all segments having the
                                   same name to form a single, contiguous
                                   segment.

 STACK                             Concatenates all segments having the
 Combine Type                      Linker Action
 ────────────────────────────────────────────────────────────────────────────
STACK                             Concatenates all segments having the
                                   same name and causes the operating
                                   system to set SS:00 to the bottom and
                                   SS:SP to the top of the resulting
                                   segment. Data initialization is
                                   unreliable, as discussed below.

 COMMON                            Overlaps segments. The length of the
                                   resulting area is the length of the
                                   largest of the combined segments. Data
                                   initialization is unreliable, as
                                   discussed below.

 MEMORY                            Used as a synonym for the PUBLIC combine
                                   type.

 AT address                        Assumes address as the segment location.
                                   An AT segment cannot contain any code or
                                   initialized data, but it is useful for
 Combine Type                      Linker Action
 ────────────────────────────────────────────────────────────────────────────
                                  initialized data, but it is useful for
                                   defining structures or variables that
                                   correspond to specific far memory
                                   locations, such as a screen buffer or
                                   low memory.
                                   The AT combine type cannot be used in
                                   protected-mode programs.



 Do not place initialized data in STACK or COMMON segments. With these
 combine types, the linker overlays initialized data for each module at the
 beginning of the segment. The last module containing initialized data writes
 over any data from other modules.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 Normally, you should provide at least one stack segment (having STACK
 combine type) in a program. If no stack segment is declared, LINK displays a
 warning message. You can ignore this message if you have a specific reason
 for not declaring a stack segment. For example, you would not have a
 separate stack segment in a DOS tiny model (.COM) program, nor would you
 need a separate stack in a DLL library that used the caller's stack.
 ────────────────────────────────────────────────────────────────────────────


 Setting Segment Word Sizes (80386/486 Only)

 The use type in the SEGMENT directive specifies the segment word size on the
 80386/486 processors. Segment word size determines the default operand and
 address size of all items in a segment.

 The 80386/486 can operate in 16-bit or 32-bit mode.

 The size attribute can be USE16, USE32, or FLAT. If the 80386 or 80486
 processor has been selected with the .386 or .486 directive, and this
 directive precedes .MODEL, then USE32 is the default. This attribute
 specifies that items in the segment are addressed with a 32-bit offset
 rather than a 16-bit offset. If .MODEL precedes the .386 or .486 directive,
 USE16 is the default. To make USE32 the default, put .386 or .486 before
 .MODEL. You can override the USE32 default with the USE16 attribute.


 ────────────────────────────────────────────────────────────────────────────
 NOTE
 Mixing 16-bit and 32-bit segments in the same program is possible but
 usually is necessary only in systems programming.
 ────────────────────────────────────────────────────────────────────────────


 Setting Segment Order with Class Type

 Segments of the same class are grouped together in the executable file.

 The optional class type in the SEGMENT directive helps control segment
 ordering. Two segments with the same name are not combined if their class is
 different. The linker arranges segments so that all segments identified with
 a given class type are next to each other in the executable file. However,
 within a particular class, the linker orders segments in the order
 encountered. The .ALPHA, .SEQ, or .DOSSEG directive determines this order in
 each .OBJ file. The most common application for specifying a class type is
 to place all code segments first in the executable file.


 2.3.2  Controlling the Segment Order

 The assembler normally positions segments in the object file in the order in
 which they appear in source code. The linker, in turn, processes object
 files in the order in which they appear on the command line. Within each
 object file, the linker outputs segments in the order they appear, subject
 to any group, class, and .DOSSEG requirements.

 You can usually ignore segment ordering. However, it is important whenever
 you want certain segments to appear at the beginning or end of a program or
 when you make assumptions about which segments are next to each other in
 memory. For tiny model (.COM) programs, code segments must appear first in
 the executable file, because execution must start at the address 100h.


 Segment Order Directives

 You can control the order in which segments appear in the executable program
 with three directives. The default, .SEQ, arranges segments in the order in
 which they are declared.

 The .ALPHA directive specifies alphabetical segment ordering within a
 module. .ALPHA is provided for compatibility with early versions of the IBM
 assembler. If you have trouble running code from older books on assembly
 language, try using .ALPHA.

 The .DOSSEG directive specifies the DOS segment-ordering convention. It
 places segments in the standard order required by Microsoft languages. Do
 not use .DOSSEG in a module to be called from another module.

 The .DOSSEG directive orders segments in this order:


   1.  Code segments

   2.  Data segments, in this order:

       a.  Segments not in class BSS or STACK

       b.  Class BSS segments

       c.  Class STACK segments



 When you declare two or more segments to be in the same class, the linker
 automatically makes them contiguous. This rule overrides the
 segment-ordering directives. (See "Setting Segment Order with Class Type" in
 the previous section for more about segment classes.)


 Linker Control

 Most of the segment-ordering techniques (class names, .ALPHA, .SEQ) control
 the order in which the assembler outputs segments. Usually, you are more
 interested in the order in which segments appear in the executable file. The
 linker controls this order.

 The linker processes object files in the order in which they appear on the
 command line. Within each module, it then outputs segments in the order
 given in the object file. If the first module defines segments DSEG and
 STACK and the second module defines CSEG, then CSEG is output last. If you
 want to place CSEG first, there are two ways to do so.

 .DOSSEG handles segment ordering.

 The simpler method is to use .DOSSEG. This directive is output as a special
 record to the object file linker, and it tells the linker to use the
 Microsoft segment-ordering convention. This convention overrides
 command-line order of object files, and it places all segments of class
 'CODE'  first. (See Section 2.3.1, "Defining Segments with the SEGMENT
 Directive.")

 The other method is to define all the segments as early as possible (in an
 include file, for example, or in the first module). These definitions can be
 "dummy segments"─that is, segments with no content. The linker observes the
 segment ordering given, then later combines the empty segments with segments
 in other modules that have the same name.

 For example, you might include the following at the start of the first
 module of your program or in an include file:

   _TEXT   SEGMENT WORD PUBLIC 'CODE'
   _TEXT   ENDS
   _DATA   SEGMENT WORD PUBLIC 'DATA'
   _DATA   ENDS
   CONST   SEGMENT WORD PUBLIC 'CONST'
   CONST   ENDS
   STACK   SEGMENT PARA STACK 'STACK'
   STACK   ENDS

 Later in the program, the order in which you write  _TEXT,  _DATA, or other
 segments does not matter because the ultimate order is controlled by the
 segment order defined in the include file.


 2.3.3  Setting the ASSUME Directive for Segment Registers

 Many of the assembler instructions assume a default segment. For example,
 JMP assumes the segment associated with the CS register, PUSH and POP assume
 the segment associated with the SS register, and MOV instructions assume the
 segment associated with the DS register.


 The assembler must know the location of segment addresses.

 When the assembler needs to reference an address, it must know what segment
 contains the address. It finds this by using the default segment or group
 addresses assigned with the ASSUME directive. The syntax is

   ASSUME  segregister:seglocation [[,segregister:seglocation]]
   ASSUME  dataregister:qualifiedtype [[,dataregister:qualifiedtype]]
   ASSUME  register:ERROR [[,register:ERROR]]
   ASSUME  [[register:»NOTHING
   [[, register: NOTHING]]

 The seglocation must be the name of the segment or group that is to be
 associated with segregister. Subsequent instructions that assume a default
 register for referencing labels or variables automatically assume that if
 the default segment is segregister, the label or variable is in the
 seglocation. Beginning with MASM 6.0, the assembler automatically sets CS to
 have the address of the current code segment. Therefore, you do not need to
 include

   ASSUME CS : MY_CODE

 at the beginning of your program if you want the current segment associated
 with CS.

 ────────────────────────────────────────────────────────────────────────────
 NOTE
 Using the ASSUME directive to tell the assembler which segment to associate
 with a segment register is not the same as telling the processor. The ASSUME
 directive  affects only assembly-time assumptions. You may need to use
 instructions to change run-time assumptions. Initializing segment registers
 at run time is discussed in Section 3.1.1.1, "Informing the Assembler about
 Segment Values."
 ────────────────────────────────────────────────────────────────────────────

 The ASSUME directive can define a segment for each of the segment registers.
 The segregister can be CS, DS, ES, or SS (and FS and GS on the 80386/486).
 The seglocation must be one of the following:


   ■   The name of a segment defined in the source file with the SEGMENT
       directive

   ■   The name of a group defined in the source file with the GROUP
       directive

   ■   The keyword NOTHING, ERROR, or FLAT

   ■   A SEG expression (see Section 3.2.2, "Immediate Operands")

   ■   A string equate (text macro) that evaluates to a segment or group name
       (but not a string equate that evaluates to a SEG expression)


 It is legal to combine assumes to FLAT with assumes to specific segments.
 Combinations might be necessary in operating-system code that handles both
 16- and 32-bit segments.

 The keyword NOTHING cancels the current segment assumptions. For example,
 the statement ASSUME NOTHING cancels all register assumptions made by
 previous ASSUME statements.

 The ASSUME directive can be used anywhere in your program.

 Usually, a single ASSUME statement defines all four segment registers at the
 start of the source file. However, you can use the ASSUME directive at any
 point to change segment assumptions.

 Using the ASSUME directive to change segment assumptions is often equivalent
 to changing assumptions with the segment-override operator (:) (see Section
 3.2.3, "Direct Memory Operands"). The segment-override operator is more
 convenient for one-time overrides, whereas the ASSUME directive may be more
 convenient if previous assumptions must be overridden for a sequence of
 instructions.

 You can also prevent the use of a register with

   ASSUME  SegRegister : ERROR

 The assembler does an  ASSUME CS:ERROR  when you use simplified directives
 to create data segments, effectively preventing instructions or code labels
 from appearing in a data segment.

 See Section 3.3.2 for information on other applications of ASSUME.


 2.3.4  Defining Segment Groups

 A group is a collection of segments totalling not more than 64K in 16-bit
 mode. Each code or data item in the group can be addressed relative to the
 beginning of the group through DS or SS.

 Segments within a group can be treated as if they shared the same segment
 address.

 A group lets you develop separate segments for different kinds of data and
 then combine these into one segment (a group) for all the data. Using a
 group can save you from having to continually reload segment registers to
 access different segments. As a result, the program uses fewer instructions
 and runs faster.

 The most common example of a group is the specially named group for near
 data, DGROUP. In the Microsoft segment model, several segments (_DATA, _BSS,
 CONST, and STACK) are combined into a single group called DGROUP. Microsoft
 high-level languages place all near data segments in this group. (By
 default, the stack is placed here, too.) The .MODEL directive automatically
 defines DGROUP. The DS register normally points to the beginning of the
 group, giving you relatively fast access to all data in DGROUP.

 The syntax of the group directive is

   name GROUP segment [[,segment]]...

 The name labels the group. It can refer to a group that was previously
 defined. This feature lets you add segments to a group one at a time. For
 example, if MYGROUP  was previously defined to include  ASEG  and  BSEG,
 then the statement

   MYGROUP GROUP   CSEG

 is perfectly legal. It simply adds  CSEG  to the group  MYGROUP;  ASEG  and
 BSEG  are not removed.

 Each segment can be any valid segment name (including a segment defined
 later in source code), with one restriction: a segment cannot belong to more
 than one group.

 The GROUP directive does not affect the order in which segments of a group
 are loaded. You can place any number of 16-bit segments in a group as long
 as the total size does not exceed 65,536 bytes. If the processor is in
 32-bit mode, the maximum size is four gigabytes. You need to make sure that
 non-grouped segments do not get placed between grouped segments in such a
 way that the size of the group exceeds 64K or 4 gigabytes. Neither can you
 place a 16-bit and a 32-bit segment in the same group.


 2.4  Related Topics in Online Help

 In addition to information covered in this chapter, information on the
 following topics can be found in online help.

 Topic                                 Access
 ────────────────────────────────────────────────────────────────────────────
 Memory models                         Choose "Memory Models" from the list
                                       of tables on the "MASM 6.0 Contents"
                                       screen

 @Model, @CodeSize, @DataSize          Choose "Predefined Symbols" from the
                                       "MASM 6.0 Contents" screen

 Calling conventions                   From the MASM Index, choose "Calling
                                       Convention"

 Coprocessor Directives                From the "MASM 6.0 Contents" screen,
                                       choose "Directives"; then choose
                                       "Processor Selection"

 Simplified and full (complete)        From the "MASM 6.0 Contents" screen,
 segment control                       choose "Directives"; then choose
                                       "Simplified Segment Control" or
                                       "Complete Segment Control"







 Chapter 3  Using Addresses and Pointers
 ────────────────────────────────────────────────────────────────────────────

 Most processor and operating-system modes require the use of segmented
 addresses to access the code and data for MASM applications. The address of
 the code or data in a segment is relative to an address in a segment
 register. You can also use pointers to access data in MASM programs.

 The first section of this chapter describes how to initialize default
 segment registers to access near and far addresses. The next section
 describes how to use the available addressing modes to access the code and
 data. It also describes the related operators, syntax, and displacements.

 The third section of this chapter explains how to use the TYPEDEF directive
 to declare pointers (variables containing addresses) and the ASSUME
 directive to give the assembler information about registers containing
 pointers. This section also shows you how to do typical pointer operations
 and how to write code that works for pointer variables in any memory model.



 3.1  Programming Segmented Addresses

 Before you use segmented addresses in your programs, you need to initialize
 the segment registers. The initialization process depends on the registers
 used and on your choice of simplified segment directives or full segment
 definitions. The simplified segment directives (introduced in Section 2.2)
 handle most of the initialization process for you. This section explains how
 to inform the assembler and the processor of segment addresses, and how to
 access the near and far code and data in those segments.


 3.1.1  Initializing Default Segment Registers

 The segmented architecture of the 8086-family of processors does not require
 you to specify two addresses every time you access memory. As Chapter 2,
 "Organizing MASM Segments," explains, the 8086 family of processors uses a
 system of default segment registers to simplify access to the most commonly
 used data and code.

 The segment registers DS, SS, and CS are normally initialized to default
 segments at the beginning of a program. If you write the main module in a
 high-level language, the compiler initializes the segment registers. If you
 write the

 main module in assembly language, you must initialize them yourself. Follow
 these two steps to initialize segments:


   1.  Tell the assembler which segment is associated with a register. The
       assembler must know the default segments at assembly time.

   2.  Tell the processor which segment is associated with a register by
       writing the necessary code to load the correct segment value into the
       segment register on the processor.


 These steps are discussed separately in the following sections.


 3.1.1.1  Informing the Assembler about Segment Values

 Use ASSUME to inform the assembler about default segments.

 The first step in initializing segments is to tell the assembler which
 segment to associate with a register. You do this with the ASSUME directive.
 If you use simplified segment directives, the assembler generates the
 appropriate ASSUME statements automatically. If you use full segment
 definitions, you must code the ASSUME statements for registers other than CS
 yourself. (ASSUME can also be used on general-purpose registers, as
 explained in Section 3.3.2, "Defining Register Types with ASSUME.")

 With simplified segment directives, the .STARTUP directive and the start-up
 code initialize DS to be equal to SS (unless you specify FARSTACK), which
 allows default data to be accessed through either SS or DS. This can improve
 efficiency in the code generated by compilers. The "DS equals SS" convention
 may not work with certain applications, such as memory-resident programs in
 DOS and multithread programs in OS/2. The code generated for .STARTUP is
 shown in Section 2.2.6, "Starting and Ending Code with .STARTUP and .EXIT."
 You can use similar code to set DS equal to SS in programs using full
 segment definitions.

 Here is an example using full segment definitions; it is equivalent to the
 ASSUME statement generated with simplified segment directives in small model
 with NEARSTACK:

   ASSUME cs:_TEXT, ds:DGROUP, ss:DGROUP

 In the example above, DS and SS are part of the same segment group. It is
 also possible to have different segments for data and code, and to use
 ASSUME to set ES, as shown below:

   ASSUME cs:MYCODE, ds:MYDATA, ss:MYSTACK, es:OTHER

 Correct use of the ASSUME statement can help find addressing errors. With
 .CODE, the assembler assumes CS to the current segment. When you use the
 simplified segment directives .DATA, .DATA?, .CONST, .FARDATA, or .FARDATA?,
 the assembler automatically assumes CS to ERROR. This prevents

 instructions from appearing in these segments. If you use full segment
 definitions, you can accomplish the same by placing  ASSUME CS:ERROR  in a
 data segment.

 With either simple or full segments, you can cancel the control of an ASSUME
 statement by assuming NOTHING. No assumptions is the default condition. For
 example, you cancel the assumption for ES above with the following
 statement:

   ASSUME es:NOTHING

 Prior to the .MODEL statement (or in its absence), the assembler sets the
 ASSUME statement for DS, ES, and SS to the current segment.


 3.1.1.2  Informing the Processor about Segment Values

 The second step in initializing segments is to inform the processor of
 segment values at run time. How segment values are initialized at run time
 differs for each segment register and depends on your use of simplified
 segment directives or full segment definitions and on the operating system.


 Specifying a Starting Address - The CS segment register and the IP
 (instruction pointer) register are initialized automatically if you use the
 .STARTUP directive with simplified segment directives. If you use full
 segment definitions, you must specifically set a label in the code segment
 at the instruction you want executed first. Then provide that label as an
 argument to the END directive. Both CS and IP are set at load time to the
 start address the linker gets from the END directive:

   _TEXT   SEGMENT WORD PUBLIC 'CODE
           ORG     100h    ; Use this declaration for .COM files only
   start:                  ; First instruction here
           .
           .
           .
   _TEXT   ENDS
           END     start   ; Name of starting label

 The operating system automatically resolves the value of CS:IP at load time.
 The label specified as the start address becomes the initial value of IP. In
 an executable (.EXE) file, the start address is encoded into the header and
 is initialized by the operating system at load time. In a .COM file, the
 initial IP is always assumed to be 100h. Therefore, you must use the ORG
 directive to set the start address to 100h. CS and IP cannot be directly
 modified except through jump, call, and interrupt instructions.

 DS is initialized automatically under OS/2, but you must initialize it for
 DOS.

 Initializing DS - The DS register is automatically initialized to the
 correct value (DGROUP) if you use .STARTUP or if you are writing a program
 for OS/2. If you do not use .STARTUP with DOS, you must initialize DS using
 the following instructions:

   mov     ax, DGROUP
           mov     ds, ax

 The initialization requires two instructions because the segment name is a
 constant and the assembler does not allow a constant to be loaded directly
 to a segment register. The example above loads DGROUP, but you can load any
 valid segment or group.

 SS and SP are initialized automatically.

 Initializing SS and SP - The SS and SP registers are initialized
 automatically if you use the .STACK directive with simplified segments or if
 you define a segment that has the STACK combine type with full segment
 definitions. Using the STACK directive initializes SS to the stack segment.
 If you want SS to be equal to DS, use .STARTUP or its equivalent. (See
 "Combining Segments" in Section 2.3.1.) For an executable file, the values
 are encoded into the executable header and resolved at link time. For a .COM
 file, SS is initialized to the first address of the 64K program segment and
 SP is initialized to 0FFFEh.

 If you do not need to access far data in your program, you do not need to
 initialize the ES register, although you can do so. Use the same technique
 as for the DS register. You can initialize SS to a far stack in the same
 way.


 3.1.2  Near and Far Addresses

 Addresses which have an implied segment name or segment registers associated
 with them are called "near addresses." Addresses which have an explicit
 segment associated with them are called "far addresses." The assembler
 handles near and far code automatically, as described below. You must
 specify how to handle far data.

 The Microsoft segment model puts all near data and the stack in a group
 called DGROUP. Near code is put in a segment called _TEXT. Each module's far
 code or far data is placed in a separate segment. This convention is
 described in Section 2.3.2, "Controlling the Segment Order."

 The assembler cannot determine the address for some program components,
 which are said to be relocatable. The assembler generates a fixup record and
 the linker provides the address once the location of all segments has been
 determined. Usually a relocatable operand references a label, but there are
 exceptions. Examples in the next two sections include information about the
 relocatability of near and far data.

 Near Code - Control transfers within near code do not require changes to
 segment registers. The processor automatically handles changes to the offset
 in the IP register when control-flow instructions such as JMP, CALL, and RET
 are used. The statement

   call    nearproc        ; Change code offset

 changes the IP register to the new address but leaves the segment unchanged.
 When the procedure returns, the processor resets IP to the offset of the
 next instruction after the call.

 Far Code - The processor automatically handles segment register changes when
 dealing with far code. The statement

   call    farproc         ; Change code segment and offset

 automatically moves the segment and offset of the  farproc  procedure to the
 CS and IP registers. When the procedure returns, the processor sets CS to
 the original code segment and sets IP to the offset of the next instruction
 after the call.

 Near Data - Near data can usually be accessed directly. That is, a segment
 register already holds the correct segment for the data item. The term "near
 data" is often used to refer to the data in the DGROUP group.

 After the first initialization of the DS and SS registers, these registers
 normally point into DGROUP. If you modify the contents of either of these
 registers during the execution of the program, the register may need to be
 reloaded prior to being used for addressing DGROUP data.

 If a stack variable is accessed directly through BP or SP, the SS register
 is the default. Otherwise, the default is DS:

   nearvar WORD    0
           .
           .
           .
           mov     ax, nearvar ; Access near data through DS or SS
           mov     ax, [bp+6]  ; Access near data through SS

 In this example,  nearvar  is a relocatable label. The assembler does not
 know where the memory for  nearvar  will be allocated. The linker provides
 the address at link time. The expression  [bp+6]  is not relocatable. The
 linker does not need to provide an address for this expression.

 Far Data - To read or modify a far address, a segment register must point to
 the segment of the data. This requires two steps. First load the segment
 (normally either ES or DS) with the correct value, and then (optionally) set
 an assume of the segment register to the segment of the address (or to
 NOTHING).

 ────────────────────────────────────────────────────────────────────────────
 NOTE
 In flat model (OS/2 2.x), far addresses are rarely used. By default, all
 addressing is relative to the initial values of the segment registers. Thus,
 this section on far addressing does not apply to most flat model programs.
 ────────────────────────────────────────────────────────────────────────────

 You can initialize ES.

 One method commonly used to access far data is to initialize the ES segment
 register. This example shows two ways to do this:

   ; First method
           mov     ax, SEG farvar  ; Load segment of the far address
           mov     es, ax
           mov     ax, es:farvar   ; Provide an explicit segment
                                   ;  override on the addressing
   ; Second method
           mov     ax, SEG farvar2 ; Load the segment of the
                                   ;  far address
           mov     ex, ax
           ASSUME  ES:SEG farvar2  ; Tell the assembler that ES points
                                   ;  to the segment containing farvar2
           mov     ax, farvar2     ; The assembler provides the ES
                                   ;  override since it knows that
                                   ;  the label is addressable

 After loading the segment of the address into the ES segment register, you
 can either explicitly override the segment register so that the addressing
 is correct (method 1) or allow the assembler to insert the override for you
 (method 2). The assembler uses ASSUME statements to determine which segment
 register can be used to address a segment of memory. To use the segment
 override operator, the left operand must be a segment register, not a
 segment name. (See Section 3.2.3 for more information on segment overrides.)


 If an instruction needs a segment override, the resulting code is slightly
 larger and slower, since the override must be encoded into the instruction.
 However, the resulting code may still be smaller than the code for multiple
 loads of the default segment register for the instruction.

 The DS, SS, FS, and GS segment registers (FS and GS are available only on
 the 80386/486 processors) may also be used to provide for addressing through
 other segments.

 If a program uses ES to access far data, it need not restore ES when
 finished (unless the program uses flat model). Some compilers require that
 you restore ES before returning to a module written in a high-level
 language.

 You can reinitialize DS.

 For a series of memory accesses to far data, you can reinitialize DS to the
 far data and then restore DS when you are finished. Use the ASSUME directive
 to let the assembler know that DS is no longer associated with the default
 data segment, as shown below:

   push    ds               ; Save original segment
           mov     ax, SEG fararray ; Move segment into data register
           mov     ds, ax           ; Initialize segment register
           ASSUME  ds:SEG fararray  ; Tell assembler where data is
           mov     ax, fararray[0]  ; Direct access faster
           mov     dx, fararray[2]  ; (A relocatable expression)
           .
           .
           .
           pop     ds               ; Restore segment
           ASSUME  ds:@DATA         ;  and default assumption

 The additional overhead of saving and restoring the DS register in this data
 access method may be worthwhile to avoid repeated segment overrides.

 If a program changes DS to access far data, it should restore DS when
 finished. This allows procedures to assume that DS is the segment for near
 data. This is a convention used in many compilers, including Microsoft
 compilers.

 Relocatable Data - The memory expression  es:farvar  is a relocatable memory
 expression, since the assembler cannot determine the address at assembly
 time.

 Since no label is referenced, you may expect

   mov ax, _myseg:0

 to be nonrelocatable (in small model). However, in this case,  _myseg:0  is
 a location in a local module whose memory location is dependent on the link
 order, so  mov ax, _myseg:0  is relocatable.

 A group name is also an immediate constant representing the beginning of the
 group. The first three expressions below are relocatable expressions; the
 fourth is not.

   mov ax, DGROUP             ; Relocatable
           mov ax, @data              ; Relocatable
           mov ax, mygroup            ; Relocatable
           mov ax, ds:0               ; Not relocatable


 3.2  Specifying Addressing Modes

 The 8086 family of processors recognizes four kinds of instruction operands:
 register, immediate, direct memory, and indirect memory. Each type of
 operand corresponds to a different addressing mode.

 The four types of operands are summarized in the following list and
 described at length in the rest of this section.

 Operand Type                      Addressing Mode
 ────────────────────────────────────────────────────────────────────────────
 Register                          An 8-bit or 16-bit register on the
                                   8086-80486; can also be 32-bit on the
                                   80386/486

 Immediate                         A constant value contained in the
                                   instruction itself

 Direct memory                     A fixed location in memory

 Indirect memory                   A memory location determined at run time
                                   by using the address stored in one or
                                   two registers and a constant



 3.2.1  Register Operands

 A register operand specifies that the value in a particular register is an
 operand. Code for the register or registers used in operands is encoded into
 the instruction at assembly time.

 Register operands can be used anywhere you need an operand. The following
 examples show typical register operands:

   mov     bx, 10          ; Load constant to BX
           add     ax, bx          ; Add AX and BX
           jmp     di              ; Jump to the address in DI

 Register operands have a specific use related to addresses.

 An offset stored in a base or index register is often used as a pointer into
 memory. An offset can be stored in one of the base or index registers; the
 register can then be used as an indirect memory operand (see Section 3.2.4).
 For example:

   mov     [bx], dl ; Store DL in indirect memory operand
           inc     bx       ; Increment register operand
           mov     [bx], dl ; Store DL in new indirect memory operand

 This example moves the value in DL to two consecutive bytes of a memory
 location pointed to by BX. Any instruction that changes the register value
 also changes the data item pointed to by the register.


 3.2.2  Immediate Operands

 An immediate operand is a constant value that is specified at assembly time.
 It can be a constant or the result of a constant expression. Immediate
 values are usually encoded into the internal representation of the
 instruction at assembly time. These are typical examples:

   mov     cx, 20          ; Load constant to register
           add     var, 1Fh        ; Add hex constant to variable
           sub     bx, 25 * 80     ; Subtract constant expression

 The OFFSET Operator - Address constants are a special case of immediate
 operand and consist of an offset or segment value. The OFFSET operator
 specifies the offset of a memory location, as shown below:

   mov     bx, OFFSET var  ; Load offset address

 For information on differences between MASM 5.1 behavior and MASM 6.0
 behavior related to OFFSET, see Appendix A.

 An OFFSET expression is resolved at link time.

 Since segments in different modules may be combined into a single segment,
 the true base of the segment is not known. Thus, the offset cannot be
 resolved until link time and  var  is a relocatable immediate.

 The SEG Operator - The SEG operator specifies the segment of a memory
 location:

   mov     ax, SEG farvar  ; Load segment address
           mov     es, ax

 A SEG expression is resolved at load time.

 The actual value of a particular segment is never known until the program is
 loaded into memory. Constant segments are encoded into the header of the
 executable file at link time. Executable files in the DOS .COM format (tiny
 model) cannot contain relocatable segment expressions.

 When you use the SEG operator with a variable that is not external, MASM 6.0
 returns the address of the frame (the segment, group, or segment register)
 if one has been explicitly set. Otherwise, it returns the group if one has
 been specified. In the absence of a defined group, SEG returns the segment
 where the variable is defined.

 For external variables that are not defined in a segment, the linker fills
 in the segment portion of the address, which may be a segment or group.

 This behavior can be changed with the /Zm command-line option or with the
 OPTION OFFSET:SEGMENT statement (see Appendix A, "Differences between MASM
 6.0 and 5.1"). Section 1.3.2 introduces the OPTION directive.


 3.2.3  Direct Memory Operands

 A direct memory operand specifies the data at a given address. The address
 and size of the data are encoded into the internal representation of the
 instruction. However, the instruction acts on the contents of the address,
 not the address itself. You must usually specify the size of these operands
 so that the instruction knows how much memory to operate on.

 The offset value of a direct memory operand is not resolved until link time,
 and the segment must always be in a segment register at run time. The
 assembler automatically handles address resolution.

 You usually represent a direct memory operand in source code as a symbolic
 name previously declared with a data directive such as BYTE, as illustrated
 below:

   .DATA?          ; Segment for uninitialized data
   var     BYTE   ?        ; Reserve one byte at current address
                           ;  and assign this address to var
           .CODE
           .
           .
           .
           mov     var, al ; Load contents of byte register into
                              address specified by var

 Any location in memory can be a direct memory operand as long as a size is
 specified and the location is fixed. The data at the address can change, but
 the address cannot. By default, instructions that use direct memory
 addressing use the DS register. You can create an expression that points to
 a memory location using any of the following operators:

 Operator Name                     Symbol
 Plus
 ────────────────────────────────────────────────────────────────────────────
 Minus                             -
 Index                             [ ]
 Structure member                  .
 Segment override                  :

 These operators are discussed in more detail below.

 Several operators can be used in expressions that evaluate to direct memory
 operands.

 Plus and Minus - The result of combining a memory operand and a constant
 number with the plus or minus operator is a direct memory operand. However,
 the result of combining two memory operands with the minus operator is an
 immediate operand. For example:

   memvar  EQU     array + 5       ; Address five bytes beyond
   array
   immexp  EQU     mem1 - mem2     ; Distance between addresses

 The second expression is legal only if both addresses are in the same
 segment.

 The expression  mem1 - mem2  is not relocatable, since the reference to the
 two labels represents a difference in addresses (offsets). The linker does
 not need to know about the labels in this statement.

 Index - The index operator (brackets enclosing an index value) specifies the
 register or registers for indirect operands. It should contain a constant
 index when used with direct memory operands. It is equivalent to the plus
 operator. For example, the following statements are the same:

   mov     ax, array[5]
           mov     ax, array+5

 Any direct memory operand can be enclosed in the index operator. The
 following are equivalent:

   mov     ax, var
           mov     ax, [var]

 Some programmers prefer to enclose the operand in brackets to show that the
 contents, not the address, are used.

 Structure Field - The structure operator (a period) accesses elements of a
 structure. A field within a structure variable can be accessed as a direct
 memory operand:

   mov     bx, structvar.field1

 The address of the structure operand is the sum of the offsets of  structvar
  and  field1. See Section 5.2, "Structures and Unions," for more information
 about structures.

 Segment Override - The segment override operator (a colon) specifies a
 segment portion of the address that is different from the default segment.
 When used with instructions, this operator can apply to segment registers or
 segment names:

   mov     ax, es:farvar           ; Use segment override

 The assembler will not generate a segment override if the default segment is
 explicitly provided. Thus, the following two statements are equivalent:

   mov     [bx], ax
           mov     ds:[bx], ax

 A segment name override or the segment override operator forces the operand
 to be an address expression.

   mov     WORD PTR FARSEG:0, ax   ; Segment name override
           mov     WORD PTR es:100h, ax    ; Legal and equivalent
           mov     WORD PTR es:[100h], ax  ;  expressions
   ;       mov     WORD PTR [100h], ax     ; Illegal, not an address

 As the example shows, a constant expression cannot be an address expression
 unless it has a segment override.


 3.2.4  Indirect Memory Operands

 Like direct memory operands, indirect memory operands specify the contents
 of a given address. However, the processor calculates the address at run
 time by referring to the contents of registers. Since values in the
 registers can change at run time, indirect memory operands provide dynamic
 access to memory.

 Indirect memory operands make possible run-time operations such as pointer
 indirection and dynamic indexing of array elements, including indexing of
 multidimensional arrays.

 Strict rules govern which registers can be used for indirect memory operands
 under 16-bit versions of the 8086-based processors. The rules change
 significantly for 32-bit processors starting with the 80386. However, the
 new rules apply only to code that does not need to be backward compatible.

 This section first discusses features of indirect operands in either mode.
 Then it explains the specific 16-bit rules and 32-bit rules separately.


 3.2.4.1  Indirect Operands with 16- and 32-Bit Registers

 Some rules and options for indirect memory operands always apply, regardless
 of the size of the register. For example, you must always specify the
 register and operand size for indirect memory operands. But you can use
 various syntaxes to indicate an indirect memory operand. This section
 describes the rules that apply to both 16-bit and 32-bit register modes.

 Certain rules govern the use of base and index registers.

 Specifying Indirect Memory Operands - The index operator specifies the
 register or registers for indirect operands. The processor uses the data
 pointed to by the register. For example, the following instruction moves the
 word-sized data at the address contained in DS:BX into AX:

   mov     ax, WORD PTR [bx]

 When you specify more than one register, the processor adds the two
 addresses together to determine the effective address (the address of the
 data to operate on):

   mov     ax, [bx+si]

 An indirect memory operand can have a displacement.

 Specifying Displacements - You can specify an address displacement─ a
 constant value to add to the effective address. A direct memory specifier is
 the most common displacement:

   mov     ax, table[si]

 In the relocatable expression above, the displacement  table  is the base
 address of an array; SI holds an index to an array element. The SI value is
 calculated at run time, often in a loop. The element loaded into AX depends
 on the value of SI at the time the instruction is executed.

 Each displacement can be an address or numeric constant. If there is more
 than one displacement, the assembler adds them together at assembly time and
 encodes the total displacement. For example, in the statement

   table   WORD    100 DUP (0)
           .
           .
           .
           mov     ax, table[bx][di]+6

 both  table  and  6  are displacements. The assembler adds the value of
 table  to  6  to get the total displacement. However, this statement is not
 legal:

   mov ax, mem1[si] + mem2

 Indirect memory operands must always have a size.

 Specifying Operand Size - Indirect memory operands must always have a
 specified size. Often the size is specified by the size of the identifier.
 In the example above, the size of the  table  array determines the operand
 size. If an indirect memory operand is used with a register operand, the
 register size determines the size of the memory object:

   mov     ax, [bx]        ; Size is 2 bytes - same as
   AX
           mov     table[bx], 0    ; Size is 2 bytes - from size
                                   ;  of table

 If there is no address or register operand, the size must be given
 specifically with the PTR operator, as shown below:

   inc     WORD PTR [bx]           ; Word size
           mov     BYTE PTR [bp+6], 0      ; Byte size

 Syntax Options - The assembler allows a variety of syntaxes for indirect
 memory operands. However, all registers must be inside brackets. You can
 enclose each register in its own pair of brackets, or you can place the
 registers in the same pair of brackets separated by a plus operator (+). All
 the following variations are legal and equivalent:

   mov     ax, table[bx][di]
           mov     ax, table[di][bx]
           mov     ax, table[bx+di]
           mov     ax, [table+bx+di]
           mov     ax, [bx][di]+table

 All of these statements move the value in  table  indexed by  BX+DI  into
 AX.

 Registers pointing into arrays must be zero-based and scaled for the size of
 the array.

 Scaling Indexes - The value of index registers pointing into arrays must
 often be adjusted for zero-based arrays and scaled according to the size of
 the array items. For a word array, the item number must be multiplied by two
 (shifted left two places). When you are using 16-bit registers, scaling must
 be done with separate instructions, as shown below:

   mov     bx, 5           ; Get sixth element (adjust
   for 0)
           shl     bx, 1           ; Scale by two (word size)
           inc     wtable[bx]      ; Increment sixth element in table

 When using 32-bit registers on the 80386/486 processor, you can include
 scaling in the operand, as described in Section 3.2.4.3, "Indirect Memory
 Operands with 32-Bit Registers."

 Accessing Structure Elements  - The structure member operator can be used in
 indirect memory operands to access structure elements. In this example, the
 structure member operator loads the  year  field of the fourth element of
 the  students  array into AL:

   STUDENT STRUCT
     grade WORD    ?
     name  BYTE    20 DUP (?)
     year  BYTE    ?
   STUDENT ENDS

   students        STUDENT  < >
           .
           .                               ; Assume array initialized
           .                               ;  earlier
           mov     bx, OFFSET students     ; Point to array of students
           mov     ax, 4                   ; Get fourth element
           mov     di, SIZE STUDENT        ; Get size of STUDENT
           mul     di                      ; Multiply size times
                                           ;  elements to point to
                                           ;  current element
                                           ; Load field from element:
           mov     al, (STUDENT PTR[bx+di]).year

 See Section 5.2 for more information on MASM structures.


 3.2.4.2  Indirect Memory Operands with 16-Bit Registers

 For 8086-based computers and DOS, you must follow the strict indexing rules
 established for the 8086 processor. Only four registers are allowed─BP, BX,
 SI, and DI─and those only in certain combinations.

 BP and BX are base registers. SI and DI are index registers. You can use
 either a base or an index register by itself. But if you combine two
 registers, one must be a base and one an index. Here are legal and illegal
 forms:

   mov     ax, [bx+di]     ; Legal
           mov     ax, [bx+si]     ; Legal
           mov     ax, [bp+di]     ; Legal
           mov     ax, [bp+si]     ; Legal
   ;       mov     ax, [bx+bp]     ; Illegal - two base registers
   ;       mov     ax, [di+si]     ; Illegal - two index registers

 Table 3.1 shows the modes in which registers can be used to specify indirect
 memory operands.

 Table 3.1  Indirect Addressing Modes with 16-Bit Registers

╓┌─────────────────────┌────────────────────────┌────────────────────────────╖
 Mode                  Syntax                   Effective Address
 ────────────────────────────────────────────────────────────────────────────
 Register indirect     [BX]                     Contents of register
                       [BP]
                       [DI]
 Mode                  Syntax                   Effective Address
 ────────────────────────────────────────────────────────────────────────────
                      [DI]
                       [SI]

 ────────────────────────────────────────────────────────────────────────────

 Base or index         displacement[BX]         Contents of register plus
                       displacement[BP]         displacement
                       displacement[DI]
                       displacement[SI]

 ────────────────────────────────────────────────────────────────────────────

 Base plus index       [BX][DI]                 Contents of base register
                       [BP][DI]                 plus contents of index
                       [BX][SI]                 register
                       [BP][SI]

 ────────────────────────────────────────────────────────────────────────────

 Mode                  Syntax                   Effective Address
 ────────────────────────────────────────────────────────────────────────────

 Base plus index with  displacement[BX][DI]     Sum of base register, index
 displacement          displacement[BP][DI]     register, and displacement
                       displacement[BX][SI]
                       displacement[BP][SI]

 ────────────────────────────────────────────────────────────────────────────



 Different combinations of registers and displacements have different
 timings, as shown in the Macro Assembler Reference.


 3.2.4.3  Indirect Memory Operands with 32-Bit Registers

 Instructions for the 80386/486 processor can be given in two segment
 modes─16-bit and 32-bit. Indirect memory operands are different in each
 mode. The segment mode is independent of the register size; you can use
 32-bit registers in either mode.

 In 16-bit mode, the 80386/486 operates in the mode used by all other
 8086-based processors, with one difference: you can use 32-bit registers. If
 the 80386/486 processor is enabled (with the .386 or .486 directive), 32-bit
 general-purpose registers are available in either segment mode. Using them
 eliminates many of the limitations of 16-bit indirect memory operands. Using
 80386/486 features can make your DOS programs run faster and more
 efficiently if you are willing to sacrifice backward compatibility with
 other processors.

 In 32-bit mode, an offset address can be up to four gigabytes. (Segments are
 still represented in 16 bits.) This effectively eliminates size restrictions
 on each segment, since few programs need four gigabytes of memory. OS/2 2.x
 uses 32-bit mode and flat model, which spans all segments. XENIX 386 uses
 32-bit mode with multiple segments.

 Any general-purpose 32-bit register can be used as either the base or the
 index.

 80386/486 Enhancements - On the 80386/486, the processor allows any
 general-purpose 32-bit register to be used as either the base or the index
 register (except ESP, which can be a base but not an index). The same
 register can also be used as both the base and index, but you cannot combine
 16-bit and 32-bit registers. Several examples are shown below:

   add     edx, [eax]              ; Add double
           mov     dl, [esp+10]            ; Add byte from stack
           dec     WORD PTR [edx][eax]     ; Decrement word
           cmp     ax, array[ebx][ecx]     ; Compare word from array
           jmp     FWORD PTR table[ecx]    ; Jump into pointer table

 The index register can have a scaling factor of 1, 2, 4, or 8.

 Scaling Factors - With 80386/486 registers, the index register can have a
 scaling factor of 1, 2, 4, or 8. Any register except ESP can be the index
 register and can have a scaling factor. Specify the scaling factor by using
 the multiplication operator (*) adjacent to the register.

 You can use scaling to index into arrays with different sizes of elements.
 For example, the scaling factor is 1 for byte arrays (no scaling needed), 2
 for word arrays, 4 for doubleword arrays, and 8 for quadword arrays. There
 is no performance penalty for using a scaling factor. Scaling is illustrated
 in the following examples:

   mov     eax, darray[edx*4]     ; Load double of double
   array
           mov     eax, [esi*8][edi]      ; Load double of quad array
           mov     ax, wtbl[ecx+2][edx*2] ; Load word of word array

 Scaling is also necessary on earlier processors, but it must be done with
 separate instructions before the indirect memory operand is used, as
 described in Section 3.2.4.2, "Indirect Memory Operands with 16-Bit
 Registers."

 The number of registers and the scaling factor affect base and index
 registers.

 The default segment register is SS if the base register is EBP or ESP; it is
 DS for all other base registers. If two registers are used, only one can
 have a scaling factor. The register with the scaling factor is defined as
 the index register. The other register is defined as the base. If scaling is
 not used, the first register is the base. If only one register is used, it
 is considered the base for deciding the default segment unless it is scaled.
 The following examples illustrate how to determine the base register:

   mov   eax, [edx][ebp*4] ; EDX base (not scaled - seg
   DS)
           mov   eax, [edx*1][ebp] ; EBP base (not scaled - seg SS)
           mov   eax, [edx][ebp]   ; EDX base (first - seg DS)
           mov   eax, [ebp][edx]   ; EBP base (first - seg SS)
           mov   eax, [ebp*2]      ; EBP base (only - seg SS)

 Mixing 16-Bit and 32-Bit Registers - Statements can mix 16-bit and 32-bit
 registers if the register use is correct. For example, the following
 statement is legal for either 16-bit or 32-bit segments:

   mov     eax, [bx]

 This statement moves the 32-bit value pointed to by BX into the EAX
 register. Although BX is a 16-bit pointer, it can still point into a 32-bit
 segment.

 However, the following statement is never legal, since the CX register
 cannot be used as a 16-bit pointer (although ECX can be used as a 32-bit
 pointer):

   ;       mov     eax, [cx]       ; illegal

 Operands that mix 16-bit and 32-bit registers are also illegal:

   ;       mov     eax, [ebx+si]   ; illegal

 The following statement is legal in either mode:

   mov     bx, [eax]

 This statement moves the 16-bit value pointed to by EAX into the BX
 register. This works fine in 32-bit mode. However, in 16-bit mode, moving a
 32-bit pointer into a 16-bit segment is illegal. If EAX contains a 16-bit
 value (the top half of the 32-bit register is 0), the statement works.
 However, if the top half of the EAX register is not 0, the operand points
 into a part of the segment that doesn't exist, and this generates an error.
 If you use 32-bit registers as indexes in 16-bit mode, you must make sure
 that the index registers contain valid 16-bit addresses.


 3.3  Accessing Data with Pointers and Addresses

 In high-level languages, a "pointer" (or pointer variable) is an address
 that is stored in a variable. Assembly language also uses pointer variables,
 but the term "pointer" has a wider use. The indirect memory operands
 discussed in the previous section can be thought of as pointers stored in
 registers.

 An address can be stored in a pointer variable for later use. Program
 procedures (including OS/2 systems calls) frequently pass pointer variables
 onto the stack to transfer data between the calling program and the called
 procedure.

 A pointer variable must be transferred to registers before it can be used.

 Regardless of the reason for maintaining it, a pointer variable to data
 cannot in itself be directly used in MASM statements. (Pointers to code can
 be used directly.) It must first be loaded into registers as an indirect
 memory operand.

 There is a difference between a far address and a far pointer. A "far
 address" is the address of a variable located in a far data segment. A "far
 pointer" is a variable that can specify both a segment and an offset. Like
 any other variable, a pointer variable can be located in either the default
 (near) data segment or in a far segment.

 Previous versions of MASM allow pointer variables but provide little support
 for them. In previous versions, any address loaded into a variable can be
 considered a pointer, as in the following statements:

   Var     BYTE    0               ; Variable
   npVar   WORD    Var             ; Near pointer to variable
   fpVar   DWORD   Var             ; Far pointer to variable

 If a variable is initialized to the name of another variable, the
 initialized variable is a pointer, as shown in the example above. However,
 in previous versions of MASM, the CodeView debugger recognizes  npVar  and
 fpVar  as word and doubleword variables. CodeView does not treat them as
 pointers, nor does it recognize the type of data they point to (bytes, in
 the example).

 The new directive TYPEDEF and the new capabilities of ASSUME make it easier
 to manage pointers in registers and variables. These directives are
 discussed in the next two sections. Basic pointer and address operations are
 covered in Section 3.3.3.


 3.3.1  Defining Pointer Types with TYPEDEF

 Once defined, a TYPEDEF is considered the same as an intrinsic type.

 You can define types for pointer variables using the TYPEDEF directive. A
 type so defined is considered the same as the intrinsic types provided by
 the assembler and can be used in the same contexts. The syntax for TYPEDEF
 when used to define pointers is

   typename TYPEDEF «distance» PTR  qualifiedtype

 The typename is the name assigned to the new type. The distance can be NEAR,
 FAR, or any distance modifier. The qualifiedtype can be any previously
 intrinsic or defined MASM type, or a type previously defined with TYPEDEF.
 (See Section 1.2.6, "Data Types," for a full definition of qualifiedtype.)

 Here are some examples of user-defined types:

   PBYTE   TYPEDEF      PTR BYTE   ; Pointer to bytes
   NPBYTE  TYPEDEF NEAR PTR BYTE   ; Near pointer to bytes
   FPBYTE  TYPEDEF FAR  PTR BYTE   ; Far pointer to bytes
   PWORD   TYPEDEF      PTR WORD   ; Pointer to words
   NPWORD  TYPEDEF NEAR PTR WORD   ; Near pointer to words
   FPWORD  TYPEDEF FAR  PTR WORD   ; Far pointer to words

   PPBYTE  TYPEDEF      PTR PBYTE  ; Pointer to pointer to bytes
                                   ;  (in C, an array of strings)
   PVOID   TYPEDEF      PTR        ; Pointer to any type of data

   STRUCT  PERSON                  ; Structure type
     name  BYTE    20 DUP (?)
     num   WORD    ?
   PERSON  ENDS
   PPERSON TYPEDEF      PTR PERSON ; Pointer to structure type

 The distance of a pointer can either be set specifically or determined
 automatically by the memory model (set by .MODEL) and the segment size (16
 or 32 bits). If you don't use .MODEL, near pointers are the default.

 In 16-bit mode, a near pointer is two bytes that contain the offset of the
 object pointed to. A far pointer requires four bytes, and it contains both
 the offset and the segment. In 32-bit mode, a near pointer is four bytes and
 a far pointer is six bytes. If you specify the distance with NEAR or FAR,
 the default distance of the current segment size is used. You can use
 NEAR16, NEAR32, FAR16, and FAR32 to override the defaults set by the current
 segment size. In flat model, NEAR is the default.

 A pointer type created with TYPEDEF can be used to declare pointer
 variables. Here are some examples using the pointer types defined above:

   ; Type declarations
   Array   WORD    25 DUP (0)
   Msg     BYTE    "This is a string", 0
   pMsg    PBYTE   Msg             ; Pointer to string
   pArray  PWORD   Array           ; Pointer to word array
   npMsg   NPBYTE  Msg             ; Near pointer to string
   npArray NPWORD  Array           ; Near pointer to word array
   fpArray FPWORD  Array           ; Far pointer to word array
   fpMsg   FPBYTE  Msg             ; Far pointer to string

   S1      BYTE    "first", 0      ; Some strings
   S2      BYTE    "second", 0
   S3      BYTE    "third", 0
   pS123   PBYTE   S1, S2, S3, 0   ; Array of pointers to strings
   ppS123  PPBYTE  pS123           ; A pointer to pointers to strings

   Andy    PERSON  <>              ; Structure variable
   pAndy   PPERSON Andy            ; Pointer to structure variable

                                   ; Procedure prototype

   EXTERN  ptrArray:PBYTE          ; External variable
   Sort    PROTO   pArray:PBYTE    ; Parameter for prototype

   ; Parameter for procedure
   Sort    PROC    pArray:PBYTE
           LOCAL   pTmp:PBYTE      ; Local variable
           .
           .
           .
           ret
   Sort    ENDP

 Once defined, pointer types can be used in any context where intrinsic types
 are allowed.


 3.3.2  Defining Register Types with ASSUME

 Beginning with MASM 6.0, you can use the ASSUME directive with
 generalpurpose registers to specify that a register is a pointer to a
 certain size of object. For example:

   ASSUME  bx:PTR WORD     ; BX is word pointer until further
                                   ;  notice
           inc     [bx]            ; Increment word pointed to by BX
           add     bx, 2           ; Point to next word
           mov     [bx], 0         ; Word pointed to by BX = 0
           .
           .                       ; Other pointer operations with BX
           .
           ASSUME  bx:NOTHING      ; Cancel assumptions

 In this example, BX is specified to be a pointer to a word. After a sequence
 of using BX as a pointer, the assumption is cancelled by assuming NOTHING.

 Without the assumption to PTR WORD, many instructions need a size specifier.
 The INC and MOV statements from the examples above would have to be written
 like this to specify the sizes of the memory operands:

   inc     WORD PTR [bx]
           mov     WORD PTR [bx], 0

 When you have used ASSUME, attempts to use the register for other purposes
 generate assembly errors. In the example above, while the PTR WORD
 assumption is in effect, any use of BX inconsistent with its ASSUME
 declaration generates an error. For example,

   ;       mov     al, [bx]       ; Can't move word to byte register

 You can also use the PTR operator to override defaults:

   mov     ax, BYTE PTR [bx] ; Legal

 Similarly, you can use ASSUME to prevent the use of a register as a pointer
 or even to disable a register:

   ASSUME  bx:WORD, dx:ERROR
   ;       mov     al, [bx] ; Error - BX is an integer, not a pointer
   ;       mov     ax, dx   ; Error - DX disabled

 See Section 2.3.3 for information on using ASSUME with segment registers.


 3.3.3  Basic Pointer and Address Operations

 You can do these basic operations with pointers and addresses:


   ■   Initialize a pointer variable by storing an address in it

   ■   Load an address into registers, directly or from a pointer


 The sections in the rest of this chapter describe variations of these tasks
 with both pointers and addresses. The examples in these sections assume that
 you have previously defined the following pointer types with the TYPEDEF
 directive:

   PBYTE   TYPEDEF      PTR BYTE   ; Pointer to bytes
   NPBYTE  TYPEDEF NEAR PTR BYTE   ; Near pointer to bytes
   FPBYTE  TYPEDEF FAR  PTR BYTE   ; Far pointer to bytes


 3.3.3.1  Initializing Pointer Variables

 Let the assembler initialize pointer variables when possible.

 If the value of a pointer is known at assembly time, the assembler can
 initialize it automatically so that no processing time is wasted on the task
 at run time. The following example illustrates how to do this:

   Msg     BYTE    "String", 0
   pMsg    PBYTE   Msg

 If a pointer variable can be conditionally defined to one of several
 constant addresses, initialization must be delayed until run time. The
 technique is different for near pointers than for far pointers, as shown
 below:

   Msg1    BYTE    "String1"
   Msg2    BYTE    "String2"
   npMsg   NPBYTE  ?
   fpMsg   FPBYTE  ?
           .
           .
           .
           mov     npMsg, OFFSET Msg1              ; Load near pointer

           mov     WORD PTR fpMsg[0], OFFSET Msg2  ; Load far offset
           mov     WORD PTR fpMsg[2], SEG Msg2     ; Load far segment

 If you know that the segment for a far pointer is currently in a register,
 you can load it directly:

   mov     WORD PTR fpMsg[2], ds           ; Load segment
   of
                                                   ;  far pointer

 Dynamic Addresses - Often the address to be initialized is dynamic. You know
 the register or registers containing the address, and you want to save them
 in a variable for later use. Typical situations include memory allocated by
 DOS (see interrupt 21h function 48h in online help) and addresses found by
 the SCAS or CMPS instructions (see Section 5.1.3.1). The technique for
 saving dynamic addresses is illustrated below:

   ; Dynamically allocated buffer
   fpBuf   FPBYTE  0               ; Initialize so offset will be zero
           .
           .
           .
           mov     ah, 48h         ; Allocate memory
           mov     bx, 10h         ; Request 16 paragraphs
           int     21h             ; Call DOS
           jc      error           ; Return segment in AX
           mov     WORD PTR fpBuf[2], ax   ; Load segment
           .                               ;  (offset is already 0)
           .
           .
   error:                          ; Handle error

 There are several options for copying pointers.

 Copying Pointers - Sometimes one pointer variable must be initialized by
 copying from another. Here are two ways to copy a far pointer:

   fpBuf1  FPBYTE  ?
   fpBuf2  FPBYTE  ?
           .
           .
           .
   ; Copy through registers is faster, but requires a spare register
           mov     bx, WORD PTR fpBuf1[0]
           mov     WORD PTR fpBuf2[0], bx
           mov     bx, WORD PTR fpBuf1[2]
           mov     WORD PTR fpBuf2[2], bx

   ; Copy through stack is slower, but does not use a register
           push    WORD PTR fpBuf1[0]
           push    WORD PTR fpBuf1[2]
           pop     WORD PTR fpBuf2[2]
           pop     WORD PTR fpBuf2[0]

 Pointers passed as procedure arguments are pushed onto the stack.

 Pointers as Arguments - When a pointer is passed as an argument to a
 procedure, it must be pushed onto the stack. The procedure then sets up a
 stack frame so that it can access the arguments from the stack. This
 technique is discussed in detail in Section 7.3.2, "Passing Arguments on the
 Stack." Pushing a pointer is illustrated below:

   ; Push a far pointer (segment always pushed first)
           push    WORD PTR fpMsg[2]       ; Push segment
           push    WORD PTR fpMsg[0]       ; Push offset

 Pushing an address is somewhat different:

   ; Push a far address as a far pointer
           mov     ax, SEG fVar    ; Load and push segment
           push    ax
           mov     ax, OFFSET fVar ; Load and push offset
           push    ax

 On the 80186 and later processors, you can shorten pushing a constant to one
 step:

   push    SEG fVar        ; Push segment
           push    OFFSET fVar     ; Push offset


 3.3.3.2  Loading Addresses into Registers

 Loading an address into a pair of registers is one of the most common tasks
 in assembly-language programming. You cannot do processing work with a
 constant address or a pointer variable until the address is loaded into
 registers.

 Certain register pairs have standard uses.

 You often load addresses into particular segment:offset pairs. The following
 pairs have specific uses:

 Segment:Offset Pair        Standard Use
 ────────────────────────────────────────────────────────────────────────────
 DS:SI                      Source for string operations
 ES:DI                      Destination for string operations
 DS:DX                      Input for DOS functions
 ES:BX                      Output from DOS functions

 In addition, you can use ES:SI, DS:DI, DS:BX, or any segment:offset pair for
 your own indirect memory operands. You can use SS:BP with a displacement to
 access procedure arguments or local variables in procedures.

 Addresses from Data Segments - For near addresses, you need only load the
 offset; the segment is assumed as SS for stack-based data and as DS for
 other data. You must load both segment and offset for far pointers.

 Here is an example of loading an address to DS:BX from a near data segment:


   .DATA
   Msg     BYTE    "String"
           .
           .
           .
           mov     bx, OFFSET Msg  ; Load address to BX
                                   ;  (DS already loaded)

 If the data is in a far data segment, it is loaded like this:

   .FARDATA
   Msg     BYTE    "String"
           .
           .
           .
           mov     ax, SEG Msg     ; Load address to ES:BX
           mov     es, ax
           mov     bx, OFFSET Msg

 Stack Variables - The technique for loading the address of a stack variable
 is significantly different from the technique for loading near addresses.
 You may need to put the correct segment value into ES for string operations.
 The following example illustrates how to load the address of a local (stack)
 variable to ES:DI:

   Task    PROC
           LOCAL   Arg[4]:BYTE

           push    ss      ; Since it's stack-based, segment is SS
           pop     es      ; Copy SS to ES
           lea     di, Arg ; Load offset to DI

 Use LEA to load the offset of an indirect memory operand.

 The local variable in this case actually evaluates to SS:[BP-4]. This is an
 offset from the stack frame (described in Section 7.3.2, "Passing Arguments
 on the Stack"). Since you cannot use the OFFSET operator to get the offset
 of an indirect memory operand, you must use the LEA (Load Effective Address)
 instruction.

 Use MOV and OFFSET to load the offset of a direct memory operand.

 Direct Memory Operands - To get the address of a direct memory operand, you
 can use the MOV instruction with OFFSET or the LEA instruction. MASM 6.0
 automatically optimizes the LEA statement by generating the smaller and
 faster code, as shown in this example:


   lea     si, Msg        ; If you code this statement,
           mov     si, OFFSET Msg ;  MASM 6.0 generates this code

 The LEA instruction can be used to determine the address of indirect memory
 operands, as shown below.

   lea     si, [bx]        ; Legal - LEA required for indirect
    ;      mov     si, OFFSET [bx] ; Illegal - no OFFSET on indirect

 Far Pointers  - Use the LES and LDS instructions to load far pointers. Use
 the MOV instruction to load a near pointer. The following example shows how
 to load a far pointer to ES:DI and a near pointer to SI (assuming DS as the
 segment):

   InBuf   BYTE    20 DUP (1)
   OutBuf  BYTE    20 DUP (0)

   npIn    NPBYTE  InBuf
   fpOut   FPBYTE  OutBuf
           .
           .
           .
           les     di, fpOut      ; Load far pointer to ES:DI

           mov     si, npIn       ; Load near pointer to SI (assume DS)

 Copying between Segment Pairs - Copying from one register pair to another is
 complicated by the fact that you cannot copy one segment register directly
 to another. Two methods are shown below. Timings are for the 8088 processor:


   ; Copy DS:SI to ES:DI, generating smaller code
           push    ds              ; 1 byte, 14 clocks
           pop     es              ; 1 byte, 12 clocks
           mov     di, si          ; 2 bytes, 2 clocks

   ; Copy DS:SI to ES:DI, generating faster code
           mov     di, ds          ; 2 bytes, 2 clocks
           mov     es, di          ; 2 bytes, 2 clocks
           mov     di, si          ; 2 bytes, 2 clocks


 3.3.3.3  Model-Independent Techniques

 Use conditional assembly to write memory-model independent code.

 Often you may want to write code that is memory-model independent. If you
 are writing libraries that must be available for different memory models,
 you can use conditional assembly to handle different sizes of pointers. You
 can use the predefined symbols @DataSize and @Model to test the current
 assumptions.

 Use conditional assembly to handle pointers that have no specified distance.


 You can use conditional assembly to write code that works with pointer
 variables that have no specified distance. The predefined symbol @DataSize
 tests the pointer size for the current memory model:

   Msg1    BYTE    "String1"
   pMsg    PBYTE   ?
           .
           .
           .
           IF      @DataSize
           mov     WORD PTR pMsg[0], OFFSET Msg1   ; Load far offset
           mov     WORD PTR pMsg[2], SEG Msg1      ; Load far segment
           ELSE
           mov     pMsg, OFFSET Msg1               ; Load near pointer
           ENDIF

 In the following example, a procedure receives as an argument a pointer to a
 word variable. The code inside the procedure uses @DataSize to determine
 whether the current memory model supports far or near data. It loads and
 processes the data accordingly:

   ; Procedure that receives an argument by reference
   mul8    PROC    arg:PTR WORD

           IF      @DataSize
           les     bx, arg     ; Load far pointer to ES:BX
           mov     ax, es:[bx] ; Load the data pointed to
           ELSE
           mov     bx, arg     ; Load near pointer to BX (assume DS)
           mov     ax, [bx]    ; Load the data pointed to
           ENDIF
           shl     ax, 1       ; Multiply by 8
           shl     ax, 1
           shl     ax, 1
           ret
   mul8    ENDP

 If you have many routines, writing the conditionals for each case can be
 tedious. The following conditional statements generate the proper
 instructions and segment overrides automatically.

   ; Equates for conditional handling of pointers
           IF @DataSize
   lesIF   TEXTEQU    <les>
   ldsIF   TEXTEQU    <lds>
   esIF    TEXTEQU    <es:>
           ELSE
   lesIF   TEXTEQU    <mov>
   ldsIF   TEXTEQU    <mov>
   esIF    TEXTEQU    <>
           ENDIF

 Once you define these conditionals, you can use them to simplify code that
 must handle several types of pointers. This next example rewrites the above
 mul8  procedure to use conditional code.

   mul8    PROC    arg:PTR WORD

           lesIF   bx, arg         ; Load pointer to BX or ES:BX
           mov     ax, esIF [bx]   ; Load the data from [BX] or ES:[BX]
           shl     ax, 1           ; Multiply by 8
           shl     ax, 1
           shl     ax, 1
           ret
   mul8    ENDP

 The conditional statements from the examples above can be defined once in an
 include file and used whenever you need to handle pointers.


 3.4  Related Topics in Online Help

 In addition to information covered in this chapter, information on the
 following topics can be found in online help.

╓┌─────────────────────────────────────┌─────────────────────────────────────╖
 Topics                                Access
 ────────────────────────────────────────────────────────────────────────────
 LROFFSET, THIS                        From the "MASM 6.0 Contents" screen,
                                       choose "Operators"; then choose
                                       "Address"

 LFS, LGS, and LSS                     From the "MASM 6.0 Contents" screen,
 Topics                                Access
 ────────────────────────────────────────────────────────────────────────────
LFS, LGS, and LSS                     From the "MASM 6.0 Contents" screen,
                                       choose "Processor Instructions";
                                       then choose "Data
                                       Transfer"

 ALIGN, EVEN, ORG                      From the "MASM 6.0 Contents" screen,
                                       choose "Directives"; then choose
                                       "Miscellaneous"

 NEAR, NEAR16, NEAR32, FAR16, FAR32,   From the "MASM 6.0 Contents" screen,
 and TYPE                              choose "Operators"; then choose
                                       "Type and Size"

 PTR                                   From the "MASM 6.0 Contents" screen,
                                       choose "Operators"; then choose
                                       "Miscellaneous"

 PUSHCONTEXT and POPCONTEXT            Access from the Macro Assembler
                                       Index
 Topics                                Access
 ────────────────────────────────────────────────────────────────────────────
                                      Index

 ASSUME, .MODEL                        From the "MASM 6.0 Contents" screen,
                                       choose "Directives"; then choose
                                       "Simplified Segment Control"

 @DataSize, @Model                     From the "MASM 6.0 Contents" screen,
                                       choose "Predefined Symbols"








 Chapter 4  Defining and Using Integers
 ────────────────────────────────────────────────────────────────────────────

 The 8086 family of processors is designed to operate on integer data;
 therefore, most assembler statements are integer operations. Even string
 elements (discussed in Chapter 5, "Defining and Using Complex Data Types")
 are byte-sized integers to the assembler.

 This chapter covers the concepts essential for using integer variables in
 assembly-language programs. The first section shows how to declare integer
 variables. The second section describes basic integer operations including
 moving, loading, and sign-extending integers, as well as calculating with
 integers. Finally, the last section describes how to do various operations
 with integers at the bit level, such as using bitwise logical instructions
 and shifting and rotating bits.

 The complex data types introduced in the next chapter─arrays, strings,
 structures, unions, and records─use many of the integer operations
 illustrated in this chapter, since the components of complex data types are
 often integers. Floating-point operations require a different set of
 instructions and techniques. These are covered in Chapter 6, "Using
 Floating-Point and Binary Coded Decimal Numbers."


 4.1  Declaring Integer Variables

 You declare integer variables in the data segment of your program to
 allocate memory for data. The EQU and = directives define integer constants.
 Integer variables allocated with the data allocation directives can be
 initialized in several ways. MASM 6.0 provides new forms of the data
 allocation directives. This section discusses these features and explains
 how to use the SIZEOF and TYPE operators to provide information to the
 assembler about the types in your program. For information on symbolic
 integer constants, see Section 1.2.4, "Integer Constants and Constant
 Expressions."


 4.1.1  Allocating Memory for Integer Variables

 When you declare an integer variable by assigning a label to a data
 allocation directive, the assembler allocates memory space for the integer.
 The variable's name becomes a label for the memory space. The syntax is

   «name» directive initializer

 These directives, listed below, indicate the integer's size and value range.


╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Directive                         Description of Initializers
 BYTE, DB (bytes)                  Allocates unsigned numbers  from
                                   0 to 255.
 ────────────────────────────────────────────────────────────────────────────
 SBYTE (signed bytes)              Allocates signed numbers from
                                   -128 to +127.

 WORD, DW (words = 2 bytes)        Allocates unsigned numbers from
                                   0 to 65,535 (64K).

 SWORD (signed words)              Allocates signed numbers from
                                   -32,768 to +32,767.

 DWORD, DD (doublewords = 4 bytes) Allocates unsigned numbers from
                                   0 to 4,294,967,295 (4 megabytes).

 SDWORD (signed doublewords)       Allocates signed numbers from
 Directive                         Description of Initializers
 BYTE, DB (bytes)                  Allocates unsigned numbers  from
                                   0 to 255.
 ────────────────────────────────────────────────────────────────────────────
SDWORD (signed doublewords)       Allocates signed numbers from
                                   -2,147,483,648 to +2,147,483,647.

 FWORD, DF (farwords = 6 bytes)    Allocates 6-byte (48-bit) integers.
                                   These values are normally used only as
                                   pointer variables on the 80386/486
                                   processors.

 QWORD, DQ (quadwords = 8 bytes)   Allocates 8-byte integers used with
                                   8087-family coprocessor instructions.

 TBYTE, DT (10 bytes)              Allocates 10-byte (80-bit) integers if
                                   the initializer has a radix specifying
                                   the base of the number.



 See Chapter 6 for information on the REAL4, REAL8, and REAL10 directives
 that allocate real numbers.




 The assembler enforces only the size of initializers.

 MASM does not enforce the range of values assigned to an integer. If the
 value does not fit in the space allocated, however, the assembler generates
 an error.

 The SIZEOF and TYPE operators, when applied to a type, return the size of an
 integer of that type. The following list gives the size attribute associated
 with each data type.

 Data Type                       Bytes
 BYTE
 ────────────────────────────────────────────────────────────────────────────
 WORD, SWORD                     2
 DWORD, SDWORD                   3
 FWORD                           6
 QWORD                           8
 TBYTE                           10

 The SBYTE, SWORD, and SDWORD data types are new to MASM 6.0. Use of these
 signed data types tells the assembler to treat the initializers as signed
 data. It is important to use these signed types with high-level constructs
 such as .IF, .WHILE, and .REPEAT (see Section 7.2.1, "Loop-Generating
 Directives"), and with PROTO and INVOKE directives (see Sections 7.3.6,
 "Declaring Procedure Prototypes," and 7.3.7, "Calling Procedures with
 INVOKE").

 The assembler stores integers with the least significant bytes lowest in
 memory. Note that assembler listings and most debuggers show the bytes of a
 word in the opposite order─high byte first.

 Figure 4.1 illustrates the integer formats.

 (This figure may be found in the printed book.)

 TYPEDEF can define integer aliases.

 Although the TYPEDEF directive's primary purpose is to define pointer
 variables (see Section 3.3.1), you can also use TYPEDEF to create an alias
 for any integer type. For example, these declarations

   char    TYPEDEF SBYTE
   longint TYPEDEF DWORD
   float   TYPEDEF REAL4
   double  TYPEDEF REAL8

 allow you to use  char,  longint,  float, or  double  in your programs if
 you prefer the C data labels.


 4.1.2  Data Initialization

 You can initialize variables when you declare them by giving initial
 values─that is, constants or expressions that evaluate to integer constants.
 The assembler generates an error if you specify an initial value too large
 for the specified variable type. Variables can also be initialized with ? if
 there are no initial values.

 You can declare and initialize variables in one step with the data
 directives, as these examples show.

   integer         BYTE    16          ; Initialize byte to 16
   negint          SBYTE   -16         ; Initialize signed byte to -16
   expression      WORD    4*3         ; Initialize word to 12
   signedexp       SWORD   4*3         ; Initialize signed word to 12
   empty           QWORD   ?           ; Allocate uninitialized long
                                       ;  integer
                   BYTE    1,2,3,4,5,6 ; Initialize six unnamed bytes
   long            DWORD   4294967295  ; Initialize doubleword to
                                       ;  4,294,967,295
   longnum         SDWORD  -2147433648 ; Initialize signed doubleword

                                       ;  to -2,147,433,648
   tb              TBYTE   2345t       ; Initialize 10-byte binary
                                       ;  number

 See Section 5.1, "Arrays and Strings," for information on arrays and on
 using the DUP operator to allocate initializer lists.

 Once you have declared integer variables in your program, you can use them
 in integer operations such as adding, moving, loading, and exchanging. The
 next section describes these operations.


 4.2  Integer Operations

 You often need to copy, move, exchange, load, and sign-extend integer
 variables in your MASM code. This section shows how to do these operations
 as well as how to add, subtract, multiply, and divide integers; push and pop
 integers onto the stack; and do bit-level manipulations with logical, shift,
 and rotate instructions.

 The PTR operator tells the assembler the size of the operand.

 Since MASM instructions require operands to be the same size, you may need
 to operate on data in a size other than the size originally declared. The
 PTR operator lets you do this. For example, you can use the PTR operator to
 access the high-order word of a DWORD-size variable. The syntax for the PTR
 operator is

   type PTR expression

 where the PTR operator forces expression to be treated as having the type
 specified. An example of this use is

   .DATA
   num     DWORD   0
           .CODE

   mov     ax, WORD PTR num[0] ; Loads a word-size value
   from
           mov     dx, WORD PTR num[2] ;  a doubleword variable

 You might choose not to use PTR, in contrast to this example. In that case,
 trying to move  num[0]  into AX generates an error.


 4.2.1  Moving and Loading Integers

 The primary instructions for moving integers from operand to operand and
 loading them into registers are MOV (Move), XCHG (Exchange), XLAT
 (Translate), CWD (Convert Word to Double), and CBW (Convert Byte to Word).


 4.2.1.1  Moving Integers

 The most common method of moving data, the MOV instruction, can be thought
 of as a copy instruction, since it always copies the source operand to the
 destination operand. Immediately after a MOV instruction, both the source
 and destination operands contain the same value.

 The statements in the following example illustrate each type of memory move
 that can be performed with a single instruction. Note that you cannot move
 memory operands to memory operands in one operation.

   ; Immediate value moves
           mov     ax, 7       ; Immediate to register
           mov     mem, 7      ; Immediate to memory direct
           mov     mem[bx], 7  ; Immediate to memory indirect
   ; Register moves
           mov     mem, ax     ; Register to memory direct
           mov     mem[bx], ax ; Register to memory indirect
           mov     ax, bx      ; Register to register
           mov     ds, ax      ; General register to segment
                               ;  register

   ; Direct memory moves
           mov     ax, mem     ; Memory direct to register
           mov     ds, mem     ; Memory to segment register

   ; Indirect memory moves
           mov     ax, mem[bx] ; Memory indirect to register
           mov     ds, mem[bx] ; Memory indirect to segment register

   ; Segment register moves
           mov     mem, ds     ; Segment register to memory
           mov     mem[bx], ds ; Segment register to memory indirect
           mov     ax, ds      ; Segment register to general
                               ;  register

 This next example shows several common types of moves that require two
 instructions.

   ; Move immediate to segment register
           mov     ax, DGROUP  ; Load immediate to general register
           mov     ds, ax      ; Store general register to segment
                               ;  register

   ; Move memory to memory
           mov     ax, mem1    ; Load memory to general register
           mov     mem2, ax    ; Store general register to memory

   ; Move segment register to segment register
           mov     ax, ds      ; Load segment register to general
                               ;  register
           mov     es, ax      ; Store general register to segment
                               ;  register

 The MOVSX and MOVZX instructions for the 80386/486 processors extend and
 copy values in one step. See Section 4.2.1.4, "Extending Signed and Unsigned
 Integers."


 4.2.1.2  Exchanging Integers

 The XCHG (Exchange) instruction exchanges the data in the source and
 destination operands. Data can be exchanged between registers or between
 registers and memory, but not from memory to memory:

   xchg    ax, bx       ; Put AX in BX and BX in AX
           xchg    memory, ax   ; Put "memory" in AX and AX in "memory"
   ;       xchg    mem1, mem2   ; Illegal- can't exchange between
                                ;  memory location

 In some circumstances, register-to-register moves are faster with XCHG than
 with MOV. If speed is important in your programs, check the Reference to
 find the fastest clock speeds for various operand combinations allowed with
 MOV  and XCHG.


 4.2.1.3  Translating Integers from Tables

 The XLAT (Translate) instruction loads data from a table into memory. The
 instruction is useful for translating bytes from one coding system to
 another. The syntax is

   XLAT[[B]] [[[[segment:]]memory]]

 XLAT and XLATB are synonyms.

 The BX register must contain the address of the start of the table. By
 default, the DS register contains the segment of the table, but you can use
 a segment override to specify a different segment. Also, you need not give
 the operand except when specifying a segment override. (See Section 3.2.3,
 "Direct Memory Operands," for information about the segment override
 operator.)

 Before the XLAT instruction executes, the AL register should contain a value
 that points into the table (the start of the table is position 0). After the
 instruction executes, AL contains the table value pointed to. For example,
 if AL contains 7, the assembler puts the eighth byte of the table in the AL
 register.

 This example, illustrating XLAT, looks up hexadecimal characters in a table
 to convert an eight-bit binary number to a string representing a hexadecimal
 number.

   ; Table of hexadecimal digits
   hex     BYTE    "0123456789ABCDEF"
   convert BYTE    "You pressed the key with ASCII code "
   key     BYTE    ?,?,"h",13,10,"$"
           .CODE
           .
           .
           .
           mov     ah, 8               ; Get a key in AL
           int     21h                 ; Call DOS
           mov     bx, OFFSET hex      ; Load table address
           mov     ah, al              ; Save a copy in high byte
           and     al, 00001111y       ; Mask out top character
           xlat                        ; Translate
           mov     key[1], al          ; Store the character
           mov     cl, 12              ; Load shift count
           shr     ax, cl              ; Shift high character into
                                       ;  position
           xlat                        ; Translate
           mov     key, al             ; Store the character
           mov     dx, OFFSET convert  ; Load message
           mov     ah, 9               ; Display character
           int     21h                 ; Call DOS


 4.2.1.4  Extending Signed and Unsigned Integers

 Since moving data to a different-sized register is illegal, you must
 "sign-extend" integers to convert signed data to a larger register or
 register pair.

 Sign-extending means copying the sign bit of the unextended operand to all
 bits of the extended operand. The instructions in the following list
 sign-extend values as shown. They work only on signed values in the
 accumulator register.

 Instruction    Function
 ────────────────────────────────────────────────────────────────────────────
 CBW            Convert byte to word
 CWD            Convert word to doubleword
 CWDE           Convert word to doubleword extended (80386/486 only)
 CDQ            Convert doubleword to quadword (80386/486 only)

 On the 80386/486, the CWDE instruction converts a signed 16-bit value in AX
 to a signed 32-bit value in EAX. The CDQ instruction converts a signed
 32-bit value in EAX to a signed 64-bit value in the EDX:EAX register pair.

 This example converts signed integers using CBW, CWD, CWDE, and CDQ.

   .DATA
   mem8    SBYTE   -5
   mem16   SWORD   -5
   mem32   SDWORD  -5
           .CODE
           .
           .
           .
           mov     al, mem8    ; Load 8-bit -5 (FBh)
           cbw                 ; Convert to 16-bit -5 (FFFBh) in AX

           mov     ax, mem16   ; Load 16-bit -5 (FFFBh)
           cwd                 ; Convert to 32-bit -5 (FFFF:FFFBh)
                               ;  in DX:AX
           mov     ax, mem16   ; Load 16-bit -5 (FFFBh)
           cwde                ; Convert to 32-bit -5 (FFFFFFFBh)
                               ;  in EAX
           mov     eax, mem32  ; Load 32-bit -5 (FFFFFFFBh)
           cdq                 ; Convert to 64-bit -5
                               ;  (FFFFFFFF:FFFFFFFBh) in EDX:EAX

 Conversion instructions do not operate on unsigned numbers.

 The procedure is different for unsigned values. Unsigned values are extended
 by filling the upper bits with zeros rather than by sign extension. Because
 the sign-extend instructions do not work on unsigned integers, you must set
 the value of the higher register to zero.

 This example shows sign extension for unsigned numbers.

   .DATA
   mem8    BYTE    251
   mem16   WORD    251
           .CODE
           .
           .
           .
           mov     al, mem8  ; Load 251 (FBh) from 8-bit memory
           sub     ah, ah    ; Zero upper half (AH)

           mov     ax, mem16 ; Load 251 (FBh) from 16-bit memory
           sub     dx, dx    ; Zero upper half (DX)

 The 80386/486 processors provide instructions that move and extend a value
 to a larger data size in a single step. MOVSX moves a signed value into a
 register and sign-extends it. MOVZX moves an unsigned value into a register
 and zeroextends it.

   ; 80386/486 instructions
           movzx   dx, bl      ; Load unsigned 8-bit value into
                               ;  16-bit register and zero-extend

 These special 80386 and 80486 instructions usually execute much faster than
 the equivalent 8086-80286 instructions.


 4.2.2  Pushing and Popping Stack Integers

 A stack is an area of memory for storing data temporarily. Unlike other
 segments that store data starting from low memory, the stack stores data in
 reverse order─starting from high memory. Data is always pushed or popped
 from the top of the stack. The data on the stack can be the calling
 addresses of procedures or interrupts, procedure arguments, or any operands,
 flags, or registers your program needs to store temporarily.

 At first, the stack is an uninitialized segment of a finite size. As data is
 added to the stack at run time, the stack grows downward from high memory to
 low memory. When items are removed from the stack, it shrinks upward from
 low to high memory.


 4.2.2.1  Saving Operands on the Stack

 PUSH and POP always operate on word-sized data.

 The PUSH instruction stores a two-byte operand on the stack. The POP
 instruction retrieves a previously pushed value. When a value is pushed onto
 the stack, the assembler decreases the SP (Stack Pointer) register by 2. On
 8086-based processors, the SP register always points to the top of the
 stack. The PUSH and POP instructions use the SP register to keep track of
 the current position.

 When a value is popped off the stack, the assembler increases the SP
 register by 2. Although the stack always contains word values, the SP
 register points to byte addresses. Thus, SP changes in multiples of two.
 When a PUSH or POP instruction executes in a 32-bit code segment (one with
 USE32 use type), the assembler transfers a four-byte value, and ESP changes
 in multiples of four.

 ────────────────────────────────────────────────────────────────────────────
 NOTE
 The 8086 and 8088 processors differ from later Intel processors in how they
 push and pop the SP register. If you give the statement  push sp  with the
 8086 or 8088, the word pushed is the word in SP after the push operation.
 ────────────────────────────────────────────────────────────────────────────

 Figure 4.2 illustrates how pushes and pops change the SP register.

 (Please refer to the printed book.)

 (This figure may be found in the printed book.)

 On the 8086, PUSH and POP take only registers or memory expressions as their
 operands. The other processors allow an immediate value to be an operand for
 PUSH. For example, the following statement is legal on the 80186-80486
 processors:

   push     7              ; 3 clocks on 80286

 That statement is faster than these equivalent statements, which are
 required on the 8088 or 8086:

   mov     ax, 7           ; 2 clocks plus
           push    ax              ; 3 clocks on 80286

 There are two ways to clean up the stack.

 Words are popped off the stack in reverse order: the last item pushed is the
 first popped. To return the stack to its original status, you can do the
 same number of pops as pushes. You can subtract the correct number of words
 from the SP register if you want to restore the stack without using the
 values on it.

 To reference operands on the stack, keep in mind that the values pointed to
 by the BP (Base Pointer) and SP registers are relative to the SS (Stack
 Segment) register. The BP register is often used to point to the base of a
 frame of reference (a stack frame) within the stack.

 This example shows how you can access values on the stack using indirect
 memory operands with BP as the base register.

   push    bp              ; Save current value of BP
           mov     bp, sp          ; Set stack frame
           push    ax              ; Push first;  SP = BP - 2
           push    bx              ; Push second; SP = BP - 4
           push    cx              ; Push third;  SP = BP - 6
           .
           .
           .
           mov     ax, [bp-6]      ; Put third in AX
           mov     bx, [bp-4]      ; Put second in BX
           mov     cx, [bp-2]      ; Put first in CX
           .
           .
           .
           add     sp, 6           ; Restore stack pointer
                                   ;  two bytes per push
           pop     bp              ; Restore BP

 Creating labels for stack variables makes code easier to read.

 If you use these stack values often in your program, you may want to give
 them labels. For example, you can use TEXTEQU to create a label such as
 count TEXTEQU <bp-6>. Now you can replace the  mov ax, [bp - 6]  statement
 in the example above with  mov ax, count. Section 9.1, "Text Macros," gives
 more information about the TEXTEQU directive.


 4.2.2.2  Saving Flags on the Stack

 Flags can be pushed and popped onto the stack with the PUSHF and POPF
 instructions. You can use these instructions to save the status of flags
 before a procedure call and then to restore the original status after the
 procedure. You can also use them within a procedure to save and restore the
 flag status of the caller. The 32-bit versions of these instructions are
 PUSHFD and POPFD.


 This example saves the flags register before calling the  systask
 procedure:

   pushf
           call    systask
           popf

 If you do not need to store the entire flag register, you can use the LAHF
 instruction to manually load and store the status of the lower byte of the
 flag register in the AH register. (You need to save AH before making a
 procedure call.) SAHF restores the value.


 4.2.2.3  Saving Registers on the Stack (80186-80486 Only)

 Starting with the 80186 processor, the PUSHA and POPA instructions push or
 pop all the general-purpose registers with only one instruction. These
 instructions save the status of all registers before a procedure call and
 then restore them after the return. Using PUSHA and POPA is significantly
 faster and takes fewer bytes of code than pushing and popping each register
 individually.

 The processor pushes the registers in the following order: AX, CX, DX, BX,
 SP, BP, SI, and DI. The SP word pushed is the value before the first
 register is pushed.

 The processor pops the registers in the opposite order. The 32-bit versions
 of these instructions are PUSHAD and POPAD.


 4.2.3  Adding and Subtracting Integers

 You can use the ADD, ADC, INC, SUB, SBB, and DEC instructions for adding,
 incrementing, subtracting, and decrementing values in single registers. You
 can also combine them to handle larger values that require two registers for
 storage.


 4.2.3.1  Adding and Subtracting Integers Directly

 The ADD, INC (Increment), SUB, and DEC (Decrement) instructions operate on
 8- and 16-bit values on the 8086-80286 processors, and on 8-, 16-, and
 32-bit values on the 80386/486 processors. They can be combined with the ADC
 and SBB instructions to work on 32-bit values on the 8086 and 64-bit values
 on the 80386/486 processors (see Section 4.2.3.2).


 These instructions have two requirements:


   1.  If there are two operands, only one operand can be a memory operand.

   2.  If there are two operands, both must be the same size.


 PTR allows you to operate on data in sizes different from its declared type.


 To meet the second requirement, you can use the PTR operator to force an
 operand to the size required (see Section 4.2, "Integer Operations"). For
 example, if  Buffer  is an array of bytes and BX points to an element of the
 array, you can add a word from  Buffer  with

   add     ax, WORD PTR Buffer[bx] ; Adds a word from the
                                           ;  byte variable

 The next example shows 8-bit signed and unsigned addition and subtraction.

   DATA
   mem8    BYTE    39
           .CODE

   ; Addition

                            ;                    signed    unsigned
           mov     al, 26   ; Start with register   26       26
           inc     al       ; Increment              1        1
           add     al, 76   ; Add immediate         76     + 76
                            ;                     ----     ----
                            ;                      103      103
           add     al, mem8 ; Add memory            39     + 39
                            ;                     ----     ----
           mov     ah, al   ; Copy to AH          -114      142
                                                  +overflow
           add     al, ah   ; Add register                  142
                            ;                              ----
                            ;                                28+carry

   ; Subtraction

                            ;                   signed    unsigned
           mov     al, 95   ; Load register         95       95
           dec     al       ; Decrement             -1       -1
           sub     al, 23   ; Subtract immediate   -23      -23
                            ;                     ----     ----
                            ;                       71       71
           sub     al, mem8 ; Subtract memory     -122     -122
                            ;                     ----     ----
                            ;                      -51      205+sign

           mov     ah, 119  ; Load register        119
           sub     al, ah   ;  and subtract        -51
                            ;                     ----
                            ;                       86+overflow

 The INC and DEC instructions treat integers as unsigned values and do not
 update the carry flag for signed carries and borrows.


 Your programs must include error-recovery for overflows and carries.

 When the sum of eight-bit signed operands exceeds 127, the processor sets
 the overflow flag. (The overflow flag is also set if both operands are
 negative and the sum is less than or equal to -128.) Placing a JO (Jump on
 Overflow) or INTO (Interrupt on Overflow) instruction in your program at
 this point can transfer control to error-recovery statements. When the sum
 exceeds 255, the processor sets the carry flag. A JC (Jump on Carry)
 instruction at this point can transfer control to error-recovery statements.


 In the subtraction example above, the processor sets the sign flag if the
 result goes below 0. At this point, you can use a JS (Jump on Sign)
 instruction to transfer control to error-recovery statements.


 4.2.3.2  Adding and Subtracting in Multiple Registers

 You can add and subtract numbers larger than the register size on your
 processor with the ADC (Add with Carry) and SBB (Subtract with Borrow)
 instructions. If the operations prior to an ADC or SBB instruction do not
 set the carry flag, these instructions are identical to ADD and SUB. When
 you operate on large values in more than one register, use ADD and SUB for
 the least significant part of the number and ADC or SBB for the most
 significant part.

 The following example illustrates multiple-register addition and
 subtraction. You can also use this technique with 64-bit operands on the
 80386/486 processors.


   .DATA
   mem32   DWORD   316423
   mem32a  DWORD   316423
   mem32b  DWORD   156739
           .CODE
           .
           .
           .
   ; Addition
           mov     ax, 43981               ; Load immediate     43981
           sub     dx, dx                  ;  into DX:AX
           add     ax, WORD PTR mem32[0]   ; Add to both     + 316423
           adc     dx, WORD PTR mem32[2]   ;  memory words     ------
                                           ; Result in DX:AX   360404

   ; Subtraction
           mov     ax, WORD PTR mem32a[0]  ; Load mem32        316423
           mov     dx, WORD PTR mem32a[2]  ;  into DX:AX
           sub     ax, WORD PTR mem32b[0]  ; Subtract low    - 156739
           sbb     dx, WORD PTR mem32b[2]  ;  then high        ------
                                           ; Result in DX:AX   159684

 For 32-bit registers on the 80386/486, only two steps are necessary. If your
 program needs to be assembled for more than one processor, you can assemble
 the statements conditionally, as shown in this example:

   .DATA
   mem32   DWORD   316423
   mem32a  DWORD   316423
   mem32b  DWORD   156739
   p386    TEXTEQU (@Cpu AND 08h)
           .CODE
           .
           .
           .
   ; Addition
           IF      p386
           mov     eax, 43981  ; Load immediate
           add     eax, mem32  ; Result in EAX
           ELSE
           .
           .       ; do steps in previous example
           .
           ENDIF

   ; Subtraction
           IF      p386
           mov     eax, mem32a ; Load memory
           sub     eax, mem32b ; Result in EAX
           ELSE
           .
           .       ; do steps in previous example
           .
           ENDIF

 Since the status of the carry flag affects the results of calculations with
 ADC and SUB, be sure to turn off the carry flag with the CLC (Clear Carry
 Flag) instruction or use ADD for the first calculation when appropriate.


 4.2.4  Multiplying and Dividing Integers

 The 8086 family of processors uses different multiplication and division
 instructions for signed and unsigned integers. Multiplication and division
 instructions also have special requirements depending on the size of the
 operands and the processor the code runs on.


 4.2.4.1  Using Multiplication Instructions

 The MUL instruction multiplies unsigned numbers. IMUL multiplies signed
 numbers. For both instructions, one factor must be in the accumulator
 register (AL for 8-bit numbers, AX for 16-bit numbers, EAX for 32-bit
 numbers). The other factor can be in any single register or memory operand.
 The result overwrites the contents of the accumulator register.

 Multiplying two 8-bit numbers produces a 16-bit result returned in AX.
 Multiplying two 16-bit operands yields a 32-bit result in DX:AX. The
 80386/486 processor handles 64-bit products in the same way in the EDX:EAX
 pair.

 This example illustrates multiplication of signed 16- and 32-bit integers.

   .DATA
   mem16   SWORD   -30000
           .CODE
           .
           .
           .
   ; 8-bit signed multiply
           mov     al, 23     ; Load AL                     23
           mov     bl, 24     ; Load BL                   * 24
           mul     bl         ; Multiply BL              -----
                              ; Product in AX              552
                              ;  overflow and carry set

   ; 16-bit unsigned multiply
           mov     ax, 50     ; Load AX                     50
                              ;                         -30000
           imul    mem16      ; Multiply memory          -----
                              ; Product in DX:AX      -1500000
                              ;  overflow and carry set

 A nonzero number in the upper half of the result (AH for byte, DX or EDX for
 word) sets the overflow and carry flags.

 On the 80186-80486 processors, the IMUL instruction supports three different
 operand combinations. The first syntax option allows for 16-bit multipliers
 producing a 16-bit product or 32-bit multipliers for 32-bit products on the
 80386/486. The result overwrites the destination. The syntax for this
 operation is

   IMUL register16, immediate

 Multiplication by an immediate operand is possible on the 80386/486.

 The second syntax option specifies three operands for IMUL. The first
 operand must be a 16-bit register operand, the second a 16-bit memory or
 register operand, and the third a 16-bit immediate operand. IMUL multiplies
 the memory (or register) and immediate operands and stores the product in
 the register operand with this syntax:

 IMUL register16, memory16 | register16, immediate

 For the 80386/486 only, a third option for IMUL allows an additional operand
 for multiplication of a register value by a register or memory value. This
 is the syntax:

   IMUL register,{register | memory}

 The destination can be any 16-bit or 32-bit register. The source must be the
 same size as the destination.

 In all of these options, products too large to fit in 16 or 32 bits set the
 overflow and carry flags. The following examples show these three options
 for IMUL.

   imul    dx, 456     ; Multiply DX times 456 on 80186-80486
           imul    ax, [bx],6  ; Multiply the value pointed to by BX
                               ;  by 6 and put the result in AX

           imul    dx, ax      ; Multiply DX times AX on 80386
           imul    ax, [bx]    ; Multiply AX by the value pointed to
                               ;  by BX on 80386

 The IMUL instruction with multiple operands can be used for either signed or
 unsigned multiplication, since the 16-bit product is the same in either
 case. To get a 32-bit result, you must use the single-operand version of MUL
 or IMUL.


 4.2.4.2  Using Division Instructions

 The DIV instruction divides unsigned numbers, and IDIV divides signed
 numbers. Both return a quotient and a remainder.

 Table 4.1 summarizes the division operations. The dividend is the number to
 be divided, and the divisor is the number to divide by. The quotient is the
 result. The divisor can be in any register or memory location except the
 registers where the quotient and remainder are returned.

 Table   4.1 Division Operations

 Size of        Dividend       Size of
 Operand        Register       Divisor        Quotient  Remainder
 ────────────────────────────────────────────────────────────────────────────
 16 bits        AX             8 bits         AL        AH

 32 bits        DX:AX          16 bits        AX        DX

 64 bits        EDX:EAX        32 bits        EAX       EDX
 (80386
 and 80486)

 ────────────────────────────────────────────────────────────────────────────


 Unsigned division does not require careful attention to flags. The following
 examples illustrate signed division, which can be more complex.

   .DATA
   mem16   SWORD   -2000
   mem32   SDWORD  500000
           .CODE
           .
           .
           .
   ; Divide 16-bit unsigned by 8-bit
           mov     ax, 700               ; Load dividend      700
           mov     bl, 36                ; Load divisor DIV    36
           div     bl                    ; Divide BL       ------
                                         ; Quotient in AL      19
                                         ; Remainder in AH          16

   ; Divide 32-bit signed by 16-bit
           mov     ax, WORD PTR mem32[0] ; Load into DX:AX
           mov     dx, WORD PTR mem32[2] ;                 500000
           idiv    mem16                 ;              DIV -2000
                                         ; Divide memory   ------
                                         ; Quotient in AX    -250
                                         ; Remainder in DX           0

   ; Divide 16-bit signed by 16-bit
           mov     ax, WORD PTR mem16    ; Load into AX     -2000
           cwd                           ; Extend to DX:AX
           mov     bx,-421               ;               DIV -421
           idiv    bx                    ; Divide by BX     -----
                                         ; Quotient in AX       4
                                         ; Remainder in DX        -316

 If the dividend and divisor are the same size, sign-extend or zero-extend
 the dividend so that it is the length expected by the division instruction.
 See Section 4.2.1.4, "Extending Signed and Unsigned Integers."


 4.3  Manipulating Integers at the Bit Level

 The instructions introduced so far in this chapter accessed integers at the
 byte or word level. The logical, shift, and rotate instructions described in
 this section, however, access the individual bits of the integers. You can
 use logical instructions to evaluate characters and do other text and screen
 operations. The shift and rotate instructions do similar tasks by shifting
 and rotating bits through registers. This section discusses some
 applications of these bit-level operations.


 4.3.1  Logical Operations

 The logical instructions─AND, OR, XOR, and NOT─operate on each bit in one
 operand and on the corresponding bit in the other. The following list shows
 how each instruction works. Except for NOT, these instructions require two
 integers of the same size.


 Instruction                       Sets a Bit to 1 under These Conditions
 ────────────────────────────────────────────────────────────────────────────
 AND                               Both corresponding bits in the operands
                                   have the value 1.

 OR                                Either of the corresponding bits in the
                                   operands has the value 1.

 XOR                               Either, but not both, of the
                                   corresponding bits in the operands has
                                   the value 1.

 NOT                               The corresponding bit in the operand is
                                   0. (This instruction takes only one
                                   operand.)


 ────────────────────────────────────────────────────────────────────────────
 NOTE
 Do not confuse logical instructions with the logical operators, which
 perform these operations at assembly time, not run time. Although the names
 are the same, the assembler recognizes the difference from context.
 ────────────────────────────────────────────────────────────────────────────

 The following example shows the result of the AND, OR, XOR, and NOT
 instructions operating on a value in the AX register and in a mask. A mask
 is a binary or hexadecimal number with appropriate bits set for the intended
 operation.

   mov     ax, 035h   ; Load value                  00110101
           and     ax, 0FBh   ; Clear bit 2             AND 11111011
                              ;                             --------
                              ; Value is now 31h            00110001
           or      ax, 016h   ; Set bits 4,2,1          OR  00010110
                              ;                             --------
                              ; Value is now 37h            00110111
           xor     ax, 0ADh   ; Toggle bits 7,5,3,2,0   XOR 10101101
                              ;                             --------
                              ; Value is now 9Ah            10011010
           not     ax         ; Value is now 65h            01100101

 Use AND, OR, and XOR to set or clear specific bits.

 You can use the AND instruction to clear the value of specific bits
 regardless of their current settings. To do this, put the target value in
 one operand and a mask of the bits you want to clear in the other. The bits
 of the mask should be 0 for any bit positions you want to clear and 1 for
 any bit positions you want to remain unchanged.

 You can use the OR instruction to force specific bits to 1 regardless of
 their current settings. The bits of the mask should be 1 for any bit
 positions you want to set and 0 for any bit positions you want to remain
 unchanged.

 You can use the XOR instruction to toggle the value of specific bits
 (reverse them from their current settings). This instruction sets a bit to 1
 if the corresponding bits are different or to 0 if they are the same. The
 bits of the mask should be 1 for any bit positions you want to toggle and 0
 for any bit positions you want to remain unchanged.

 The following examples show an application for each of these instructions.
 The code illustrating the AND instruction converts a "y" or "n" read from
 the keyboard to uppercase, since bit 5 is always clear in uppercase letters.
 In the example for OR, the first statement is faster and uses fewer bytes
 than  cmp bx, 0. When the operands for XOR are identical, each bit cancels
 itself, producing 0.

   ; Converts characters to uppercase
           mov     ah, 7           ; Get character without echo
           int     21h
           and     al, 11011111y   ; Convert to uppercase by clearing

                                   ;  bit 5
           cmp     al, 'Y'         ; Is it Y?
           je      yes             ; If so, do Yes actions
           .                       ;  else do No actions
           .
   yes:    .

   ; Compares operand to 0
           or      bx, bx          ; Compare to 0
                                   ;  2 bytes, 2 clocks on 8088
           jg      positive        ; BX is positive
           jl      negative        ; BX is negative
                                   ; else BX is zero

   ; Sets a register to 0
           xor     cx, cx          ; 2 bytes, 3 clocks on 8088
           sub     cx, cx          ; 2 bytes, 3 clocks on 8088
           mov     cx, 0           ; 3 bytes, 4 clocks on 8088

 On the 80386 and 80486, the BSF (Bit Scan Forward) and the BSR (Bit Scan
 Reverse) instructions perform operations similar to those of the logical
 instructions. They scan the contents of a register to find the first-set or
 last-set bit. You can use BSF or BSR to find the position of a set bit in a
 mask or to check if a register value is 0.


 4.3.2  Shifting and Rotating Bits

 The 8086-based processors provide a complete set of instructions for
 shifting and rotating bits. Shift instructions move bits a specified number
 of places to the right or left. The last bit in the direction of the shift
 goes into the carry flag, and the first bit is filled with 0 or with the
 previous value of the first bit.

 Rotate instructions also move bits a specified number of places to the right
 or left. For each bit rotated, the last bit in the direction of the rotate
 operation moves into the first bit position at the other end of the operand.
 With some variations, the carry bit is used as an additional bit of the
 operand. Figure 4.3 illustrates the eight variations of shift and rotate
 instructions for eight-bit operands. Notice that SHL and SAL are identical.


 (This figure may be found in the printed book.)

 All shift instructions use the same format. Before the instruction executes,
 the destination operand contains the value to be shifted; after the
 instruction executes, it contains the shifted operand. The source operand
 contains the number of bits to shift or rotate. It can be the immediate
 value 1 or the CL register. The 8088 and 8086 processors do not accept any
 other values or registers with these instructions.


 The shift instruction allows you to change masks during program execution.

 Masks for logical instructions can be shifted to new bit positions. For
 example, an operand that masks off a bit or group of bits can be shifted to
 move the mask to a different position, allowing you to mask off a different
 bit each time the mask is used. This technique, illustrated in the following
 example, is useful only if the mask value is unknown until run time.


   .DATA
   masker  BYTE    00000010y   ; Mask that may change at run time
           .CODE
           .
           .
           .
           mov     cl, 2       ; Rotate two at a time
           mov     bl, 57h     ; Load value to be changed 01010111y
           rol     masker, cl  ; Rotate two to left       00001000y
           or      bl, masker  ; Turn on masked values    ---------
                               ; New value is 05Fh        01011111y
           rol     masker, cl  ; Rotate two more          00100000y
           or      bl, masker  ; Turn on masked values    ---------
                               ; New value is 07Fh        01111111y

 Starting with the 80186 processor, you can use eight-bit immediate values
 larger than 1 as the source operand for shift or rotate instructions, as
 shown below:

   shr     bx, 4   ;  9 clocks, 3 bytes on 80286

 The following statements are equivalent if the program must run on the 8088
 or 8086 processor:

   mov     cl, 4   ;  2 clocks, 3 bytes on 80286
           shr     bx, cl  ;  9 clocks, 2 bytes on 80286
                           ; 11 clocks, 5 bytes


 4.3.3  Multiplying and Dividing with Shift Instructions

 You can use the shift and rotate instructions (SHR, SHL, SAR, and SAL) for
 multiplication and division. Shifting an integer right by one bit has the
 effect of dividing by two; shifting left by one bit has the effect of
 multiplying by two. You can take advantage of shifts to do fast
 multiplication and division by powers of two. For example, shifting left
 twice multiplies by four, shifting left three times multiplies by eight, and
 so on.

 Use SHR (Shift Right) to divide unsigned numbers. You can use SAR (Shift
 Arithmetic Right) to divide signed numbers, but SAR rounds numbers down─IDIV
 always rounds up. Division using SAR must adjust for this difference.
 Multiplication by shifting is the same for signed and unsigned numbers, so
 you can use either SAL or SHL.

 Use shifts instead of MUL or DIV to optimize your code.

 Since the multiply and divide instructions are very slow on the 8088 and
 8086 processors, using shifts instead can often speed operations by a factor
 of 10 or more. For example, on the 8088 or 8086 processor, these statements
 take only four clocks:

   sub     ah, ah    ; Clear AH
           shl     ax, 1     ; Multiply byte in AL by 2

 The following statements produce the same results, but take between 74 and
 81 clocks on the 8088 or 8086. The same statements take 15 clocks on the
 80286 and between 11 and 16 clocks on the 80386.

   mov     bl, 2     ; Multiply byte in AL by 2
           mul     bl

 You can put multiplication and division operations in macros so they can be
 changed if the constants in a program change, as shown in the two macros
 below.

   mul_10  MACRO   factor       ; Factor must be unsigned
           mov     ax, factor   ; Load into AX
           shl     ax, 1        ; AX = factor * 2
           mov     bx, ax       ; Save copy in BX
           shl     ax, 1        ; AX = factor * 4
           shl     ax, 1        ; AX = factor * 8
           add     ax, bx       ; AX = (factor * 8) + (factor * 2)
           ENDM                 ; AX = factor * 10

   div_512 MACRO   dividend     ; Dividend must be unsigned
           mov     ax, dividend ; Load into AX
           shr     ax, 1        ;  AX = dividend / 2 (unsigned)
           xchg    al, ah       ; xchg is like rotate right 8
                                ;  AL = (dividend / 2) / 256
           cbw                  ; Clear upper byte
           ENDM                 ;  AX = (dividend / 512)

 Since RCR and RCL use the carry flag, clear it before multiple-register
 shifts.

 If you need to shift a value that is too large to fit in one register, you
 can shift each part separately. The RCR (Register Carry Right) and RCL
 (Register Carry Left) instructions carry values from the first register to
 the second by passing the leftmost or rightmost bit through the carry flag.


 This example shifts a multiword value.


   .DATA
   mem32    DWORD  500000
           .CODE

   ; Divide 32-bit unsigned by 16
           mov     cx, 4                ; Shift right 4        500000
   again:  shr     WORD PTR mem32[2], 1 ; Shift into carry  DIV    16
           rcr     WORD PTR mem32[0], 1 ; Rotate carry in      ------
           loop    again                ;                       31250

 Since the carry flag is treated as part of the operand (it's like using a
 nine-bit or 17-bit operand), the flag value before the operation is crucial.
 The carry flag can be set by a previous instruction, but you can also set it
 directly by using the CLC (Clear Carry Flag), CMC (Complement Carry Flag),
 and STC (Set Carry Flag) instructions.


 On the 80386 and 80486, an alternate method for multiplying quickly by
 constants takes advantage of the LEA (Load Effective Address) instruction
 and the scaling of indirect memory operands. By using a 32-bit value as both
 the index and the base register in an indirect memory operand, you can
 multiply by the constants 2, 3, 4, 5, 8, and 9 more quickly than you can by
 using the MUL instruction. LEA calculates the offset of the source operand
 and stores it into the destination register, EBX, as this example shows:

   lea     ebx, [eax*2]        ; EBX = 2 * EAX
           lea     ebx, [eax*2+eax]    ; EBX = 3 * EAX
           lea     ebx, [eax*4]        ; EBX = 4 * EAX
           lea     ebx, [eax*4+eax]    ; EBX = 5 * EAX
           lea     ebx, [eax*8]        ; EBX = 8 * EAX
           lea     ebx, [eax*8+eax]    ; EBX = 9 * EAX

 Section 3.2.4.3, "Indirect Memory Operands with 32-Bit Registers," discusses
 scaling of 80386 indirect memory operands, and Section 3.3.3.2, "Loading
 Addresses into Registers," introduces LEA.

 This chapter has covered the integer operations you use in your MASM
 programs. The next chapter looks at more complex data types─arrays, strings,
 structures, unions, and records. Many of the operations presented in this
 chapter can also be applied to the data structures discussed in Chapter 5,
 "Defining and Using Complex Data Types."


 4.4  Related Topics in Online Help

 Online help features additional information about the topics discussed in
 this chapter. From the "MASM 6.0 Contents" screen for MASM online help,
 select the following topics:

╓┌─────────────────────────────────────┌─────────────────────────────────────╖
 Topic                                 Access
 ────────────────────────────────────────────────────────────────────────────
 BYTE, WORD, ...                       Choose "Directives" and then "Data
                                       Allocation"

 Bitwise logical operations            Choose "Operators" and then from the
                                       list of operators, choose "Logical
                                       and Shift"

 Location counter                      Choose "Predefined Symbols" for
                                       information on the $ symbol
 Topic                                 Access
 ────────────────────────────────────────────────────────────────────────────
                                      information on the $ symbol

 BSF, BSR, SHLD, SHRD, and SET         From the "Processor Instructions"
 condition                             categories, choose "Logical and
                                       Shift"

 LES,  LFS,  LGS                       From the "Processor Instructions"
                                       categories, choose "Data Transfer"

 .RADIX directive                      Choose "Directives" and then choose
                                       "Miscellaneous"

 MOD                                   Choose "Operators," and then
                                       "Arithmetic"

 OPATTR, .TYPE, HIGH, LOW, HIGHWORD,   Choose "Operators," then
 and  LOWWORD                          "Miscellaneous"

 OPTION EXPR32,                        Choose "Directives," and then
 Topic                                 Access
 ────────────────────────────────────────────────────────────────────────────
OPTION EXPR32,                        Choose "Directives," and then
 OPTION EXPR16,                        "OPTION"








 Chapter 5  Defining and Using Complex Data Types
 ────────────────────────────────────────────────────────────────────────────

 With the complex data types available in MASM 6.0─arrays, strings, records,
 structures, and (new to version 6.0) unions─you can access data either as a
 unit or as individual elements that make up the unit. The individual
 elements of complex data types are often the integer types discussed in
 Chapter 4, "Defining and Using Integers."

 Section 5.1 first discusses how to declare, reference, and initialize arrays
 and strings. This section summarizes the general steps needed to process
 arrays and strings and describes the MASM instructions for moving,
 comparing, searching, loading, and storing operations.

 Section 5.2 covers similar information for structures and unions: how to
 declare structure and union types, how to define structure and union
 variables, and how to reference structures and unions and their fields.

 Section 5.3 explains how to declare record types, define record variables,
 and use record operators.

 All three sections also describe how to use the LENGTHOF, SIZEOF, and TYPE
 operators with each complex data type.


 5.1  Arrays and Strings

 An assembly-language array is a sequence of fixed-size variables. A string
 is an array of characters. You can access the elements in an array or string
 relative to the first element.

 This section explains and illustrates the essential ways to handle arrays
 and strings in your programs. It covers arrays first, beginning with the two
 ways to declare an array and continuing with how to reference it. The
 section then explains the special requirements for declaring and
 initializing a string. Finally, it describes the processing of arrays and
 strings.


 5.1.1  Declaring and Referencing Arrays

 You can declare an array in two ways: you can specify a list of array
 elements, or you can use the DUP operator to specify a group of identical
 elements.

 To declare an array, you must supply a label name, a type, and a series of
 elements separated by commas. You can access each element of an array
 relative to the first. In the examples below,  warray  and  xarray  are
 arrays.

   warray  WORD    1, 2, 3, 4
   xarray  DWORD   OFFFh, OAAAh

 The assembler stores the elements consecutively in memory, with the first
 address referenced by the label name.

 Initializer lists can be longer than one line.

 Beginning with MASM 6.0, initializer lists of array declarations can span
 multiple lines. The first initializer must appear on the same line as the
 data type, all entries must be initialized, and, if you want the array to
 continue to the new line, the line must end with a comma. These examples
 show legal multiple-line array declarations:

   big             BYTE    21, 22, 23, 24, 25,
                           26, 27, 28

   somelist        WORD    10,
                           20,
                           30

 If you do not want to use the new LENGTHOF and SIZEOF operators discussed
 later in this section, then an array may span more than one logical line,
 although a separate type declaration is needed on each logical line:

   var1    BTYE    10, 20, 30
           BYTE    40, 50, 60
           BYTE    70, 80, 90


 The DUP Operator

 You can also declare an array with the DUP operator. This operator can be
 used with any of the data allocation directives described in Section 4.1.1.
 In the syntax

   count DUP (initialvalue [[,initialvalue]]...)

 the count value sets the number of times to repeat the last initialvalue.
 Each initial value is evaluated only once and can be any expression that
 evaluates to an integer value, a character constant, or another DUP
 operator. The initial value (or values) must always be placed within
 parentheses. For example, the statement

   barray  BYTE    5 DUP (1)

 allocates the integer  1  five times for a total of five bytes.

 The following examples show various ways to use the DUP operator to allocate
 data elements.

   array   DWORD   10 DUP (1)                    ; 10 doublewords
                                                 ;  initialized to 1
   buffer  BYTE    256 DUP (?)                   ; 256-byte buffer

   masks   BYTE    20 DUP (040h, 020h, 04h, 02h) ; 80-byte buffer
                                                 ;  with bit masks
   three_d DWORD   5 DUP (5 DUP (5 DUP (0)))     ; 125 doublewords
                                                 ;  initialized to 0


 Referencing Arrays

 Once an array is defined, you can refer to its first element by typing the
 array name (no brackets required). The array name refers to the first object
 of the given type in the list of initial values.

 If  warray  has been defined as

   warray WORD 2, 4, 6, 8, 10

 then referencing  warray  in your program refers to the first word─the word
 containing  2.

 To refer to the next element (in an array of words), use either of these two
 forms, each of which refers to the array element two bytes past the
 beginning of warray:

   warray+2
   warray[2]

 This element can be used as you would any data item:

   mov     ax, warray[2]
           push    warray+2

 When used with a variable name, brackets only add a number to the address.
 If  warray  refers to the address  2400h, then  warray[2]  refers to the
 address  2402h. The BOUND instruction (80186-80486 only) can be used to
 verify that an index value is within the bounds of an array.

 Array indexes are not scaled. The index is a distance in bytes.

 In assembly language, array indexes are zero-based and unscaled. The number
 within brackets always represents an absolute distance in bytes. In
 practical terms, the fact that indexes are unscaled means that if an element
 is larger than one byte, you must multiply the index of the element by its
 size (in the example above,  2), and then add the result to the address of
 the array. Thus, the expression  warray[4]  represents the third element,
 which is four bytes past the beginning of the array. Similarly, the
 expression  warray[6]  represents the fourth element.

 You can also determine an index at run time:

   mov     si, cx          ; CX holds index value
           shl     si, 7           ; Scale for word referencing
           mov     ax, warray[si]  ; Move element into AX

 The offset required to access an array element can be calculated with the
 following formula:

 nth element of array = array[(n-1) * size of element]


 LENGTHOF, SIZEOF, and TYPE for Arrays

 When applied to arrays, the LENGTHOF, SIZEOF, and TYPE operators return
 information about the length and size of the array and about the type of the
 initializers.

 The LENGTHOF operator returns the number of items in the definition. It can
 be applied only to an integer label. This is useful for determining the
 number of elements you need to process in an array of integers. For an array
 or string label, SIZEOF returns the number of bytes used by the initializers
 in the definition. TYPE returns the size of the elements of the array. These
 examples illustrate these operators:

   array   WORD    40 DUP (5)

   larray  EQU     LENGTHOF array    ; 40 elements
   sarray  EQU     SIZEOF   array    ; 80 bytes
   tarray  EQU     TYPE     array    ;  2 bytes per element

   num     DWORD   4, 5, 6, 7,
                   8, 9, 10, 11

   lnum    EQU     LENGTHOF num      ;  8 elements
   snum    EQU     SIZEOF   num      ; 32 bytes
   tnum    EQU     TYPE     num      ;  4 bytes per element

   warray  WORD    40 DUP (40 DUP (5))

   len     EQU     LENGTHOF warray   ; 1600 elements
   siz     EQU     SIZEOF   warray   ; 3200 bytes
   typ     EQU     TYPE     warray   ;    2 bytes per element


 5.1.2  Declaring and Initializing Strings

 A string is an array of bytes. Initializing a string like  "Hello, there"
 allocates and initializes one byte for each character in the string. An
 initialized string can be no longer than 255 characters.

 Strings declared with types other than BYTE must fit the memory space
 allocated.

 For data directives other than BYTE, a string may initialize only a single
 element. This element must be short enough to fit into the specified size
 and conform to the expression word size in effect (see Section
 1.2.4,"Integer Constants and Constant Expressions"), as shown in these
 examples:

   wstr    WORD    "OK"
   dstr    DWORD   "ADCD"  ; Legal under EXPR32 only

 As with arrays, string initializers can span multiple lines. The line must
 end with a comma if you want the string to continue to the next line.

   str1    BYTE    "This is a long string that does not ",
                   "fit on one line."

 You can also have an array of pointers to strings. For example:

   PBYTE   TYPEDEF PTR BYTE
           .DATA
   msg1    BYTE    "Operation completed successfully."
   msg2    BYTE    "Unknown command"
   msg3    BYTE    "File not found"
   pmsg1   PBYTE   msg1
   pmsg2   BPBYTE  msg2
   pmsg3   PBYTE   msg3

   errors  WORD    pmsg1, pmsg2, pmsg3    ; An array of pointers
                                          ;  to strings

 Strings must be enclosed in single (') or double (") quotation marks. To put
 a single quotation mark inside a string enclosed by single quotation marks,
 use two single quotation marks. Likewise, if you need quotation marks inside
 a string enclosed by double quotation marks, use two sets. These examples
 show the various uses of quotation marks:

   char    BYTE    'a'
   message BYTE    "That's the message."       ; That's the message.
   warn    BYTE    'Can''t find file.'         ; Can't find file.
   string  BYTE    "This ""value"" not found." ; This "value"

                                                 not found.

 You can always use single quotation marks inside a string enclosed by double
 quotation marks, as the initialization for  message  shows, and vice versa.



 The ? Initializer

 The actual values stored when you use ? depend on the other data in your
 program.

 You do not have to initialize all elements in an array to a value. If there
 is no initial value, you can initialize the array elements with the ?
 operator. The ? operator either is treated as a zero or causes a byte to be
 left unspecified in the object file. Object files contain records for
 initialized data. An unspecified byte left in the object file means that no
 records contain initialized data for that address.

 The actual values stored in arrays allocated with ? depend on certain
 conditions. The ? initializer is treated as a zero in a DUP statement that
 contains initializers in addition to the ? initializer. An unspecified byte
 is left in the object file if the ? initializer does not appear in a DUP
 statement, or if the DUP statement contains only ? initializers for nested
 DUP statements.


 Length-Specified Strings

 Often there are reasons to know the length of a string. To use the DOS
 functions for writing to a file, for example, CX must contain the length of
 the string before the interrupt is called, as shown in this example.

   msg     BYTE    "This is a length-specified string"
           .
           .
           .
           mov     ah, 40h
           mov     bx, 1
           mov     cx, LENGTHOF msg
           mov     dx, OFFSET msg
           int     21h

 Some high-level languages also expect strings passed to procedures to have a
 certain format. For example, Pascal procedures require the first byte of a
 string passed as a parameter to contain the length of the string. You can
 write this length into the first byte with

   msg     BYTE    LENGTHOF msg - 1, "This is a Pascal string"

 Interfacing with high-level languages requires special techniques with
 strings.

 Other languages such as Basic have string descriptions─a kind of structure
 containing both the length and the address of the string. For example, this
 structure  DESC  could be used in a procedure accessed from Basic:

   DESC    STRUCT
     len   WORD    ?       ; Length of string1
     off   WORD    ?       ; Offset of string1
   DESC    ENDS

   string1 BYTE    "This string goes in a string descriptor"
   msg     DESC    {LENGTHOF string1, string1}

 See Section 5.2, "Structures and Unions."


 Null-Terminated and $-Terminated Strings

 Null-terminated and $-terminated strings have a special use with DOS
 functions. Strings in modules shared with C need to end with a null
 character (0).

   str1    BYTE    "This string ends with a null character", 0

 DOS file names also require a null character at the end. This example opens
 a file named  "MYFILE.ASM".

   name1   BYTE    "MYFILE.ASM", 0
           .
           .
           .
           mov     ah, 3Dh
           mov     dx, OFFSET name1
           int     21h

 DOS function 9 requires a string to end with a dollar sign ($) so that it
 can recognize the end of the string to write to the screen, as shown in this
 example.

   msg     BYTE    "This is a dollar-terminated string$"
           .
           .
           .
           mov     ah, 09h
           mov     dx, OFFSET msg
           int     21h


 LENGTHOF, SIZEOF, and TYPE for Strings

 Because the assembler considers strings as simply arrays of byte elements,
 the LENGTHOF and SIZEOF operators return the same values for strings as they
 do for arrays, as illustrated in this example. The TYPE operator considers
 msg  to be one data unit and returns 1.

   msg     BYTE    "This string extends ",
                   "over three ",
                   "lines."

   lmsg    EQU     LENGTHOF msg      ; 37 elements
   smsg    EQU     SIZEOF   msg      ; 37 bytes
   tmsg    EQU     TYPE     msg      ;  1 byte per element


 5.1.3  Processing Arrays and Strings

 The 8086-family instruction set has seven string instructions for fast and
 efficient processing of entire strings and arrays. The term "string" in
 "string instructions" refers to a sequence of elements, not just character
 strings. These instructions work directly only on arrays of bytes and words
 on the 8086-80486 and on arrays of bytes, words, and doublewords on the
 80386 and 80486. Processing larger elements must be done indirectly with
 loops.


 The following list gives capsule descriptions of the five instructions
 discussed in this section. Two additional instructions not described here
 are the INS and OUTS instructions that transfer values to and from a memory
 port.

 Instruction   Description
 ────────────────────────────────────────────────────────────────────────────
 MOVS          Copies a string from one location to another
 STOS          Stores values from the accumulator register to a string
 CMPS          Compares values in one string with values in another
 LODS          Loads values from a string to the accumulator register
 SCAS          Scans a string for a specified value

 All of these instructions use registers in a similar way and have a similar
 syntax. Most are used with the repeat instruction prefixes REP, REPE (or
 REPZ), and REPNE (or REPNZ). REPZ is a synonym for REPE (Repeat While Equal)
 and REPNZ is a synonym for REPNE (Repeat While Not Equal).


 This section first explains the general procedures for using all string
 instructions. It then illustrates each instruction with an example.


 5.1.3.1  Overview of String Operations

 The string instructions have specific requirements for the location of
 strings and the use of registers. To operate on any string, follow these
 three steps:

 All string operations follow three basic steps.


   1.  Set the direction flag to indicate the direction in which you want to
       process the string. The STD instruction sets the flag, while CLD
       clears it.

       If the direction flag is clear, the string is processed upward (from
       low addresses to high addresses, which is from left to right through
       the string). If the direction flag is set, the string is processed
       downward (from high addresses to low addresses, or from right to
       left). Under DOS, the direction flag is normally clear if your program
       has not changed it.

   2.  Load the number of iterations for the string instruction into the CX
       register.

       If you want to process a 100-byte string, move 100 into CX. If you
       wish the string instruction to terminate conditionally (for example,
       during a search when a match is found), load the maximum number of
       iterations that can be performed without an error.

   3.  Load the starting offset address of the source string into DS:SI and
       the start-ing address of the destination string into ES:DI. Some
       string instructions take only a destination or source, not both (see
       Table 5.1).

       Normally, the segment address of the source string should be DS, but
       you can use a segment override to specify a different segment for the
       source operand. You cannot override the segment address for the
       destination string. Therefore, you may need to change the value of ES.
       See Section 3.1 for information on changing segment registers.

 ────────────────────────────────────────────────────────────────────────────
 NOTE
 Although you can use a segment override on the source operand, a segment
 override combined with a repeat prefix can cause problems in certain
 situations on all processors except the 80386/486. If an interrupt occurs
 during the string operation, the segment override is lost and the rest of
 the string operation processes incorrectly. Segment overrides can be used
 safely when interrupts are turned off or with an 80386/486
 processor.───────────────────────────────────────────────────────────────────



 You can adapt these steps to the requirements of any particular string
 operation. The syntax for the string instructions is:

   «prefix» CMPS «segmentregister:»
   source, «ES:» destination
                  LODS «segmentregister:» source
   «prefix» MOVS «ES:» destination,
   «segmentregister:» source
   «prefix» SCAS «ES:» destination
   «prefix» STOS «ES:« destination

 Some instructions have special forms for byte, word, or doubleword operands.
 If you use the form of the instruction that ends in B (BYTE), W (WORD), or D
 (DWORD) with LODS, SCAS, and STOS, the assembler knows whether the element
 is in the AL, AX, or EAX register. Therefore, these instruction forms do not
 require operands.

 Table 5.1 lists each string instruction with the type of repeat prefix it
 uses and indicates whether the instruction works on a source, a destination,
 or both.

 Table 5.1  Requirements for String Instructions

╓┌─────────────┌───────────────┌───────────────────┌─────────────────────────╖
 Instruction   Repeat Prefix   Source/Destination  Register Pair
 ────────────────────────────────────────────────────────────────────────────
 MOVS          REP             Both                DS:SI, ES:DI
 SCAS          REPE/REPNE      Destination         ES:DI
 CMPS          REPE/REPNE      Both                DS:SI, ES:DI
 LODS          None            Source              DS:SI
 STOS          REP             Destination         ES:DI
 INS           REP             Destination         ES:DI
 OUTS          REP             Source              DS:SI
 ────────────────────────────────────────────────────────────────────────────


 The instruction automatically increments DI or SI.

 The repeat prefix causes the instruction that follows it to repeat for the
 number of times specified in the count register or until a condition becomes
 true. After each iteration, the instruction increments or decrements SI and
 DI so that it points to new array elements. The string instructions work on
 these elements. The direction flag determines whether SI and DI are
 incremented (flag clear) or decremented (flag set). The size of the
 instruction determines whether SI and DI are altered by one, two, or four
 bytes each time.

 These are the conditions that determine the number of repetitions specified
 by a prefix.

 Prefix                            Description
 ────────────────────────────────────────────────────────────────────────────
 REP                               Repeats instruction CX times

 REPE, REPZ                        Repeats instruction CX times, or as long
                                   as elements are equal, whichever is
                                   fewer

 REPNE, REPNZ                      Repeats instruction CX times, or as long
                                   as elements are not equal, whichever is
                                   fewer


 The prefixes apply to only one string instruction at a time. To repeat a
 block of instructions, use a loop construction (see Section 7.2, "Loops").

 At run time, if a string instruction is preceded by a repeat sequence, the
 processor takes the following steps:


   1.  Checks the CX register and exits if CX is 0. If the REPE prefix is
       used, the loop exits if the zero flag is set; if REPNE is used, the
       loop exits if the zero flag is clear.

   2.  Performs the string operation once.

   3.  Increases SI and/or DI if the direction flag is clear. Decreases SI
       and/or DI if the direction flag is set. The amount of increase or
       decrease is 1 for byte operations, 2 for word operations, and 4 for
       doubleword operations (80386/486 only).

   4.  Decrements CX (no flags are modified).

   5.  Checks the zero flag at this point if the REPE or REPNE prefix is used
       (for SCAS or CMPS). If the repeat condition does not hold, execution
       proceeds to the next instruction.

   6.  Proceeds to the next iteration and repeats from step 1.


 At loop end, SI and DI point to the element immediately after the match.

 When the repeat loop ends, SI (or DI) points to the position following a
 match (when using SCAS or CMPS), so you need to decrement or increment DI or
 SI to point to the element where the match occurred.

 Although string instructions (except LODS) are most often used with repeat
 prefixes, they can also be used by themselves. In this case, the SI and/or
 DI registers are adjusted as specified by the direction flag and the size of
 operands. However, you must decrement the CX register and set up a loop for
 the repeated action.


 5.1.3.2  String Instructions

 To use the 8086-family string instructions, apply the steps outlined in the
 previous section. Examples in this section illustrate each instruction.

 You can also use the techniques in this section with structures and unions,
 since arrays and strings can be fields in structures and unions (see Section
 5.2).

 Moving Array Data - The MOVS instruction copies data from one area of memory
 to another. To move data, first load the count and the source and
 destination addresses into the appropriate registers. Then use REP with the
 MOVS instruction.

   .MODEL  small
           .DATA
   source  BYTE    10 DUP ('0123456789')
   destin  BYTE    100 DUP (?)
           .CODE
           mov     ax, @data           ; Load same segment
           mov     ds, ax              ;  to both DS
           mov     es, ax              ;  and ES
           .
           .
           .
           cld                         ; Work upward
           mov     cx, LENGTHOF source ; Set iteration count to 100
           mov     si, OFFSET source   ; Load address of source
           mov     di, OFFSET destin   ; Load address of destination
           rep     movsb               ; Move 100 bytes

 Storing Data in Arrays - The STOS instruction stores a specified value in
 each position of a string. The string is the destination, so it must be
 pointed to by ES:DI. The value to store must be in the accumulator.

 This example stores the character  'a'  in each byte of a 100-byte string.
 Notice that it does this by storing 50 words rather than 100 bytes. This
 makes the code faster by reducing the number of iterations. To fill an odd
 number of bytes, you would have to adjust for the last byte.

   .MODEL  small, C
           .DATA
   destin  BYTE    100 DUP (?)
   ldestin EQU     (LENGTHOF destin) / 2
           .CODE
           .                           ; Assume ES = DS
           .
           .
           cld                         ; Work upward
           mov     ax, 'aa'            ; Load character to fill
           mov     cx, ldestin         ; Load length of string
           mov     di, OFFSET destin   ; Load address of destination
           rep     stosw               ; Store 'aa' into array

 Comparing Arrays - The CMPS instruction compares two strings and points to
 the address after which a match or nonmatch occurs. If the values are the
 same, the zero flag is set. Either string can be considered as the
 destination or the source unless a segment override is used.

 This example using CMPSB assumes that the strings are in different segments.
 Both segments must be initialized to the appropriate segment register.

   .MODEL  large, C
           .DATA
   string1 BYTE    "The quick brown fox jumps over the lazy dog"
           .FARDATA
   string2 BYTE    "The quick brown dog jumps over the lazy fox"
   lstring EQU     LENGTHOF string2
           .CODE
           mov     ax, @data           ; Load data segment
           mov     ds, ax              ;  into DS
           mov     ax, @fardata        ; Load far data segment
           mov     es, ax              ;  into ES
           .
           .
           .
           cld                         ; Work upward
           mov     cx, lstring         ; Load length of string
           mov     si, OFFSET string1  ; Load offset of string1
           mov     di, OFFSET string2  ; Load offset of string2
           repe    cmpsb               ; Compare
           jcxz    allmatch            ; CX is 0 if no nonmatch
           .
           .
           .
   allmatch:                           ; Special case for all match

 Loading Data from Arrays - The LODS instruction loads a value from a string
 into a register. The string is the source; the value is in the accumulator.
 This instruction normally is not used with a repeat instruction prefix,
 since something must be done with each element before going on to the next.

 The code in this example loads, processes, and displays each byte in a
 string of bytes.

   .DATA
   info    BYTE    0, 1, 2, 3, 4, 5, 6, 7, 8, 9
   linfo   WORD    LENGTHOF info
           .CODE
           .
           .
           .
           cld                       ; Work upward
           mov     cx, linfo         ; Load length
           mov     si, OFFSET info   ; Load offset of source
           mov     ah, 2             ; Display character function

   get:
           lodsb                     ; Get a character
           add     al, '0'           ; Convert to ASCII
           mov     dl, al            ; Move to DL
           int     21h               ; Call DOS to display character
           loop    get               ; Repeat

 Searching Arrays - The SCAS instruction scans a string for a specified
 value. As the loop executes, this instruction compares the value pointed to
 by DI with the value in the accumulator. If values are the same, the zero
 flag is set.

 After a REPNE SCAS, the zero flag is cleared if no match was found. After a
 REPE SCAS, the zero flag is set if all values matched.

 This example assumes that ES is not the same as DS and that the address of
 the string is stored in a pointer variable. The LES instruction loads the
 far address of the string into ES:DI.

   .DATA
   string  BYTE    "The quick brown fox jumps over the lazy dog"
   pstring PBYTE   string             ; Far pointer to string
   lstring EQU     LENGTHOF string    ; Length of string
           .CODE
           .
           .
           .
           cld                        ; Work upward
           mov     cx, lstring        ; Load length of string
           les     di, pstring        ; Load address of string
           mov     al, 'z'            ; Load character to find
           repne   scasb              ; Search
           jcxz    notfound           ; CX is 0 if not found
           .                          ; ES:DI points to character
           .                          ;  after first 'z'
           .
   notfound:                          ; Special case for not found


 5.2  Structures and Unions

 A structure is a group of possibly dissimilar data types and variable
 declarations that can be accessed as a unit or by any of its components. The
 fields within the structure can have different sizes and data types.

 Unions are identical to structures, except that the fields of a union
 overlap in memory, which allows you to define different data formats for the
 same memory space. Unions can store different types of data depending on the
 situation. They can also store data as one data type and retrieve it as
 another data type.

 Whereas each field in a structure has an offset relative to the first byte
 of the structure, all the fields in a union start at the same offset. The
 size of a structure is the sum of its components, while the size of a union
 is the length of the longest field.

 A MASM structure is similar to a struct in the C language, a STRUCTURE in
 FORTRAN, and a RECORD in Pascal. Unions in MASM are similar to unions in C
 and FORTRAN, and to variant records in Pascal.

 Follow these steps when using structures and unions:


   1.  Declare a structure (or union) type.

   2.  Define one or more variables having that type.

   3.  Reference the fields directly or indirectly with the field (dot)
       operator.


 You can use the entire structure or union variable or just the individual
 fields as operands in assembler statements. This section explains the
 allocating, initializing, and nesting of structures and unions.

 MASM 6.0 extends the functionality of structures and also makes some changes
 to MASM 5.1 behavior. You can still retain MASM 5.1 behavior if you prefer
 by specifying OPTION OLDSTRUCTS in your program. See Section 1.3.2 for
 information about the OPTION directive, and Section 5.2.3 for information
 about referencing structures and unions.


 5.2.1  Declaring Structure and Union Types

 When you declare a structure or union type, you create a template for data
 that contains the sizes and, optionally, the initial values for fields in
 the structure or union but that allocates no memory.

 The STRUCT keyword marks the beginning of a type declaration for a
 structure. (STRUCT and STRUC are synonyms.) STRUCT and UNION type
 declarations have the following format:

   name {STRUCT | UNION}  «alignment»
   «,NONUNIQUE »
   fielddeclarations
   name ENDS

 The fielddeclarations are a series of one or more variable declarations. You
 can declare default initial values individually or with the DUP operator
 (see Section 5.2.2, "Defining Structure and Union Variables"). Section
 5.2.3, "Referencing Structures, Unions, and Fields," explains the NONUNIQUE
 keyword. Structures and unions can also be nested in MASM 6.0 (see Section
 5.2.4).


 Initializing Fields

 If you provide initializers for the fields of a structure or union when you
 declare the type, these initializers become the default value for the fields
 when you define a variable of that type. Section 5.2.2 explains default
 initializers.

 When you initialize the fields of a union type, the type and value of the
 first field become the default value and type for the union. In this example
 of an initialized union declaration, the default type for the union is
 DWORD:

   DWB     UNION
     d     DWORD   00FFh
     w     WORD    ?
     b     BYTE    ?
   DWB     ENDS

 If the size of the first member is less than the size of the union, the
 assembler initializes the rest of the union to zeros. When initializing
 strings in a type, make sure the initial values are long enough to
 accommodate the largest possible string.


 Field Names

 Structure and union field names in MASM 6.0 must be unique within a given
 nesting level because they represent the offset from the beginning of the
 structure to the corresponding field.

 A nested structure has its own level.

 In MASM 6.0, a label and a structure field may have the same name, but not a
 text macro and a field name. Also, field names between structures need not
 be unique. Field names do need to be unique if you place OPTION M510 or
 OPTION OLDSTRUCTS in your code or use the /Zm option from the command line,
 since versions of MASM prior to 6.0 require unique field names (see Appendix
 A).


 Alignment Value and Offsets for Structures

 Data access to structures is faster on aligned fields than on unaligned
 fields. Therefore, alignment gains speed at the cost of space. Alignment
 improves access on 16-bit processors but makes no difference on code
 executing on an 8-bit 8088 processor.

 The way the assembler aligns structure fields determines the amount of space
 required to store a variable of that type. Each field in a structure has an
 offset relative to 0. If you specify an alignment in the structure
 declaration (or with the /Zpn command-line option), the offset for each
 field may be modified by the alignment (or n).

 The only values accepted for alignment are 1, 2, and 4. The default is 1. If
 the type declaration includes an alignment, the fields are aligned to the
 minimum of the field's size and the alignment. Any padding required to reach
 the correct offset for the field is added prior to allocating the field. The
 padding consists of zeros and always precedes the field.

 If the number of bytes in the field is greater than the alignment value, the
 element will be padded such that the offset of the element is divisible by
 the alignment value. If the number of bytes is greater than or equal to the
 alignment value, the offset of the element is padded such that it is
 divisible by the element size.

 The size of the structure must also be evenly divisible by the structure
 alignment value, so zeros may be added at the end of the structure.

 If neither the alignment nor the /Zp command-line option is used, the offset
 is incremented by the size of each data directive. This is the same as a
 default alignment equal to 1. The alignment specified in the type
 declaration overrides the /Zp command-line option.

 These examples show how offsets are determined:

   STUDENT2    STRUCT  2   ; Alignment value is 2
     score     WORD    1   ; Offset is  0
     id        BYTE    2   ; Offset is  2
     year      DWORD   3   ; Offset is  4; one byte padding added
     sname     BYTE    4   ; Offset is  8
   STUDENT2    ENDS

 One byte of padding is added at the end of the first byte-sized field.
 Otherwise the offset of the  year  field would be 3, which is not divisible
 by the alignment value of 2. The size of this structure is now 9 bytes.
 Since 9 is not evenly divisible by 2, one byte of padding is added at the
 end of  student2.

   STUDENT4    STRUCT  4            ; Alignment value is 4
     sname     BYTE    1            ; Offset is  0
     score     WORD    10 DUP (100) ; Offset is  2
     year      BYTE    2            ; Offset is 22; 1 byte padding
                                    ;  added so offset of next field
                                    ;  is divisible by 4
     id        DWORD   3            ; Offset is 24
   STUDENT4    ENDS

 The alignment value affects memory allocation of structure variables.

 The alignment value affects the alignment of structure variables, so adding
 an alignment value affects memory usage. This feature provides compatibility
 with structures in Microsoft C.

 With MASM 6.0, C programmers can use the H2INC utility to translate C
 structures to MASM (see Chapter 16).


 5.2.2  Defining Structure and Union Variables

 Once you have declared a structure or union type, variables of that type can
 be defined. For each variable defined, memory is allocated in the current
 segment in the format declared by the type. The syntax for defining a
 structure or union variable is:

   [[name]] typename < [[initializer
   [[,initializer]]...]] >

   [[name]] typename { [[initializer
   [[,initializer]]...]] }

   [[name]] typename constant
   DUP ({ [[initializer [[,initializer]]...]]
   })

 The name is the label assigned to the variable. If no name is given, the
 assembler allocates space for the variable but does not give it a symbolic
 name. The typename is the name of a previously declared structure or union
 type.

 An initializer can be given for each field. The type of each initializer
 must be the type of the corresponding field defined in the type declaration.
 For unions, the type of the initializer must be the same as the type for the
 first field. An initialization list can also be repeated using the DUP
 operator.

 The list of initializers can be broken only after a comma unless you use a
 line continuation character (\) at the end of the line. The last curly brace
 or angle bracket must appear on the same line as the last initializer. You
 can also use the line continuation character to extend a line as shown in
 the  Item4 declaration below. Angle brackets and curly braces can be
 intermixed in an initialization as long as they match. This example using
 the  ITEMS  structure illustrates the options for initializing lists:

   ITEMS       STRUCT
     Iname     BYTE      'Item Name'
     Inum      WORD      ?
     ITYPE     UNION
       oldtype BYTE      0
       newtype WORD      ?
     ENDS
   ITEMS       ENDS
           .
           .
           .
           .DATA
   Item1   ITEMS   < >              ; Accepts default initializers
   Item2   ITEMS   { }              ; Accepts default initializers
   Item3   ITEMS   <'Bolts', 126>   ; Overrides default value of first
                                    ;  2 fields; use default of
                                    ;  the third field
   Item4   ITEMS   { \
                    'Bolts',        ; Item name
                    126 \           ; Part number
                   }

 The angle brackets or curly braces are required even if no initial value is
 given, as in  Item1  and  Item2  in the example. If initial values are given
 for more than one field, the values must be separated by commas, as shown in
  Item3.

 You need not initialize all fields in a structure. If an initial value is
 blank, the assembler automatically uses the default initial value of the
 field, which was originally provided in the structure type declaration. If
 there is no default value, the field is undefined.

 For nested structures or unions (see Section 5.2.4), however, these are
 equivalent:

   Item5   ITEMS   {'Bolts', ,     }
   Item6   ITEMS   {'Bolts', , { } }

 A variable and an array of union type  WB  look like this:

   WB      UNION
     w     WORD    ?
     b     BYTE    ?
   WB      ENDS

   num     WB      {0Fh}                       ; Store 0Fh
   array   WB      (40 / SIZEOF WB) DUP ({2})  ; Allocates and
                                               ;  initializes 10 unions

 (This figure may be found in the printed book.)

 In MASM 6.0, control structures (such as IF, macros, and directives) are
 also allowed within structure and union declarations.


 Arrays as Field Initializers

 Default initializers for string or array fields set the size for the field.


 The length of the array that can override the contents of a field in a
 variable definition is fixed by the size of the initializer. The override
 cannot contain more elements than the default. Specifying fewer override
 array elements changes the first n values of the default where n is the
 number of values in the override. The rest of the array elements take their
 default values from the initializer.


 Strings as Field Initializers

 If the override is shorter, the assembler pads the override with spaces to
 equal the length of the initializer. If the initializer is a string and the
 override value is not a string, the override value must be enclosed in angle
 brackets or curly braces.

 A string may be used to override any member of type BYTE (or SBYTE). The
 string does not need to be enclosed in angle brackets or curly braces unless
 mixed with other override methods.

 The string fields for structure variables are the length defined by the type
 declaration.

 If a structure has an initialized string field or an array of bytes, any new
 string assigned to a variable of the field that is smaller than the default
 is padded with spaces. The assembler adds four spaces at the end of  'Bolts'
  in the variables of type  ITEMS  above. The  Iname  field in the  ITEMS
 structure cannot contain a field initializer longer than  'Item Name'.


 Structures as Field Initializers

 Initializers for structure variables must be enclosed in curly braces or
 angle brackets, but you can specify overrides with fewer elements than the
 defaults.

 This example illustrates the use of default values with structures as field
 initializers:

   DISKDRIVES      STRUCT
     a1            BYTE ?
     b1            BYTE ?
     c1            BYTE ?
   DISKDRIVES      ENDS

   INFO            STRUCT
     buffer        BYTE    100 DUP (?)
     crlf          BYTE    13, 10
     query         BYTE    'Filename: ' ; String <= can override
     endmark       BYTE    36
     drives        DISKDRIVES <0, 1, 1>
   INFO            ENDS

   info1   INFO    { , , 'Dir' }

   ; Illegal since name in query field is too long
   ; and a string cannot initialize a field defined with DUP:
   ; info2  INFO    {"TESTFILE", , "DirectoryName",}

   lotsof  INFO    { , , 'file1', , {0,0,0} },
                   { , , 'file2', , {0,0,1} },
                   { , , 'file3', , {0,0,2} }

 The diagram below shows how the assembler stores  info1.

 (This figure may be found in the printed book.)

 The initialization for  drives  gives default values for all three fields of
 the structure. The fields left blank in  info1  use the default values for
 those fields. The  info2  declaration is illegal since  "DirectoryName"  is
 longer than the initial string for that field, and the  "TESTFILE"  string
 cannot initialize a field defined with DUP.


 Arrays of Structures and Unions

 You can define an array of structures using the DUP operator (see Section
 5.1.1, "Declaring and Referencing Arrays") or by creating a list of
 structures. For example, you can define an array of structure variables like
 this:

   Item7   ITEMS    30 DUP ({,,{10}})

 The  Item7  array defined here has 30 elements of type  ITEMS, with the
 third field of each element (the union) initialized to  10.

 You can also list array elements as shown in this example:

   Item8   ITEMS    {'Bolts', 126, 10},
                    {'Pliers',139, 10},
                    {'Saws',  414, 10}


 Structure Redefinition

 The assembler generates an error for a structure redefinition unless all of
 the following are the same:


   ■   Field names

   ■   Offsets of named fields

   ■   Initialization lists

   ■   Field alignment value


 Additionally, all fields must be present and at the same offset.


 LENGTHOF, SIZEOF, and TYPE for Structures

 The size of a structure determined by SIZEOF is the offset of the last
 field, plus the size of the last field, plus any padding required for proper
 alignment (see Section 5.2.1 for information about alignment). This example,
 using the data declarations above, shows how to use the LENGTHOF, SIZEOF,
 and TYPE operators with structures:

   INFO            STRUCT
     buffer        BYTE    100 DUP (?)
     crlf          BYTE    13, 10
     query         BYTE    'Filename: '
     endmark       BYTE    36
     drives        DISKDRIVES <0, 1, 1>
   INFO            ENDS

   info1   INFO    { , , 'Dir' }
   lotsof  INFO    { , , 'file1', , {0,0,0} },
                   { , , 'file2', , {0,0,1} },
                   { , , 'file3', , {0,0,2} }

   sinfo1  EQU     SIZEOF    info1  ; 116 = number of bytes in

                                    ;  initializers
   linfo1  EQU     LENGTHOF  info1  ; 1 = number of items
   tinfo1  EQU     TYPE      info1  ; 116 = same as size

   slotsof EQU     SIZEOF    lotsof ; 116 * 3 = number of bytes in
                                    ;  initializers
   llotsof EQU     LENGTHOF  lotsof ; 3 = number of items
   tlotsof EQU     TYPE      lotsof ; 116 = same as size for structure

                                    ;  of type INFO


 LENGTHOF, SIZEOF, and TYPE for Unions

 The size of a union determined by SIZEOF is the size of the longest field
 plus any padding required. The length of a union variable determined by
 LENGTHOF equals the number of initializers defined inside angle brackets or
 curly braces. TYPE returns a value indicating the type of the longest field.


   DWB     UNION
     d     DWORD   ?
     w     WORD    ?
     b     BYTE    ?
   DWB     ENDS

   num     DWB     {0FFFFh}
   array   DWB     (100 / SIZEOF DWB) DUP ({0})

   snum    EQU     SIZEOF   num      ; = 4
   lnum    EQU     LENGTHOF num      ; = 1
   tnum    EQU     TYPE     num      ; = 4
   sarray  EQU     SIZEOF   array    ; = 100 (4*25)
   larray  EQU     LENGTHOF array    ; = 25
   tarray  EQU     TYPE     array    ; = 4


 5.2.3  Referencing Structures, Unions, and Fields

 Like other variables, structure variables can be accessed by name. You can
 access fields within structure variables with this syntax:

   variable.field

 In MASM 6.0, references to fields must always be fully qualified, with both
 the structure or union name and the dot operator preceding the field name.
 Also, in MASM 6.0, the dot operator can be used only with structure fields,
 not as an alternative to the plus operator; nor can the plus operator be
 used as an alternative to the dot operator.

 This example shows several ways to reference the fields of a structure
 called  date.

   DATE    STRUCT                            ; Defines structure
   type
     month BYTE    ?
     day   BYTE    ?
     year  WORD    ?
   DATE    ENDS

   yesterday       DATE    {9, 30, 1987}     ; Declare structure
                                             ;  variable
           .
           .
           .
           mov     al, yesterday.day         ; Use structure variables
           mov     bx, OFFSET yesterday      ; Load structure address
           mov     al, (DATE PTR [bx]).month ; Use as indirect operand
           mov     al, [bx].date.month       ; This is necessary if
                                             ;  month were already a

                                             ;  field in a different
                                             ;  structure

 Under OPTION M510 or OPTION OLDSTRUCTS, unique structure names do not need
 to be qualified. See Section 1.3.2 for information on the OPTION directive.


 If the NONUNIQUE keyword appears in a structure definition, all fields of
 the structure must be fully qualified when referenced, even if the OPTION
 OLDSTRUCTS directive appears in the code. Also, in MASM 6.0, all references
 to a field must be qualified.

 Even if the initialized union is the size of a WORD or DWORD, members of
 structures or unions are accessible only through the field's names.

 In the following example, the two MOV statements show how you can access the
 elements of an array of structures.

   WB      UNION
     w     WORD    ?
     b     BYTE    ?
   WB      ENDS

   array   WB      (100 / SIZEOF WB) DUP ({0})

           mov     array[12].w, 40
           mov     array[32].b,  2

 (This figure may be found in the printed book.)

 The  WB  union cannot be used directly as a WORD variable. However, you can
 define a union containing both the structure and a WORD variable and access
 either field. (The next section discusses nested structures and unions.)

 You can use unions to access the same data in more than one form. For
 example, one application of structures and unions is to simplify the task of
 reinitializing a far pointer. If you have a far pointer declared as

   FPWORD  TYPEDEF FAR PTR WORD

           .DATA
   BoxB    FPWORD ?
   BoxA    FPWORD ?
   BoxB2   uptr   < >

 you must follow these steps to point  BoxB  to  BoxA:

   mov     bx,  OFFSET BoxA
           mov     WORD PTR BoxB[2], ds
           mov     WORD PTR BoxB, bx

 When you do this, you must remember whether the segment or the offset is
 stored first. However, if your program contains this union:

   uptr      UNION
     dwptr   FPWORD   0
     STRUCT
       offs  WORD     0
       segm  WORD     0
     ENDS
   uptr      ENDS

 you can initialize a far pointer with these steps:

   mov     BoxB2.segm, ds
           mov     BoxB2.offs, bx
           lds     si, BoxB2.dwptr

 This code moves the segment and the offset into the pointer and then moves
 the pointer into a register with the other field of the union. Although this
 technique does not reduce the code size, it avoids confusion about the order
 for loading the segment and offset.


 5.2.4  Nested Structures and Unions

 Structures and unions in MASM 6.0 can be nested in several ways. This
 section explains how to refer to the fields in a nested structure or union.
 The example below illustrates the four techniques for nesting and how to
 reference the fields. Note the syntax for nested structures. The discussion
 of these techniques follows the example.

   ITEMS           STRUCT
     Inum          WORD    ?
     Iname         BYTE    'Item Name'
   ITEMS           ENDS

   INVENTORY       STRUCT
     UpDate        WORD    ?
     oldItem       ITEMS   { \
                           ?,
                           'AF8' \       ; Named variable of
                           }             ;  existing structure
                   ITEMS   { ?, '94C' }  ; Unnamed variable of
                                         ;  existing type
     STRUCT ups                          ; Named nested structure
       source      WORD    ?
       shipmode    BYTE    ?
     ENDS
     STRUCT                              ; Unnamed nested structure
       f1          WORD    ?
       f2          WORD    ?
     ENDS
   INVENTORY       ENDS

           .DATA

   yearly  INVENTORY       { }

   ; Referencing each type of data in the yearly structure:

           mov     ax, yearly.oldItem.Inum
           mov     yearly.ups.shipmode, 'A'
           mov     yearly.Inum, 'C'
           mov     ax, yearly.f1

 To nest structures and unions, you can use any of these techniques:


   ■   The field of a structure or union can be a named variable of an
       existing structure or union type, as in the  oldItem  field. The field
       names in  oldItem  are not unique, so the full field names must be
       used when referencing those fields in the statement

       mov     ax, yearly.oldItem.Inum


   ■   To declare a named structure or union inside another structure or
       union, give the STRUCT or UNION keyword first and then define a label
       for it. Fields of the nested structure or union must always be
       qualified, as shown in this example:

       mov     yearly.ups.shipmode, 'A'


   ■   As shown in the  Items  field of  Inventory, you can also use unnamed
       variables of existing structures or unions inside another structure or
       union. In this case you can reference its fields directly, as shown in
       this example:

       mov     yearly.Inum, 'C'
                      mov     ax, yearly.f1



 Offsets of nested structures are relative to the nested structure, not the
 root structure. In the example above, the offset of  yearly.ups.shipmode  is
  (current address of yearly) + 8 + 2. It is relative to the  ups  structure,
 not the  yearly  structure.


 5.3  Records

 Records are similar to structures, except that fields in records are bit
 strings. Each bit field in a record variable can be used separately in
 constant operands or expressions. The processor cannot access bits
 individually at run time, but it can access bit fields with instructions
 that manipulate bits.

 Record fields are bits, not bytes or words.

 Records are bytes, words, or doublewords in which the individual bits or
 groups of bits are considered fields. In general, the three steps for using
 record variables are the same as those for other complex data types:


   1.  Declare a record type.

   2.  Define one or more variables having the record type.

   3.  Reference record variables using shifts and masks.


 Once defined, the record variable can be used as an operand in assembler
 statements.

 This section explains the record declaration syntax and the use of the MASK
 and WIDTH operators. It also shows a few applications of record variables
 and constants.


 5.3.1  Declaring Record Types

 A record type creates a template for data with the sizes and, optionally,
 the initial values for bit fields in the record, but it does not allocate
 memory space for the record.

 The RECORD directive declares a record type for an 8-bit, 16-bit, or 32-bit
 record that contains one or more bit fields. The maximum size is based on
 the expression word size. See OPTION EXPR16 and OPTION EXPR32 in Section
 1.3.2. The syntax is

   recordname RECORD field [[,field]]...

 The field declares the name, width, and initial value for the field. The
 syntax for each field is:

   fieldname:width[[=expression]]

 Global labels, macro names, and record field names must all be unique, but
 record field names can have the same names as structure field names or
 global labels. Width is the number of bits in the field, and expression is a
 constant giving the initial (or default) value for the field. Record
 definitions can span more than one line if the continued lines end with
 commas.

 If expression is given, it declares the initial value for the field. The
 assembler generates an error message if an initial value is too large for
 the width of its field.

 The assembler shifts bits in a record to the right if all bits are not used.


 The first field in the declaration always goes into the most significant
 bits of the record. Subsequent fields are placed to the right in the
 succeeding bits. If the fields do not total exactly 8, 16, or 32 bits as
 appropriate, the entire record is shifted right, so the last bit of the last
 field is the lowest bit of the record. Unused bits in the high end of the
 record are initialized to 0.

 The following example creates a byte record type  color  having four fields:
  blink,  back,  intense, and  fore. The contents of the record type are
 shown after the example. Since no initial values are given, all bits are set
 to 0. Note that this is only a template maintained by the assembler. No data
 is created.

   COLOR   RECORD  blink:1, back:3, intense:1, fore:3

 (This figure may be found in the printed book.)

 The next example creates a record type  cw  having six fields. Each record
 declared with this type occupies 16 bits of memory. Initial (default) values
 are given for each field. They can be used when data is declared for the
 record. The bit diagram after the example shows the contents of the record
 type.

   CW      RECORD  r1:3=0, ic:1=0, rc:2=0, pc:2=3, r2:2=1, masks:6=63

 (This figure may be found in the printed book.)


 5.3.2  Defining Record Variables

 Once you have declared a record type, you can define record variables of
 that type. For each variable, memory is allocated to the object file in the
 format declared by the type. The syntax is

   [[name]] recordname <[[initializer
   [[,initializer]]...]] > <$IAngle
   brackets (<< \ra);records>

   [[name]] recordname {
   [[initializer [[,initializer]]...]]
   }

   [[name]] recordname constant
    DUP ( [[initializer [[,initializer]]...]]
   )

 The recordname is the name of a record type that was previously declared by
 using the RECORD directive.

 A fieldlist for each field in the record can be a list of integers,
 character constants, or expressions that correspond to a value compatible
 with the size of the field. Curly braces or angle brackets are required even
 if no initial value is given.

 If you use the DUP operator (see Section 5.1.1, "Declaring and Referencing
 Arrays") to initialize multiple record variables, only the angle brackets
 and initial values, if given, need to be enclosed in parentheses. For
 example, you can define an array of record variables with

   xmas    COLOR   50 DUP ( <1, 2, 0, 4> )

 You do not have to initialize all fields in a record. If an initial value is
 blank, the assembler automatically stores the default initial value of the
 field. If there is no default value, the assembler clears each bit in the
 field.

 The definition in the example below creates a variable named  warning  whose
 type is given by the record type  color. The initial values of the fields in
 the

 variable are set to the values given in the record definition. The initial
 values override any default record values, had any been given in the
 declaration.

   COLOR   RECORD  blink:1,back:3,intense:1,fore:3 ; Record
                                                   ;  declaration
   warning COLOR   <1, 0, 1, 4>                    ; Record
                                                   ;  definition

 (This figure may be found in the printed book.)


 LENGTHOF, SIZEOF, and TYPE with Records

 The SIZEOF and TYPE operators applied to a record name return the number of
 bytes used by the record. SIZEOF for a record variable returns the number of
 bytes used by the variable. You cannot use LENGTHOF with record types, but
 you can with the variables of that type. LENGTHOF returns the number of
 items in an initializer. The record can be used as an operand. The value of
 the operand is a bit mask of the defined record. This example illustrates
 these points.

   ; Record definition
   ; 9 bits stored in 2 bytes
   RGBCOLOR        RECORD  red:3,  green:3,  blue:3

           mov     ax, RGBCOLOR          ; Equivalent to "mov ax,
                                         ;  01FFh"
   ;       mov     ax, LENGTHOF RGBCOLOR ; Illegal since LENGTHOF can
                                         ;  apply only to data label
           mov     ax, SIZEOF   RGBCOLOR ; Equivalent to "mov ax, 2"

           mov     ax, TYPE     RGBCOLOR ; Equivalent to "mov ax, 2"


   ; Record instance
   ; 8 bits stored in 1 byte
   RGBCOLOR2       RECORD red:3, green:3, blue:2
   rgb             RGBCOLOR2 <1, 1, 1>     ; Initialize to 025h

           mov     ax, RGBCOLOR2           ; Equivalent to "mov ax,
                                           ;  00FFhh"
           mov     ax, LENGTHOF rgb        ; Equivalent to "mov ax,
   1"
           mov     ax, SIZEOF   rgb        ; Equivalent to "mov ax,
   1"
           mov     ax, TYPE     rgb        ; Equivalent to "mov ax,
   1"


 5.3.3  Record Operators

 The WIDTH operator (which is used only with records) returns the width in
 bits of a record or record field. The MASK operator returns a bit mask for
 the bit positions occupied by the given record field. A bit in the mask
 contains a 1 if that bit corresponds to a bit field. The example below shows
 how to use MASK and WIDTH.

   .DATA
   COLOR           RECORD  blink:1, back:3, intense:1, fore:3
   message         COLOR   <1, 5, 1, 1>
   wblink  EQU     WIDTH blink         ; "wblink"     = 1
   wback   EQU     WIDTH back          ; "wback"      = 3
   wintense EQU    WIDTH intense       ; "wintense"   = 1
   wfore   EQU     WIDTH fore          ; "wfore"      = 3
   wcolor  EQU     WIDTH color         ; "wcolor"     = 8
           .CODE
           .
           .
           .
           mov     ah, message         ; Load initial   0101 1001
           and     ah, NOT MASK back   ; Turn off   AND 1000 1111
                                       ; "back"         ---------
                                       ;                0000 1001
           or      ah, MASK blink      ; Turn on     OR 1000 0000
                                       ; "blink"        ---------
                                       ;                1000 1001
           xor     ah, MASK intense    ; Toggle     XOR 0000 1000
                                       ; "intense"      ---------
                                       ;                1000 0001
           .
           IF      (WIDTH color) GE 8  ; If color is 16 bit, load
           mov     ax, message         ;  into 16-bit register
           ELSE                        ; else
           mov     al, message         ;  load into low 8-bit register
           xor     ah, ah              ;  and clear high 8-bits
           ENDIF

 This example illustrates several ways in which record fields can be used as
 operands and in expressions.

   ; Rotate "back" of "cursor" without changing other
   values

               mov     al, cursor       ; Load value from memory
               mov     ah, al           ; Save a copy for work      1101
   1001=ah/al
               and     al, NOT MASK back; Mask out old bits     AND
   1000 1111=mask
                                        ;  to save old cursor       ---------
                                        ;                           1000
   1001=al
               mov     cl, back         ; Load bit position
               shr     ah, cl           ; Shift to right            0000
   1101=ah
               inc     ah               ; Increment                 0000
   1110=ah

               shl     ah, cl           ; Shift left again          1110
   0000=ah
               and     ah, MASK back    ; Mask off extra bits   AND
   0111 0000=mask
                                        ;  to get new cursor        ---------
                                        ;                           0110
   0000 ah
               or      ah, al           ; Combine old and new    OR
   1000 1001 al
                                        ;                           ---------
               mov     cursor, ah       ; Write back to memory      1110
   1001 ah

 Record variables are often used with the logical operators to perform
 logical operations on the bit fields of the record, as in the previous
 example using the MASK operator.


 5.4  Related Topics in Online Help

 In addition to information on all the instructions and directives mentioned
 in this chapter, information on the following topics can be found in online
 help, starting at the "MASM 6.0 Contents" screen:

 Topic                             Access
 ────────────────────────────────────────────────────────────────────────────
 INS, OUTS                         Choose "Processor Instructions" and then
                                   "System and I/O Access"

 LABEL                             Choose "Directives" and then "Code
                                   Labels"

 RECORD, UNION, STRUCT, MASK, ORG   Choose "Directives" and then choose
 , WIDTH, and ALIGN                "Complex Data Types"

 SHRD, SHLD, BSF, and BSR          From "Processor Instructions," choose
                                   "Logical and Shifts"

 BOUND                             From "Processor Instructions," choose
                                   "Data
                                   Transfer"








 Chapter 6  Using Floating-Point and Binary Coded Decimal Numbers
 ────────────────────────────────────────────────────────────────────────────

 MASM requires different techniques for handling floating-point (real)
 numbers and binary coded decimal (BCD) numbers than for handling integers.
 You have two choices for working with real numbers─a math coprocessor or
 emulation routines.

 Math coprocessors─the 8087, 80287, and 80387 chips─work with the main
 processor to handle real-number calculations. The 80486 processor performs
 floating-point operations directly. All information in this chapter
 pertaining to the 80387 coprocessor applies to the 80486 processor as well.


 This chapter begins with a summary of the directives and formats of
 floating-point data; you need to use these to allocate memory storage and
 initialize variables before you can work with floating-point numbers.

 The chapter then explains how to use a math coprocessor for floating-point
 operations. It covers these areas:


   ■   The architecture of the registers

   ■   The operands for the coprocessor instruction formats

   ■   The coordination of coprocessor and main processor memory access

   ■   The basic groups of coprocessor instructions─for loading and storing
       data, doing arithmetic calculations, and controlling program flow


 The next main section describes emulation libraries. With the emulation
 routines provided with all Microsoft high-level languages, you can use
 coprocessor instructions as though your computer had a math coprocessor.
 However, some coprocessor instructions are not handled by emulation, as this
 section explains.

 Finally, because math coprocessor and emulation routines can also operate on
 BCD numbers, this chapter discusses the instruction set for these numbers.


 6.1  Using Floating-Point Numbers

 Before using floating-point data in your program, you need to allocate the
 memory storage for the data. You can then initialize variables either as
 real numbers in decimal form or as encoded hexadecimals. The assembler
 stores allocated data in 10-byte IEEE format. This section looks at
 floating-point declarations and floating-point data formats.


 6.1.1  Declaring Floating-Point Variables and Constants

 You can allocate real constants using the REAL4, REAL8, and REAL10
 directives. The list below shows the size of the floating-point number each
 of these directives allocates.

 Directive       Size
 ────────────────────────────────────────────────────────────────────────────
 REAL4           Short (32-bit) real numbers
 REAL8           Long (64-bit) real numbers
 REAL10          10-byte (80-bit) real numbers and BCD numbers

 The possible ranges for floating-point variables are given in Table 6.1.

 Table   6.1 Ranges of Floating-Point Variables

                      Significant
 Data Type     Bits   Digits             Approximate Range
 ────────────────────────────────────────────────────────────────────────────
 Short real    32     6-7                ±1.18 x 10-38 to ±3.40 x 10(38)

 Long real     64     15-16              ±2.23 x 10-308 to ±1.79 x 10(308)

 10-byte real  80     19                 ±3.37 x 10-4932 to ±1.18 x 10
                                         (4932)

 ────────────────────────────────────────────────────────────────────────────


 With previous versions of MASM, the DD, DQ, and DT directives could be used
 to allocate real constants. These directives are still supported by MASM
 6.0, but this means that the variables are integers rather than
 floating-point values. Although this makes no difference in the assembly
 code, CodeView displays the values incorrectly.

 There are two forms for specifying floatingpoint numbers.

 You can specify floating-point constants either as decimal constants or as
 encoded hexadecimal constants. You can express decimal real-number constants
 in the form

   [[+ | -]] integer.[[fraction]][[E]][[[[+
   | -]]exponent]]

 For example, the numbers  2.523E1  and  -3.6E-2  are written in the correct
 decimal format. These numbers can be used as initializers for real-number
 variables.

 Digits of real numbers are always evaluated as base 10. During assembly, the
 assembler converts real-number constants given in decimal format to a binary
 format. The sign, exponent, and mantissa of the real number are encoded as
 bit fields within the number.

 You can also specify the encoded format directly with hexadecimal digits
 (0-9 plus A-F). The number must begin with a decimal digit (0-9) and a
 leading zero if necessary, and end with the real-number designator (R). It
 cannot be signed.

 For example, the hexadecimal number  3F800000r  can be used as an
 initializer for a doubleword-sized variable.

 The maximum range of exponent values and the number of digits required in
 the hexadecimal number depend on the directive. The number of digits for
 encoded numbers used with REAL4, REAL8, and REAL10 must be 8, 16, and 20
 digits, respectively. If the number has a leading zero, the number must be
 9, 17, or 21 digits.

 Examples of decimal constant and hexadecimal specifications are shown here:


   ; Real numbers
   short   REAL4    25.23              ; IEEE format
   double  REAL8    2.523E1            ; IEEE format
   tenbyte REAL10   2523.0E-2          ; 10-byte real format

   ; Encoded as hexadecimals
   ieeeshort       REAL4    3F800000r             ; 1.0 as IEEE short
   ieeedouble      REAL8    3FF0000000000000r     ; 1.0 as IEEE long
   temporary       REAL10   3FFF8000000000000000r ; 1.0 as 10-byte
                                                  ;  real

 Section 6.1.2, "Storing Numbers in Floating-Point Format," explains the IEEE
 formats--the way the assembler actually stores the data.

 Pascal or C programmers may prefer to create language-specific TYPEDEF
 declarations, as illustrated in this example:

   ; C-language specific
   float           TYPEDEF REAL4
   double          TYPEDEF REAL8
   long_double     TYPEDEF REAL10
   ; Pascal-language specific
   SINGLE          TYPEDEF REAL4
   DOUBLE          TYPEDEF REAL8
   EXTENDED        TYPEDEF REAL10

 For applications of TYPEDEF other than aliasing, see Section 3.3.1,
 "Defining Pointer Types with TYPEDEF."


 6.1.2  Storing Numbers in Floating-Point Format

 The assembler stores real numbers in the IEEE format.

 The assembler stores the floating-point variables in the IEEE format. MASM
 6.0 does not support .MSFLOAT and Microsoft binary format, which are
 available in previous versions.

 Figure 6.1 illustrates the IEEE format for encoding short (four-byte), long
 (eight-byte), and 10-byte real numbers. Although this figure places the
 most-significant bit first for illustration, low bytes actually appear first
 in memory.

 (This figure may be found in the printed book.)

 This is how the parts of a real number are stored in the IEEE format:


   1.  Sign bit (0 for positive or 1 for negative) in the upper bit of the
       first byte.

   2.  Exponent in the next bits in sequence (8 bits for a short real number,
       11 bits for a long real number, and 15 bits for a 10-byte real
       number).

   3.  Mantissa in the remaining bits. The first bit is always assumed to be
       1. The length is 23 bits for short real numbers, 52 bits for long real
       numbers, and 63 bits for 10-byte reals.


 The exponent field represents a multiplier 2n. To accommodate negative
 exponents (such as 2-6), the value in the exponent field is biased; that is,
 the actual exponent is determined by subtracting the appropriate bias value
 from the value in the exponent field. For example, the bias for short reals
 is 127. If the value in the exponent field is 130, the exponent represents a
 value of 2130-127, or 23. The bias for long reals is 1,023. The bias for
 10-byte reals is 16,383.

 Notice that the 10-byte real format stores the integer part of the mantissa.
 This differs from the 4-byte and 8-byte formats, in which the integer part
 is implicit.

 Once you have declared floating-point data for your program, you can use
 coprocessor or emulator instructions to access the data. The next section
 focuses on the coprocessor architecture, instructions, and operands required
 for floating-point operations.


 6.2  Using a Math Coprocessor

 When used with real numbers, packed BCD numbers, or long integers,
 coprocessors (the 8087, 80287, 80387, and 80486) calculate many times faster
 than the 8086-based processors. The coprocessor handles data with its own
 registers. The organization of these registers reflects four possible
 formats for using operands (as explained in Section 6.2.2, "Instruction and
 Operand Formats").

 This section also describes how the coprocessor performs various tasks:
 transferring data to and from the coprocessor, coordinating processor and
 coprocessor operations, and controlling program flow.


 6.2.1  Coprocessor Architecture

 The coprocessor accesses memory as the CPU does, but it has its own data and
 control registers--eight data registers organized as a stack and seven
 control registers similar to the 8086 flag registers. The coprocessor's
 instruction set provides direct access to these registers.

 The eight coprocessor data registers form a stack.

 The eight 80-bit data registers of the 8087-based coprocessors are organized
 as a stack although they need not be used as a stack. As data items are
 pushed into the top register, previous data items move into higher-numbered
 registers, which are lower on the stack. Register 0 is the top of the stack;
 register 7 is the bottom. The syntax for specifying registers is shown
 below:

   ST «(number)»

 The number must be a digit between 0 and 7 or a constant expression that
 evaluates to a number from 0 to 7. ST is another way to refer to ST(0).

 All coprocessor data is stored in registers in the 10-byte real format. The
 registers and the register format are shown in Figure 6.2.

 (This figure may be found in the printed book.)

 Internally, all calculations are done on numbers of the same type. Since
 10-byte real numbers have the greatest precision, lower-precision numbers
 are guaranteed not to lose precision as a result of calculations. The
 instructions that transfer values between the main memory and the
 coprocessor automatically convert numbers to and from the 10-byte real
 format.


 6.2.2  Instruction and Operand Formats

 Because of the stack organization of registers, you can consider registers
 either as elements on a stack or as registers much like 8086-family
 registers. Table 6.2 lists the four main groups of coprocessor instructions
 and the general syntax for each. The names given to the instruction format
 reflect the way the instruction uses the coprocessor registers. The
 instruction operands are placed in the coprocessor data registers before the
 instruction executes.

 Table 6.2  Coprocessor Operand Formats

 Instruction                           Implied Operands
 Format             Syntax                               Example
 ────────────────────────────────────────────────────────────────────────────
 Classical stack    Faction            ST, ST(1)         fadd

 Memory             Faction memory     ST                fadd memloc

 Register           Faction  ST(num),  ─                 fadd st(5), st
                    ST                                   fadd st, st(3)
                    Faction  ST, ST(
                    num)

 Register pop       FactionP  ST(num    ─                 faddp st(4), st
                    ), ST

 ────────────────────────────────────────────────────────────────────────────


 All coprocessor instructions begin with F.

 You can easily recognize coprocessor instructions because, unlike all
 8086-family instruction mnemonics, they start with the letter F. Coprocessor
 instructions can never have immediate operands and, with the exception of
 the FSTSW instruction, they cannot have processor registers as operands.


 6.2.2.1  Classical-Stack Format

 Instructions in the classical-stack format treat the coprocessor registers
 like items on a stack─thus its name. Items are pushed onto or popped off the
 top elements of the stack. Since only the top item can be accessed on a
 traditional stack, there is no need to specify operands. The first (top)
 register (and the second if the instruction needs two operands) is always
 assumed.

 In coprocessor arithmetic operations, the top of the stack (ST) is the
 source operand and the second register [ST(1)] is the destination. The
 result of the operation goes into the destination operand, and the source is
 popped off the stack. The result is left at the top of the stack.

 Instructions that load constants are one example of instructions that
 require the classical-stack format. In this case, the constant created by
 the instruction is the implied source, and the top of the stack is the
 destination.

 This example illustrates the classical-stack format, and Figure 6.3 shows
 the status of the register stack after each instruction:

   fld1               ; Push 1 into first position
               fldpi              ; Push pi into first position
               fadd               ; Add pi and 1 and pop

 (This figure may be found in the printed book.)


 6.2.2.2  Memory Format

 Instructions using the memory format, such as data transfer instructions,
 also treat coprocessor registers like items on a stack. However, with this
 format, items are pushed from memory onto the top element of the stack or
 popped from the top element to memory. You must specify the memory operand.


 Some coprocessor instructions operate on integers or BCDs.

 Some instructions that use the memory format specify how a memory operand is
 to be interpreted─as an integer (I) or as a binary coded decimal (B). The
 letter I or B follows the initial F in the syntax. For example, FILD
 interprets its operand as an integer and FBLD interprets its operand as a
 BCD number. If the instruction name does not include a type letter, the
 instruction works on real numbers.

 You can also use memory operands in calculation instructions that operate on
 two values (see Section 6.2.4, "Using Coprocessor Instructions"). The memory
 operand is always the source. The stack top (ST) is always the implied
 destination. The result of the operation replaces the destination without
 changing its stack position, as shown in this example and Figure 6.4:

   .DATA
   m1       REAL4  1.0
   m2       REAL4  2.0
            .CODE
            .
            .
            .
            fld    m1      ; Push m1 into first position
            fld    m2      ; Push m2 into first position
            fadd   m1      ; Add m2 to first position
            fstp   m1      ; Pop first position into m1
            fst    m2      ; Copy first position to m2

 (This figure may be found in the printed book.)


 6.2.2.3  Register Format

 Instructions using the register format treat coprocessor registers as
 registers rather than as stack elements. Instructions that use this format
 require two register operands; one of them must be the stack top (ST).

 In the register format, specify all operands by name. The first operand is
 the destination; its value is replaced with the result of the operation. The
 second operand is the source; it is not affected by the operation. The stack
 position of the operands does not change.

 The only instructions using the register operand format are the FXCH
 instruction and the arithmetic instructions that do calculations on two
 values. With the FXCH instruction, the stack top is implied and need not be
 specified, as shown in this example and Figure 6.5:

   fadd    st(1), st   ; Add second position to first -
                               ;  result goes in second position
           fadd    st, st(2)   ; Add first position to third -
                               ;  result goes in first position
           fxch    st(1)       ; Exchange first and second positions

 (This figure may be found in the printed book.)


 6.2.2.4  Register-Pop Format

 The register-pop format treats coprocessor registers as a modified stack.
 The source register must always be the stack top. Specify the destination
 with the register's name.

 Instructions with this format place the result of the operation into the
 destination operand, and the stack top pops off the stack. The effect is
 that both values being operated on are lost and the result of the operation
 is saved in the specified destination register. The register-pop format is
 used only for instructions that do calculations on two values, as in this
 example and Figure 6.6:

   faddp   st(2), st ; Add first and third positions and
   pop -
                             ;  first position destroyed;
                             ;  third moves to second and holds result

 (This figure may be found in the printed book.)


 6.2.3  Coordinating Memory Access

 The math coprocessor works simultaneously with the main processor. However,
 since the coprocessor cannot handle device input or output, data originates
 in the main processor.

 The processor and coprocessor exchange data through memory.

 The main processor and the coprocessor have their own registers, which are
 completely separate and inaccessible to each other. They usually exchange
 data through memory, since memory is available to both.

 When using the coprocessor, follow these three steps:


   1.  Load data from memory to coprocessor registers.

   2.  Process the data.

   3.  Store the data from coprocessor registers back to memory.


 Step 2, processing the data, can occur while the main processor is handling
 other tasks. Steps 1 and 3 must be coordinated with the main processor so
 that the processor and coprocessor do not try to access the same memory at
 the same time; otherwise, problems of coordinating memory access can occur.
 Since the processor and coprocessor work independently, they may not finish
 working on memory in the order in which you give instructions. Two potential
 timing conflicts can occur; they are handled in different ways.

 One timing conflict results if a coprocessor instruction follows a processor
 instruction. The processor may have to wait until the coprocessor finishes
 if the next processor instruction requires the result of the coprocessor's
 calculation. You do not have to write your code to avoid this conflict,
 however. The assembler coordinates this timing automatically for the 8088
 and 8086 processors, and the processor coordinates it automatically on the
 80186-80486 processors. This is the first case shown in the example later in
 this section.

 Another conflict results if a processor instruction that accesses memory
 follows a coprocessor instruction that accesses the same memory. The
 processor can try to load a variable that is still being used by the
 coprocessor. You need careful synchronization to control the timing, and
 this synchronization is not automatic on the 8087 coprocessor. For code to
 run correctly on the 8087, you must include the WAIT or FWAIT instruction
 (they are mnemonics for the same instruction) to ensure that the coprocessor
 finishes before the processor begins, as shown in the second example. In
 this situation, the processor does not generate the FWAIT instruction
 automatically.

   ; Processor instruction first - No wait needed
           mov     WORD PTR mem32[0], ax   ; Load memory
           mov     WORD PTR mem32[2], dx
           fild    mem32                   ; Load to register

   ; Coprocessor instruction first - Wait needed (for 8087)
           fist    mem32                   ; Store to memory
           fwait                           ; Wait until coprocessor
                                           ;  is done
           mov     ax, WORD PTR mem32[0]   ; Move to register
           mov     dx, WORD PTR mem32[2]

 When generating code for the 8087 coprocessor, the assembler automatically
 inserts a WAIT instruction before the coprocessor instruction. However, if
 you use the .286 or .386 directive, the compiler assumes that the
 coprocessor instructions are for the 80287 or 80387 and does not insert the
 WAIT instruction.

 If your code does not need to run on an 8086 or 8088 processor, you can make
 your programs shorter and more efficient by using the .286 or .386
 directive.


 6.2.4  Using Coprocessor Instructions

 The 8087 family of coprocessors has separate instructions for each of the
 following operations:


   ■   Loading and storing data

   ■   Doing arithmetic calculations

   ■   Controlling program flow


 The following sections explain the available instructions and show how to
 use them for each of the operations listed above. See Section 6.2.2,
 "Instruction and Operand Formats," for general syntax information.


 6.2.4.1  Loading and Storing Data

 Data-transfer instructions transfer data between main memory and the
 coprocessor registers or between different coprocessor registers. Two
 principles govern data transfers:


   ■   The choice of instruction determines whether a value in memory is
       considered an integer, a BCD number, or a real number. The value is
       always considered a 10-byte real number once it is transferred to the
       coprocessor.

   ■   The size of the operand determines the size of a value in memory.
       Values in the coprocessor always take up 10 bytes.


 Load commands transfer data, and store commands remove data.

 You can transfer data to stack registers using load commands. These commands
 push data onto the stack from memory or from coprocessor registers. Store
 commands remove data. Some store commands pop data off the register stack
 into memory or coprocessor registers; others simply copy the data without
 changing it on the stack.

 If you use constants as operands, you cannot load them directly into
 coprocessor registers. You must allocate memory and initialize a variable to
 a constant value. That variable can then be loaded by using one of the load
 instructions listed below.

 A few special instructions are provided for loading certain constants. You
 can load 0, 1, pi, and several common logarithmic values directly. Using
 these instructions is faster and often more precise than loading the values
 from initialized variables.

 All instructions that load constants have the stack top as the implied
 destination operand. The constant to be loaded is the implied source
 operand.

 The coprocessor data area, or parts of it, can also be moved to memory and
 later loaded back. You may want to do this to save the current state of the
 coprocessor before executing a procedure. After the procedure ends, restore
 the previous status. Saving coprocessor data is also useful when you want to
 modify coprocessor behavior by writing certain data to main memory,
 operating on the data with 8086-family instructions, and then loading it
 back to the coprocessor data area.

 You can use the following instructions for transferring numbers to and from
 registers:

╓┌──────────────────────┌────────────────────────────────────────────────────╖
 Instruction(s)         Description
 ────────────────────────────────────────────────────────────────────────────
 Instruction(s)         Description
 ────────────────────────────────────────────────────────────────────────────
 FLD, FST, FSTP         Loads and stores real numbers
 FILD, FIST, FISTP      Loads and stores binary integers
 FBLD                   Loads BCD
 FBSTP                  Stores BCD
 FXCH                   Exchanges register values
 FLDZ                   Pushes 0 into ST
 FLD1                   Pushes 1 into ST
 FLDPI                  Pushes the value of pi into ST
 FLDCW mem2byte         Loads the control word into the coprocessor
 F«N»STCW mem2byte      Stores the control word in memory
 FLDENV mem14byte       Loads environment from memory
 F«N»STENV mem14byte    Stores environment in memory
 FRSTOR mem94byte       Restores state from memory
 F«N»SAVE mem94byte     Saves state in memory
 FLDL2E                 Pushes the value of log2e into ST
 FLDL2T                 Pushes log210 into ST
 FLDLG2                 Pushes log102 into ST
 FLDLN2                 Pushes loge2 into ST


 The following example and Figure 6.7 illustrate some of these instructions:


   .DATA
   m1      REAL4   1.0
   m2      REAL4   2.0
           .CODE
           fld     m1      ; Push m1 into first item
           fld     st(2)   ; Push third item into first
           fst     m2      ; Copy first item to m2
           fxch    st(2)   ; Exchange first and third items
           fstp    m1      ; Pop first item into m1

 (This figure may be found in the printed book.)


 6.2.4.2  Doing Arithmetic Calculations

 Most of the coprocessor instructions for doing arithmetic operations have
 several forms, depending on the operand used. You do not need to specify the
 operand type in the instruction if both operands are stack registers, since
 register values are always 10-byte real numbers. The arithmetic instructions
 are listed below. In most cases, the result replaces the destination
 register.

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Instruction                       Description
 ────────────────────────────────────────────────────────────────────────────
 FADD                              Adds the source and destination

 FSUB                              Subtracts the source from the
                                   destination

 FSUBR                             Subtracts the destination from the
                                   source

 FMUL                              Multiplies the source and the
                                   destination

 FDIV                              Divides the destination by the source

 Instruction                       Description
 ────────────────────────────────────────────────────────────────────────────

 FDIVR                             Divides the source by the destination

 FABS                              Sets the sign of ST to positive

 FCHS                              Reverses the sign of ST

 FRNDINT                           Rounds ST to an integer

 FSQRT                             Replaces the contents of ST with its
                                   square root

 FSCALE                            Multiplies the stack-top value by 2 to
                                   the power contained in ST(1)

 FPREM                             Calculates the remainder of ST divided
                                   by ST(1)




 80387 Only

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Instruction                       Description
 ────────────────────────────────────────────────────────────────────────────
 FSIN                              Calculates the sine of the value in ST

 FCOS                              Calculates the cosine of the value in ST

 FSINCOS                           Calculates the sine and cosine of the
                                   value in ST

 FPREM1                            Calculates the partial remainder by
                                   performing modulo division on the top
                                   two stack registers

 FXTRACT                           Breaks a number down into its exponent
                                   and mantissa and pushes the mantissa
                                   onto the register stack
 Instruction                       Description
 ────────────────────────────────────────────────────────────────────────────
                                  onto the register stack

 F2XM1                             Calculates 2(x)-1

 FYL2X                             Calculates Y * log2 X

 FYL2XP1                           Calculates Y * log2 (X+1)

 FPTAN                             Calculates the tangent of the value in
                                   ST

 FPATAN                            Calculates the arctangent of the ratio Y
                                   /X

 F«N»INIT                          Resets the coprocessor and restores all
                                   the default conditions in the control
                                   and status words

 F«N»CLEX                          Clears all exception flags and the busy
 Instruction                       Description
 ────────────────────────────────────────────────────────────────────────────
F«N»CLEX                          Clears all exception flags and the busy
                                   flag of the status word

 FINCSTP                           Adds 1 to the stack pointer in the
                                   status word

 FDECSTP                           Subtracts 1 from the stack pointer in
                                   the status word

 FFREE                             Marks the specified register as empty



 The following example illustrating several arithmetic instructions solves
 quadratic equations. It does no error checking and fails for some values
 because it attempts to find the square root of a negative number. You could
 revise the code using the FTST (Test for Zero) instruction to check for a
 negative number or 0 before the square root is calculated. If  b2 - 4ac  is
 negative or 0, the code can jump to routines that handle these two special
 cases.

   .DATA
   a       REAL4   3.0
   b       REAL4   7.0
   cc      REAL4   2.0
   posx    REAL4   0.0
   negx    REAL4   0.0

           .CODE
           .
           .
           .
   ; Solve quadratic equation - no error checking
   ; The formula is: -b +/- squareroot(b2 - 4ac) / (2a)
           fld1               ; Get constants 2 and 4
           fadd    st,st      ; 2 at bottom
           fld     st         ; Copy it
           fmul    a          ; = 2a

           fmul    st(1),st   ; = 4a
           fxch               ; Exchange
           fmul    cc         ; = 4ac

           fld     b          ; Load b
           fmul    st,st      ; = b2
           fsubr              ; = b2 - 4ac
                              ; Negative value here produces error
           fsqrt              ; = square root(b2 - 4ac)
           fld     b          ; Load b
           fchs               ; Make it negative
           fxch               ; Exchange

           fld     st         ; Copy square root
           fadd    st,st(2)   ; Plus version = -b + root(b2 -
   4ac)
           fxch               ; Exchange
           fsubp   st(2),st   ; Minus version = -b - root(b2 -
   4ac)

           fdiv    st,st(2)   ; Divide plus version
           fstp    posx       ; Store it
           fdivr              ; Divide minus version
           fstp    negx       ; Store it

 The examples in online help contain an enhanced version of this procedure.


 6.2.4.3  Controlling Program Flow

 The math coprocessors have several instructions that set control flags in
 the status word. The 8087-family control flags can be used with conditional
 jumps to direct program flow in the same way that 8086-family flags are
 used. Since the coprocessor does not have jump instructions, you must
 transfer the status word to memory so that the flags can be used by
 8086-family instructions.

 An easy way to use the status word with conditional jumps is to move its
 upper byte into the lower byte of the processor flags, as shown in this
 example:

   fstsw   mem16      ; Store status word in memory
           fwait              ; Make sure coprocessor is done
           mov     ax, mem16  ; Move to AX
           sahf               ; Store upper word in flags

 The SAHF (Store AH into Flags) instruction in the example above transfers AH
 into the low bits of the flags register.

 You can save several steps by loading the status word directly to AX on the
 80287 with the FSTSW and FNSTSW instructions. This is the only case in which
 data can be transferred directly between processor and coprocessor
 registers, as shown in this example:

   fstsw   ax

 The coprocessor control flags and their relationship to the status word are
 described in Section 6.2.4.4, "Control Registers."

 The 8087-family coprocessors provide several instructions for comparing
 operands and testing control flags. All these instructions compare the stack
 top (ST) to a source operand, which may either be specified or implied as
 ST(1).

 The compare instructions affect the C3, C2, and C0 control flags, but not
 the C1 flag. Table 6.3 shows the flags set for each possible result of a
 comparison or test.

 Table 6.3  Control-Flag Settings after Comparison or Test

 After FCOM        After FTEST                         C3  C2  C0
 ────────────────────────────────────────────────────────────────────────────
 ST > source       ST is positive                      0   0   0
 ST < source       ST is negative                      0   0   1
 ST = source       ST is 0                             1   0   0
 Not comparable    ST is NAN or  projective infinity   1   1   1
 ────────────────────────────────────────────────────────────────────────────

 Variations on the compare instructions allow you to pop the stack once or
 twice and to compare integers and zero. For each instruction, the stack top
 is always the implied destination operand. If you do not give an operand,
 ST(1) is the implied source. With some compare instructions, you can specify
 the source as a memory or register operand.

 All instructions summarized in the following list have implied operands:
 either ST as a single-destination operand or ST as the destination and ST(1)
 as the source. These are the instructions for comparing and testing flags.

 Some instructions have a wait version and a no-wait version. The no-wait
 versions have N as the second letter.

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Instruction                       Description
 ────────────────────────────────────────────────────────────────────────────
 FCOM                              Compares the stack top to the source.
                                   The
                                   source and destination are unaffected by
                                   the comparison.

 FTST                              Compares ST to 0.

 FCOMP                             Compares the stack top to the source and
                                   then pops the stack.

 FUCOM, FUCOMP, FUCOMPP            Compare the source to ST and set the
                                   condition codes of the status word
 Instruction                       Description
 ────────────────────────────────────────────────────────────────────────────
                                  condition codes of the status word
                                   according to the result (80386/486 only).

 F«N»STSW mem2byte                 Stores the status word in memory.

 FXAM                              Sets the value of the control flags
                                   based on the type of the number in ST.

 FPREM                             Finds a correct remainder for large
                                   operands. It uses the C2 flag to
                                   indicate whether the remainder returned
                                   is partial (C2 is set) or complete (C2
                                   is clear). (If the bit is set, the
                                   operation should be repeated. It also
                                   returns the least-significant three bits
                                   of the quotient in C0, C3, and C1.)

 FNOP                              Copies the stack top onto itself, thus
                                   padding the executable file and taking
 Instruction                       Description
 ────────────────────────────────────────────────────────────────────────────
                                  padding the executable file and taking
                                   up processing time without having any
                                   effect on registers or memory.

 FDISI, FNDISI, FENI, FNENI        Enables or disables interrupts (8087
                                   only).

 FSETPM                            Sets protected mode. Requires a .286P or
                                   .386P directive (80287, 80387, and 80486
                                   only).



 The following example illustrates some of these instructions. Notice how
 conditional blocks are used to enhance 80287 code.

   .DATA
   down    REAL4   10.35      ; Sides of a rectangle
   across  REAL4   13.07
   diamtr  REAL4   12.93      ; Diameter of a circle
   status  WORD    ?
   P287    EQU     (@Cpu AND 00111y)
           .CODE
           .
           .
           .
   ; Get area of rectangle
           fld     across     ; Load one side
           fmul    down       ; Multiply by the other

   ; Get area of circle:  Area = PI * (D/2)2
           fld1               ; Load one and
           fadd    st, st     ;  double it to get constant 2
           fdivr   diamtr     ; Divide diameter to get radius
           fmul    st, st     ; Square radius
           fldpi              ; Load pi
           fmul               ; Multiply it

   ; Compare area of circle and rectangle
           fcompp             ; Compare and throw both away
           IF      p287
           fstsw   ax         ; (For 287+, skip memory)
           ELSE
           fnstsw  status     ; Load from coprocessor to memory
           mov     ax, status ; Transfer memory to register
           ENDIF
           sahf               ; Transfer AH to flags register
           jp      nocomp     ; If parity set, can't compare
           jz      same       ; If zero set, they're the same
           jc      rectangle  ; If carry set, rectangle is bigger
           jmp     circle     ;  else circle is bigger

   nocomp:                    ; Error handler
           .
           .
           .
   same:                      ; Both equal
           .
           .
           .
   rectangle:                 ; Rectangle bigger
           .
           .
           .
   circle:                    ; Circle bigger

 Additional instructions for the 80387/486 are FLDENVD and FLDENVW for
 loading the environment; FNSTENVD, FNSTENVW, FSTENVD, and FSTENVW for
 storing the environment state; FNSAVED, FNSAVEW, FSAVED, and FSAVEW for
 saving the coprocessor state; and FRSTORD and FRSTORW for restoring the
 coprocessor state.

 The size of the code segment, not the operand size, determines the number of
 bytes loaded or stored with these instructions. The instructions ending with
 W store the 16-bit form of the control register data, and the instructions
 ending with D store the 32-bit form. For example, in 16-bit mode FSAVEW
 saves the 16-bit control register data. If you need to store the 32-bit form
 of the control register data, use FSAVED.


 6.2.4.4  Control Registers

 Some of the flags of the seven 16-bit control registers control coprocessor
 operations, while others maintain the current status of the coprocessor. In
 this sense, they are much like the 8086-family flags registers (see Figure
 6.8).

 (This figure may be found in the printed book.)

 Of the control registers, only the status word register is commonly used
 (the others are used mostly by systems programmers). The format of the
 status word register is shown in Figure 6.9, which shows how the coprocessor
 control flags align with the processor flags. C3 overwrites the zero flag,
 C2 overwrites the parity flag, and C0 overwrites the carry flag. C1
 overwrites an undefined bit, so it cannot be used directly with conditional
 jumps, although you can use the TEST instruction to check C1 in memory or in
 a register. The status word register also overwrites the sign and
 auxiliary-carry flags, so you cannot count on their being unchanged after
 the operation.

 (This figure may be found in the printed book.)


 6.3  Using Emulator Libraries

 If you do not have a math coprocessor or an 80486 processor, you can do most
 floating-point operations by writing assembly-language procedures and
 accessing the emulator from a high-level language. All Microsoft high-level
 languages come with the emulator library.

 However, you cannot use a Microsoft emulator library with stand-alone
 assembler programs, since the library depends on the high-level-language
 start-up code.

 With emulator libraries, you can use most floating-point instructions.

 To use the emulator, first write the procedure using coprocessor
 instructions. Then assemble it using the /FPi option of your compiler.
 Finally, link it with your high-level-language modules. In MASM 6.0 you can
 enter options in the Programmer's WorkBench (PWB) environment, or you can
 use the OPTION EMULATOR in your source code.

 In emulation mode, the assembler generates instructions for the linker that
 the Microsoft emulator can use. The form of the OPTION directive in the
 example below tells the assembler to use emulation mode. This option
 (introduced in Section 1.3.2) can be defined only once in a module.

   OPTION EMULATOR

 Emulator libraries do not allow for all of the coprocessor instructions. The
 following floating-point instructions are not emulated:

 (This figure may be found in the printed book.)

 The set of emulated instructions is different under OS/2 2.x. If you use a
 coprocessor instruction that is not emulated, your program generates a
 run-time error when it tries to execute the unemulated instruction.

 See Chapter 20, "Mixed-Language Programming," for information about writing
 assembly-language procedures for high-level languages.


 6.4  Using Binary Coded Decimal Numbers

 Binary coded decimal (BCD) numbers allow calculations on large numbers
 without rounding errors. The 8087-family coprocessors can do fast
 calculations with packed BCD numbers. See Section 6.4.2.2 for details. The
 8086-family processors can also do some calculations with packed BCD
 numbers, but the process is slower and more complicated. See Section 6.4.2
 for details.

 This section explains how to define BCD numbers and then how to use them in
 calculations.


 6.4.1  Defining BCD Constants and Variables

 Unpacked BCD numbers are made up of bytes containing a single decimal digit
 in the lower four bits of each byte. Packed BCD numbers are made up of bytes
 containing two decimal digits: one in the upper four bits and one in the
 lower four bits. The leftmost digit holds the sign (0 for positive, 1 for
 negative).

 Packed BCD numbers are encoded in the 8087 coprocessor's packed BCD format.
 They can be up to 18 digits long, packed two digits per byte. The assembler
 zero-pads BCDs initialized with fewer than 18 digits. Digit 20 is the sign
 bit, and digit 19 is reserved.

 The TBYTE directive allocates packed BCD constants.

 When you define an integer constant with the TBYTE directive and the current
 radix is decimal (t), the assembler interprets the number as a packed BCD
 number.

 The syntax for specifying packed BCDs is exactly the same as for other
 integers.

   pos1    TBYTE   1234567890  ; Encoded as 00000000001234567890h
   neg1    TBYTE   -1234567890 ; Encoded as 80000000001234567890h

 Unpacked BCD numbers are stored one digit to a byte, with the value in the
 lower four bits. They can be defined using the BYTE directive. For example,
 an unpacked BCD number could be defined and initialized as shown below:

   unpackedr      BYTE    1,5,8,2,5,2,9   ; Initialized to 9,252,851
   unpackedf      BYTE    9,2,5,2,8,5,1   ; Initialized to 9,252,851

 Least-significant digits can come either first or last, depending on how you
 write the calculation routines that handle the numbers.


 6.4.2  Calculating with BCDs

 When you use the processor to calculate with BCDs, the result is not correct
 unless you use the ASCII-adjust instructions to convert the result into the
 valid BCD integer.


 6.4.2.1  Unpacked BCD Numbers

 Instructions for unpacked BCDs allow accurate BCD calculations.

 To do processor arithmetic on unpacked BCD numbers, you must do the
 eight-bit arithmetic calculations on each digit separately and assign the
 result to the AL register. After each operation, use the corresponding BCD
 instruction to adjust the result. The ASCII-adjust instructions do not take
 an operand. They always work on the value in the AL register.

 When a calculation using two one-digit values produces a two-digit result,
 the AAA, AAS, AAM, and AAD instructions put the first digit in AL and the
 second in AH. If the digit in AL needs to carry to or borrow from the digit
 in AH, the instructions set the carry and auxiliary carry flags.

 These instructions get their names from Intel mnemonics that use the term
 "ASCII" to refer to unpacked BCD numbers and "decimal" to refer to packed
 BCD numbers. The four ASCII-adjust instructions for unpacked BCDs are
 described below:

 Instruction                       Description
 ────────────────────────────────────────────────────────────────────────────
 AAA                               Adjusts after an addition operation.

 AAS                               Adjusts after a subtraction operation.

 AAM                               Adjusts after a multiplication operation.
                                   Always use with MUL, not with IMUL.

 AAD                               Adjusts before a division operation.
                                   Unlike other BCD instructions, AAD
                                   converts a BCD value to a binary value
                                   before the operation. After the
                                   operation, use AAM to adjust the
                                   quotient. The remainder is lost. If you
                                   need the remainder, save it in another
                                   register before adjusting the quotient.
                                   Then move it back to AL and adjust if
                                   necessary.


 The following examples show how to use each of these instructions in BCD
 addition, subtraction, multiplication, and division.

   ; To add 9 and 3 as BCDs:
           mov     ax, 9       ; Load 9
           mov     bx, 3       ;  and 3 as unpacked BCDs
           add     al, bl      ; Add 09h and 03h to get 0Ch
           aaa                 ; Adjust 0Ch in AL to 02h,
                               ;  increment AH to 01h, set carry
                               ; Result 12 (unpacked BCD in AX)

   ; To subtract 4 from 13:
           mov     ax, 103h    ; Load 13
           mov     bx, 4       ;  and 4 as unpacked BCDs
           sub     al, bl      ; Subtract 4 from 3 to get FFh (-1)
           aas                 ; Adjust 0FFh in AL to 9,
                               ;  decrement AH to 0, set carry
                               ; Result 9 (unpacked BCD in AX)

   ; To multiply 9 times 3:
           mov     ax, 903h    ; Load 9 and 3 as unpacked BCDs
           mul     ah          ; Multiply 9 and 3 to get 1Bh
           aam                 ; Adjust 1Bh in AL
                               ;  to get 27 (unpacked BCD in AX)

   ; To divide 25 by 2:
           mov     ax, 205h    ; Load 25
           mov     bl, 2       ;  and 2 as unpacked BCDs
           aad                 ; Adjust 0205h in AX
                               ;  to get 19h in AX
           div     bl          ; Divide by 2 to get
                               ;  quotient 0Ch in AL
                               ;  remainder 1 in AH
           aam                 ; Adjust 0Ch in AL
                               ;  to 12 (unpacked BCD in AX)
                               ;  (remainder destroyed)

 If you process multidigit BCD numbers in loops, each digit is processed and
 adjusted in turn.


 6.4.2.2  Packed BCD Numbers

 Packed BCD numbers are made up of bytes containing two decimal digits: one
 in the upper four bits and one in the lower four bits. The 8086-family
 processors provide instructions for adjusting packed BCD numbers after
 addition and subtraction. You must write your own routines to adjust for
 multiplication and division.

 To do processor calculations on packed BCD numbers, you must do the
 eight-bit arithmetic calculations on each byte separately. The result should
 always be in the AL register. After each operation, use the corresponding
 BCD instruction to adjust the result. The decimal-adjust instructions do not
 take an operand. They always work on the value in the AL register.

 The 8086-family processors provide DAA (Decimal Adjust after Addition) and
 DAS (Decimal Adjust after Subtraction) for adjusting packed BCD numbers
 after addition and subtraction.

 These examples show DAA and DAS used for adding and subtracting BCDs.

   ;To add 88 and 33:
           mov     ax, 8833h   ; Load 88 and 33 as packed BCDs
           add     al, ah      ; Add 88 and 33 to get 0BBh
           daa                 ; Adjust 0BBh to 121 (packed BCD:)
                               ;  1 in carry and 21 in AL

   ;To subtract 38 from 83:
           mov     ax, 3883h   ; Load 83 and 38 as packed BCDs
           sub     al, ah      ; Subtract 38 from 83 to get 04Bh
           das                 ; Adjust 04Bh to 45 (packed BCD:)
                               ;  0 in carry and 45 in AL

 Unlike the ASCII-adjust instructions, the decimal-adjust instructions never
 affect AH. The assembler sets the auxiliary carry flag if the digit in the
 lower four bits carries to or borrows from the digit in the upper four bits,
 and it sets the carry flag if the digit in the upper four bits needs to
 carry to or borrow from another byte.

 Multidigit BCD numbers are usually processed in loops. Each byte is
 processed and adjusted in turn.


 6.5  Related Topics in Online Help

 In addition to information on the instructions and directives mentioned in
 this chapter, information on the following topics can be found in online
 help, starting from the "MASM 6.0 Contents" screen.

 Topic                             Access
 ────────────────────────────────────────────────────────────────────────────
 Control registers                 Choose "Language Overview," and then
                                   choose "Coprocessor Status Word,"
                                   "Coprocessor
                                   Control Word," or "Coprocessor
                                   Environment"

 ML options                        Choose "ML Command Line"

 Coprocessor instructions          Choose "Coprocessor Instructions"

 MATHDEMO.ASM                      Choose "Example Code" and then "Map of
                                   Demos"









 Chapter 7  Controlling Program Flow
 ────────────────────────────────────────────────────────────────────────────

 Very few programs actually execute all lines sequentially from .STARTUP to
 .EXIT. Rather, complex program logic and efficiency dictate that you control
 the flow of your program─jumping from one point to another, repeating an
 action until a condition is reached, and passing control to procedures. This
 chapter describes various means for controlling program flow and several
 features that simplify coding program-control constructs.

 The first section covers jumps from one point in the program to another. It
 explains how MASM 6.0 optimizes both unconditional and conditional jumps
 under certain circumstances, so that you do not have to specify every
 attribute. The section also describes instructions you can use to test
 conditional jumps.

 The next section describes loop and decision structures that repeat actions
 or evaluate conditions. They discuss some new MASM directives, such as
 .WHILE and .REPEAT, that generate appropriate compare, loop, and jump
 instructions for you, and the new .IF, .ELSE, and .ELSEIF directives that
 generate jump instructions.

 A number of improvements to procedure automation are covered in Section 7.3.
 These include extended functionality for PROC, a PROTO directive that lets
 you write procedure prototypes similar to those used in C, an INVOKE
 directive that automates parameter passing, and new options for the
 stack-frame setup inside procedures.

 Finally, the last section explains how to pass control to an interrupt
 routine.


 7.1  Jumps

 Jumps are the most direct method for changing program control from one
 location to another. At the processor level, jumps work by changing the
 value of the IP (Instruction Pointer) register from the address of the
 current instruction to a target address, by changing the CS register for far
 jumps, and by changing the CS register for far jumps. The many forms of the
 jump instructions handle jumps based on conditions, flags, and bit settings.


 This section first describes unconditional jumps, including the new jump
 optimization features of MASM 6.0 and the use of indirect operands to
 specify the jump's destination and to construct jump tables. The section
 then discusses conditional jumps─extending jumps, jumps based on bit or flag
 status, anonymous jumps, labels for jump targets, and decision directives
 that generate conditional jumps.


 7.1.1  Unconditional Jumps

 Jumps in assembler programs are either conditional or unconditional. The
 assembler executes conditional jumps only when the jump condition is true.
 You use the JMP instruction to jump unconditionally to a specified address.
 Its single operand contains the target address, which can be short, near, or
 far.

 Unconditional jumps are often used to skip over code that should not be
 executed, as shown in this example.

   ; Handle one case
   label1: .
           .
           .
           jmp continue
   ; Handle second case
   label2: .
           .
           .
           jmp continue
           .
           .
           .
   continue:

 The distance of the target from the jump instruction and the size of the
 operand determine the assembler's encoding of the instruction. The larger
 the distance, the more bytes the assembler uses to code the instruction. In
 previous versions of MASM, unconditional NEAR jumps sometimes generate
 inefficient code. Unspecified FAR jumps result in phase errors.


 7.1.1.1  Jump Optimizing

 Beginning with MASM 6.0, the assembler determines the smallest encoding
 possible for the direct unconditional jump. You do not specify a distance
 operator, so you do not have to determine the correct distance of the jump.
 If you do specify a distance, however, and it is too short, the assembler
 generates an error. A specified distance that is too long causes a less
 efficient jump to be generated than the assembler would generate if the
 distance had not been specified.

 MASM 6.0 optimizes jumps if the following conditions are met:


   ■   You do not specify SHORT, NEAR, FAR, NEAR16, NEAR32, FAR16, FAR32, or
       PROC as the distance of the target.

   ■   The target of the jump is not external and is in the same segment as
       the jump instruction. If the target is in a different segment (but in
       the same group), it is treated as if external.


 If these two conditions are met, MASM uses the instruction, distance, and
 size of the operand to determine how best to optimize the encoding for the
 jump. No syntax changes are necessary.

 ────────────────────────────────────────────────────────────────────────────
 NOTE
 This information about jump optimizing also applies to conditional jumps on
 the 80386/486.
 ────────────────────────────────────────────────────────────────────────────


 7.1.1.2  Indirect Operands

 Indirect operands specify a register or data memory location that holds the
 address of the jump's destination. Indirect operands differ from the
 operands of direct jumps by being a memory expression instead of an
 immediate expression. For indirect jumps, you can specify the encoding for
 the instruction by giving the size (WORD, DWORD, or FWORD) attributes for
 the operand.

 The default rules are based on the .MODEL and the default segment size.

   jmp     [bx]          ; Uses .MODEL and segment size
                                 ;  defaults
           jmp     WORD PTR [bx] ; A NEAR16 indirect call

 If the indirect operand is a register, the jump is always a NEAR16 jump for
 a 16-bit register, and FAR32 for a 32-bit register:

   jmp     bx            ; NEAR16 jump
           jmp     ebx           ; FAR32  jump

 A DWORD indirect operand, however, is an ambiguous case:

   jmp     DWORD PTR [var] ; A NEAR32 jump in a 32-bit
   segment;
                                   ;  a FAR16 jump in a 16-bit segment

 In this case, you must define a type with TYPEDEF to specify the indirect
 operand.

   NFP     TYPEDEF PTR NEAR32
   FFP     TYPEDEF PTR FAR16
           jmp     NFP PTR [var] ; NEAR32 indirect jump
           jmp     FFP PTR [var] ; FAR16  indirect jump

 You can use an unconditional jump as a form of conditional jump by
 specifying the address in a register or indirect memory operand. Also, you
 can use indirect memory operands to construct jump tables that work like C
 switch statements,

 Pascal CASE statements, or Basic ON GOTO, ON GOSUB, or SELECT CASE
 statements, as shown in this example:

   NPVOID  TYPEDEF NEAR PTR VOID
           .DATA
   ctl_tbl NPVOID  extended,    ; Null key (extended code)
                   ctrla,       ; Address of CONTROL-A key routine
                   ctrlb        ; Address of CONTROL-B key routine
           .CODE
           .
           .
           .
           mov     ah, 8h       ; Get a key
           int     21h
           cbw                  ; Stretch AL into AX
           mov     bx, ax       ; Copy
           shl     bx, 1        ; Convert to address
           jmp     ctl_tbl[bx]  ; Jump to key routine

   extended:
           mov     ah, 8h       ; Get second key of extended key
           int     21h
           .                    ; Use another jump table
           .                    ;  for extended keys
           .
           jmp     next
   ctrla:  .                    ; CONTROL-A code here
           .
           .
           jmp     next
   ctrlb:  .                    ; CONTROL-B code here
           .
           .
           jmp     next
           .
           .
   next:   .                    ; Continue

 In this example, the indirect memory operands point to addresses of routines
 for handling different keystrokes.


 7.1.2  Conditional Jumps

 The most common way to transfer control in assembly language is with a
 conditional jump. This is a two-step process: first test the condition, and
 then jump if the condition is true or continue if it is false.

 The conditional jump instructions check flag status.

 Conditional-jump instructions (except JCXZ) use the status of one or more
 flags as their condition. Thus, any statement that sets a flag under
 specified conditions can be the test statement. The most common test
 statements use the CMP or TEST instructions. The jump statement can be any
 one of 31 conditional-jump instructions. Conditional-jump instructions take
 a single operand containing the target address.


 7.1.2.1  Jump Extending

 In earlier versions of MASM, the NEAR and FAR operators cannot be used with
 conditional jumps on the 8086-80286 processors. MASM 6.0 automatically
 expands the jump instruction to include an unconditional jump to the
 destination, as long as a distance or size other than SHORT is specified or
 implicitly required from the operands. That is, MASM now generates the code
 that previously you had to write.

 Conditional jumps cannot refer to labels more than 128 bytes away.
 Therefore, in versions of MASM prior to 6.0, they are often combined with
 unconditional jumps, which have no such limitation. For example, the
 following statement is valid as long as  target  is not far away:

   ; Jump to target less than 128 bytes away
           jz      target  ; If previous operation resulted in
                           ;  zero, jump to target

 However, once  target  becomes too distant, the following sequence is
 necessary to enable a longer jump. Note that this sequence is logically
 equivalent to the example above:

   ; Jumps to distant targets previously required two steps
           jnz     skip    ; If previous operation result is
                           ;  NOT zero, jump to "skip"
           jmp     target  ;  Otherwise, jump to target
   skip:

 If the instruction is any of the conditional-jump instructions (except JCXZ
 and JECXZ ) and the target is greater than 128 bytes or is in a far segment,
 then jump-extending for an instruction such as  je target  generates two
 instructions to replace it:


   1.  The logical negation of the jump instruction, with a destination that
       skips over the second line it generates

   2.  An unconditional jump to the target destination


 For example, if  target  is more than 128 bytes away, MASM generates these
 lines of code for  je target:

   jne $ + 2 + (length in bytes of the next instruction)
            jmp NEAR PTR target

 Now the conditional jump executes correctly.

 The assembler generates this same code sequence if you specify the distance
 with NEAR PTR, FAR PTR, or SHORT. Therefore,

   jz      NEAR PTR target

 becomes

   jne     $ + 5
            jmp     NEAR PTR target

 even if  target  is nearby.

 When  skip  is more than 128 bytes away, this example

   mov     ax, cx
           jz      skip     ; Skip is more than 128 bytes away
           .
           .                ; (additional code here)
           .
   skip:

 generates code that looks like this:

   7327:0000 8BC1           MOV       AX,CX
   7327:0002 7503           JNZ       0007
   7327:0004 E9C000         JMP       00C7
   7327:0007                (more code here)

 MASM 6.0 enables this jump expansion feature by default, but you can turn it
 off with the NOLJMP form of the OPTION directive. See Section 1.3.2 for
 information about the OPTION directive.

 If the assembler generates code to extend a conditional jump, it issues a
 level 3 warning saying that the conditional jump has been lengthened. You
 can set the warning level to 1 for development and to level 3 for a final
 optimizing pass to see if you can shorten jumps by reorganizing.

 If you specify the distance for the jump and the target is out of range for
 that distance, a "Jump out of Range" error results.

 Since the JCXZ and JECXZ instructions do not have logical negations,
 expansion of the jump instruction to handle targets with unspecified
 distances cannot be performed for those instructions. Therefore the distance
 must always be short.

 The size and distance of the target operand determines the encoding for
 conditional or unconditional jumps to externals or targets in different
 segments. The new jump-extending and optimization features do not apply in
 this case.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 Conditional jumps on the 80386 and 80486 processors can be to targets up to
 32K bytes away, so jump extension occurs only for targets greater than that
 distance.
 ────────────────────────────────────────────────────────────────────────────


 7.1.2.2  Jumps Based on Comparisons

 The CMP instruction is specifically designed to test for conditional jumps.
 It does not change the destination operand─it compares two values without
 changing either of them. Instructions that change operands (such as SUB or
 AND) can also be used to test conditions.

 SUB and CMP set the same flags.

 Internally, the CMP instruction is the same as the SUB instruction, except
 that CMP does not change the destination operand. Both set flags according
 to the result that the subtraction generates.

 Table 7.1 lists conditional-jump instructions for each comparison
 relationship and shows the flags that are tested to see if the relationship
 is true. Note the difference in instructions depending on the sign of the
 operands. Some of these are equivalent to instructions listed in the
 previous section.

 Table   7.1 Conditional-Jump Instructions Used after Compare Instruction

╓┌──────────────┌──────────────┌──────────────┌──────────────┌───────────────╖
 Jump           Signed         Flags Tested   Unsigned       Flags Tested
 Condition      Compare        (Jump if True)  Compare        (Jump if True)
 ────────────────────────────────────────────────────────────────────────────
 =    (Equal)   JE             ZF = 1         JE             ZF = 1

 (Not equal)    JNE            ZF = 0         JNE            ZF = 0

 >    (Greater  JG or JNLE     ZF = 0 and     JA or JNBE     CF = 0 and
 than)                         SF = 0F                       ZF = 0

 <=   (Less     JLE or JNG     ZF = 1 or      JBE or JNA     CF = 1 or
 than                          SF  0F                        ZF = 1
        or
 equal to)

 <    (Less     JL or JNGE     SF  0F         JB or JNAE     CF = 1
 than)

 >=  (Greater   JGE or JNL     SF = 0F        JAE or JNB     CF = 0
 Jump           Signed         Flags Tested   Unsigned       Flags Tested
 Condition      Compare        (Jump if True)  Compare        (Jump if True)
 ────────────────────────────────────────────────────────────────────────────
>=  (Greater   JGE or JNL     SF = 0F        JAE or JNB     CF = 0
 than
        or
 equal to)

 ────────────────────────────────────────────────────────────────────────────



 In the CMP instruction, the mnemonic names always refer to the relationship
 of the first operand to the second operand. For instance, in this example JG
 tests whether the first operand is greater than the second.

   cmp     ax, bx  ; Compares ax and bx
           jg      contin  ; Equivalent to:  If ( ax > bx ) goto
                           ;  contin
           jl      next    ; Equivalent to:  If ( ax < bx ) goto next

 Several conditional instructions have two names. For example, JG and JNLE
 (Jump if Not Less or Equal) are equivalent. You can use whichever name seems
 more mnemonic in context.


 7.1.2.3  Testing Bits and Jumping

 Using CMP is not the only way to check a condition prior to a jump. You can
 also check the status of bits in the operands using the TEST instruction.
 This instruction tests for conditions prior to jumps by comparing specific
 bits rather than entire operands. Jump execution depends on whether certain
 bits are on or off.

 Pairs of operands cannot be both registers or both memory locations.

 The TEST instruction is the same as the AND instruction, except that TEST
 changes neither operand. If the result of the operation is 0, the zero flag
 is set, but the 0 is not actually written to the destination operand. The
 following example shows an application of TEST.

   .DATA
   bits    BYTE    ?
           .CODE
           .
           .
           .
   ; If bit 2 or bit 4 is set, then call task_a
                                ; Assume "bits" is 0D3h    11010011
           test    bits, 10100y ; If 2 or 4 is set     AND 00010100
           jz      skip1        ;                          --------
           call    task_a       ; Then call task_a         00010000
   skip1:                       ; Jump taken
           .
           .
           .
   ; If bits 2 and 4 are clear, then call task_b
                                ; Assume "bits" is 0E9h    11101001
           test    bits, 10100y ; If 2 and 4 are clear AND 00010100
           jnz     skip2        ;                          --------
           call    task_b       ; Then call task_b         00000000
   skip2:                       ; Jump taken

 Generally, when you use TEST, one of the operands is a mask in which the
 bits to be tested are the only bits set. The other operand contains the
 value to be tested. If all the bits set in the mask are clear in the operand
 being tested, the zero flag is set. If any of the flags set in the mask are
 also set in the operand, the zero flag is cleared.


 7.1.2.4  Jumping Based on Flag Status

 Your code can jump based on the condition of flags rather than on the
 relationships of operands. Use the following conditional-jump instructions:


╓┌───────────────────┌───────────────────────────────────────────────────────╖
 Instruction         Jumps if
 ────────────────────────────────────────────────────────────────────────────
 JO                  The overflow flag is set

 JNO                 The overflow flag is clear

 JC                  The carry flag is set (same as JB)
 Instruction         Jumps if
 ────────────────────────────────────────────────────────────────────────────
JC                  The carry flag is set (same as JB)

 JNC                 The carry flag is clear (same as JAE)

 JZ                  The zero flag is set (same as JE)

 JNZ                 The zero flag is clear (same as JNE)

 JS                  The sign flag is set

 JNS                 The sign flag is clear

 JP                  The parity flag is set

 JNP                 The parity flag is clear

 JPE                 Parity is even (parity flag set)

 JPO                 Parity is odd (parity flag clear)
 Instruction         Jumps if
 ────────────────────────────────────────────────────────────────────────────
JPO                 Parity is odd (parity flag clear)

 JCXZ                CX is 0

 JECXZ               ECX is 0
 (80386/486 only)



 The following example shows two ways to use the instructions from the list
 above:

   ; Uses JO to handle overflow condition
           add     ax, bx      ; Add two values
           jo      overflow    ; If value too large, adjust

   ; Uses JNZ to check for zero as the result of subtraction
           sub     ax, bx      ; Subtract
           jnz     skip        ; If the result is not zero, continue
           call    zhandler    ; Else do special case


 7.1.2.5  Anonymous Labels

 Anonymous labels are alternatives to named labels.

 Coding jumps in assembly language requires that you invent many label names.
 One alternative to continually thinking up new label names is using
 anonymous labels, which you can use anywhere in your program. But because
 anonymous labels do not provide meaningful names, they are best used for
 conditionally testing a few lines of code. You should mark major divisions
 of a program with actual named labels.

 Use two at signs (@) followed by a colon (:) as an anonymous label. To jump
 to the nearest preceding anonymous label, use @B (back) in the jump
 instruction's operand field; to jump to the nearest following anonymous
 label, use @F (forward) in the operand field.

 The jump in the example below uses an anonymous label:

   ; DX is 20, unless CX is less than -20, then make DX 30
           mov     dx, 20
           cmp     cx, -20
           jge     @F
           mov     dx, 30
   @:

 The items  @B  and  @F  always refer to the nearest occurrences of @:, so
 there is never any conflict between different anonymous labels.


 7.1.2.6  Decision Directives

 The high-level structures you can use for decision-making are the .IF,
 .ELSEIF, and .ELSE statements. These directives generate conditional jumps.
 The expression following the .IF directive is evaluated, and if true, the
 following instructions are executed until the next .ENDIF, .ELSE, or .ELSEIF
 directive is reached. The .ELSE statements execute if the expression is
 false. Using the .ELSEIF directive puts a new expression to be evaluated
 inside the alternative part of the original .IF statement. The syntax is

   .IF condition1
   statements
   «.ELSEIF condition2
   statements»
   «.ELSE
   statements»
   .ENDIF

 The decision structure

   .IF     cx = 20
           mov     dx, 20
           .ELSE
           mov     dx, 30
           .ENDIF

 generates this code:

   .IF  cx == 20
   0017  83 F9 14        *         cmp    cx, 014h
   001A  75 05           *         jne    @C0001
   001C  BA 0014                   mov    dx, 20
                              .ELSE
   001F  EB 03           *         jmp    @C0003
   0021                  *@C0001:
   0021  BA 001E                   mov   dx, 30
                              .ENDIF
   0024                  *@C0003:


 7.2  Loops

 Loops repeat an action until a termination condition is reached. This
 condition can be a counter or the result of an expression's evaluation. MASM
 6.0 offers many ways to set up loops in your programs. The following list
 compares MASM loop structures.

 Instructions                      Action
 ────────────────────────────────────────────────────────────────────────────
 LOOP                              Automatically decrements CX. When CX = 0,
                                   the loop ends. The top of the loop
                                   cannot be greater than 128 bytes from
                                   the LOOP instruction. (This is true for
                                   all LOOP instructions.)

 LOOPE, LOOPZ, LOOPNE, LOOPNZ      Loops while equal (or not equal). Checks
                                   CX and a condition. The loop ends when
                                   the condition is true. Set CX to a
                                   number out of range if you don't want a
                                   count to control the loop.

 JCXZ, JECXZ                       Branches to a label only if CX = 0 (ECX
                                   on the 80386). Useful for testing
                                   condition of CX before beginning loop.
                                   If CX = 0 before entering the loop, CX
                                   decrements to -1 on the first iteration
                                   and then must be decremented 65,535
                                   times before it reaches 0 again. Unlike
                                   conditional-jump instructions, which can
                                   jump to either a near or a short label
                                   under the 80386 or 80486, the loop
                                   instructions JCXZ and JECXZ always jump
                                   to a short label.

 Conditional jumps                 Acts only if certain conditions met.
                                   Necessary if several conditions must be
                                   tested. See Section 7.1.2, "Conditional
                                   Jumps."

 The following examples illustrate these loop constructions.

   ; The LOOP instruction: For 200 to 0 do task
           mov     cx, 200         ; Set counter
   next:   .                       ; Do the task here
           .
           .
           loop    next            ; Do again
                                   ; Continue after loop

   ; The LOOPNE instruction: While AX is not 'Y', do task
           mov     cx, 256         ; Set count too high to interfere
   wend:   .                       ; But don't do more than 256 times
           .                       ; Some statements that change AX
           .
           cmp     al, 'Y'         ; Is it Y or too many times?
           loopne  wend            ; No? Repeat
                                   ; Yes? Continue

   ; Using JCXZ:  For 0 to CX do task
                                   ; CX counter set previously
           jcxz    done            ; Check for 0
   next:   .                       ; Do the task here
           .
           .
           loop    next            ; Do again
   done:                           ; Continue after loop


 7.2.1  Loop-Generating Directives

 These directives are new to MASM 6.0.

 The high-level control structures new to MASM 6.0 generate loop structures
 for you. These new directives are similar to the while and repeat loops of C
 or Pascal. They can make your assembly programs less repetitive and easier
 to code, as well as easier to read. The assembler generates the appropriate
 assembly code. The .BREAK and .CONTINUE directives are also implemented to
 interrupt loop execution. These directives are summarized in the following
 list:

 Directives                        Action
 ────────────────────────────────────────────────────────────────────────────
 .WHILE, .ENDW                     The statements between .WHILE condition
                                   and .ENDW execute while the condition is
                                   true.

 .REPEAT, .UNTIL                   The loop executes at least once and
                                   continues until the condition given
                                   after .UNTIL is true. Generates
                                   conditional jumps.

 .REPEAT, .UNTILCXZ                Compares label to an expression and
                                   generates appropriate loop instructions.


 These constructs work much as they do in a high-level language such as C or
 Pascal. Keep in mind the following points:


   ■   These directives generate appropriate processor instructions. They are
       not new instructions.

   ■   They require proper use of signed and unsigned data declarations.


 These directives cause a set of instructions to execute based on the
 evaluation of some condition. This condition can be an expression that
 evaluates to a negative or nonnegative value, an expression using the binary
 operators in C (&&, ||, or !), or the state of a flag. See Section 7.2.2.1
 for more information about expression operators.

 The evaluation of the condition requires the assembler to know if the
 operands in the condition are signed or unsigned. To state explicitly that a
 named memory location contains a signed integer, use the signed data
 allocation directives: SBYTE, SWORD, and SDWORD.


 7.2.1.1  .WHILE Loops

 As with while loops in C or Pascal, the test condition for .WHILE is checked
 before the statements inside the loop execute. If the test condition is
 false, the loop does not execute. While the condition is true, the
 statements inside the loop repeat.

 Use the .ENDW directive to mark the end of the .WHILE loop. When the
 condition becomes false, program execution begins at the first statement
 following the .ENDW directive. The .WHILE directive generates appropriate
 compare and jump statements. The syntax is

 .WHILE condition statements .ENDW

 For example, this loop copies one buffer to another until a `$' character
 (marking the end of the string) is found:

   .DATA
           buf1    BYTE "This is a string",'$'
           buf2    BYTE 100 DUP (?)
           .CODE
           sub     bx, bx            ; Zero out bx
           .WHILE  (buf1[bx] != '$')
           mov     al, buf1[bx]      ; Get a character
           mov     buf2[bx], al      ; Move it to buffer 2
           inc     bx                ; Count forward
           .ENDW


 7.2.1.2  .REPEAT Loops

 MASM's .REPEAT directive allows for loop constructions like the do loop of C
 and the REPEAT loop of Pascal. The loop executes until the condition
 following the .UNTIL (or .UNTILCXZ) directive becomes true. Since the
 condition is checked at the end of the loop, the loop always executes at
 least once. The .REPEAT directive generates conditional jumps. The syntax
 is:

   .REPEAT
   statements
   .UNTIL condition

   .REPEAT
   statements
   .UNTILCXZ «condition»

 A condition is optional with .UNTILCXZ.

 where condition can also be expr1 == expr2 or expr1 != expr2. When two
 conditions are used, expr2 can be an immediate expression, a register, or
 (if expr1 is a register) a memory location.

 For example, the following code fills up a buffer with characters typed at
 the keyboard. The loop ends when the ENTER key (character 13) is pressed:

   .DATA
   buffer  BYTE    100 DUP (0)
           .CODE
           sub     bx, bx             ; Zero out bx
           .REPEAT
           mov     ah, 01h
           int     21h                ; Get a key
           mov     buffer[bx], al     ; Put it in the buffer
           inc     bx                 ; Increment the count
           .UNTIL (al == 13)          ; Continue until al is 13

 The .UNTIL directive generates conditional jumps, but the .UNTILCXZ
 directive generates a LOOP instruction, as shown by the listing file code
 for these examples. In a listing file, assembler-generated code is preceded
 by an asterisk.

   ASSUME  bx:PTR SomeStruct

               .REPEAT
          *@C0001:
                    inc    ax
               .UNTIL  ax==6
          *         cmp    ax, 006h
          *         jne    @C0001

   .REPEAT
          *@C0003:
                    mov    ax, 1
               .UNTILCXZ
          *         loop   @C0003

               .REPEAT
          *@C0004:
               .UNTILCXZ   [bx].field != 6
          *         cmp    [bx].field, 006h
          *         loope  @C0004


 7.2.1.3  .BREAK and .CONTINUE Directives

 .BREAK and .CONTINUE interrupt loop execution.

 The .BREAK and .CONTINUE directives can be used to terminate a .REPEAT or
 .WHILE loop prematurely. These directives allow an optional .IF clause for
 conditional breaks. The syntax is

   .BREAK «.IF condition»
   .CONTINUE «.IF condition»

 Note that .ENDIF is not used with the .IF forms of .BREAK and .CONTINUE in
 this context. The .BREAK and .CONTINUE directives work the same way as the
 break and continue instructions in C. Execution continues at the instruction
 following the .UNTIL, .UNTILCXZ, or .ENDW of the nearest enclosing loop.

 Instead of causing the loop execution to end as .BREAK does, .CONTINUE
 causes loop execution to jump directly to the code that evaluates the loop
 condition of the nearest enclosing loop.

 The following loop accepts only the keys in the range `0' to `9' and
 terminates when ENTER is pressed.

   .WHILE 1                ; Loop forever
           mov     ah, 08h         ; Get key without echo
           int     21h
           .BREAK .IF al == 13     ; If ENTER, break out of the loop
           .CONTINUE .IF (al < '0') || (al > '9')
                                   ; If not a digit, continue looping
           mov     dl, al          ; Save the character for processing
           mov     ah, 02h         ; Output the character
           int     21h
           .ENDW

 If you assemble the source code above with the /Fl and /Sg command-line
 options and then view the results in the listing file, you would see this
 code:

   .WHILE 1
    0017                    *@C0001:
    0017  B4 08                       mov    ah, 08h
    0019  CD 21                       int    21h
                               .BREAK .IF al == 13
    001B  3C 0D             *         cmp    al, 00Dh
    001D  74 10             *         je     @C0002
                               .CONTINUE .IF (al  '0') || (al  '9')
    001F  3C 30             *         cmp    al, '0'
    0021  72 F4             *         jb     @C0001
    0023  3C 39             *         cmp    al, '9'
    0025  77 F0             *         ja     @C0001
    0027  8A D0                       mov    dl, al
    0029  B4 02                       mov    ah, 02h
    002B  CD 21                       int    21h
                               .ENDW
    002D  EB E8             *         jmp    @C0001
    002F                    *@C0002:

 The high-level control structures can be nested. That is, .REPEAT or .WHILE
 loops can contain .REPEAT or .WHILE loops as well as .IF statements.

 If the code generated by a .WHILE loop, .REPEAT loop, or .IF statement
 generates a conditional or unconditional jump, MASM uses the jump extension
 and jump optimization techniques described in Sections 7.1.1, "Unconditional
 Jumps," and 7.1.2, "Conditional Jumps," to encode the jump appropriately.


 7.2.2  Writing Loop Conditions

 You can express the conditions of the .IF, .REPEAT, and .WHILE directives
 using relational operators, and you can express the attributes of the
 operand with the PTR operator. To write loop conditions, you also need to
 know how the assembler evaluates the operators and operands in the
 condition. This section explains the operators, attributes, precedence
 level, and expression evaluation order for the conditions used with
 loop-generating directives.


 7.2.2.1  Expression Operators

 The binary relational operators in MASM 6.0 high-level control structures
 are listed below. The same binary operators are used in C. These operators
 generate MASM compare, test, and conditional jump instructions.

╓┌──────────────────────┌────────────────────────────────────────────────────╖
 Operator               Meaning
 ────────────────────────────────────────────────────────────────────────────
 ==                     Equal
 !=                     Not equal
 >                      Greater than
 >=                     Greater than or equal to
 <                      Less than
 <=                     Less than or equal to
 &                      Bit test
 !                      Logical NOT
 &&                     Logical AND
 ||                     Logical OR
 Operator               Meaning
 ────────────────────────────────────────────────────────────────────────────
||                     Logical OR


 A condition without operators (other than !) tests for nonzero as it does in
 C. For example,  .WHILE (x)  is the same as  .WHILE (x != 0), and  .WHILE
 (!x)  is the same as  .WHILE (x == 0).

 Flag names can be operands in a condition.

 You can also use the flag names (ZERO?, CARRY?, OVERFLOW?, SIGN?, and
 PARITY?) as operands in conditions with the high-level control structures as
 in .WHILE (CARRY?). The particular flag set determines the outcome of the
 condition. Use flag names when you want to generate the compare or other
 instructions that set the flags.


 7.2.2.2  Signed and Unsigned Operands

 Registers, constants, and memory locations are unsigned by default.

 Expression operators generate unsigned jumps by default. However, if either
 side of the operation is signed, then the entire operation is considered
 signed. The default for the operands in registers, constants, and named
 memory locations is also to be unsigned.

 You can use the PTR operator to tell the assembler that a particular operand
 in a register or constant is a signed number, as in these examples:

   .WHILE  SWORD PTR [bx] <= 0
           .IF     SWORD PTR mem1 >  0

 Without the PTR operator, the assembler would treat the contents of BX as an
 unsigned value.

 You can also specify the size attributes of operands in memory locations
 with SBYTE, SWORD, and SDWORD, for use with .IF, .WHILE, and .REPEAT.

   .DATA
   mem1    SBYTE   ?
   mem2    WORD    ?
           .IF     mem1 > 0
           .WHILE  mem2 < bx
           .WHILE  SWORD PTR ax < count


 7.2.2.3  Precedence Level

 As with C, you can concatenate conditions with the && operator for AND, the
 || operator for OR, and the ! operator for negate. The precedence level is
 !, &&, and ||, with ! having the highest precedence. Like expressions in
 high-level languages, associativity is evaluated left to right.


 7.2.2.4  Expression Evaluation

 The assembler evaluates conditions created with high-level control
 structures according to short-circuit evaluation. If the evaluation of a
 particular condition automatically determines the final result (such as a
 condition that evaluates to false in a compound statement concatenated with
 AND), the evaluation does not continue.

 For example, in this .WHILE statement,

   .WHILE (ax > 0) && (WORD PTR [bx] == 0)

 the assembler evaluates the first condition. If this condition is false
 (that is, if AX is less than or equal to 0), the evaluation is finished. The
 second condition is not checked and the loop does not execute, because a
 compound condition containing a && requires both expressions to be true for
 the entire condition to be true.





 7.3  Procedures

 Organizing your code into procedures that execute specific tasks divides
 large programs into manageable units, allows for separate testing, and makes
 code more efficient for repetitive tasks.

 Assembly-language procedures are comparable to functions in C; subprograms,
 functions, and subroutines in Basic; procedures and functions in Pascal; or
 subroutines and functions in FORTRAN.

 Two instructions control the use of assembly-language procedures; CALL
 pushes the return address onto the stack and transfers control to a
 procedure, and RET pops the return address off the stack and returns control
 to that location.

 The PROC and ENDP directives mark the beginning and end of a procedure.
 Additionally, PROC can automatically


   ■   Preserve register values that should not change but that the procedure
       might otherwise alter

   ■   Set up a local stack pointer, so that you can access parameters and
       local variables placed on the stack

   ■   Adjust the stack when the procedure ends


 Sections 7.3.1 through 7.3.3 give information on techniques for calling
 procedures and accessing parameters. Sections 7.3.4 through 7.3.5 show how
 to allocate and access local variables and parameters.

 Sections 7.3.6 and 7.3.7 introduce new directives in MASM 6.0 to further
 automate calling procedures and passing arguments. The PROTO directive
 allows you to declare prototypes for your procedures. INVOKE handles
 procedure calls and stack cleanup. Section 7.3.8 describes the automatic
 stack setup and cleanup generated with PROC.


 7.3.1  Defining Procedures

 Procedures require a label at the start of the procedure and a return at the
 end. Procedures are normally defined by using the PROC directive at the
 start of the procedure and the ENDP directive at the end. The RET
 instruction is normally placed immediately before the ENDP directive. The
 assembler makes sure that the distance of the RET instruction matches the
 distance defined by the PROC directive. The basic syntax for PROC is

   label PROC [[NEAR|FAR]]
         .
         .
         .
   RET [[constant]]
   label ENDP

 The CALL instruction pushes the address of the next instruction in your code
 onto the stack and passes control to a specified address. The syntax is

   CALL {label | register | memory}

 The operand contains a value calculated at run time. Since that operand can
 be a register, direct memory operand, or indirect memory operand, you can
 write call tables similar to the jump table illustrated in Section 7.1.1.2.


 Calls can be near or far. Near calls push only the offset portion of the
 calling address and therefore must be within the same segment or group. You
 can specify the type for the target operand, but if you do not, MASM uses
 the declared distance (NEAR or FAR) for operands that are labels and for the
 size of register or memory operands. Then the assembler encodes the call
 appropriately, as it does with unconditional jumps (see Sections 7.1.1,
 "Unconditional Jumps," and 7.1.2, "Conditional Jumps").

 MASM 6.0 optimizes a call to a far label when the label is in the current
 segment by generating the code for a near call, saving one byte.

 You can define procedures without PROC and ENDP, but if you do, you must
 make sure that the size of the CALL matches the size of the RET. You can
 specify the RET instruction as RETN (Return Near) or RETF (Return Far) to
 override the default size:

   call    NEAR PTR task ; Call is declared near
           .                     ; Return comes to here
           .
           .
   task:                         ; Procedure begins with near label
           .
           .                     ; Instructions go here
           .
           retn                  ; Return declared near

 The syntax for RETN and RETF is

   label: | label NEAR
   statements
   RETN [[constant]]

   label LABEL FAR
   statements
   RETF [[constant]]

 The RET instruction (and its RETF and RETN variations) allows an optional
 constant operand that specifies a number of bytes to be added to the value
 of the SP register after the return. This operand adjusts for arguments
 passed to the procedure before the call, as shown in the example in Section
 7.3.4, "Using Local Variables."

 Incorrect size for RET can cause your program to fail.

 When you define procedures without PROC and ENDP, you must make sure that
 calls have the same size as corresponding returns. For example, RETF pops
 two words off the stack. If a NEAR call is made to a procedure with a far
 return, not only is the popped value meaningless, but the stack status may
 cause the execution to return to a random memory location, resulting in
 program failure.

 There is an also an extended PROC syntax that automates many of the details
 of accessing arguments and saving registers. See Section 7.3.3, "Declaring
 Parameters with the PROC Directive."


 7.3.2  Passing Arguments on the Stack

 Each time you call a procedure, you may want it to operate on different
 data. This data, called "arguments," can be passed in various ways. For
 example, arguments can be passed to a procedure in registers or in
 variables. However, the

 most common method of passing arguments is to use the stack. Microsoft
 languages have specific conventions for passing arguments. Chapter 20,
 "Mixed-Language Programming," explains these conventions for
 assembly-language modules shared with modules from high-level languages.

 This section describes how a procedure accesses the arguments passed to it
 on the stack. Each argument is accessed as an offset from BP. However, if
 you use the PROC directive to declare parameters, the assembler calculates
 these offsets for you and lets you refer to parameters by name. The next
 section, "Declaring Parameters with the PROC Directive," explains how to use
 PROC this way.

 This example shows how to pass arguments to a procedure. The procedure
 expects to find those arguments on the stack. As this example shows,
 arguments must be accessed as offsets of BP.

   ; C-style procedure call and definition

           mov     ax, 10     ; Load and
           push    ax         ;  push constant as third argument
           push    arg2       ; Push memory as second argument
           push    cx         ; Push register as first argument
           call    addup      ; Call the procedure
           add     sp, 6      ; Destroy the pushed arguments
           .                  ;  (equivalent to three pops)
           .
           .
   addup   PROC    NEAR       ; Return address for near call
                              ;  takes two bytes
           push    bp         ; Save base pointer - takes two bytes
                              ;  so arguments start at fourth byte
           mov     bp, sp     ; Load stack into base pointer
           mov     ax, [bp+4] ; Get first argument from
                              ;  fourth byte above pointer
           add     ax, [bp+6] ; Add second argument from
                              ;  sixth byte above pointer
           add     ax, [bp+8] ; Add third argument from
                              ;  eighth byte above pointer
           mov     sp, bp
           pop     bp         ; Restore BP
           ret                ; Return result in AX
   addup   ENDP

 Figure 7.1 shows the stack condition at key points in the process.

 (This figure may be found in the printed book.)

 Starting with the 80186 processor, the ENTER and LEAVE instructions simplify
 the stack setup and restore instructions at the beginning and end of
 procedures.

 However, ENTER uses a lot of time. It is necessary only with nested,
 statically scoped procedures. Thus, a Pascal compiler may sometimes generate
 ENTER. The LEAVE instruction, on the other hand, is an efficient way to do
 the stack cleanup. LEAVE reverses the effect of the last ENTER instruction
 by restoring BP and SP to their values before the procedure call.


 7.3.3  Declaring Parameters with the PROC Directive

 With the PROC directive, you can specify registers to be saved, define
 parameters to the procedure, and assign symbol names to parameters (rather
 than as offsets from BP). This section describes how to use the PROC
 directive to automate the parameter-accessing techniques described in the
 last section.

 For example, the diagram below shows a valid PROC statement for a procedure
 called from C. It takes two parameters,  var1  and  arg1, and uses (and must
 save) the DI and SI registers:

 (This figure may be found in the printed book.)

 The syntax for PROC is

   label PROC [[attributes]]
   [[USES reglist]] [[, parameter[[:tag]]...
   ]]

 The following list describes the parts of the PROC directive.

 Argument                          Description
 ────────────────────────────────────────────────────────────────────────────
 label                             The name of the procedure.

 attributes                        Any of several attributes of the
                                   procedure, including the distance,
                                   langtype, and visibility of the
                                   procedure. The syntax for attributes is
                                   given in Section 7.3.3.1.

 reglist                           A list of registers following the USES
                                   keyword that the procedure uses and that
                                   should be saved on entry. Registers in
                                   the list must be separated by blanks or
                                   tabs, not by commas. The assembler
                                   generates prologue code to push these
                                   registers onto the stack. When you exit,
                                   the assembler generates epilogue code to
                                   pop the saved register values off the
                                   stack.

 parameter                         The list of parameters passed to the
                                   procedure on the stack. The list can
                                   have a variable number of parameters.
                                   See the discussion below for the syntax
                                   of parameter. This list can be longer
                                   than one line if the continued line ends
                                   with a comma.


 This diagram shows a valid PROC definition that uses several attributes:

 (This figure may be found in the printed book.)


 7.3.3.1  Attributes

 The syntax for the attributes field is

   «distance» «langtype» «visibility»
   «<prologuearg>»

 The list below explains each of these options.

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Argument                          Description
 ────────────────────────────────────────────────────────────────────────────
 distance                          Controls the form of the RET instruction
                                   generated. Can be NEAR or FAR. If
                                   distance is not specified, it is
                                   determined from the model declared with
 Argument                          Description
 ────────────────────────────────────────────────────────────────────────────
                                  determined from the model declared with
                                   the .MODEL directive. For TINY, SMALL,
                                   COMPACT, and FLAT, NEAR is assumed. For
                                   MEDIUM, LARGE, and HUGE, FAR is assumed.
                                   For 80386/486 programming with 16- and
                                   32-bit segments, NEAR16, NEAR32, FAR16,
                                   or FAR32 can be specified.

 langtype                          Determines the calling convention used
                                   to access param-
                                   eters and restore the stack. The BASIC,
                                   FORTRAN, and PASCAL langtypes convert
                                   procedure names to uppercase, place the
                                   last parameter in the parameter list
                                   lowest on the stack, and generate a RET,
                                   which adjusts the stack upward by the
                                   number of bytes in the argument list.

                                   The C and STDCALL langtype prefixes an
 Argument                          Description
 ────────────────────────────────────────────────────────────────────────────
                                  The C and STDCALL langtype prefixes an
                                   underscore to the procedure name when
                                   the procedure's scope is PUBLIC or
                                   EXPORT and places the first parameter
                                   lowest on the stack. SYSCALL is
                                   equivalent to the C calling convention
                                   with no underscore prefixed to the
                                   procedure's name. STDCALL uses caller
                                   stack cleanup when :VARARG is specified;
                                   otherwise the called routine must clean
                                   up the stack (see Chapter 20).

 visibility                        Indicates whether the procedure is
                                   available to other modules. The
                                   visibility can be PRIVATE, PUBLIC, or
                                   EXPORT. A procedure name is PUBLIC
                                   unless it is explicitly declared as
                                   PRIVATE. If the visibility is EXPORT,
                                   the linker places the procedure's name
 Argument                          Description
 ────────────────────────────────────────────────────────────────────────────
                                  the linker places the procedure's name
                                   in the export table for segmented
                                   executables. EXPORT also enables PUBLIC
                                   visibility.

                                   You can explicitly set the default
                                   visibility with the
                                   OPTION directive. OPTION PROC:PUBLIC
                                   sets the default to public. See Section
                                   1.3.2 for more information.

 prologuearg                       Specifies the arguments that affect the
                                   generation of prologue and epilogue code
                                   (the code MASM generates when it
                                   encounters a PROC directive or the end
                                   of a procedure). See Section 7.3.8 for
                                   an explanation of prologue and epilogue
                                   code.

 Argument                          Description
 ────────────────────────────────────────────────────────────────────────────




 7.3.3.2  Parameters

 The parameters are separated from the reglist by a comma if there is a list
 of registers. In the syntax:

   parmname [[:tag»

 parmname is the name of the parameter. The tag can be either the
 qualifiedtype or the keyword VARARG. However, only the last parameter in a
 list of parameters can use the VARARG keyword. The qualifiedtype is
 discussed in Section 1.2.6, "Data Types." An example showing how to
 reference VARARG parameters appears later in this section. Procedures can be
 nested if they do not have parameters or USES register lists. This diagram
 shows a procedure definition with one parameter definition.

 (This figure may be found in the printed book.)

 The following example shows the procedure in Section 7.3.2, "Passing
 Arguments on the Stack," rewritten to use the extended PROC functionality.
 Prior to the procedure call, you must push the arguments onto the stack
 unless you use INVOKE (see Section 7.3.7, "Calling Procedures with INVOKE").


   addup   PROC NEAR C,
           arg1:WORD, arg2:WORD, count:WORD
           mov     ax, arg1
           add     ax, count
           add     ax, arg2
           ret
   addup   ENDP

 If the arguments for a procedure are pointers, the assembler does not
 generate any code to get the value or values that the pointers reference;
 your program must still explicitly treat the argument as a pointer. (See
 Chapter 3, "Using Addresses and Pointers," for more information about using
 pointers.)

 In the example below, even though the procedure declares the parameters as
 near pointers, you still must code two MOV instructions to get the values of
 the parameters─the first MOV gets the address of the parameters, and the
 second MOV gets the parameter.

   ; Call from C as a FUNCTION returning an integer

           .MODEL medium, c
           .CODE
   myadd   PROC   arg1:NEAR PTR WORD, arg2:NEAR PTR WORD

           mov     bx, arg1     ; Load first argument
           mov     ax, [bx]
           mov     bx, arg2     ; Add second argument
           add     ax, [bx]

           ret

   myadd   ENDP
           END

 You can use conditional-assembly directives to make sure that your pointer
 parameters are loaded correctly for the memory model. For example, the
 following version of  myadd  treats the parameters as FAR parameters if
 necessary:

   .MODEL  medium, c       ; Could be any model
           .CODE
   myadd   PROC    arg1:PTR WORD,   arg2:PTR WORD

           IF      @DataSize
           les     bx, arg1        ; Far parameters
           mov     ax, es:[bx]
           les     bx, arg2
           add     ax, es:[bx]
           ELSE
           mov     bx, arg1        ; Near parameters
           mov     ax, [bx]
           mov     bx, arg2
           add     ax, [bx]
           ENDIF

           ret
   myadd   ENDP

           END


 7.3.3.3  Using VARARG

 In the PROC statement, you can append the :VARARG keyword to the last
 parameter to indicate that a variable number of arguments can be passed if
 you use the C, SYSCALL, or STDCALL calling conventions (see Section 20.1). A
 label must precede :VARARG so that the arguments can be accessed as offsets
 from the variable name given. This example illustrates VARARG:

   addup3  PROTO NEAR C, argcount:WORD, arg1:VARARG

           invoke  addup3, 3, 5, 2, 4

   addup3  PROC    NEAR C, argcount:WORD, arg1:VARARG
           sub     ax, ax        ; Clear work register
           sub     si, si

           .WHILE  argcount > 0  ; Argcount has number of arguments
           add     ax, arg1[si]  ; Arg1 has the first argument
           dec     arg1          ; Point to next argument
           inc     si
           inc     si
           .ENDW

           ret                   ; Total is in AX
   addup3  ENDP

 Passing non-default-sized pointers in the VARARG portion of the parameter
 list can be done by explicitly passing the segment portion and the offset
 portion of the address separately.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 When you use the extended PROC features and the assembler encounters a RET
 instruction, it automatically generates instructions to pop saved registers,
 remove local variables from the stack, and, if necessary, remove parameters.
 It generates this code for each RET instruction it encounters. You can
 reduce code size by having only one return and jumping to it from various
 locations.
 ────────────────────────────────────────────────────────────────────────────


 7.3.4  Using Local Variables

 In high-level languages, local variables are visible only within a
 procedure. In Microsoft languages, these variables are usually stored on the
 stack. In assembly-language programs, you can also have local variables.
 These variables should not be confused with labels or variable names that
 are local to a module, as described in Chapter 8, "Sharing Data and
 Procedures among Modules and Libraries."

 This section outlines the standard methods for creating local variables. The
 next section shows how to use the LOCAL directive to make the assembler
 automatically generate local variables. When you use this directive, the
 assembler generates the same instructions as those used in this section but
 handles some of the details for you.

 If your procedure has relatively few variables, you can usually write the
 most efficient code by placing these values in registers. Local (stack) data
 is more efficient when you have a large amount of local data for the
 procedure.

 Local variables are stored on the stack.

 To use local variables you must save stack space for the variable at the
 start of the procedure. The variable can then be accessed by its position in
 the stack. At the end of the procedure, you need to restore the stack
 pointer, which restores the memory used by local variables.

 This example subtracts two bytes from the SP register to make room for a
 local word variable. This variable can then be accessed as  [bp-2].

   push    ax                 ; Push one argument
           call    task               ; Call
           .
           .
           .

   task    PROC    NEAR
           push    bp                 ; Save base pointer
           mov     bp, sp             ; Load stack into base pointer
           sub     sp, 2              ; Save two bytes for local
                                      ;  variable
           .
           .
           .
           mov     WORD PTR [bp-2], 3 ; Initialize local variable
           add     ax, [bp-2]         ; Add local variable to AX
           sub     [bp+4], ax         ; Subtract local from argument
           .                          ; Use [bp-2] and [bp+4] in
           .                          ;  other operations
           .
           mov     sp, bp             ; Clear local variables
           pop     bp                 ; Restore base
           ret     2                  ; Return result in AX and pop
   task    ENDP                       ;  two bytes to clear parameter

 Notice that the instruction  mov sp,bp  at the end of the procedure restores
 the original value of SP. The statement is required only if the value of SP
 is changed inside the procedure (usually by allocating local variables). The
 argument passed to the procedure is removed with the RET instruction.
 Contrast this to the example in Section 7.3.2, "Passing Arguments on the
 Stack," in which the calling code adjusts the stack for the argument.

 Figure 7.2 shows the state of the stack at key points in the process.

 (This figure may be found in the printed book.)


 7.3.5  Creating Local Variables Automatically

 Section 7.3.4 described how to create local variables on the stack. This
 section shows you how to automate the process with the LOCAL directive.

 The LOCAL directive generates code to set up the stack for local variables.


 You can use the LOCAL directive to save time and effort when working with
 local variables. When you use this directive, simply list the variables you
 want to create, giving a type for each one. The assembler calculates how
 much space is required on the stack. It also generates instructions to
 properly decrement SP (as described in the previous section) and to reset SP
 when you return from the procedure.

 When you create local variables this way, your source code can then refer to
 each local variable by name rather than as an offset of the stack pointer.
 Moreover, the assembler generates debugging information for each local
 variable.

 The procedure in the previous section can be generated more simply with the
 following code:

   task    PROC    NEAR    arg:WORD
           LOCAL   loc:WORD
           .
           .
           .
           mov     loc, 3    ; Initialize local variable
           add     ax, loc   ; Add local variable to AX
           sub     arg, ax   ; Subtract local from argument
           .                 ; Use "loc" and "arg" in other operations
           .
           .
           ret
   task    ENDP

 The LOCAL directive must be on the line immediately following the PROC
 statement. It cannot be used after the first instruction in a procedure. The
 LOCAL directive has the following syntax:

   LOCAL vardef [[, vardef]]...

 Each vardef defines a local variable. A local variable definition has this
 form:

   label[[ [count] ]][[:qualifiedtype]]

 These are the parameters in local variable definitions:

 Argument                          Description
 ────────────────────────────────────────────────────────────────────────────
 label                             The name given to the local variable.
                                   You can use this name to access the
                                   variable.

 count                             The number of elements of this name and
                                   type to allocate on the stack. You can
                                   allocate a simple array on the stack
                                   with count. The brackets around count
                                   are required. If this field is omitted,
                                   one data object is assumed.

 qualifiedtype                     A simple MASM type or a type defined
                                   with other types and attributes. See
                                   Section 1.2.6, "Data Types," for more
                                   information.


 If the number of local variables exceeds one line, you can place a comma at
 the end of the first line and continue the list on the next line. Another
 method is to use several consecutive LOCAL directives.

 You must initialize local variables.

 The assembler does not initialize local variables. Your program must include
 code to perform any necessary initializations. For example, the following
 code fragment sets up a local array and initializes it to zero:

   arraysz EQU     20

   aproc   PROC    USES di
           LOCAL   var1[arraysz]:WORD, var2:WORD
           .
           .
           .
   ; Initialize local array to zero
           push    ss
           pop     es              ; Set ES=SS
           lea     di, var1        ; ES:DI now points to array
           mov     cx, arraysz     ; Load count
           sub     ax, ax
           rep     stosw           ; Store zeros
   ; Use the array...
           .
           .
           .
           ret
   aproc   ENDP

 Even though you can reference stack variables by name, the assembler treats
 them as offsets from BP, and they are not visible outside the procedure. In
 this procedure,  array  is a local variable.

   index   EQU   10
   test    PROC  NEAR
   LOCAL   array[index]:WORD
           .
           .
           .
           mov     bx, index
   ;       mov     array[bx], 5           ; Not legal!

 The second MOV statement may appear to be legal, but since  array  is an
 offset of BP, this statement is the same as

   ;       mov [bp + bx + arrayoffset], 5   ; Not legal!

 BP and BX can be added only to SI and DI. This example would be legal,
 however, if the index value were moved to SI or DI. This type of error in
 your program can be difficult to find unless you keep in mind that local
 variables in procedures are offsets of BP.


 7.3.6  Declaring Procedure Prototypes

 MASM 6.0 provides a new directive, INVOKE, to handle many of the details
 important to procedure calls, such as pushing parameters according to the
 correct calling conventions. In order to use INVOKE, the procedure called
 must have previously been declared with a PROC statement, an EXTERNDEF (or
 EXTERN) statement, or a TYPEDEF. You can also place a prototype defined with
 PROTO before the INVOKE if the procedure type does not appear before the
 INVOKE. Procedure prototypes defined with PROTO inform the assembler of
 types and numbers of arguments so the assembler can check for errors and
 provide automatic conversions when INVOKE calls the procedure.


 Place prototypes after data declarations or in a separate include file.

 Prototypes in MASM perform the same function as prototypes in the C language
 and other high-level languages. A procedure prototype includes the procedure
 name, the types, and (optionally) the names of all parameters the procedure
 expects. Prototypes are usually placed at the beginning of an assembly
 program or in a separate include file. They are especially useful for
 procedures called from other modules and other languages, enabling the
 assembler to check for unmatched parameters. If you write routines for a
 library, you may want to put prototypes into an include file for all the
 procedures used in that library. See Chapter 8, "Sharing Data and Procedures
 among Modules and Libraries," for more information about using include
 files.

 Declaring procedure prototypes is optional. You can use the PROC directive
 and the CALL instruction, as shown in the previous section.

 In MASM 6.0, using the PROTO directive is one way to define procedure
 prototypes. The syntax for a prototype definition is the same as for a
 procedure declaration (see Section 7.3.3, "Declaring Parameters with the
 PROC Directive"), except that you do not include the list of registers,
 prologuearg list, or the scope of the procedure.

 Also, the PROTO keyword precedes the langtype and distance attributes. The
 attributes (like C and FAR) are optional, but if not specified, the defaults
 are based on any .MODEL or OPTION LANGUAGE statement. The names of the
 parameters are also optional, but you must list parameter types. A label
 preceding :VARARG is also optional in the prototype but not in the PROC
 statement.

 If a PROTO and a PROC for the same function appear in the same module, they
 must match in attribute, number of parameters, and parameter types. The
 easiest way to create prototypes with PROTO for your procedures is to write
 the procedure and then copy the first line (the line that contains the PROC
 keyword) to a location in your program that follows the data declarations.
 Change PROC to PROTO and remove the USES reglist, the prologuearg field, and
 the visibility field. It is important that the prototype follow the
 declarations for any types used in it to avoid any forward references used
 by the parameters in the prototype.

 The prototype defined with PROTO statement and the PROC statement for two
 procedures are given below.

   ;  Procedure prototypes

   addup     PROTO NEAR C argcount:WORD, arg2:WORD, arg3:WORD

   myproc    PROTO FAR C, argcount:WORD, arg2:VARARG

   ; Procedure declarations

   addup     PROC NEAR C, argcount:WORD, arg2:WORD, arg3:WORD

   myproc    PROC FAR C PUBLIC <callcount> USES di si,
             argcount:WORD,
             arg2:VARARG

 When you call a procedure with INVOKE, the assembler checks the arguments
 given by INVOKE against the parameters expected by the procedure. If the
 data types of the arguments do not match, MASM either reports an error or
 converts the type to the expected type. These conversions are explained in
 the next section.


 7.3.7  Calling Procedures with INVOKE

 INVOKE generates a sequence of instructions that push arguments and call a
 procedure. This helps maintain code if arguments or langtype for a procedure
 is changed. INVOKE generates procedure calls and automatically handles the
 following tasks:


   ■   Converts arguments to the expected types

   ■   Pushes arguments on the stack in the correct order

   ■   Cleans up the stack when the procedure returns


 If arguments do not match in number or if the type is not one the assembler
 can convert, an error results.

 If VARARG is an option in a procedure, INVOKE can pass arguments in addition
 to those in the parameter list without generating an error or warning. The
 extra arguments must be at the end of the INVOKE argument list. All other
 arguments must match in number and type.

 The syntax for INVOKE is

   INVOKE   expression  «, arguments»

 where expression can be the procedure's label or an indirect reference to a
 procedure, and arguments can be an expression, a register pair, or an
 expression preceded with ADDR. (The ADDR operator is discussed below.)

 Procedures that have these procedure prototypes

   addup   PROTO NEAR C argcount:WORD, arg2:WORD, arg3:WORD

   myproc  PROTO FAR C, argcount:WORD, arg2:VARARG

 and these procedure declarations

   addup   PROC NEAR C, argcount:WORD, arg2:WORD, arg3:WORD

   myproc  PROC FAR C PUBLIC <callcount> USES di si,
           argcount:WORD,
           arg2:VARARG

 may have INVOKE statements that look like this:

   INVOKE  addup,   ax, x,  y
           INVOKE  myproc,  bx, cx, 100, 10

 The assembler can convert some arguments and parameter type combinations so
 that the correct type can be passed. The signed or unsigned qualities of the
 arguments in the INVOKE statements determine how the assembler converts them
 to the types expected by the procedure.

 The  addup  procedure, for example, expects parameters of type WORD, but the
 arguments passed by INVOKE to the  addup  procedure can be any of these
 types:


   ■   BYTE, SBYTE, WORD, or SWORD

   ■   An expression whose type is specified with the PTR operator to be one
       of those types

   ■   An 8-bit or 16-bit register

   ■   An immediate expression in the range -32K to +64K

   ■   A NEAR PTR


 If the type is smaller than that expected by the procedure, MASM widens the
 argument to match.


 7.3.7.1  Widening Arguments

 For INVOKE to correctly handle type conversions, you must use the signed
 data types for any signed assignments. This list shows the cases in which
 MASM widens an argument to match the type expected by a procedure's
 parameters.

 Type Passed                       Type Expected
 ────────────────────────────────────────────────────────────────────────────
 BYTE, SBYTE                       WORD, SWORD, DWORD, SDWORD

 WORD, SWORD                       DWORD, SDWORD

 When possible, MASM widens arguments to match parameter types.

 The assembler generates instructions such as XOR and CBW to perform the
 conversion. You can see these generated instructions in the listing file by
 using the /Sg command-line option. The assembler can extend a segment if far
 data is expected, and it can convert the type given in the list to the types
 expected. If the assembler cannot convert the type, however, it generates an
 error.


 7.3.7.2  Detecting Errors

 When the assembler widens arguments, it may require the use of a register
 that could overwrite another argument.

 For example, if a procedure with the C calling convention is called with
 this INVOKE statement,

   INVOKE  myprocA, ax, cx, 100, arg

 where  arg  is a BYTE variable and  myproc  expects four arguments of type
 WORD, the assembler widens and then pushes the variable with this code:

   mov     al, DGROUP:arg
           xor     ah, ah
           push    ax

 As a result, the assembler generates code that also uses the AX register and
 therefore overwrites the first argument passed to the procedure in AX. The
 assembler generates an error in this case, requiring you to rewrite the
 INVOKE statement for this procedure.

 The INVOKE directive uses as few registers as possible. However, widening
 arguments or pushing constants on the 8088 and 8086 requires the use of the
 AX register, and sometimes the DX register or the EAX and EDX on the
 80386/486. This means that the content of AL, AH, AX, and EAX must
 frequently be overwritten, so you should avoid using these registers to pass
 arguments. As an alternative you can use DL, DH, DX, and EDX, since these
 registers are rarely used.


 7.3.7.3  Invoking Far Addresses

 You can pass a FAR pointer in a segment::offset pair, as shown below. Note
 the use of double colons to separate the register pair. The registers could
 be any other register pair, including a pair that a DOS call uses to return
 values.

   FPWORD   TYPEDEF FAR PTR WORD
   SomeProc PROTO var1:DWORD, var2:WORD, var3:WORD

           pfaritem    FPWORD     faritem
           .
           .
           .
           les         bx, pfaritem
           INVOKE      SomeProc, ES::BX, arg1, arg2

 However, you cannot give INVOKE two arguments, one for the segment and one
 for the offset, and have INVOKE combine the two for an address.


 7.3.7.4  Passing an Address

 You can use the ADDR operator to pass the address of an expression to a
 procedure that is expecting a NEAR or FAR pointer. This example generates
 code to pass a far pointer (to  arg1) to the procedure  proc1.

   PBYTE   TYPEDEF FAR PTR BYTE
   arg1    BYTE    "This is a string"
   proc1   PROTO   NEAR C fparg:PBYTE
           .
           .
           .
   INVOKE  proc1, ADDR arg1

 See Section 3.3.1 for information on defining pointers with TYPEDEF.


 7.3.7.5  Invoking Procedures Indirectly

 You can make an indirect procedure call such as  call [bx + si]  by using a
 pointer to a function prototype with TYPEDEF, as shown in this example:

   FUNCPROTO       TYPEDEF PROTO NEAR ARG1:WORD, ARG2:WORD
   FUNCPTR         TYPEDEF PTR FUNCPROTO

           .DATA
   pfunc   FUNCPTR OFFSET proc1, OFFSET proc2

           .CODE
           mov     si, Num            ; Num contains 0 or 2
           INVOKE  FUNCPTR PTR [si]   ; Selects proc1 or proc2

 You can also use ASSUME to accomplish the same task. The ASSUME statement
 associates the type  PFUNC  with the BX register.

   ASSUME  BX:FUNCPTR
           mov     si, Num
           INVOKE  FUNCPTR PTR [bx+si]


 7.3.7.6  Checking the Code Generated

 The INVOKE directive generates code that may vary depending on the processor
 mode and calling conventions in effect. You can check your listing files to
 see the code generated by the INVOKE directive if you use the /Sg
 command-line option.


 7.3.8  Generating Prologue and Epilogue Code

 When you use the PROC directive with its extended syntax and argument list,
 the assembler automatically generates the prologue and epilogue code in your
 procedure. "Prologue code" is generated at the start of the procedure; it
 sets up a stack pointer so you can access parameters from within the
 procedure. It also saves space on the stack for local variables, initializes
 registers such as DS, and pushes registers that the procedure uses.
 Similarly, "epilogue code" is the code at the end of the procedure that pops
 registers and returns from the procedure.

 The assembler automatically generates the prologue code when it encounters
 the first instruction after the PROC directive. It generates the epilogue
 code when it encounters a RET or IRET instruction. Using the
 assembler-generated prologue and epilogue code saves you time and decreases
 the number of repetitive lines of code in your procedures.

 The generated prologue or epilogue code depends on the


   ■   Local variables defined

   ■   Arguments passed to the procedure

   ■   Current processor selected (affects epilogue code only)

   ■   Current calling convention

   ■   Options passed in the prologuearg of the PROC directive

   ■   Registers being saved


 The prologuearg list contains options specifying how the prologue or
 epilogue code should be generated. The next section explains how to use
 these options, gives the standard prologue and epilogue code, and explains
 the techniques for defining your own prologue and epilogue code.


 7.3.8.1  Using Automatic Prologue and Epilogue Code

 The standard prologue and epilogue code handles parameters and local
 variables. If a procedure does not have any parameters or local variables,
 the prologue and epilogue code that sets up and restores a stack pointer is
 omitted, unless FORCEFRAME is included in the prologuearg list. (FORCEFRAME
 is discussed later in this section.) Prologue and epilogue code also
 generates a push and pop for each register in the register list unless the
 register list is empty.

 RETN and RETF suppress epilogue code generation.

 When a RET is used without an operand, the assembler generates the standard
 epilogue code. If you do not want the standard epilogue generated, you can
 use RETN or RETF with or without operands. RET with an integer operand does
 not generate epilogue code, but it does generate the right size of return.

 In the examples below showing standard prologue and epilogue code,
 localbytes  is a variable name used in this example to represent the number
 of bytes needed on the stack for the locals declared,  parmbytes  represents
 the number of bytes that the parameters take on the stack, and  registers
 represents the list of registers to be pushed or popped.

 The standard prologue code is the same in any processor mode:

   push bp
           mov bp, sp
           sub sp, localbytes  ; if localbytes is not 0
           push registers

 The standard epilogue code is:

   pop registers
           mov sp, bp    ; if localbytes is not 0
           pop bp
           ret parmbytes ; use parmbytes only if lang is not C

 The standard prologue and epilogue code recognizes two operands passed in
 the prologuearg list, LOADDS and FORCEFRAME. These operands modify the
 prologue code. Specifying LOADDS saves and initializes DS. Specifying
 FORCEFRAME as an argument generates a stack frame even if no arguments are
 sent to the procedure and no local variables are declared. If your procedure
 has any parameters or locals, you do not need to specify FORCEFRAME.

 Specifying LOADDS generates this prologue code:

   push bp
           mov bp, sp
           sub sp, localbytes  ; if localbytes is not 0
           push ds
           mov ax, DGROUP
           mov ds, ax
           push registers

 Specifying LOADDS generates the following epilogue code:

   pop registers
           pop ds
           mov sp, bp
           pop bp
           ret parmbytes ; use parmbytes only if lang is not C


 7.3.8.2  User-Defined Prologue and Epilogue Code

 If you want a different set of instructions for prologue and epilogue code
 in your procedures, you can write macros that are executed instead of the
 standard prologue and epilogue code. For example, while you are debugging
 your procedures, you may want to include a stack check or track the number
 of times a procedure is called. You can write your own prologue code to do
 these things whenever a procedure executes. Different prologue code may also
 be necessary if you are writing applications for Microsoft Windows or any
 other environment application for DOS. User-defined prologue macros will
 respond correctly if you specify FORCEFRAME in the prologuearg of a
 procedure.

 To write your own prologue or epilogue code, the OPTION directive must
 appear in your program. It disables automatic prologue and epilogue code
 generation. When you specify

   OPTION PROLOGUE : macroname

   OPTION EPILOGUE : macroname

 the assembler calls the macro specified in the OPTION directive instead of
 generating the standard prologue and epilogue code. The prologue macro must
 be a macro function, and the epilogue macro must be a macro procedure.

 The assembler expects your prologue or epilogue macro to have this form:

   macroname  MACRO procname, /
                                                   flag, /
                                        parmbytes, /
                                        localbytes, /
                                          <reglist>, /
                                          userparms

 The following list explains the arguments passed to your macro. Your macro
 must have formal parameters to match all the actual arguments passed.

╓┌───────────┌───────────────────────────────┌───────────────────────────────╖
 Argument    Description
 ────────────────────────────────────────────────────────────────────────────
 procname    The name of the procedure.

 flag        A 16-bit flag containing the
             following information:

             Bit = Value                     Description

             Bit 0, 1, 2                     For calling conventions
                                             (000=unspecified language type,
                                             001=C, 010=SYSCALL, 011=
                                             STDCALL, 100=PASCAL, 101=
                                             FORTRAN, 110=BASIC)

             Bit 3                           Undefined (not necessarily
 Argument    Description
 ────────────────────────────────────────────────────────────────────────────
            Bit 3                           Undefined (not necessarily
                                             zero)

             Bit 4                           Set if the caller restores the
                                             stack (Use RET, not RETn)

             Bit 5                           Set if procedure is FAR

             Bit 6                           Set if procedure is PRIVATE

             Bit 7                           Set if procedure is EXPORT

             Bit 8                           Set if the epilogue was
                                             generated as a result of an
                                             IRET instruction and cleared
                                             if the epilogue was generated
                                             as a result of a RET
                                             instruction

 Argument    Description
 ────────────────────────────────────────────────────────────────────────────

             Bits 9-15                       Undefined (not necessarily
                                             zero)

 parmbytes   The byte count of all the
             parameters given in the PROC
             statement.

 localbytes  The count in bytes of all
             locals defined with the LOCAL
             directive.

 reglist     A list of the registers
             following the USES operator in
             the procedure declaration.
             This list is enclosed by angle
             brackets (< >), and each item
             is separated by commas. This
             list is reversed for epilogues.
 Argument    Description
 ────────────────────────────────────────────────────────────────────────────
            list is reversed for epilogues.

 userparms   Any argument you want to pass
             to the macro. The
             prologuearg (if there is one)
             specified in the PROC
             directive is passed to this
             argument.



 Your macro function must return the parmbytes parameter. However, if the
 prologue places other values on the stack after pushing BP and these values
 are not referenced by any of the local variables, the exit value must be the
 number of bytes for procedure locals plus any space between BP and the
 locals. Therefore parmbytes is not always equal to the bytes occupied by the
 locals.

 The following macro is an example of a user-defined prologue that counts the
 number of times a procedure is called.

   ProfilePro      MACRO procname,       \
                         flag,           \
                         bytecount,      \
                         numlocals,      \
                         regs,           \
                         macroargs

                   .DATA
   procname&count  WORD 0
                   .CODE
           inc     procname&count  ; Accumulates count of times the
                                   ;  procedure is called
           push    bp
           mov     bp, sp
                                   ; Other BP operations
           IFNB <regs>
              FOR r, regs
                 push r
              ENDM
           ENDIF
           EXITM %bytecount
   ENDM

 Your program must also include this statement before any procedures are
 called that use the prologue:

   OPTION PROLOGUE:ProfilePro

 If you define only a prologue or an epilogue macro, the standard prologue or
 epilogue code is used for the one you do not define. The form of the code
 generated depends on the .MODEL and PROC options used.

 If you want to revert to the standard prologue or epilogue code, use
 PROLOGUEDEF  or  EPILOGUEDEF  as the macroname in the OPTION statement.

   OPTION EPILOGUE:EPILOGUEDEF

 You can completely suppress prologue or epilogue generation with

   OPTION PROLOGUE:None
   OPTION EPILOGUE:None

 In this case, no user-defined macro is called, and the assembler does not
 generate a default code sequence. This state remains in effect until the
 next OPTION PROLOGUE or OPTION EPILOGUE is encountered.

 See Chapter 9 for additional information about writing macros. The
 PROLOGUE.INC file provided in the MASM 6.0 distribution disks can be used to
 create the prologue and epilogue sequences for the Microsoft C Professional
 Development System, version 6.0.


 7.4  DOS Interrupts

 In addition to jumps, loops, and procedures that alter program execution,
 interrupt routines transfer execution to a different location. In this case,
 control goes to an interrupt routine.

 You can write your own interrupt routines, either to replace an existing
 routine or to use an undefined interrupt number. You may want to replace the
 processor's divide-overflow (0h) interrupts or DOS interrupts, such as the
 critical-error (24h) and CONTROL+C (23h) handlers. The BOUND instruction
 checks array bounds and calls interrupt 5 when an error occurs. If you use
 this instruction, you need to write an interrupt handler for it.

 This section summarizes the following:


   ■   How to call interrupts

   ■   How the processor handles interrupts

   ■   How to redefine an existing interrupt routine


 The example routine in this section handles addition or multiplication
 overflow and illustrates the steps necessary for writing an interrupt
 routine. See Chapter 19, "Writing Memory-Resident Software" for additional
 information about DOS and BIOS interrupts.

 ────────────────────────────────────────────────────────────────────────────
 NOTE
 Under OS/2, system access is made through calls to the Applications Program
 Interface (API), not through interrupts. Microsoft Windows applications use
 both interrupts and API calls.
 ────────────────────────────────────────────────────────────────────────────


 7.4.1  Calling DOS and ROM-BIOS Interrupts

 Interrupts are the only way to access DOS from assembly language. They are
 called with the INT instruction, which takes one operand─an immediate value
 between 0 and 255.

 When calling DOS and ROM-BIOS interrupts, you usually need to place a
 function number in the AH register. You can use other registers to pass
 arguments to functions. Some interrupts and functions return values in
 certain registers, although register use varies for each interrupt. This
 code writes the text of  msg  to the screen.

   .DATA
   msg     BYTE    "This writes to the screen",$
           .CODE
           mov     dx, offset msg
           mov     ah, 09h
           int     21h

 When the INT instruction executes, the processor takes the following six
 steps:


   1.  Looks up the address of the interrupt routine in the interrupt
       descriptor table (also called the "interrupt vector"). This table
       starts at the lowest point in memory (segment 0, offset 0) and
       consists of four bytes (two segment and two offset) for each
       interrupt. Thus, the address of an interrupt routine equals the number
       of the interrupt multiplied by 4.

   2.  Clears the trap flag (TF) and interrupt enable flag (IF).

   3.  Pushes the flags register, the current code segment (CS), and the
       current instruction pointer (IP).

   4.  Jumps to the address of the interrupt routine, as specified in the
       interrupt descriptor table.

   5.  Executes the code of the interrupt routine until it encounters an IRET
       instruction.

   6.  Pops the instruction pointer, code segment, and flags.


 Figure 7.3 illustrates how interrupts work.

 (This figure may be found in the printed book.)

 Some DOS interrupts should not normally be called. Some (such as 20h and
 27h) have been replaced by other DOS interrupts. Others are used internally
 by DOS.


 7.4.2  Replacing or Redefining Interrupt Routines

 One interrupt routine you may want to redefine is the routine called by
 INTO. The INTO (Interrupt on Overflow) instruction is a variation of the INT
 instruction. It calls interrupt 04h when the overflow flag is set. By
 default, the routine for interrupt 4 simply consists of an IRET, so it
 returns without doing anything. Using INTO is an alternative to using JO
 (Jump on Overflow) to jump to an overflow routine.

 To replace or redefine an existing interrupt, your routine must


   ■   Replace the address in the interrupt descriptor table with the address
       of your new routine and save the old address

   ■   Provide new instructions to handle the interrupt

   ■   Restore the old address when your routine ends


 An interrupt routine can be written like a procedure by using the PROC and
 ENDP directives. The routine should always be defined as FAR and should end
 with an IRET instruction instead of a RET instruction.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 Since the assembler doesn't know whether you are going to terminate with
 RET or  IRET, you can use the full extended PROC syntax (described in
 Section 7.3.3, "Declaring Parameters with the PROC Directive") to write
 interrupt procedures. However, you should not make interrupt procedures NEAR
 or specify arguments for them. You can use the USES keyword, however, to
 correctly generate code to save and to restore a register list in interrupt
 procedures.
 ────────────────────────────────────────────────────────────────────────────

 The STI (Set Interrupt Flag) and CLI (Clear Interrupt Flag) instructions
 turn interrupts on or off. You can use CLI to turn off interrupt processing
 so that an important routine cannot be stopped by a hardware interrupt.
 After the routine has finished, use STI to turn interrupt processing back
 on. Interrupts received while interrupt processing was turned off by CLI are
 saved and executed when STI turns interrupts back on.

 MASM 6.0 provides two new forms of the IRET instruction that suppress
 epilogue sequences. This allows an interrupt to have local variables or use
 a userdefined prologue. IRETF pops a FAR16 return address, and IRETFD pops a
 FAR32 return address.

 The following example uses DOS functions to save the address of the initial
 interrupt routine in a variable and to put the address of the new interrupt
 routine in the interrupt descriptor table. Once the new address has been
 set, the new routine is called any time the interrupt is called. This new
 routine prints a message and sets AX and DX to 0.

 To replace the address in the interrupt descriptor table with the address of
 your procedure, AL needs to be loaded with 04h and AH loaded with 35, the
 Get Interrupt Vector function. The Set Interrupt Vector function requires 25
 in AH.

 Follow this example to replace an existing interrupt routine. To write an
 interrupt handler for an unused interrupt, see online help for available
 vectors.

   .MODEL LARGE, C, DOS
   FPFUNC  TYPEDEF FAR PTR
           .DATA
   msg     BYTE    "Overflow - result set to 0",13,10,"$"
   vector  FPFUNC  ?
           .CODE
           .STARTUP

           mov     ax, 3504h       ; Load interrupt 4 and call DOS
           int     21h             ;  Get Interrupt Vector function
           mov     WORD PTR vector[2],es ; Save segment
           mov     WORD PTR vector[0],bx ;  and offset

           push    ds              ; Save DS
           mov     ax, cs          ; Load segment of new routine
           mov     ds, ax
           mov     dx, OFFSET ovrflow   ; Load offset of new routine
           mov     ax, 2504h       ; Load interrupt 4 and call DOS
           int     21h             ;  Set Interrupt Vector function
           pop     ds              ; Restore
           .
           .
           .
           add     ax, bx          ; Do addition (or multiplication)
           into                    ; Call interrupt 4 if overflow
           .
           .
           .
           lds     dx, vector      ; Load original interrupt address
           mov     ax, 2504h       ; Restore interrupt number 4
           int     21h             ;  with DOS set vector function
           mov     ax, 4C00h       ; Terminate function
           int     21h

   ovrflow         PROC    FAR
                   sti             ; Enable interrupts
                                   ;  (turned off by INT)
                   mov     ah, 09h ; Display string function
                   mov     dx, OFFSET msg     ; Load address
                   int     21h     ; Call DOS
                   sub     ax, ax  ; Set AX to 0
                   sub     dx, dx  ; Set DX to 0
                   iret            ; Return
   ovrflow         ENDP
                   END

 Before your program ends, you should restore the original address by loading
 DX with the original interrupt address and using the DOS set vector function
 to store the original address at the correct location.


 7.5  Related Topics in Online Help

 Other information available online which relates to topics in this chapter
 is given in the list below:

 Topic                             Access
 ────────────────────────────────────────────────────────────────────────────
 OPTION directive                  From the "MASM 6.0 Contents" screen,
                                   choose "Directives," then choose
                                   "Miscellaneous"

 DOS and ROM-BIOS interrupts       From the list of System Resources on the
                                   "MASM 6.0 Contents" screen, choose "DOS
                                   Calls" or "BIOS Calls"

 BT, BTC, BTR, BTS                 From the "MASM 6.0 Contents" screen,
                                   choose "Processor Instructions" and then
                                   "Logical and Shifts"

 Other forms of the LOOP           From the "MASM 6.0 Contents" screen,
 instruction                       choose "Processor Instructions" and then
                                   "Control Flow"

 Processor Flag Summary            From the "MASM 6.0 Contents" screen,
                                   choose "Processor Instructions"







 Chapter 8  Sharing Data and Procedures among Modules and Libraries
 ────────────────────────────────────────────────────────────────────────────

 To use symbols and procedures in more than one module, the assembler must be
 able to recognize the shared data as global to all the modules where they
 are used. MASM 6.0 provides new techniques to simplify data-sharing and give
 a high-level interface to multiple-module programming. With these
 techniques, you can place shared symbols in include files. This makes the
 data declarations in the file available to all modules that use the include
 file.

 After an overview of the data-sharing methods, the next section of this
 chapter focuses on organizing modules and using the include file to simplify
 data-sharing. The first method allows you to create a single include file
 that works in the modules where the symbol is used as well as where it is
 defined.

 Sharing procedures and data items using the PUBLIC and EXTERN directives in
 the appropriate modules is the other method of data-sharing. The third
 section of this chapter explains how to use PUBLIC and EXTERN.

 You may also want to place commonly used routines in libraries. Section 8.4
 explains how to create program libraries and access their routines.


 8.1  Selecting Data-Sharing Methods

 If data defined in one module is to be used in the other modules of a
 multiple-module program, the data must be made public and external. MASM
 provides several methods for doing this.

 One method is to declare a symbol public (with the PUBLIC directive) in the
 module where it is defined. This makes the symbol available to other
 modules. Then place an EXTERN statement for that symbol in the rest of the
 modules that use the public symbol. This statement informs the assembler
 that the symbol is external─defined in another module.

 As an alternative, you can use the COMM directive instead of PUBLIC and
 EXTERN. However, communal variables have some limitations. You cannot depend
 on their location in memory because they are allocated by the linker, and
 they cannot be initialized.

 These two data-sharing methods are still available, but MASM 6.0 introduces
 a new directive, EXTERNDEF, that declares a symbol either public or
 external, as appropriate. EXTERNDEF simplifies the declarations for global
 (public and external) variables and encourages the use of include files.

 The next section provides further details on using include files. Section
 8.3, "Using Alternatives to Include Files," provides more information on
 PUBLIC and EXTERN.


 8.2  Sharing Symbols with Include Files

 Place statements common to all modules in include files.

 Include files can contain any valid MASM statement but typically consist of
 type and symbol declarations. The assembler inserts the contents of the
 include file into a module at the location of the INCLUDE directive. Include
 files can simplify project organization by eliminating the need to
 physically insert common declarations into more than one program or module.
 Include files are always optional. See Section 8.3 for alternatives to using
 include files.

 The first part of this section explains how to organize symbol definitions
 and the declarations that make the symbols global (available to all
 modules). It then shows how to make both variables and procedures public
 with EXTERNDEF, PROTO, and COMM. The last part of this section tells where
 to place these directives in the modules and include files.


 8.2.1  Organizing Modules

 This section summarizes the organization of declarations and definitions in
 modules and include files and the use of the INCLUDE directive.

 Include Files - Type declarations that need to be identical in every module
 should be placed in an include file. Doing so ensures consistency and can
 save programming time when updating programs. Include files should contain
 only symbol declarations and any other declarations that are resolved at
 assembly time. (See Section 1.3.1, "Generating and Running Executable
 Programs," for a list of assembly-time operations.) If the include file is
 associated with more than one module, it cannot contain statements that
 define and allocate memory for symbols unless you include the data
 conditionally (see Section 1.3.3).

 Modules - Label definitions that cause the assembler to allocate memory
 space must be defined in a module, not in an include file. If any of these
 definitions is located in the include file, it is copied into each file that
 uses the include file, creating an error.

 Include files are inserted at the location of the INCLUDE directive.

 Once you have placed public symbols in an include file, you need to
 associate that file with the main module. The INCLUDE statement is usually
 placed before data and code segments in your modules. When the assembler
 encounters an INCLUDE directive, it opens the specified file and assembles
 all its statements. The assembler then returns to the original file and
 continues the assembly process.

 The INCLUDE directive takes the form

   INCLUDE filename

 where filename is the full name or fully specified path of the include file.
 For example, the following declaration inserts the contents of the include
 file SCREEN.INC in your program:

   INCLUDE SCREEN.INC

 You must make sure that the assembler can find include files.

 The file name in the INCLUDE directive must be fully specified; no
 extensions are assumed. If a full path name is not given, the assembler
 searches first in the directory of the source file containing the INCLUDE
 directive.

 If the include file is not in the source file directory, the assembler
 searches the paths specified in the assembler's command-line option /I, or
 in PWB's Include Paths field in the MASM Option dialog box (accessed from
 the Option menu). The /I option takes this form:

   /I path

 Multiple /I options can be used to specify that multiple directives be
 searched in the order they appear on the command line. If none of these
 directories contains the desired include file, the assembler finally
 searches in the paths specified in the INCLUDE environment variable. If the
 include file still cannot be found, an assembly error occurs. The related /x
 option tells the assembler to ignore the INCLUDE environment variable for
 all subsequent assemblies.

 An include file may specify another include file. The assembler processes
 the second include file before returning to the first. Include files can be
 nested this way as deeply as desired; the only limit is the amount of free
 memory.

 Put constants used in more than one module into the include file.

 Include Files or Modules - You can use the EQU directive to create named
 constants that cannot be redefined in your program (see Section 1.2.4,
 "Integer Constants and Constant Expressions," for information about the EQU
 directive). Placing a constant defined with EQU in an include file makes it
 available to all modules that use that include file.

 Placing TYPEDEF, STRUCT, UNION, and RECORD definitions in an include file
 guarantees consistency in type definitions. If required, the variable
 instances derived from these definitions can be made public among the
 modules with EXTERNDEF declarations (see the next section). Macros
 (including macros defined with TEXTEQU) must be placed in include files to
 make them visible in other modules.

 If you elect to use full segment definitions (along with, or instead of,
 simplified definitions), you can force a consistent segment order in all
 files by defining segments in an include file. This technique is explained
 in Section 2.3.2, "Controlling the Segment Order."


 8.2.2  Declaring Symbols Public and External

 It is sometimes useful to make procedures and variables (such as large
 arrays or status flags) global to all program modules. Global variables are
 freely accessible within all routines; you do not have to explicitly pass
 them to the routines that need them.

 Variables can be made global to multiple modules in several ways. This
 section describes three ways to make them global by using the EXTERNDEF,
 PROTO, or COMM declarations within include files. Section 8.3.1 explains how
 to use the PUBLIC and EXTERN directives within modules.

 External identifiers must be unique.

 These methods make symbols global to the modules in which they are used.
 Therefore, symbols must be unique. The linker enforces this requirement.


 8.2.2.1  Using EXTERNDEF

 EXTERNDEF can appear in the defining or calling modules.

 MASM treats EXTERNDEF as a public declaration in the defining module and as
 an external declaration in accessing module(s). You can use the EXTERNDEF
 statement in your include file to make a variable common among two or more
 modules. EXTERNDEF works with all types of variables, including arrays,
 structures, unions, and records. It also works with procedures.

 As a result, a single include file can contain an EXTERNDEF declaration that
 works in both the defining module and any accessing module. It is ignored in
 modules that neither define nor access the variable. Therefore, an include
 file for a library which is used in multiple .EXE files does not force the
 definition of a symbol as EXTERN does.

 The EXTERNDEF statement takes this form:

   EXTERNDEF [[langtype]] name:qualifiedtype

 The name is the variable's identifier. The qualifiedtype is explained in
 detail in Section 1.2.6, "Data Types."

 The optional langtype specifier sets the naming conventions for the name it
 precedes. It overrides any language specified in the .MODEL directive. The
 specifier can be C, SYSCALL, STDCALL, PASCAL, FORTRAN, or BASIC. See Section
 20.1, "Naming and Calling Conventions," for information on selecting the
 appropriate langtype type.

 The diagram below shows the statements that declare an array, make it
 public, and use it in another module.

 (This figure may be found in the printed book.)

 The file position of EXTERNDEF directives is important. See Section 8.2.3,
 "Positioning External Declarations," for more information.

 The assembler does not check parameters when you call EXTERNDEF procedures.


 You can also make procedures visible by using EXTERNDEF without PROTO inside
 an include file. This method treats the procedure name as a simple
 identifier, without the parameter list, so you forgo the assembler's ability
 to check for the correct parameters during assembly.

 The method for using EXTERNDEF for procedures is the same as using it with
 variables. You can also use EXTERNDEF to make code labels global.


 8.2.2.2  Using PROTO

 When a procedure is defined in one module and called from another module, it
 must be declared public in the defining module and external in the calling
 modules; otherwise, assembly or linking errors occur.

 You have three methods for declaring a procedure public. Using PUBLIC and
 EXTERN is the only method prior to MASM 6.0. Section 8.3.1 explains the use
 of PUBLIC and EXTERN. The previous section (8.2.2.1) explains the use of
 EXTERNDEF. This section illustrates the use of PROTO.

 A PROTO (prototype) declaration in the include file establishes a
 procedure's interface in both the defining and calling modules. The PROTO
 directive automatically generates an EXTERNDEF for the procedure unless the
 procedure has been declared PRIVATE in the PROC statement. Defining a
 prototype enables type-checking for the procedure arguments.

 PROTO and INVOKE simplify procedure calls.

 Follow these steps to create an interface for a procedure defined in one
 module and called from other modules:


   1.  Place the PROTO declaration in the include file.

   2.  Define the procedure with PROC. The PROC directive declares the
       procedure PUBLIC by default.

   3.  Call the procedure with the INVOKE statement (or with CALL).


 The following example is a PROTO declaration for the far procedure
 CopyFile, which uses the C parameter-passing and naming conventions, and
 takes the arguments  filename  and  numberlines. The diagram following the
 example shows the file placement for these statements. This definition goes
 into the include file:

   CopyFile PROTO FAR C filename:BYTE, numberlines:WORD

 The procedure definition for  CopyFile  is

   CopyFile PROC FAR C USES cx, filename:BYTE, numberlines:WORD

 To call the  CopyFile  procedure, you can use this INVOKE statement:

   INVOKE   CopyFile, NameVar, 200

 (This figure may be found in the printed book.)

 See Chapter 7, "Controlling Program Flow," for descriptions, syntax, and
 examples of PROTO, PROC, and INVOKE.


 8.2.2.3  Using COMM

 Another way to share variables among modules is to add the COMM (communal)
 declaration to your include file. Since communal variables are allocated by
 the linker and cannot be initialized, you cannot depend on their location or
 sequence.

 Communal variables are supported by MASM primarily for compatibility with
 communal variables in Microsoft C. Communal variables are not used in any
 other Microsoft language, and they are not compatible with C++ and some
 other languages.

 Communal variables can reduce the size of executable files.

 COMM declares a variable external but cannot be used with code. COMM also
 instructs the linker to define the variable if it has not been explicitly
 defined in a module. The memory space for communal variables may not be
 assigned until load time, so using communal variables may reduce the size of
 your executable file.

 The COMM declaration has the syntax

   COMM [[langtype]] [[NEAR
   | FAR]] label:type«:count»

 The label is the name of the variable. The langtype sets the naming
 conventions for the name it precedes. It overrides any language specified in
 the .MODEL directive.

 If NEAR or FAR is not specified, the variable determines the default from
 the current memory model (NEAR for TINY, SMALL, COMPACT, and FLAT; FAR for
 MEDIUM, LARGE, and HUGE).

 The type can be a constant expression, but it is usually a type such as
 BYTE, WORD, or DWORD, or a structure, union, or record. If you first declare
 the type with TYPEDEF, CodeView can provide type information. The count is
 the number of elements. If no count is given, one element is assumed.

 The following example creates the common far variable  DataBlock, which is a
 1,024-element array of uninitialized signed doublewords:

   COMM FAR DataBlock:SDWORD:1024

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 C variables declared outside functions (except static variables) are
 communal unless explicitly initialized; they are the same as
 assembly-language communal variables. If you are writing assembly-language
 modules for C, you can declare the same communal variables in both C and
 MASM include files. However, communal variables in C do not have to be
 declared communal in assembler. The linker will match the EXTERN, PUBLIC,
 and COMM statements for the variable.
 ────────────────────────────────────────────────────────────────────────────

 EXTERNDEF is a flexible alternative to using COMM.

 EXTERNDEF (explained in the previous section) is more flexible than COMM
 because you can initialize variables defined with it, and you can use those
 variables in code that depends on the position and sequence of the data.


 8.2.3  Positioning External Declarations

 Although LINK determines the actual address of an external symbol, the
 assembler assumes a default segment for the symbol, based on the location of
 the external directive in the source code. You should therefore position
 EXTERN and EXTERNDEF directives according to these rules:


   ■   If you know which segment defines an external symbol, put the EXTERN
       statement in that segment.

   ■   If you know the group but not the segment, position the EXTERN
       statement outside any segment and reference the variable with the
       group name. For example, if  var1  is in DGROUP, you would reference
       the variable as  mov DGROUP:var1, 10.

   ■   If you know nothing about the location of an external variable, put
       the EXTERN statement outside any segment. You can use the SEG
       directive to access the external variable like this:

       mov ax, SEG var1
             mov es, ax
             mov ax, es:var1


   ■   If the symbol is an absolute symbol or a far code label, you can
       declare it external anywhere in the source code.


 Always close opened segments.

 Any segments opened in include files should always be closed so that
 external declarations following an include statement are not incorrectly
 placed inside a segment. Any include statements in your program should
 immediately follow the .MODEL, OPTION, and processor directives.

 For the same reason, if you want to be certain that an external definition
 is outside a segment, you can use @CurSeg. The @CurSeg predefined symbol
 returns a blank if the definition is not in a segment. For example,

   .DATA
           .
           .
           .
   @CurSeg ENDS                    ; Close segment
           EXTERNDEF var:WORD

 See Section 1.2.3, "Predefined Symbols," for information about predefined
 symbols such as @CurSeg.


 8.3  Using Alternatives to Include Files

 If your project uses only two modules (or if it is written with a version of
 MASM prior to 6.0), you may want to continue using PUBLIC in the defining
 module and EXTERN in the accessing module, and not create an include file
 for the project. The EXTERN directive can be used in an include file, but
 the include file containing EXTERN cannot be added to the module that
 contains the corresponding PUBLIC directive for that symbol. This section
 assumes that you are not using include files.


 8.3.1  PUBLIC and EXTERN

 The PUBLIC and EXTERN directives are less flexible than EXTERNDEF and PROTO
 because they are module-specific: PUBLIC must appear in the defining module
 and EXTERN must appear in the calling modules. This section shows how to use
 PUBLIC and EXTERN. Information on where to place the external declarations
 in your file is in Section 8.2.3, "Positioning External Declarations."

 The PUBLIC directive makes a name visible outside the module in which it is
 defined. This gives other program modules access to that identifier.

 The EXTERN directive performs the complementary function. It tells the
 assembler that a name referenced within a particular module is actually
 defined and declared public in another module that will be specified at link
 time.

 A PUBLIC directive can appear anywhere in a file. Its syntax is

   PUBLIC [[langtype]] name[[,
   [[langtype]] name]] ...

 The name must be the name of an identifier defined within the current source
 file. Only code labels, data labels, procedures, and numeric equates can be
 declared public.

 If you specify the langtype field here, it overrides the language specified
 by .MODEL. The langtype field can be C, SYSCALL, STDCALL, PASCAL, FORTRAN,
 or BASIC. Section 7.3.3, "Declaring Parameters with the PROC Directive," and
 Section 20.1, "Naming and Calling Conventions," provide more information on
 specifying langtype types.

 The EXTERN directive tells the assembler that an identifier is
 external─defined in some other module that will be supplied at link time.
 Its syntax is

   EXTERN «langtype» name:{ABS | qualifiedtype}

 Section 1.2.6, "Data Types," describes qualifiedtype. The ABS (absolute)
 keyword can be used only with external numeric constants. ABS causes the
 identifier to be imported as a relocatable unsized constant. This identifier
 can then be used anywhere a constant can be used. If the identifier is not
 found in another module at link time, the linker generates an error.

 In the following example, the procedure  BuildTable  and the variable  Var
 are declared public. The procedure uses the Pascal naming and data-passing
 conventions:

 (This figure may be found in the printed book.)


 8.3.2  Other Alternatives

 You can also use the directives discussed earlier (EXTERNDEF, PROTO, and
 COMM) without the include file. In this case, place the declarations to make
 a symbol global in the same module where the symbol is defined. You might
 want to use this technique if you are linking only a few modules that have
 very little data in common.


 8.4  Developing Libraries

 As you create reusable procedures, you can place them in a library file for
 convenient access. Although you can put any routine into a library, each
 library usually contains related routines. For example, you might place
 string-manipulation functions in one library, matrix calculations in
 another, and port communications in another.

 A library consists of combined object modules, each created from a single
 source file. The object module is the smallest independent unit in a
 library. If you link with one symbol in a module, you get the entire module,
 but not the entire library.

 A library can consist of two files─an include file containing necessary
 declarations and constants and a .LIB file containing procedures already
 assembled into object code.


 8.4.1  Associating Libraries with Modules

 You can choose either of two methods for associating your libraries with the
 modules that use them: you can use the INCLUDELIB directive inside your
 source files or link the modules from the command line.

 Specify library names with INCLUDELIB.

 To associate a specified library with your object code, use INCLUDELIB. You
 can add this directive to the source file to specify the libraries you want
 linked, rather than specifying them in the LINK command line. The INCLUDELIB
 syntax is

   INCLUDELIB libraryname

 The libraryname can be a file name or a complete path specification. If you
 do not specify an extension, .LIB is assumed. The libraryname is placed in
 the comment record of the object file. LINK reads this record and links with
 the specified library file.

 For example, the statement  INCLUDELIB GRAPHICS  passes a message from the
 assembler to the linker telling LINK to use library routines from the file
 GRAPHICS.LIB. If this statement is in the source file DRAW.ASM and
 GRAPHICS.LIB is in the same directory, the program can be assembled and
 linked with the following command line:

   ML DRAW.ASM

 Link libraries with command-line options.

 Without the INCLUDELIB directive, the program DRAW.ASM has to be linked with
 either of the following command lines:

   ML DRAW.ASM GRAPHICS.LIB
   ML DRAW /link GRAPHICS

 If you want to assemble and link separately, you can use

   ML /c DRAW.ASM
   LINK DRAW,,,GRAPHICS

 LINK searches in a specific order.

 If you do not specify a complete path in the INCLUDELIB statement or at the
 command line, LINK searches for the library file in the following order:


   1.  In the current directory

   2.  In any directories in the library field of the LINK command line

   3.  In any directories in the LIB environment variable


 The LIB utility provided with MASM 6.0 helps you create, organize, and
 maintain run-time libraries.


 8.4.2  Using EXTERN with Library Routines

 In some cases, EXTERN helps you limit the size of your executable file by
 specifying in the syntax an alternative name for a procedure. You would use
 this form of the EXTERN directive when declaring a procedure or symbol that
 may not need to be used.

 The syntax looks like this:

   EXTERN «langtype» name « (altname)
   » :qualifiedtype

 The addition of the altname to the syntax provides the name of an alternate
 procedure that the linker uses to resolve the external reference if the
 procedure given by name is not needed. Both name and altname must have the
 same qualifiedtype.

 When the linker encounters an external definition for a procedure that gives
 an altname, the linker finishes processing that module before it links the
 object module that contains the procedure given by name. If the program does
 not reference any symbols in the name file's object from any of the linked
 modules, the assembler uses altname to satisfy the external reference. This
 saves space because the library object module is not brought in.

 For example, assume that the contents of STARTUP.ASM include these
 statements:

   EXTERN  init(dummy)
           .
           .
           .
   dummy   PROC
           .
           .
           .                     ; A procedure definition containing
   no
           ret                   ;  executable code

   dummy   ENDP
           .
           .
           .
           call   init   ; Defined in FLOAT.OBJ

 In this example, the reference to the routine  init  (defined in FLOAT.OBJ)
 does not force the module FLOAT.OBJ to be linked into the executable file.
 If another reference causes FLOAT.OBJ to be linked into the executable file,
 then  init  will refer to the  init  label in FLOAT.OBJ. If there are no
 references which force FLOAT.OBJ to be loaded, then the alternate name for
 init(dummy)  will be used by the linker.


 8.5  Related Topics in Online Help

 In addition to information covered in this chapter, information on the
 following topics can be found in online help.

 Topic                             Access
 ────────────────────────────────────────────────────────────────────────────
 LIB                               From the "Microsoft Advisor Contents"
                                   screen, choose "LIB" from the list of
                                   Microsoft Utilities

 INCLUDE, INCLUDELIB,              From the "MASM 6.0 Contents" screen,
 EXTERNDEF, COMM, and              choose "Directives," then "Scope and
 PUBLIC                            Visibility"

 TYPEDEF                           From the "MASM 6.0 Contents" screen,
                                   choose "Directives," then "Complex Data
                                   Types"

 PROTO and INVOKE                  From the "MASM 6.0 Contents" screen,
                                   choose "Directives," then "Procedures
                                   and Code Labels"

 OPTION directive                  From the "MASM 6.0 Contents" screen,
                                   choose "Directives," then "Miscellaneous"

 @CurSeg                           From the "MASM 6.0 Contents" screen,
                                   choose "Predefined Symbols"

 PWB Options menu                  From the "Microsoft Advisor Contents"
                                   screen, choose "Programmer's WorkBench"






 Chapter 9  Using Macros
 ────────────────────────────────────────────────────────────────────────────

 A "macro" is a symbolic name you give to a series of characters (a text
 macro) or to one or more statements (a macro procedure or function). As the
 assembler evaluates each line of your program, it scans the source code for
 names of previously defined macros. When it finds one, it substitutes the
 macro text for the macro name. In this way, you can avoid writing the same
 code several places in your program.

 This chapter describes the following types of macros:


   ■   Text macros, which expand to text within a source statement

   ■   Macro procedures, which expand to one or more complete statements and
       can optionally take parameters

   ■   Repeat blocks, which generate a group of statements a specified number
       of times or until a specified condition becomes true

   ■   Macro functions, which look like macro procedures and can be used like
       text macros but which also return a value

   ■   Predefined macro functions and string directives, which perform string
       operations


 Macro processing is a text-processing mechanism that is done sequentially at
 assembly time. By the end of assembly, all macros have been expanded and the
 resulting text assembled into object code.

 This chapter shows how to use macros for simple code substitutions as well
 as how to write sophisticated macros with parameter lists and repeat loops.
 It also describes how to use these features in conjunction with local
 symbols, macro operators, and predefined macro functions.


 9.1  Text Macros

 You can give a sequence of characters a symbolic name and then use the name
 in place of the text later in the source code. The named text is called a
 text macro.

 The syntax for defining a text macro is

   name TEXTEQU <text>
   name TEXTEQU macroId | textmacro
   name TEXTEQU %constExpr


 where text is a sequence of characters enclosed in angle brackets, macroId
 is a previously defined macro function (see Section 9.6), textmacro is a
 previously defined text macro, and %constExpr is an expression that
 evaluates to text. The use of angle brackets to delimit text is discussed in
 more detail in Section 9.3.1, and the % operator is explained in Section
 9.3.2.

 Here are some examples:

   msg     TEXTEQU <Some text>         ; Text assigned to symbol
   string  TEXTEQU msg                 ; Text macro assigned to symbol
   msg     TEXTEQU <Some other text>   ; New text assigned to symbol
   value   TEXTEQU %(3 + num)          ; Text representation of
                                       ;  resolved expression assigned
                                       ;  to symbol

 In the first line, text is assigned to the symbol  msg. In the second line,
 the text of the  msg  text macro is assigned to a new text macro called
 string. In the third line, new text is assigned to  msg. The result is that
 msg  has the new text value, while  string  has the original text value. The
 fourth line assigns  7  to  value  if  num  equals  4. If a text macro
 expands to another text macro (or macro function, which is discussed in
 Section 9.6), the resulting text macro will be recursively expanded.

 Text macros are useful for naming strings of text that do not evaluate to
 integers. For example, you might use a text macro to name a floating-point
 constant or a bracketed expression. Here are some practical examples:

   pi      TEXTEQU <3.1416>            ; Floating point constant
   WPT     TEXTEQU <WORD PTR>          ; Sequence of key words
   arg1    TEXTEQU <[bp+4]>            ; Bracketed expression

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 Use of the TEXTEQU directive to define text macros is new in MASM 6.0. In
 previous versions, you can use the EQU directive for the same purpose. If
 you have old code that worked under previous versions, it should still work
 under 6.0. However, the more consistent and flexible TEXTEQU is recommended
 for new code.
 ────────────────────────────────────────────────────────────────────────────


 9.2  Macro Procedures

 If your program needs to perform the same task many times, you can avoid
 having to type the same statements each time by writing a macro procedure.
 Macro procedures (commonly called macros) can be seen as text-processing
 mechanisms that automatically generate repeated text.

 The term "macro procedure" rather than macro is used when necessary to
 distinguish between macro procedures and macro functions (a new feature of
 MASM 6.0 described in Section 9.6, "Returning Values with Macro Functions").



 9.2.1  Creating Macro Procedures

 To define a macro procedure without parameters, place the desired statements
 between the MACRO and ENDM directives:

 name MACRO statements ENDM

 For example, suppose you want a program to beep when it encounters certain
 errors. A  beep  macro can be defined as follows:

   beep    MACRO
       mov  ah, 2          ;; Select DOS Print Char function
       mov  dl, 7          ;; Select ASCII 7 (bell)
       int  21h            ;; Call DOS
   ENDM

 Macro comments must start with two semicolons instead of one.

 The double semicolons mark the beginning of macro comments. Macro comments
 appear in a listing file only at the macro's initial definition, not at the
 point where it is called and expanded. Listings are usually easier to read
 if the comments aren't always expanded. Regular comments (those with a
 single semicolon) are listed in macro expansions. Appendix C discusses
 listing files and shows examples of how macros are expanded in listings.

 Once a macro is defined, you can call it anywhere in the program by using
 the macro's name as a statement. The following example calls the  beep
 macro two times if an error flag has been set.

   .IF     error   ; If error flag is true
           beep            ;  execute macro two times
           beep
           .ENDIF

 The instructions in the macro take the place of the macro call when the
 program is assembled. This would be the resulting code (from the listing
 file):

   .IF     error
   0017  80 3E 0000 R 00   *         cmp    error, 000h
   001C  74 0C             *         je     @C0001
                              beep
   001E  B4 02               1         mov     ah, 2
   0020  B2 07               1         mov     dl, 7
   0022  CD 21               1         int     21h
                              beep
   0024  B4 02               1         mov     ah, 2
   0026  B2 07               1         mov     dl, 7
   0028  CD 21               1         int     21h
                              .ENDIF
   002A                  *@C0001:

 Contrast this with the results of defining  beep  as a procedure using the
 PROC directive and then calling it using the CALL instruction. The
 instructions of the procedure occur only once in the executable file, but
 you would also have the additional overhead of the CALL and RET
 instructions.

 Macros are usually faster than run-time procedures.

 In some cases the same task can be done with either a macro or a procedure.
 Macros are potentially faster because they have less overhead, but they
 generate the same code multiple times rather than just once.


 9.2.2  Passing Arguments to Macros

 Parameters allow macros to execute variations of a general task.

 By defining parameters for macros, you can define a general task and then
 execute variations of it by passing different arguments each time you call
 the macro. The complete syntax for a macro procedure includes a parameter
 list:

 name MACRO parameterlist statements ENDM

 The parameterlist can contain any number of parameters. Use commas to
 separate each parameter in the list. Parameter names cannot be reserved
 words unless the keyword has been disabled with OPTION NOKEYWORD, the
 compatibility modes have been set by specifying OPTION M510 (see Section
 1.3.2), or the /Zm command-line option has been set.

 To pass arguments to a macro, place the arguments after the macro name when
 you call the macro:

 macroname arglist

 All text between matching quotation marks in an arglist is considered one
 text item.

 The  beep  macro introduced in the last section used the DOS interrupt to
 write the bell character (ASCII 7). It can be rewritten with a parameter to
 specify any character to write.

   writechar MACRO char
       mov  ah, 2                  ;; Select DOS Print Char function
       mov  dl, char               ;; Select ASCII char
       int  21h                    ;; Call DOS
   ENDM

 Wherever  char  appears in the macro definition, the assembler replaces it
 with the argument in the macro call. Each time you call  writechar, you can
 print a different value:

   writechar 7             ; Causes computer to beep
           writechar 'A'           ; Writes A to screen

 If you pass more arguments than there are parameters, the additional
 arguments generate a warning (unless you use the VARARG keyword; see Section
 9.4.3). If you pass fewer arguments than the macro procedure expects,
 remaining parameters are assigned empty strings (unless default values have
 been specified). This may cause errors. For example, if you call the
 writechar  macro with no argument, it generates the following:

   mov     dl,

 The assembler generates an error for the expanded statement but not for the
 macro definition or the macro call.

 Macros can be made more flexible by leaving off macro arguments or adding
 additional ones. The next section tells some of the ways you can handle
 missing or extra arguments.


 9.2.3  Specifying Required and Default Parameters

 You can specify required and default parameters for macros.

 You can give macro parameters special attributes to make them more flexible
 and improve error handling; you can make them required, give them default
 values, or vary their number. Because variable parameters are used almost
 exclusively with the FOR directive, discussion of them is postponed until
 Section 9.4.3, "FOR Loops and Variable-Length Parameters."

 The syntax for a required parameter is

   parameter:REQ

 For example, you can rewrite the  writechar  macro to require the  char
 parameter:

   writechar MACRO char:REQ
       mov  ah, 2                  ;; Select DOS Print Char function
       mov  dl, char               ;; Select ASCII char
       int  21h                    ;; Call DOS
   ENDM

 If the call does not include a matching argument, the assembler reports the
 error in the line that contains the macro call. The effect of REQ is to
 improve error reporting.

 A default value fills in missing parameters.

 Another way to handle missing parameters is to specify a default value. The
 syntax is

   parameter:=textvalue

 Suppose that you often use  writechar  to beep by printing ASCII 7. The
 following macro definition uses an equal sign to tell the assembler to
 assume the parameter  char  is  7  unless you specify otherwise:

   writechar  MACRO char:=<7>
       mov  ah, 2                  ;; Select DOS Print Char function
       mov  dl, char               ;; Select ASCII char
       int  21h                    ;; Call DOS
   ENDM

 In this case,  char  is not required. If you don't supply a value, the
 assembler fills in the blank with the default value of  7  and the macro
 beeps when called.

 The default parameter value is enclosed in angle brackets so that the
 supplied value will be recognized as a text value. Section 9.3.1, "Text
 Delimiters (< >) and the Literal-Character Operator (!)," explains this in
 more detail.

 Missing arguments can also be handled with the IFB, IFNB, .ERRB, and .ERRNB
 directives. They are described briefly in Section 1.3.3, "Conditional
 Directives," and in online help. Here is a slightly more complex macro that
 uses some of these techniques.

   Scroll MACRO distance:REQ, attrib:=<07h>, tcol, trow, bcol,
   brow
       IFNB <tcol>             ;; Ignore arguments if blank
           mov   cl, tcol
       ENDIF
       IFNB <trow>
           mov   ch, trow
       ENDIF
       IFNB <bcol>
           mov   dl, bcol
       ENDIF
       IFNB <brow>
           mov   dh, brow
       ENDIF
       IFDIFI <attrib>, <bh>   ;; Don't move BH onto itself
           mov   bh, attrib
       ENDIF
       IF distance LE 0        ;; Negative scrolls up, positive down
           mov   ax, 0600h + (-(distance) AND 0FFh)
       ELSE
           mov   ax, 0700h + (distance AND 0FFh)
       ENDIF
       int   10h
   ENDM

 In this macro, the  distance  parameter is required. The  attrib  parameter
 has a default value of  07h  (white on black), but the macro also tests to
 make sure the corresponding argument isn't BH, since it would be inefficient
 (though legal) to load a register onto itself. The IFNB directive is used to
 test for blank arguments. These are ignored to allow the user to manipulate
 rows and columns directly in registers CX and DX at run time.

 The following are two valid ways to call the macro:

   ; Assume DL and CL already loaded
           dec     dh                   ; Decrement top row
           inc     ch                   ; Increment bottom row
           Scroll -3                    ; Scroll white on black dynamic
                                        ;  window up three lines
           Scroll 5, 17h, 2, 2, 14, 12  ; Scroll white on blue constant
                                        ;  window down five lines

 This macro can generate completely different code, depending on its
 arguments. In this sense, it is not comparable to a procedure, which always
 has the same code regardless of arguments.


 9.2.4  Defining Local Symbols in Macros

 You can make a symbol local to a macro by declaring it at the start of the
 macro with the LOCAL directive. Any identifier may be declared local.

 You can choose whether you want numeric equates and text macros to be local
 or global. If a symbol will be used only inside a particular macro, you can
 declare it local so that the name will be available for other declarations
 inside other macros or at the global level. On the other hand, it is
 sometimes convenient to define text macros and equates that are not local,
 so that their values can be shared between macros.

 If you need to use a label inside a macro, you must declare it local, since
 a label can occur only once in the source. The LOCAL directive makes a
 special instance of the label each time the macro is called. This prevents
 redefinition of the label.

 All local symbols must be declared immediately following the MACRO statement
 (although blank lines and comments may precede the local symbol). Separate
 each symbol with a comma. Comments are allowed on the LOCAL statement.
 Multiple LOCAL statements are also permitted. Here is an example macro that
 declares local labels:

   power   MACRO   factor:REQ, exponent:REQ
       LOCAL   again, gotzero      ;; Local symbols
       sub     dx, dx              ;; Clear top
       mov     ax, 1               ;; Multiply by one on first loop
       mov     cx, exponent        ;; Load count
       jcxz    gotzero             ;; Done if zero exponent
       mov     bx, factor          ;; Load factor
   again:
       mul     bx                  ;; Multiply factor times exponent
       loop    again               ;; Result in AX
   gotzero:
   ENDM

 If the labels  again  and  gotzero  were not declared local, the macro would
 work the first time it is called, but it would generate redefinition errors
 on subsequent calls. MASM implements local labels by generating different
 names for them each time the macro is called. You can see this in listing
 files. The labels in the  power  macro might be expanded to  ??0000  and
 ??0001  on the first call and to  ??0002  and  ??0003  on the second.


 9.3  Assembly Time Variables and Macro Operators

 In writing macros, you will often assign and modify values assigned to
 symbols. These symbols can be thought of as assembly-time variables. Like
 memory variables, they are symbols that represent values. But since macros
 are processed at assembly time, any symbol modified in a macro must be
 resolved as a constant by the end of assembly.

 The three kinds of assembly-time variables are:


   ■   Macro parameters

   ■   Text macros

   ■   Macro functions


 When a macro is expanded, the symbols are processed in the order shown
 above. First macro parameters are replaced with the text of their actual
 arguments. Then text macros are expanded.

 Macro parameters are similar to procedure parameters in some ways, but they
 also have important differences. In a procedure, a parameter has a type and
 a memory location. Its value can be modified within the procedure. In a
 macro, a parameter is a placeholder for the argument text. The value can
 only be assigned to another symbol or used directly; it cannot be modified.
 The macro may interpret the argument text it receives either as a numeric
 value or as a text value.

 It is important to understand the difference between text values and numeric
 values. Numeric values can be processed with arithmetic operators and
 assigned to numeric equates. Text values can be processed with macro
 functions and assigned to text macros.

 Macro operators are often helpful when processing assembly-time variables.
 Table 9.1 shows the macro operators that MASM provides:

 Table 9.1  MASM Macro Operators

 Symbol  Name                              Description
 ────────────────────────────────────────────────────────────────────────────
 < >     Text Delimiters                   Opens and closes a literal
                                           string.

 !       Literal-Character Operator        Treats the next character as a
                                           literal character, even if it
                                           would normally have another
                                           meaning.

 %       Expansion Operator                Causes the assembler to expand a
                                           constant expression or text
                                           macro.

 &       Substitution Operator             Tells the assembler to replace a
                                           macro parameter or text macro
                                           name with its
                                           actual value.

 ────────────────────────────────────────────────────────────────────────────


 The next sections explain these operators in detail.


 9.3.1  Text Delimiters (< >) and the Literal-Character Operator (!)

 The angle brackets (< >) are text delimiters. The most common reason to
 delimit a text value is when assigning a text macro. You can do this with
 TEXTEQU, as previously shown, or with the SUBSTR and CATSTR directives
 discussed in Section 9.5, "String Directives and Predefined Functions."

 By delimiting the text of macro arguments, you can pass text that includes
 spaces, commas, semicolons, and other special characters. In the following
 example, assume you have previously defined a macro called  work:

   work    <1, 2, 3, 4, 5> ; Passes one argument
                                   ;  with 15 characters
           work    1, 2, 3, 4, 5   ; Passes five arguments, each
                                   ;  with 1 character

 Since angle brackets are delimiters, you can't include them as part of a
 delimited text value. The literal-character operator (!) can be used to
 override this limitation. It forces the assembler to treat the character
 following it literally rather than as a special character.

   errstr  TEXTEQU <Expression !> 255>  ; errstr = "Expression
   > 255"

 Text delimiters also have a special use with the FOR directive, as explained
 in Section 9.4.3.


 9.3.2  Expansion Operator (%)

 The expansion operator (%) expands text macros or converts constant
 expressions into their text representations. It performs these tasks
 differently in different contexts, as discussed below.


 9.3.2.1  The Expansion Operator with Constants

 The expansion operator can be used in any context where a text value is
 expected but a numeric value is supplied. In these contexts, it can be
 thought of as a conversion operator to convert numeric values to text
 values.

 The expansion operator forces immediate evaluation of a constant expression
 and replaces it with a text value consisting of the digits of the result.
 The digits are generated in the current radix (default decimal).

 This application of the expansion operator is useful when defining a text
 macro:

   a       TEXTEQU <3 + 4>         ; a = "3 + 4"
   b       TEXTEQU %3 + 4          ; b = "7"

 When assigning text macros, numeric equates can be used in the constant
 expressions, but text macros cannot:

   num     EQU     4               ; num = 4
   numstr  TEXTEQU <4>             ; numstr = <4>
   a       TEXTEQU %3 + num        ; a = <7>
   b       TEXTEQU %3 + numstr     ; b = <7>

 The expansion operator can be used when passing macro arguments. If you want
 the value rather than the text of an expression to be passed, use the
 expansion operator. Use of the expansion operator depends on whether you
 want the expression to be evaluated inside the macro on each use, or outside
 the macro once. The following macro

   work    MACRO   arg
       mov ax, arg * 4
   ENDM

 can be called with these statements:

   work    2 + 3           ; Passes "2 + 3"
                                   ; Code: mov ax, 2 + 3 * 4 (14)
           work    %2 + 3          ; Passes 5
                                   ; Code: mov ax, 5 * 4 (20)

 Notice that because of operator precedence, results can vary depending on
 whether the expansion operator is used. Sometimes parentheses can be used
 inside the macro to force evaluation in a particular order:

   work    MACRO   arg
       mov ax, (arg) * 4
   ENDM

           work    2 + 3           ; Code: mov ax, (2 + 3) * 4 (20)
           work    %2 + 3          ; Code: mov ax, (5) * 4 (20)

 This example generates the same code regardless of whether you pass the
 argument as a value or as text, but in some cases you need to specify how
 the argument is passed.

 The value for a default argument must be text, but frequently you need to
 give a constant value. The expansion operator is one way to force the
 conversion. The following statements are equivalent:

   work    MACRO   arg:=<07h>
   work    MACRO   arg:=%07h

 The expansion operator also has several uses with macro functions. See
 Section 9.6.


 9.3.2.2  The Expansion Operator with Symbols

 When you use the expansion operator on a macro argument, any text macros or
 numeric equates in the argument are expanded:

   num     EQU     4
   numstr  TEXTEQU <4>

           work    2 + num         ; Passes "2 + num"
           work    %2 + num        ; Passes "6"
           work    2 + numstr      ; Passes "2 + numstr"
           work    %2 + numstr     ; Passes "6"

 The arguments can optionally be enclosed in parentheses. For example, these
 two statements are equivalent:

   work    %2 + num
           work    %(2 + num)


 9.3.2.3  The Expansion Operator as the First Character on a Line

 The expansion operator has a different meaning when used as the first
 character on a line. In this case, it instructs the assembler to expand any
 text macros and macro functions it finds on the rest of the line.

 This feature makes it possible to use text macros with directives such as
 ECHO, TITLE, and SUBTITLE that take an argument consisting of a single text
 value. For instance, ECHO displays its argument to the standard output
 device during assembly. Such expansion can be useful for debugging macros
 and expressions, but the requirement that its argument be a single text
 value may have unexpected results:

   ECHO    Bytes per element: %(SIZEOF array / LENGTHOF
   array)

 Instead of evaluating the expression, this line just echoes it:

   Bytes per element: %(SIZEOF array / LENGTHOF array)

 However, you can achieve the desired result by assigning the text of the
 expression to a text macro and then using the expansion operator at the
 beginning of the line to force expansion of the text macro.

   temp    TEXTEQU %(SIZEOF array / LENGTHOF array)
   %       ECHO    Bytes per element: temp

 Note that you cannot get the same results by simply putting the % at the
 beginning of the first echo line, because % expands only text macros, not
 numeric equates or constant expressions.

 Here are more examples of the use of the expansion operator at the start of
 a line:

   ; Assume memmod, lang, and os are passed in with /D option
   %   SUBTITLE  Model: memmod  Language: lang  Operating System: os

   ; Assume num defined earlier
   tnum    TEXTEQU %num
   %       .ERRE   num LE 255, <Failed because tnum !> 255>


 9.3.3  Substitution Operator (&)

 In MASM 6.0, the substitution operator (&) enables substitution of macro
 parameters, even when the parameter occurs within a larger word or within a
 quoted string. It can also be used to concatenate two macro parameters after
 they have been expanded.

 The syntax for the substitution operator looks like this:

   ¶metername&

 The operators delimiting a name always tell the assembler to substitute the
 actual argument for the name. However, the substitution operator is often
 optional. The substitution operator is not necessary when there is a space
 or separation character (comma, tab, or other operator) on that side. In the
 case of a parameter name inside a string, at least one substitution operator
 must appear.

 The rules for using the substitution operator have changed significantly
 since MASM 5.1, making macro behavior more consistent and flexible. If you
 have macros written for a previous version of MASM, you can specify the old
 behavior by using OLDMACROS or M510 with the OPTION directive (see Section
 1.3.2).

 In the macro

   work    MACRO   arg
       mov ax, &arg& * 4
   ENDM

 the & symbols tell the assembler to replace the value of  arg  with the
 corresponding argument. However, the characters on both the right and left
 are spaces. Therefore, the operators are unnecessary. The macro would
 normally be written like this:

   work    MACRO   arg
       mov ax, arg * 4
   ENDM

 The substitution operator is used for one of the following reasons:


   ■   To paste together two parameter names or a parameter name and text

   ■   To indicate that a parameter name inside double or single quotation
       marks should be expanded rather than be treated as part of the quoted
       string


 This macro illustrates both uses:

   errgen  MACRO   num, msg
       PUBLIC  err&num
       err&num BYTE    "Error &num: &msg"
   ENDM

 When called with the following arguments,

   errgen  5, <Unreadable disk>

 the macro generates this code:

   PUBLIC  err5
   err5    BYTE    "Error 5: Unreadable disk"

 In the second line of the macro, the left & symbol must be provided because
 it is adjacent to the  r  character, which is a valid identifier symbol. The
 right & symbol is not needed because there is a space to the right of the
 m. The statement pastes the text  err  to the argument value  5  to generate
 the symbol  err5.

 The substitution operator is used again inside quotation marks at the start
 of the parameter names  num  and  msg  to indicate that these names should
 be expanded. In this case, no pasting operation is necessary, so either
 operator could be omitted, but not both. The macro line could have been
 written as

   err&num BYTE    "Error num&: msg&"

 or

   err&num BYTE    "Error &num&: &msg&"

 The assembler processes substitution operators from left to right. This can
 have unexpected results when you are pasting together two macro parameters.
 For example, if  arg1  has the value  var  and  arg2  has the value  3, you
 could paste them together with this statement:

   &arg1&&arg2&    BYTE    "Text"

 Eliminating extra substitution operators, you might expect the following to
 be equivalent:

   &arg1&arg2      BYTE    "Text"

 However, this actually produces the symbol  vararg2  because in processing
 from left to right the assembler associates both the first and the second &
 symbols with the first parameter. The assembler replaces  &arg1&  by  var ,
 producing  vararg2 . The  arg2  is never evaluated. The correct abbreviation
 is

   arg1&&arg2      BYTE    "Text"

 which produces the desired symbol  var3. The symbol  arg1&&arg2  is replaced
 by  var&arg2, which is replaced by  var3.

 The substitution operator is also necessary if you want a text macro
 substituted inside quotes. For example,

   arg     TEXTEQU <hello>
   %echo   This is a string "&arg" ; Produces: This is a string "hello"
   %echo   This is a string "arg"  ; Produces: This is a string "arg"

 The substitution operator can also be used in lines beginning with the
 expansion operator (%) symbol, even outside macros (see Section 9.3.2.3).
 Text macros are always expanded in such lines, but it may be necessary to
 use the substitution operator to paste text macro names to adjacent
 characters or symbol names, as shown below:

   text    TEXTEQU <var>
   value   TEXTEQU %5
   %       ECHO    textvalue is text&&value

 This echoes the message

   textvalue is var5

 Bit-test and macro expansion statements can be confused.

 The single ampersand (&) is the bit-test operator in MASM, as it is for C.
 This operator is also used in macro expansion as the substitute operator.
 Macro substitution always occurs before evaluation of the high-level control
 structures; therefore, in ambiguous cases, the & operator is treated as a
 macro-expansion character. You can always guarantee the correct use of the
 bit-test operator by enclosing the bit-test operands in parentheses. The
 example below illustrates these two uses.

   test    MACRO   x
       .IF ax==&x      ; &x substituted with parameter value
       mov     ax, 10
       .ELSEIF ax&(x)  ; & is bitwise AND
       mov ax, 20
       .ENDIF
       ENDM


 9.4  Defining Repeat Blocks with Loop Directives

 A "repeat block" is an unnamed macro defined with a loop directive. It
 generates the statements inside the repeat block a specified number of times
 or until a given condition becomes true.

 Several loop directives are available, providing different ways of
 specifying the number of iterations. Some loop directives also provide a way
 to specify arguments for each iteration. Although the number of iterations
 is usually specified in the directive, you can use the EXITM directive to
 exit from the loop early.

 Repeat blocks can be used outside macros, but they frequently appear inside
 macro definitions to perform some repeated operation in the macro.

 This section explains the following four loop directives: REPEAT, WHILE,
 FOR, and FORC. In previous versions of MASM, REPEAT was called REPT, FOR was
 called IRP, and FORC was called IRPC. MASM 6.0 still recognizes the old
 names.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 The REPEAT and WHILE directives should not be confused with the .REPEAT and
 .WHILE directives (see Section 7.2.1, "Loop-Generating Directives"), which
 generate loop and jump instructions for run-time program control.
 ────────────────────────────────────────────────────────────────────────────


 9.4.1  REPEAT Loops

 Repeat loops are expanded at assembly time.

 The REPEAT directive is the simplest loop directive. It specifies the number
 of times to generate the statements inside the macro. The syntax is

   REPEAT constexpr
   statements
   ENDM

 The constexpr can be a constant or a constant expression, and must contain
 no forward references. Since the repeat block will be expanded at assembly
 time, the number of iterations must be known then.

 Here is an example of a repeat block used to generate data. It initializes
 an array containing sequential ASCII values for all uppercase letters.

   alpha   LABEL   BYTE            ;  Name the data generated
   letter  =       'A'             ;  Initialize counter
   REPEAT  26                      ;; Repeat for each letter
       BYTE    letter              ;; Allocate ASCII code for letter
       letter  = letter + 1        ;; Increment counter
   ENDM

 Here is another use of REPEAT, this time inside a macro:

   beep    MACRO   iter:=<3>
       mov ah, 2                   ;; Character output function
       mov dl, 7                   ;; Bell character
       REPEAT iter                 ;; Repeat number specified by macro
           int 21h                 ;; Call DOS
       ENDM
   ENDM


 9.4.2  WHILE Loops

 The WHILE directive is similar to REPEAT, but the loop continues as long as
 a given condition is true. The syntax is

   WHILE expression
   statements
   ENDM

 The expression must be a value that can be calculated at assembly time.
 Normally the expression uses relational operators, but it can be any
 expression that evaluates to zero (false) or nonzero (true). Usually, the
 condition changes during the evaluation of the macro so that the loop won't
 attempt to generate an infinite amount of code. However, you can use the
 EXITM directive to break out of the loop.

 Loops are especially useful for generating lookup tables.

 The following repeat block uses the WHILE directive to allocate variables
 initialized to calculated values. This is a common technique for generating
 lookup tables. Frequently it is faster to look up a value precalculated by
 the assembler at assembly time than to have the processor calculate the
 value at run time.

   cubes   LABEL   BYTE            ;; Name the data generated
   root    =   1                   ;; Initialize root
   cube    =   root * root * root  ;; Calculate first cube
   WHILE   cube LE 32767           ;; Repeat until result too large
       WORD   cube                 ;; Allocate cube
       root   =    root + 1        ;; Calculate next root and cube
       cube   =    root * root * root
   ENDM


 9.4.3  FOR Loops and Variable-Length Parameters

 With the FOR directive you can iterate through a list of arguments, doing
 some operation on each of them in turn. It has the following syntax:

 FOR parameter, <argumentlist> statements ENDM

 The parameter is a placeholder that will be used as the name of each
 argument inside the FOR block. The argument list must be a list of
 comma-separated arguments and must always be enclosed in angle brackets, as
 the following example illustrates:

   series  LABEL   BYTE
   FOR     arg, <1,2,3,4,5,6,7,8,9,10>
       BYTE  arg DUP (arg)
   ENDM

 On the first iteration, the  arg  parameter is replaced with the first
 argument, the value 1. On the second iteration  arg  is replaced with 2. The
 result is an array with the first byte initialized to 1, the next two bytes
 initialized to 2, the next three bytes initialized to 3, and so on.

 In this example the argument list is given specifically, but in some cases
 the list must be generated as a text macro. The value of the text macro must
 include the angle brackets.

   arglist TEXTEQU <!<3,6,9!>>     ; Generate list as text macro
   FOR  arg, arglist
       .                           ; Do something to arg
       .
       .
   ENDM

 Note the use of the literal character operator (!) to use angle brackets as
 characters, not delimiters (see Section 9.3.1).

 Variable parameter lists provide flexibility.

 The FOR directive also provides a convenient way to process macros with a
 variable number of arguments. To do this, add VARARG to the last parameter
 to indicate that a single named parameter will have the actual value of all
 additional arguments. For example, the following macro definition includes
 the three possible parameter attributes─required, default, and variable.

   work    MACRO   rarg:REQ, darg:=<5>, varg:VARARG

 The variable argument must always come last. If this macro is called with
 the statement

   work 5, , 6, 7, a, b

 the first argument is received as passed, the second is replaced by the
 default value  5, and the last four are received as the single argument  <6,
 7, a, b>. This is the same format expected by the FOR directive. The FOR
 directive discards leading spaces but recognizes trailing spaces.

 The following macro illustrates variable arguments:

   show    MACRO chr:VARARG
       mov     ah, 02h
       FOR arg, <chr>
           mov     dl, arg
           int     21h
       ENDM
   ENDM

 When called with

   show  'O', 'K', 13, 10

 the macro displays each of the specified characters one at a time.

 The parameter in a FOR loop can have the required or default attribute. The
 show  macro can be modified to make blank arguments generate errors:

   show    MACRO chr:VARARG
       mov     ah, 02h
       FOR arg:REQ, <chr>
           mov     dl, arg
           int     21h
       ENDM
   ENDM

 The macro now generates an error if called with

   show  'O',, 'K', 13, 10

 Another approach would be to use a default argument:

   show    MACRO chr:VARARG
       mov     ah, 02h
       FOR arg:=<' '>, <chr>
           mov     dl, arg
           int     21h
       ENDM
   ENDM

 Now if the macro is called with

   show  'O',, 'K', 13, 10

 it inserts the default character, a space, for the blank argument.


 9.4.4  FORC Loops

 The FORC directive is similar to FOR but takes a string of text rather than
 a list of arguments. The statements are assembled once for each character
 (including spaces) in the string, substituting a different character for the
 parameter each time through.

 The syntax looks like this:

   FORC parameter, < text>
   statements
   ENDM

 The text must be enclosed in angle brackets. The following example
 illustrates FORC:

   FORC arg, <ABCDEFGHIJKLMNOPQRSTUVWXYZ>
       BYTE  '&arg'             ;; Allocate uppercase letter
       BYTE  '&arg' + 20h       ;; Allocate lowercase letter
       BYTE  '&arg' - 40h       ;; Allocate ordinal of letter
   ENDM

 Notice that the substitution operator must be used inside the quotation
 marks to make sure that  arg  is expanded to a character rather than treated
 as a literal string.

 With earlier versions of MASM, FORC is often used for complex parsing tasks.
 A long sentence can be examined character by character. Each character is
 then either thrown away or pasted onto a token string, depending on whether
 it is a separator character. In MASM 6.0, the predefined macro functions and
 string processing directives discussed in Section 9.5 are usually more
 efficient for these tasks.


 9.5  String Directives and Predefined Functions

 Predefined macro string functions are new to MASM 6.0.

 The assembler provides the following directives for manipulating text:
 SUBSTR, INSTR, SIZESTR, and CATSTR. Each of these has a corresponding
 predefined macro function version: @SubStr, @InStr, @SizeStr, and @CatStr.

 You use the directive versions to assign a processed value to a text macro
 or numeric equate. For example, CATSTR, which concatenates a list of text
 values, can be used like this:

   num     =       7
   newstr  CATSTR  <3 + >, %num, < = > , %3 + num ; "3 + 7 = 10"

 Assignment with CATSTR and SUBSTR works like assignment with the TEXTEQU
 directive. Assignment with SIZESTR and INSTR works like assignment with the
 = operator.

 The arguments to directives must be text values. Use the expansion operator
 to make sure that constants and numeric equates are expanded to text.

 The macro function versions are similar, but their arguments must be
 enclosed in parentheses. Macro functions return text values and can be used
 in any context where text is expected. Section 9.6 tells how to write your
 own macro functions. An equivalent statement to the previous example using
 CATSTR is

   num     =       7
   newstr  TEXTEQU @CatStr( <3 + >, %num, < = > , %3 + num )

 Although the directive version is simpler in the example above, the function
 versions are often convenient because they can be used as arguments to
 string directives or to other macro functions.

 Unlike the string directives, predefined macro function names are case
 sensitive. Since MASM is not case sensitive by default, the case doesn't
 matter unless you use the /Cp command-line option.

 The following sections summarize the syntax for each of the string
 directives and functions. The explanations focus on the directives, but the
 functions work the same except where noted.


 SUBSTR

   name SUBSTR string, start«, length»
   @SubStr( string, start«, length» )

 The SUBSTR directive assigns a substring from a given string to a new
 symbol, specified by name. Start specifies the position (1-based) in string
 to start the substring. Length specifies the length of the substring. If
 length is not given, it is assumed to be the remainder of the string
 including the start character. The string

 in the SUBSTR syntax, as well as in the syntax for the other string
 directives and predefined functions, can be any textItem where textItem can
 be text enclosed in angle brackets (< >), the name of a macro, or a constant
 expression preceded by % (%constExpr).


 INSTR

   name INSTR «start,» string, substring
   @InStr( «start», string, substring
   )

 The INSTR directive searches a specified string for an occurrence of a given
 substring and assigns its position (1-based) to name. The search is case
 sensitive. Start is the position in string to start the search for
 substring. If start is not given, it is assumed to be 1 (the start of the
 string). If substring is not found, the position assigned to name is 0.

 If the INSTR directive is used, the position value is assigned to a name as
 if it were a numeric equate. If the @InStr function is used, the value is
 returned as a string of digits in the current radix.

 The @InStr function has a slightly different syntax than the INSTR
 directive. You can omit the first argument and its associated comma from the
 directive. You can leave the first argument blank with the function, but a
 blank function argument must still have a comma. For example,

   pos     INSTR   <person>, <son>

 is the same as

   pos     = @InStr( , <person>, <son> )

 The return value could also be assigned to a text macro:

   strpos  TEXTEQU @InStr( , <person>, <son> )


 SIZESTR

   name SIZESTR string
   @SizeStr( string )

 The SIZESTR directive assigns the number of characters in string to name. An
 empty string assigns a length of zero. Although the length is always a
 positive number, it is assigned as a string of digits in the current radix
 rather than as a numeric value.

 If the SIZESTR directive is used, the size value is assigned to a name as if
 it were a numeric equate. If the @SizeStr function is used, the value is
 returned as a string of digits in the current radix.


 CATSTR

   name CATSTR string«, string»...
   @CatStr( string«,  string»... )

 The CATSTR directive concatenates a list of text values specified by string
 into a single text value and assigns it to name. TEXTEQU is technically a
 synonym for CATSTR. TEXTEQU is normally used for single-string assignments,
 while CATSTR is used for multistring concatenations.

 The following example that pushes and pops one set of registers illustrates
 several uses of string directives and functions:

   ; SaveRegs - Macro to generate a push instruction for each
   ; register in argument list. Saves each register name in the
   ; regpushed text macro.
   regpushed TEXTEQU <>                    ;; Initialize empty string

   SaveRegs MACRO regs:VARARG
       FOR reg, <regs>                     ;; Push each register
           push reg                        ;;  and add it to the list
           regpushed CATSTR   <reg>, <,>, regpushed
       ENDM                                ;; Strip off last comma
       regpushed CATSTR <!<>, regpushed    ;; Mark start of list with
   <
       regpushed SUBSTR regpushed, 1, @SizeStr( regpushed )
       regpushed CATSTR regpushed, <!>>    ;; Mark end with >
   ENDM

   ; RestoreRegs - Macro to generate a pop instruction for registers
   ; saved by the SaveRegs macro. Restores one group of registers.

   RestoreRegs MACRO
       LOCAL regs
       %FOR reg, regpushed                  ;; Pop each register        pop
   reg
       ENDM
    ENDM

 Notice how the  SaveRegs  macro saves its result in the  regpushed  text
 macro for later use by the  RestoreRegs  macro. In this case, a text macro
 is used as a global variable. By contrast, the  regs  text macro is used
 only in RestoreRegs. It is declared LOCAL so that it won't take the name
 regs  from the global name space. The MACROS.INC file provided with MASM 6.0
 includes expanded versions of these same two macros.


 9.6  Returning Values with Macro Functions

 A macro function returns a text string.

 A macro function is a named group of statements that returns a value. When a
 macro function is called, its argument list must be enclosed in parentheses,
 even if the list is empty. The value returned is always text.

 Macro functions are new to MASM 6.0, as are several predefined macro
 functions for common tasks. The predefined macros include @Environ (see
 Section 1.2.3) and the string functions @SizeStr, @CatStr, @SubStr, and
 @InStr (discussed in the preceding section).

 Macro functions are defined in exactly the same way as macro procedures,
 except that a value must always be returned using the EXITM directive. Here
 is an example:

   DEFINED MACRO   symbol:REQ
       IFDEF symbol
           EXITM <-1>              ;; True
       ELSE
           EXITM <0>               ;; False
       ENDIF
   ENDM

 This macro works like the defined operator in the C language. You can use it
 to test the defined state of several different symbols with a single
 statement, as shown below:

   IF DEFINED( DOS ) AND NOT DEFINED( XENIX )
       ;; Do something
   ENDIF

 Notice that the macro returns integer values as strings of digits, but the
 IF statement evaluates numeric values or expressions. There is no conflict
 because the value returned by the macro function is seen in the statement
 exactly as if the user had typed the values directly into the program:

   IF -1 AND NOT 0


 Returning Values with EXITM

 The return value must be text, a text equate name, or the result of another
 macro function. If a function must return a numeric value (such as a
 constant, a numeric equate, or the result of a numeric expression), it must
 first convert the value to text using angle brackets or the expansion
 operator (%). The defined macro, for example, could have returned its value
 as

   EXITM   %-1

 Although macro functions can include any legal statement, they seldom need
 to include instructions. This is because a macro function is expanded and
 its value returned at assembly time, while instructions are executed at run
 time.

 Here is another example of a macro function. It uses the WHILE directive to
 calculate factorials:

   factorial   MACRO   num:REQ
       LOCAL   i, factor
       factor  =   num
       i       =   1
       WHILE   factor GT 1
           i       =   i * factor
           factor  =   factor - 1
       ENDM
       EXITM   %i
   ENDM

 The integer result of the calculation is changed to a text string with the
 expansion operator (%). The  factorial  macro can be used to define data, as
 shown below:

   var     WORD    factorial( 4 )

 The effect of this statement is to initialize  var  with the number 24 (the
 factorial of 4).


 Using Macro Functions with Variable-Length Parameter Lists

 Macro functions can enhance FOR loops.

 You can use the FOR directive to handle macro parameters with the VARARG
 attribute. Section 9.4.3 explains how to do this in simple cases where the
 variable parameters are handled sequentially, from first to last. However,
 you may sometimes need to process the parameters in reverse order or
 nonsequentially. Macro functions make these techniques possible.

 You may need to know the number of arguments in a VARARG parameter. The
 following macro functions handle this.

   @ArgCount MACRO arglist:VARARG
       LOCAL count
       count = 0
       FOR arg, <arglist>
           count = count + 1       ;; Count the arguments
       ENDM
       EXITM %count
   ENDM

 You could use this inside a macro that has a VARARG parameter, as shown
 below:

   work    MACRO args:VARARG
   %   ECHO Number of arguments is: @ArgCount( args )
   ENDM

 Another useful task might be to select an item from an argument list using
 an index to indicate which item. The following macro simplifies this.

   @ArgI MACRO index:REQ, arglist:VARARG
       LOCAL count, retstr
       retstr TEXTEQU <>            ;; Initialize count
       count  = 0                   ;; Initialize return string
       FOR arg, <arglist>
           count = count + 1
           IF count EQ index        ;; Item is found
               retstr TEXTEQU <arg> ;; Set return string
               EXITM                ;;  and exit IF
           ENDIF
       ENDM
       EXITM retstr                 ;; Exit function
   ENDM

 This function can be used as shown below:

   work    MACRO args:VARARG
   %   ECHO Third argument is: @ArgI( 3, args )
   ENDM

 Finally, you might need to process arguments in reverse order. The following
 macro returns a new argument list in reverse order.

   @ArgRev MACRO arglist:REQ
       LOCAL txt, arg
       txt TEXTEQU <>
   %   FOR arg, <arglist>
           txt CATSTR <arg>, <,>, txt      ;; Paste each onto list
       ENDM
                                           ;; Remove terminating comma
       txt SUBSTR  txt, 1, @SizeStr( %txt ) - 1
       txt CATSTR  <!<>, txt, <!>>         ;; Add angle brackets
       EXITM txt
   ENDM

 You could call this function as shown below:

   work    MACRO   args:VARARG
   %   FOR  arg, @ArgRev( <args> )   ;; Process in reverse order
           ECHO    arg
       ENDM
   ENDM

 These three macro functions are provided on the MASM distribution disk in
 the MACROS.INC include file.


 Macro Operators and Macro Functions

 This list summarizes the behavior of the expansion operator with macro
 functions.


   ■   If a macro function is not preceded by a %, it will be expanded.
       However, if it expands to a text macro or a macro function call, the
       result will not be expanded further.

   ■   If you use a macro function call as an argument for another macro
       function call, a % is not needed.

   ■   If a macro function expands to a text macro (or another macro
       function), the macro function will be recursively expanded.

   ■   If a macro function is called inside angle brackets and is preceded by
       %, it will be expanded.



 9.7  Advanced Macro Techniques

 The concept of replacing macro names with predefined macro text is simple in
 theory, but it has many implications and complications. Here is a brief
 summary of some advanced techniques you can use in macros.


 9.7.1  Nesting Macro Definitions

 Macros can define other macros or can be redefined. MASM does not process
 nested definitions until the outer macro has been called. Therefore, the
 inner macros cannot be called until the outer macro has been called. The
 nesting of macro definitions is limited only by memory.

   shifts  MACRO   opname                  ;; Macro generates macros
       opname&s    MACRO operand:REQ, rotates:=<1>
           IF rotates LE 2                 ;; One at a time is faster
               REPEAT rotate               ;;  for 2 or less
                   opname  operand, 1
               ENDM
           ELSE                            ;; Using CL is faster for
               mov     cl, rotates         ;;  more than 2
               opname  operand, cl
           ENDIF
       ENDM
   ENDM

   ; Call macro to make new macros
           shifts  ror                     ; Generates rors
           shifts  rol                     ; Generates rols
           shifts  shr                     ; Generates shrs
           shifts  shl                     ; Generates shls
           shifts  rcl                     ; Generates rcls
           shifts  rcr                     ; Generates rcrs
           shifts  sal                     ; Generates sals
           shifts  sar                     ; Generates sars

 This macro generates enhanced versions of the shift and rotate instructions.
 The macros could be called like this:

   shrs    ax, 5
           rols    bx, 3

 The macro versions handle multiple shifts by generating different code,
 depending on how many shifts are specified. The example above is optimized
 for the 8088 and 8086 processors. If you want to enhance for other
 processors, you can simply change the outer macro; it automatically changes
 all the inner macros. Code that uses the inner macros benefits from the
 enhancements but does not change so long as the macro interface doesn't
 change.


 9.7.2  Testing for Argument Type and Environment

 Macros can check the type of arguments and generate different code depending
 on what they find. For example, you can use the OPATTR operator to determine
 if an argument is a constant, a register, or a memory operand.

 If you discover a constant value, you can often optimize the code. In some
 cases, you can generate better code for 0 or 1 than for other constants. If
 the argument is a memory operand, you know nothing about the value of the
 operand, since it may change at run time. However, you may want to generate
 different code depending on the operand size and on whether it is a pointer.
 Similarly, if the operand is a register, you know nothing of its contents,
 but you may be able to optimize if you can identify a particular register
 with the IFDIFI or IFIDNI directives.

 The following example illustrates some of these techniques. It loads a
 specified address into a specified offset register. The segment register is
 assumed to be DS.

   load    MACRO reg:REQ, adr:REQ
       IF (OPATTR (adr)) AND 00010000y    ;; Register
           IFDIFI reg, adr                ;; Don't load register
               mov     reg, adr           ;;  onto itself
           ENDIF
       ELSEIF (OPATTR (adr)) AND 00000100y
           mov     reg, adr               ;; Constant
       ELSEIF (TYPE (adr) EQ BYTE) OR (TYPE (adr) EQ SBYTE)
           mov    reg, OFFSET adr         ;; Bytes
       ELSEIF (SIZE (TYPE (adr)) EQ 2
           mov    reg, adr                ;; Near pointer
       ELSEIF (SIZE (TYPE (adr)) EQ 4
           mov    reg, WORD PTR adr[0]    ;; Far pointer
           mov    ds,  WORD PTR adr[2]
       ELSE
           .ERR <Illegal argument>
       ENDIF
   ENDM

 A macro may also generate different code depending on the assembly
 environment. The predefined text macro @Cpu can be used to test for
 processor type. The following example uses the more efficient constant
 variation of the PUSH instruction if the processor is an 80186 or higher.

   IF  @Cpu AND 00000010y
       pushc  MACRO op             ;; 80186 or higher
           push op
       ENDM
   ELSE
       pushc  MACRO op             ;; 8088/8086
           mov  ax, op
           push ax
       ENDM
   ENDIF

 Note that the example generates a completely different macro for the two
 cases. This is more efficient than testing the processor inside the macro
 and conditionally generating different code. With this macro, the
 environment is checked only once; if the conditional were inside the macro
 it would be checked every time the macro is called.

 You can test the language and operating system using the @Interface text
 macro. The memory model can be tested with the @Model, @DataSize, or
 @CodeSize text macros.

 You can save the contexts inside macros with PUSHCONTEXT and POPCONTEXT. The
 options for these keywords are:

 Option            Description
 ────────────────────────────────────────────────────────────────────────────
 RADIX             Saves segment register information
 LIST              Saves listing and CREF information
 CPU               Saves current CPU and processor
 ALL               All of the above


 9.7.3  Using Recursive Macros

 Macros can call themselves. In previous versions of MASM, recursion is an
 important technique for handling variable arguments. With MASM 6.0, you can
 do this much more cleanly using the FOR directive and the VARARG attribute,
 as described in Section 9.4.3. However, recursion is still available and may
 be useful for some macros.


 9.8  Related Topics in Online Help

 In addition to information covered in this chapter, information on the
 following topics can be found in online help. From the "MASM 6.0 Contents"
 screen:

╓┌─────────────────────────────────────┌─────────────────────────────────────╖
 Topics                                Access
 ────────────────────────────────────────────────────────────────────────────
 INCLUDE                               Choose "Directives," and then "Scope
                                       and
                                       Visibility"

 GOTO, PURGE                           Choose "Directives," and then
 Topics                                Access
 ────────────────────────────────────────────────────────────────────────────
GOTO, PURGE                           Choose "Directives," and then
                                       "Macros and Iterative Blocks"

 .LISTMACRO                            Choose "Directives," and then
                                       "Listing
                                       Control"

 IFB, IFNB, IFDIFI,                    Choose "Directives," and then
 and IFIDNI                            "Conditional Assembly"

 ECHO                                  Choose "Directives," and then
                                       "Miscellaneous"

 OPATTR                                Choose "Operators," and then
                                       "Miscellaneous"

 @Cpu, @Interface, @DataSize,          Choose "Predefined Symbols"
 @Environ, and @CodeSize

 Topics                                Access
 ────────────────────────────────────────────────────────────────────────────

 PUSHCONTEXT,                          Choose "Directives" and then
 POPCONTEXT                            "Iterative Blocks"








 Chapter 10  Managing Projects with NMAKE
 ────────────────────────────────────────────────────────────────────────────

 The Microsoft Program Maintenance Utility (NMAKE) is a sophisticated command
 processor that saves time and simplifies project management. Once you
 specify which project files depend on others, NMAKE automatically executes
 the commands needed to update your project when any project file has
 changed.

 The advantage of using NMAKE instead of simple batch files is that NMAKE
 recompiles only those files that need recompiling. NMAKE doesn't waste time
 with files that haven't changed since the last build. NMAKE also has
 advanced features (such as macros) that simplify managing complex projects.


 This chapter includes examples that show how each feature of NMAKE works. In
 addition, Section 10.9, "A Sample NMAKE Description File," shows how many of
 these features work together.

 If you are using the Microsoft Programmer's WorkBench (PWB) to build your
 project, PWB automatically creates a description file (called a "makefile"
 in the PWB documentation) and calls NMAKE to run the file. You may want to
 read this chapter if you intend to build your program outside of PWB or if
 you want to understand or modify a description file created by PWB.

 A utility called NMK allows you to use NMAKE to manage your project under
 DOS (or in a DOS session under OS/2). Section 10.11, "Using NMK," explains
 when and how to use NMK.

 If you are familiar with MAKE, the predecessor to NMAKE, be sure to read
 Section 10.10, "Differences between NMAKE and MAKE." These utilities differ
 in several important respects.


 10.1  Overview of NMAKE

 NMAKE works by looking at the last times and dates of modification for a
 "target" file and its "dependents" and then comparing them. A target is
 usually a file you want to create, such as an executable file. A dependent
 is usually a file from which a target is created, such as a source file. A
 target is "out-of-date" if any of its dependents has changed more recently
 than the target.

 ────────────────────────────────────────────────────────────────────────────
 WARNING

 For NMAKE to work properly, the date and time setting on your system must be
 consistent relative to previous settings. If you set the date and time each
 time you start the system, be careful to set it accurately. If your system
 stores a setting, be certain that the battery is working.
 ────────────────────────────────────────────────────────────────────────────

 When you run NMAKE, it reads a "description file" that you supply. The
 description file consists of one or more description blocks. Each
 description block typically lists a target, the target's dependents, and the
 commands that build the target. NMAKE compares the last time the targets
 changed to the last time the dependents changed. If the modification time of
 any dependents is the same or later than the time of the target, NMAKE
 updates the target by executing the command or commands listed in the
 description block.

 NMAKE's main purpose is to help you update applications quickly and simply.
 However, it can execute any DOS or OS/2 command, so it is not limited to
 compiling and linking. NMAKE can also make backups, move files, and perform
 other project-management tasks that you ordinarily do at the
 operating-system prompt.


 10.2  Running NMAKE

 You invoke NMAKE with the following syntax:

   NMAKE [[options]] [[macros]]
   [[targets]]

 The options field lists NMAKE options, which are described in Section 10.4,
 "Command-Line Options."

 The macros field lists macro definitions, which allow you to change text in
 the description file. The syntax for macros is described in "User-Defined
 Macros" in Section 10.3.4.1, "Macros."

 The targets field lists targets to build. NMAKE rebuilds only the targets
 listed on the command line. If you don't specify any targets, NMAKE builds
 only the first target in the description file. (This behavior departs
 significantly from that of MAKE. See Section 10.10, "Differences between
 NMAKE and MAKE.")

 NMAKE follows the instructions you specify in a description file.

 NMAKE searches the current directory for the name of a description file you
 specify with the /F option. It halts and displays an error message if the
 file does not exist. If you do not use the /F option to specify a
 description file, NMAKE searches the current directory for a description
 file named MAKEFILE. If MAKEFILE does not exist, NMAKE checks the command
 line for target files and tries to build them using predefined inference
 rules (either default or defined in TOOLS.INI). This feature lets you use
 NMAKE without a description file (as long as NMAKE has a predefined
 inference rule for the target). If the command line does not specify any
 target files, NMAKE halts and displays an error message.


 Example

   NMAKE /S "program=sample" sort.exe search.exe

 This command supplies four arguments: an option (/S), a macro definition
 ("program=sample"), and two target specifications (sort.exe  and
 search.exe).

 The command does not specify a description file, so NMAKE looks for the
 default description file, MAKEFILE. The /S option tells NMAKE not to display
 the commands as they are executed. (See Section 10.4, "Command-Line
 Options.") The macro definition performs a text substitution throughout the
 description file, replacing every instance of  program  with  sample. The
 target specifications tell NMAKE to update the targets SORT.EXE and
 SEARCH.EXE.


 10.3  NMAKE Description Files

 The most important parts of a description file are the description blocks,
 which tell NMAKE how to build your project's target files. A description
 file can also contain comments, macros, inference rules, and directives.
 This section describes the elements of description files.


 10.3.1  Description Blocks

 Description blocks form the heart of the description file. Figure 10.1
 illustrates a typical NMAKE description block, including the three sections:
 targets, dependents, and commands.

 (This figure may be found in the printed book.)


 10.3.1.1  Targets

 The target is the file that you want to build.

 The targets section of the dependency line lists one or more files to build.
 The line that lists targets and dependents is called the "dependency line."


 The example in Figure 10.1 tells NMAKE how to build a single target,
 MYAPP.EXE, if it is missing or out-of-date. Although single targets are
 common, you can also list multiple targets in a single dependency line; you
 must separate each target name with a space. If the name of the last target
 before the colon (:) is one character long, put a space between the name and
 the colon, so NMAKE won't interpret the character as a drive specification.


 A target can appear in only one dependency line when specified as shown
 above. To update a target using more than one description block, specify two
 consecutive colons (::) between targets and dependents. For details, see
 Section 10.3.1.8, "Specifying a Target in Multiple Description Blocks."

 The target is usually a file, but it can also be a "pseudotarget," a name
 that lets you build groups of files or execute a group of commands. For more
 information, see Section 10.3.2, "Pseudotargets."


 10.3.1.2  Dependents

 A dependent is a file used to build a target.

 The dependents section of the description block lists one or more files from
 which the target is built. A colon (:) separates it from the targets
 section. The example in Figure 10.1 lists three dependents after MYAPP.EXE:


   myapp.exe : myapp.obj another.obj myapp.def

 You can also specify the directories in which NMAKE should search for a
 dependent. Enclose one or more directory names in braces ( { } ). Separate
 multiple directories with a semicolon ( ; ). The syntax for a directory
 specification is

   {directory[[;directory...]]}dependent


 Example

 The following dependency line tells NMAKE to search the current directory
 first, then the specified directories:

   forward.exe : {\src\alpha;d:\proj}pass.obj

 In the line above, the target, FORWARD.EXE, has one dependent, PASS.OBJ. The
 directory list specifies two directories:

   {\src\alpha;d:\proj}

 NMAKE first searches for PASS.OBJ in the current directory. If PASS.OBJ
 isn't there, NMAKE searches the \ SRC \ ALPHA directory, then the D:\ PROJ
 directory. If NMAKE cannot find a dependent in the current directory or a
 listed directory, it looks for a description block with a dependency line
 containing PASS.OBJ as a target, and uses the commands in that description
 block to create PASS.OBJ. If NMAKE cannot find such a description block, it
 looks for an inference rule that describes how to create the dependent. (See
 Section 10.3.5, "Inference Rules.")


 10.3.1.3  Dependency Line

 The dependency line in Figure 10.1 tells NMAKE to rebuild the target
 MYAPP.EXE whenever MYAPP.OBJ, ANOTHER.OBJ, or MYAPP.DEF has changed more
 recently than MYAPP.EXE.

 The object files in the dependency list above would never be newer than the
 executable file (unless you had recompiled the source code before running
 NMAKE). So NMAKE checks to see if the object files themselves are targets in
 other dependency lists, and if any dependents in those lists are targets
 elsewhere, and so on.

 NMAKE continues moving through all dependencies this way to build a
 "dependency tree" that specifies all the steps required to fully update the
 target. If NMAKE then finds any dependents in the tree that are newer than
 the target, NMAKE updates the appropriate files and rebuilds the target.


 10.3.1.4  Commands

 The commands section can contain one or more commands.

 The commands section of the description block lists the commands that NMAKE
 should use to build the target. You can use any command that can be executed
 from the command line. The example in Figure 10.1 tells NMAKE to build
 MYAPP.EXE using the following LINK command:

       link myapp another.obj, , NUL, os2, myapp

 Notice that the line is indented. NMAKE uses indentation to distinguish
 between a dependency line and a command line. A command line must be
 indented at least one space or tab. The dependency line must not be indented
 (it cannot start with a space or tab).

 Many targets are built with a single command, but you can place more than
 one command after the dependency line, each on a separate line, as shown in
 Figure 10.1.

 A long command can span several lines if each line ends with a backslash ( \
 ). A backslash at the end of a line is equivalent to a space on the command
 line. For example, the command

   echo abcd\
   efgh

 is equivalent to the command

   echo abcd efgh

 You can also place a command at the end of a dependency line. Use a
 semicolon (;) to separate the command from the rightmost dependent, as in

   project.exe : project.obj ; link project;

 OS/2 allows multiple commands on one command line.

 OS/2 allows you to combine two or more commands on a single command line
 with an ampersand (&). For example, the following command line is legal in
 an OS/2 description file:

   DIR & COPY sample.exe backup.exe

 A slight restriction is imposed on the use of the CD, CHDIR, and SET
 commands in OS/2 description files. NMAKE executes these commands itself
 rather than passing them to OS/2. Therefore, if any of these commands is the
 first command on a line, the remaining commands are not executed because
 they aren't passed to OS/2.

 The following multiple-command line does not display the directory listing
 because DIR is preceded by a CD command:

       CD \mydir & DIR

 To use CD, CHDIR, or SET in a description block, place these commands on
 separate lines:

   CD \mydir
       DIR

 NMAKE interprets a percent symbol (%) within a command line as the start of
 a file specifier. To use a literal percent symbol in a command line, specify
 it as a double percent symbol (%%). (See Section 10.3.8, "Extracting
 Filename Components.")


 10.3.1.5  Wild Cards

 You can use DOS and OS/2 wild-card characters (* and ?) to specify target
 and dependent filenames. NMAKE expands the wild cards when analyzing
 dependencies and when building targets. For example, the following
 description block links all files having the .OBJ extension in the current
 directory:

   project.exe : *.obj
       LINK $*.obj;


 10.3.1.6  Command Modifiers

 Command modifiers are special prefixes attached to the command. They provide
 extra control over the commands in a description block. You can use more
 than one modifier for a single command. Table 10.1 describes the three NMAKE
 command modifiers.

 Table 10.1  Command Modifiers

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Character                         Action
 ────────────────────────────────────────────────────────────────────────────
 @                            Prevents NMAKE from displaying the
                                   command as it executes. In the example
                                   below, the at sign (@) suppresses
                                   display of the ECHO command line:

                                   sort.exe : sort.obj
                                        @ECHO Now sorting.

                                   The output of the ECHO command is not
                                   suppressed.

 Character                         Action
 ────────────────────────────────────────────────────────────────────────────

 -«number»                         Turns off error checking for the command.
                                   Spaces and tabs can appear before the
                                   command. If the dash is followed by a
                                   number, NMAKE checks the exit code
                                   returned by the command and stops if the
                                   code is greater than the number. No
                                   space or tab can appear between the dash
                                   and number. (See Section 10.12, "Using
                                   Exit Codes with NMAKE.")

                                   In the following example, if the program
                                    sample  returns an exit code, NMAKE
                                   does not stop but continues to execute
                                   commands; if sort  returns an exit code
                                   greater than 5, NMAKE stops:

                                   light.lst : light.txt
                                     -sample light.txt
 Character                         Action
 ────────────────────────────────────────────────────────────────────────────
                                    -sample light.txt
                                      -5 sort light.txt

 !                                 Executes the command for each dependent
                                   file if the command preceded by the
                                   exclamation point uses the predefined
                                   macros $** or $?. (See Section 10.3.4,
                                   "Macros.") The $** macro refers to all
                                   dependent files in the description block.
                                   The $? macro refers to all dependent
                                   files in the description block that have
                                   a more recent modification time than the
                                   target. For example,

                                   print : one.txt two.txt three.txt
                                        !print $** lpt1:

                                   generates the following commands:

 Character                         Action
 ────────────────────────────────────────────────────────────────────────────

                                   print one.txt lpt1:
                                   print two.txt lpt1:
                                   print three.txt lpt1:

 ────────────────────────────────────────────────────────────────────────────




 10.3.1.7  Using Special Characters as Literals

 You may need to specify as a literal character one of the characters that
 NMAKE uses for a special purpose. These characters are

 : ; # ( ) $ ^ \ { } ! @  ─

 To use one of these characters literally, place a caret (^) in front of it.
 For example, suppose you define a macro that ends with a backslash:

   exepath=c:\bin\

 The line above is intended to define a macro named  exepath  with the value
 c:\bin\. But the second backslash has an unintended side effect. Since the
 backslash is NMAKE's line-continuation character, the line actually defines
 exepath  as  c:\bin, followed by whatever appears on the next line of the
 description file. You can avoid this problem by placing a caret in front of
 the second backslash:

   exepath=c:\bin^\

 You can also use a caret to insert a literal newline character in a string
 or macro:

   XYZ=abc^
   def

 The caret tells NMAKE to interpret the newline character as part of the
 macro, not a line break. Note that this effect differs from using a
 backslash ( \ ) to continue a line. A newline character that follows a
 backslash is replaced with a space.

 NMAKE ignores carets that precede characters other than the special
 characters listed above. The line

   ign^ore : these ca^rets

 is interpreted as

   ignore : these carets

 A caret within a quoted string is treated as a literal caret character.


 10.3.1.8  Specifying a Target in Multiple Description Blocks

 You can specify a target in more than one description block by placing two
 colons (::) after the target. This feature is useful for building a complex
 target, such as a library, that contains components created with different
 commands. For example,

   target.lib :: a.asm b.asm c.asm
       ML a.asm b.asm c.asm
       LIB target -+a.obj -+b.obj -+c.obj;
   target.lib :: d.c e.c
       CL /c d.c e.c
       LIB target -+d.obj -+e.obj;

 Both description blocks update the library named TARGET.LIB. If any of the
 assembly-language files have changed more recently than the library, NMAKE
 executes the commands in the first block to assemble the source files and
 update the library. Similarly, if any of the C-language files have changed,
 NMAKE executes the second group of commands to compile the C files and
 update the library.

 If you use a single colon in the example above, NMAKE issues an error
 message. It is legal, however, to use single colons if the target appears in
 only one block. In this case, dependency lines are cumulative. For example,


   target : jump.bas
   target : up.c
       echo Building target...

 is equivalent to

   target : jump.bas up.c
       echo Building target...

 No commands can appear between cumulative dependency lines, but blank lines,
 comment lines, macro definitions, and directives can appear.


 10.3.2  Pseudotargets

 A "pseudotarget" is similar to a target, but it is not a file. It is a name
 used as a label for executing a group of commands. In the following example,
  UPDATE  is a pseudotarget.

   UPDATE : *.*
       !COPY $** a:\product

 NMAKE always considers the pseudotarget to be out-of-date. In the previous
 example, NMAKE copies all the dependent files to the specified drive and
 directory.

 Like target names, pseudotarget names are not case sensitive.


 10.3.3  Comments

 You can place comments in a description file by preceding them with a number
 sign (#):

   # Comment on line by itself
   OPTIONS = /MAP  # Comment on macro's line
   all.exe : one.obj two.obj  # Comment on dependency line
       link $(OPTIONS) one.obj two.obj;

 A comment extends to the end of the line in which it appears. Command lines
 (and dependency lines containing commands) cannot contain comments.

 To specify a literal #, precede it with a caret (^ ), as in the following:

   DEF=^#define #Macro representing a C preprocessing directive


 10.3.4  Macros

 Macros offer a convenient way to replace a particular string in the
 description file with another string. Macros are useful for a variety of
 tasks, including the following:


   ■   Creating a single description file that works for several projects.
       You can define a macro that replaces a dummy filename in the
       description file with the specific filename for a particular project.

   ■   Controlling the options NMAKE passes to the compiler or linker. When
       you specify options in a macro, you can change options throughout the
       description file in a single step.


 You can define your own macros or use predefined macros. This section
 describes user-defined macros first.


 10.3.4.1  User-Defined Macros

 You can define a macro with this syntax:

   macroname=string

 The macroname can be any combination of letters, digits, and the underscore
 ( _ ) character. Macro names are case sensitive. NMAKE interprets  MyMacro
 and  MYMACRO  as different macro names.

 The string can be any sequence of zero or more characters. (A string of zero
 characters is called a "null string." A string consisting only of spaces,
 tabs, or both is also considered a null string.) For example,

   linkcmd=LINK /map

 defines a macro named  linkcmd  and assigns it the string  LINK /map.

 You can define macros in the description file, on the command line, in a
 command file (see Section 10.5, "NMAKE Command File"), or in TOOLS.INI (see
 Section 10.6, "The TOOLS.INI File"). Each macro defined in the description
 file must appear on a separate line. The line cannot start with a space or
 tab.

 When you define a macro in the description file, NMAKE ignores spaces on
 either side of the equal sign. The string itself can contain embedded
 spaces. You do not need to enclose string in quotation marks (if you do,
 they become part of the string).

 Slightly different rules apply when you define a macro on the command line
 or in a command file. The command-line parser treats spaces as argument
 delimiters. Therefore, the string itself, or the entire macro, must be
 enclosed in double quotation marks if it contains embedded spaces. All three
 forms of the following command-line macro are legal and equivalent:

   NMAKE program=sample
   NMAKE "program=sample"
   NMAKE "program = sample"

 The macro  program  is passed to NMAKE, with an assigned value of  sample.

 If the string contains spaces, either the string or the entire macro must
 appear within quotes. Either form of the following command-line macro is
 allowed:

   NMAKE linkcmd="LINK /map"
   NMAKE "linkcmd=LINK /map"

 However, the following form of the same macro is not allowed. It contains
 spaces that are not enclosed by quotation marks:

   NMAKE linkcmd = "LINK /map"

 A macro name can be given a null value. Both of the following definitions
 assign a null value to the macro  linkoptions:

   NMAKE linkoptions=
   NMAKE linkoptions="    "

 A macro name can be "undefined" with the !UNDEF preprocessing directive (see
 Section 10.3.7, "Preprocessing Directives"). Assigning a null value to a
 macro name does not undefine it; the name is still defined, but with a null
 value.

 A macro can be followed by a comment, using the syntax described in the
 preceding section on comments.


 10.3.4.2  Using Macros

 Use a macro by enclosing its name in parentheses preceded by a dollar sign
 ($). For example, you can use the  linkcmd  macro defined above by
 specifying

   $(linkcmd)

 NMAKE replaces every occurrence of  $(linkcmd)  with  LINK /map.

 The following description file defines and uses three macros:

   program=sample
   L=LINK
   options=

   $(program).exe : $(program).obj
       $(L)  $(options)  $(program).obj;

 NMAKE interprets the description block as

   sample.exe : sample.obj
       LINK sample.obj;

 NMAKE replaces every occurrence of  $(program)  with  sample, every instance
 of  $(L)  with  LINK, and every instance of  $(options)  with a null string.


 An undefined macro is replaced by a null string.

 If you use as a macro a name that has never been defined, or was undefined,
 NMAKE treats that name as a null string. No error occurs.

 To use the dollar sign ($) as a literal character, specify two dollar signs
 ($$).

 The parentheses are optional if macroname is a single character. For
 example,  $L  is equivalent to  $(L). However, parentheses are recommended
 for consistency.


 10.3.4.3  Special Macros

 NMAKE provides several special macros to represent various filenames and
 commands. One use for these macros is in predefined inference rules. (See
 Section 10.3.5.4.) Like user-defined macro names, special macro names are
 case sensitive. For example, NMAKE interprets  CC  and  cc  as different
 macro names.

 Tables 10.2 through 10.5 summarize the four categories of special macros.
 The filename macros offer a convenient representation of filenames from a
 dependency line; these are listed in Table 10.2. The recursion macros,
 listed in Table 10.3, allow you to call NMAKE from within your description
 file. Tables 10.4 and 10.5 describe the command macros and options macros
 that make it convenient for you to invoke the Microsoft language compilers.


 The filename macros conveniently represent filenames from the dependency
 line.

 Table 10.2 lists macros that are predefined to represent file names. As with
 all one-character macros, these do not need to be enclosed in parentheses.
 (The $$@ and $** macros are exceptions to the parentheses rule for macros;
 they do not require parentheses even though they contain two characters.)
 Note that the macros in Table 10.2 represent filenames as you have specified
 them in the dependency line, and not the full specification of the filename.


 Table 10.2  Filename Macros

╓┌────────────────┌──────────────────────────────────────────────────────────╖
 Macro
 Reference        Meaning
 ────────────────────────────────────────────────────────────────────────────
 $@          The current target's full name, as currently specified.
                  This is not necessarily the full path name.

 Macro
 Reference        Meaning
 ────────────────────────────────────────────────────────────────────────────

 $*               The current target's full name minus the file extension.

 $**              The dependents of the current target.

 $?               The dependents that have a more recent modification time
                  than the current target.

 $$@         The target that NMAKE is currently evaluating. You can
                  use this macro only to specify a dependent.

 $<               The dependent file that has a more recent modification
                  time than the current target (evaluated only for
                  inference rules).

 ────────────────────────────────────────────────────────────────────────────



 The example below uses the $? macro, which represents all dependents that
 are more recent than the target. The ! command modifier causes NMAKE to
 execute a command once for each dependent in the list (see Table 10.1). As a
 result, the LIB command is executed up to three times, each time replacing a
 module with a newer version.

   trig.lib : sin.obj cos.obj arctan.obj
       !LIB trig.lib -+$?;

 In the next example, NMAKE updates files in another directory by replacing
 them with files of the same name from the current directory. The $@ macro is
 used to represent the current target's full name:

   #Files in objects directory depend on versions in current directory
   DIR=c:\objects
   $(DIR)\globals.obj : globals.obj
       COPY globals.obj $@
   $(DIR)\types.obj : types.obj
       COPY types.obj $@
   $(DIR)\macros.obj : macros.obj
       COPY macros.obj $@

 Macro modifiers specify parts of the predefined filename macros.

 You can append one of the modifiers in the following list to any of the
 filename macros to extract part of a filename. If you add one of these
 modifiers to the macro, you must enclose the macro name and the modifier in
 parentheses.

 Modifier                          Resulting Filename Part
 ────────────────────────────────────────────────────────────────────────────
 D                                 Drive plus directory

 B                                 Base name

 F                                 Base name plus extension

 R                                 Drive plus directory plus base name

 For example, assume that $@ has the value C:\SOURCE\PROG\SORT.OBJ. The
 following list shows the effect of combining each modifier with $@:

 Macro Reference                   Value
 ────────────────────────────────────────────────────────────────────────────
 $(@D)                             C:\SOURCE\PROG

 $(@F)                             SORT.OBJ

 $(@B)                             SORT

 $(@R)                             C:\SOURCE\PROG\SORT

 If $@ has the value SORT.OBJ without a preceding directory, the value of
 $(@R) is just SORT, and the value of $(@D) is a dot (.) to represent the
 current directory.

 Recursion macros let you use NMAKE to call NMAKE.

 Table 10.3 lists three macros that you can use when you want to call NMAKE
 recursively from within a description file.

 Table 10.3  Recursion Macros

 Macro
 Reference        Meaning
 ────────────────────────────────────────────────────────────────────────────
 $(MAKE)          The name used to call NMAKE recursively. The line on
                  which it appears is executed even if the /N command-line
                  option is specified.

 $(MAKEDIR)       The directory from which NMAKE is called.

 $(MAKEFLAGS)     The NMAKE options currently in effect. This macro is
                  passed automatically when you call NMAKE recursively. You
                  cannot redefine this macro. Use the preprocessing
                  directive !CMDSWITCHES to update the MAKEFLAGS macro.
                  (See Section 10.3.7, "Preprocessing Directives.")

 ────────────────────────────────────────────────────────────────────────────


 To call NMAKE recursively, use the command

   $(MAKE) /$(MAKEFLAGS)

 The MAKE macro is useful for building different versions of a program. The
 following description file calls NMAKE recursively to build targets in the
 \VERS1 and \VERS2 directories.

   all : vers1 vers2

   vers1 :
       cd \vers1
       $(MAKE)
       cd ..

   vers2 :
       cd \vers2
       $(MAKE)
       cd ..

 The example changes to the \VERS1 directory and then calls NMAKE
 recursively, causing NMAKE to process the file MAKEFILE in that directory.
 Then it changes to the \VERS2 directory and calls NMAKE again, processing
 the file MAKEFILE in that directory.

 You can add options to the ones already in effect for NMAKE by following the
 MAKE macro with the options in the same syntax as you would specify them on
 the command line. You can also pass the name of a description file with the
 /F option instead of using a file named MAKEFILE.

 Deeply recursive build procedures can exhaust NMAKE's run-time stack,
 causing an error. If this occurs, use the EXEHDR utility to increase NMAKE's
 run-time stack. The following command, for example, gives NMAKE.EXE a stack
 size of 16,384 (0x4000) bytes:

   exehdr /stack:0x4000 nmake.exe

 Command macros are shortcut calls to Microsoft compilers.

 NMAKE defines several macros to represent commands for Microsoft products.
 (See Table 10.4.) You can use these macros as commands in a description
 block, or invoke them using a predefined inference rule. (See Section
 10.3.5, "Inference Rules.") You can redefine these macros to represent part
 or all of a command line, including options.

 Table 10.4  Command Macros

╓┌────────────────┌─────────────────────────────────────────┌────────────────
 Macro Reference  Command Action                            Predefined Value
 ─────────────────────────────────────────────────────────────────────────────
 $(AS)            Invokes the Microsoft Macro               AS=ml
                  Assembler

 $(BC)            Invokes the Microsoft Basic               BC=bc
                  Compiler

 $(CC)            Invokes the Microsoft C Compiler          CC=cl

 $(COBOL)         Invokes the Microsoft COBOL Compiler      COBOL=cobol

 $(FOR)           Invokes the Microsoft FORTRAN             FOR=fl
                  Compiler

 $(PASCAL)        Invokes the Microsoft Pascal              PASCAL=pl
                  Compiler
 Macro Reference  Command Action                            Predefined Value
 ─────────────────────────────────────────────────────────────────────────────
                 Compiler

 $(RC)            Invokes the Microsoft Resource Compiler   RC=rc

 ─────────────────────────────────────────────────────────────────────────────



 Options macros pass preset options to Microsoft compilers.

 The macros in Table 10.5 are used by NMAKE to represent options to be passed
 to the commands for Microsoft languages. By default, these macros are
 undefined. You can define them to mean the options you want to pass to the
 commands. Whether or not they are defined, the macros are used automatically
 in the predefined inference rules. If the macros are undefined, or if they
 are defined to be null strings, a null string is generated in the command
 line. (See Section 10.3.5.4, "Predefined Inference Rules.")

 Table 10.5  Options Macros

╓┌─────────────────────────┌─────────────────────────────────────────────────╖
 Macro Reference           Passed to
 ────────────────────────────────────────────────────────────────────────────
 $(AFLAGS)                 Microsoft Macro Assembler
 $(BFLAGS)                 Microsoft Basic Compiler
 $(CFLAGS)                 Microsoft C Compiler
 $(COBFLAGS)               Microsoft COBOL Compiler
 $(FFLAGS)                 Microsoft FORTRAN Compiler
 $(PFLAGS)                 Microsoft Pascal Compiler
 $(RFLAGS)                 Microsoft Resource Compiler
 ────────────────────────────────────────────────────────────────────────────



 10.3.4.4  Substitution within Macros

 You can replace text in a macro as well as in the description file.

 Just as macros allow you to substitute text in a description file, you can
 also substitute text within a macro itself. The substitution is temporary;
 it applies only to the current use of the macro and does not modify the
 original macro definition. Use the following form:

   $(macroname:string1=string2)

 Every occurrence of string1 is replaced by string2 in the macro macroname.
 Do not put any spaces or tabs between macroname and the colon. Spaces
 between the colon and string1 or between string1 and the equal sign are part
 of string1. Spaces between the equal sign and string2 or between string2 and
 the right parenthesis are part of string2. If string2 is a null string, all
 occurrences of string1 are deleted from the macroname macro.

 Macro substitution is case sensitive. This means that the case as well as
 the characters in string1 must exactly match the target string in the macro,
 or the substitution is not performed. It also means that the string2
 substitution is exactly as specified.


 Example 1

 The following description file illustrates macro substitution:

   SOURCES = project.for one.for two.for

   project.exe : $(SOURCES:.for=.obj)
       LINK  $**;

   COPY : $(SOURCES)
       !COPY $** c:\backup

 The predefined macro $** stands for the names of all the dependent files
 (see Table 10.2).

 If you invoke the example file with a command line that specifies both
 targets,

   NMAKE project.exe copy

 NMAKE executes the following commands:

   LINK project.obj one.obj two.obj;
   COPY project.for c:\backup
   COPY one.for c:\backup
   COPY two.for c:\backup

 The macro substitution does not alter the  SOURCES  macro definition.
 Rather, it replaces the listed characters. When NMAKE builds the target
 PROJECT.EXE, it gets the definition for the predefined macro $** (the
 dependent list) from the dependency line, which specifies the macro
 substitution in  SOURCES.

 The same is true for the second target,  COPY. In this case, however, no
 macro substitution is requested, so  SOURCES  retains its original value,
 and $** represents the names of the FORTRAN source files. (In the example
 above, the target  COPY  is a pseudotarget; Section 10.3.2 describes
 pseudotargets.)


 Example 2

 If the macro OBJS is defined as

   OBJS=ONE.OBJ TWO.OBJ THREE.OBJ

 with exactly one space between each object name, you can replace each space
 in the defined value of OBJS with a space, followed by a plus sign, followed
 by a newline, by using

   $(OBJS: = +^
   )

 The caret (^) tells NMAKE to treat the end of the line as a literal newline
 character. This example is useful for creating response files.


 10.3.4.5  Substitution within Predefined Macros

 You can also substitute text in any predefined macro except $$@. The
 principle is the same as for other macros. The command in the following
 description block substitutes within a predefined macro. Note that even
 though $@ is a singlecharacter macro, the substitution makes it a
 multi-character macro invocation, so it must be enclosed in parentheses.

   target.abc : depend.xyz
       echo $(@:targ=blank)

 If dependent  depend.xyz  has a later modification time than target
 target.abc, then NMAKE executes the command

   echo blanket.abc

 The example uses the predefined macro $@, which equals the full name of the
 current target (target.abc). It substitutes  blank  for  targ  in the
 target, resulting in  blanket.abc.


 10.3.4.6  Inherited Macros

 When NMAKE executes, it inherits macro definitions equivalent to every
 environment variable. The inherited macro names are converted to uppercase.


 Inherited macros can be used like other macros. You can also redefine them.
 The following example redefines the inherited macro  PATH:

   PATH = c:\tools\bin

   sample.exe : sample.obj
       LINK sample;

 Inherited macros take their definitions from environment variables.

 No matter what value the environment variable PATH had before, it has the
 value  c:\tools\bin  when NMAKE executes the LINK command in this
 description block. Redefining the inherited macro does not affect the
 original environment variable; when NMAKE terminates, PATH still has its
 original value.

 Inherited macros have one restriction: in a recursive call to NMAKE, the
 only macros that are preserved are those defined on the command line or in
 environment variables. Macros defined in the description file are not
 inherited when NMAKE is called recursively. To pass a macro to a recursive
 call:


   ■   Use the SET command before the recursive call to set the variable for
       the entire NMAKE session.

   ■   Define the macro on the command line for the recursive call.


 The /E option causes macros inherited from environment variables to override
 any macros with the same name in the description file.


 10.3.4.7  Precedence among Macro Definitions

 If you define the same macro name in more than one place, NMAKE uses the
 macro with the highest precedence. The precedence from highest to lowest is
 as follows:


   1.  A macro defined on the command line

   2.  A macro defined in a description file or include file

   3.  An inherited environment-variable macro

   4.  A macro defined in the TOOLS.INI file

   5.  A predefined macro such as CC and AS



 10.3.5  Inference Rules

 Inference rules are templates that define how a file with one extension is
 created from a file with a different extension. When NMAKE encounters a
 description block that has no commands, it searches for an inference rule
 that matches the extensions of the target and dependent files. Similarly, if
 a dependent file doesn't exist, NMAKE looks for an inference rule that shows
 how to create the missing dependent from another file with the same base
 name.

 Inference rules tell NMAKE how to create files with a specific extension.

 Inference rules provide a convenient shorthand for common operations. For
 instance, you can use an inference rule to avoid repeating the same command
 in several description blocks. You can define your own inference rules or
 use predefined inference rules.

 ────────────────────────────────────────────────────────────────────────────
 NOTE
 An inference rule is useful only when a target and dependent have the same
 base name, and have a one-to-one correspondence. For example, you cannot
 define an inference rule that replaces several modules in a library, because
 the modules would have different base names than the target library.
 ────────────────────────────────────────────────────────────────────────────

 Inference rules can exist only for dependents with extensions that are
 listed in the .SUFFIXES directive. (For information on the .SUFFIXES
 directive, see Section 10.3.6, "Directives.") NMAKE searches in the current
 or specified directory for a file whose base name matches the target and
 whose extension is listed in the .SUFFIXES list. If it finds such a file, it
 applies the inference rule that matches the extensions of the target and the
 located file.

 The .SUFFIXES list specifies an order of priority for NMAKE to use when
 searching for files. If more than one file is found, and thus more than one
 rule matches a dependency line, NMAKE searches the .SUFFIXES list and uses
 the rule whose extension appears earlier in the list. For example, the
 dependency line

   project.exe :

 can be matched to several predefined inference rules and possibly one or
 more user-defined rules, all of which describe a command for creating an
 .EXE file. NMAKE uses the inference rule corresponding to the first matching
 file it finds.


 10.3.5.1  Inference Rule Syntax

 An inference rule has the following syntax:

   .fromext.toext:
      commands

 The first line lists two extensions: fromext extension represents the
 filename extension of a dependent file, and toext represents the extension
 of a target file. Extensions are not case sensitive.

 The second line of the inference rule gives the command to create a target
 file of toext from a dependent file of fromext. Use the same rules for
 commands in inference rules as in description blocks. (See Section 10.3.1,
 "Description Blocks.")


 10.3.5.2  Inference Rule Search Paths

 The inference-rule syntax described above tells NMAKE to look for the
 specified files in the current directory. You can also specify directories
 to be searched by NMAKE when it looks for files with the extensions fromext
 and toext. An inference rule that specifies paths has the following syntax:


   {frompath}.fromext {topath}.toext:
       commands

 NMAKE searches in the frompath directory for files with the fromext
 extension. It uses commands to create files with the toext extension in the
 topath directory, if the fromext file has a later modification time than the
 toext file.

 The paths in the inference rule must exactly match the paths explicitly
 specified in the dependency line of a description block.

 If you use a path on one element of the inference rule, you must use paths
 on both. You can specify the current directory for either element by using
 the operating system notation for the current directory, which is a dot (.),
 or by specifying an empty pair of braces.

 You can specify only one path for each element in an inference rule. To
 specify more than one path, repeat the inference rule with the alternate
 path.


 10.3.5.3  User-Defined Inference Rules

 You can define inference rules in the description file or in TOOLS.INI (see
 Section 10.6, "The TOOLS.INI File"). An inference rule lists two file
 extensions and one or more commands.


 Example 1

 The following inference rule tells NMAKE how to build a .OBJ file from a .C
 file:

   .c.obj:
       CL /c $<

 In this example, the predefined macro $< represents the name of a dependent
 that has a more recent modification time than the target.

 NMAKE applies this inference rule to the following description block:

   sample.obj :

 The description block lists only a target, SAMPLE.OBJ. Both the dependent
 and the command are missing. However, given the target's base name and
 extension, plus the inference rule, NMAKE has enough information to build
 the target.

 NMAKE first looks for a file with the same base name as the target and with
 one of the extensions in the .SUFFIXES list. If SAMPLE.C exists (and no
 files with higher-priority extensions exist), NMAKE compares its time to
 that of SAMPLE.OBJ. If SAMPLE.C has changed more recently, NMAKE compiles it
 using the CL command listed in the inference rule:

   CL /c sample.c


 Example 2

 The following inference rule compares a .C file in the current directory
 with the corresponding .OBJ file in another directory:

   {.}.c{c:\objects}.obj:
       cl /c $<;

 The path for the .C file is represented by a dot. A path for the dependent
 extension is required because one is specified for the target extension.

 This inference rule matches a dependency line containing the same
 combination of paths, such as:

   c:\objects\test.obj : test.c

 This rule does not match a dependency line such as:

   test.obj : test.c

 In this case, NMAKE uses the predefined inference rule .c.obj when building
 the target.


 10.3.5.4  Predefined Inference Rules

 NMAKE provides predefined inference rules containing commands for creating
 object, executable, and resource files. Table 10.6 describes the predefined
 inference rules.

 Table 10.6  Predefined Inference Rules

╓┌──────────┌─────────────────────────────────────┌──────────────────────────╖
 Rule       Command                               Default Action
 ────────────────────────────────────────────────────────────────────────────
 .asm.obj   $(AS) $(AFLAGS) /c $*.asm             ML /c $*.ASM
 .asm.exe   $(AS) $(AFLAGS) $*.asm                ML $*.ASM
 .bas.obj   $(BC) $(BFLAGS) $*.bas;               BC $*.BAS;
 .c.obj     $(CC) $(CFLAGS) /c $*.c               CL /c $*.C
 .c.exe     $(CC) $(CFLAGS) $*.c                  CL $*.C
 .cbl.obj   $(COBOL) $(COBFLAGS) $*.cbl;          COBOL $*.CBL;
 .cbl.exe   $(COBOL) $(COBFLAGS) $*.cbl, $*.exe;  COBOL $*.CBL, $*.EXE;
 .for.obj   $(FOR) /c $(FFLAGS) $*.for            FL /c $*.FOR
 .for.exe   $(FOR) $(FFLAGS) $*.for               FL $*.FOR
 .pas.obj   $(PASCAL) /c $(PFLAGS) $*.pas         PL /c $*.PAS
 .pas.exe   $(PASCAL) $(PFLAGS) $*.pas            PL $*.PAS
 .rc.res    $(RC) $(RFLAGS) /r $*                 RC /r $*
 ────────────────────────────────────────────────────────────────────────────


 For example, assume you have the following description file:

   sample.exe :

 This description block lists a target without any dependents or commands.
 NMAKE looks at the target's extension (.EXE) and searches for an inference
 rule that describes how to create an .EXE file. Table 10.6 shows that more
 than one inference rule exists for building an .EXE file. NMAKE looks for a
 file in the current or specified directory that has the same base name as
 the target  sample  and one of the extensions in the .SUFFIXES list. For
 example, if a file called SAMPLE.FOR exists, NMAKE applies the  .for.exe
 inference rule. If more than one file with the base name SAMPLE is found,
 NMAKE applies the inference rule for the extension listed earliest in the
 .SUFFIXES list. In this example, if both SAMPLE.C and SAMPLE.FOR exist,
 NMAKE uses the  .c.exe inference rule to compile SAMPLE.C and links the
 resulting file SAMPLE.OBJ to create SAMPLE.EXE.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 By default, the options macros such as CFLAGS shown in Table 10.5 are
 undefined. As explained in Section 10.3.4.2, "Using Macros," this causes no
 problem; NMAKE replaces an undefined macro with a null string. Because the
 predefined options macros are included in the inference rules, you can
 define these macros and have their assigned values passed automatically to
 the predefined inference rules. The predefined inference rules are listed in
 Table 10.6.
 ────────────────────────────────────────────────────────────────────────────


 10.3.5.5  Precedence among Inference Rules

 If the same inference rule is defined in more than one place, NMAKE uses the
 rule with the highest precedence. The precedence from highest to lowest is


   1.  An inference rule defined in the description file. If more than one,
       the last one applies.

   2.  An inference rule defined in the TOOLS.INI file. If more than one, the
       last one applies.

   3.  A predefined inference rule.


 User-defined inference rules always override predefined inference rules.
 NMAKE uses a predefined inference rule only if no user-defined inference
 rule exists for a given target and dependent.

 If two inference rules could produce a target with the same extension, NMAKE
 uses the inference rule whose dependent's extension appears first in the
 .SUFFIXES list. See Table 10.7 in the next section, "Directives."


 10.3.6  Directives

 The directives in Table 10.7 provide additional control of NMAKE operations.
 You can use them in a description file outside of a description block or in
 the TOOLS.INI file. The four directives listed in the table are case
 sensitive and must appear in all uppercase letters. (Preprocessing
 directives are not case sensitive; see Section 10.3.7, "Preprocessing
 Directives.")

 Table 10.7  Directives

 Directive                         Action
 ────────────────────────────────────────────────────────────────────────────
 .IGNORE :                         Ignores exit codes returned by programs
                                   called from the description file. This
                                   directive has the same effect as
                                   invoking NMAKE with the /I option.

 .PRECIOUS : target...             Tells NMAKE not to delete targets if the
                                   commands that build them quit or are
                                   interrupted. Overrides the NMAKE default,
                                   which is to delete the target if
                                   building was interrupted by CTRL+C or
                                   CTRL+BREAK.

 .SILENT :                         Does not display lines as they are
                                   executed. This directive has the same
                                   effect as invoking NMAKE with the /S
                                   option.

 .SUFFIXES : list                  Lists file suffixes for NMAKE to try
                                   when building a target file for which no
                                   dependents are specified. This list is
                                   used together with inference rules. See
                                   Section
                                   10.3.5, "Inference Rules."

 ────────────────────────────────────────────────────────────────────────────


 The .IGNORE and .SILENT directives affect the file from their location
 onward. Location within the file does not matter for the .PRECIOUS and
 .SUFFIXES directives; they affect the entire description file.

 NMAKE refers to the value of the .SUFFIXES directive when using inference
 rules. When NMAKE finds a target without dependents, it searches the current
 directory for a file with the same base name as the target and a suffix from
 list. If NMAKE finds such a file, and if an inference rule applies to the
 file, then NMAKE treats the file as a dependent of the target. The order of
 the suffixes in the list defines the order in which NMAKE searches for the
 file. The list is predefined as follows:

   .SUFFIXES : .exe .obj .asm .c .bas .cbl .for .pas .res .rc

 To add additional suffixes to the end of the list, specify  .SUFFIXES :
 followed by the additional suffixes. To clear the list, specify  .SUFFIXES :
  by itself. To change the list order or to specify an entirely new list,
 clear the list and specify a new  .SUFFIXES :  setting.


 10.3.7  Preprocessing Directives

 NMAKE preprocessing directives are similar to compiler preprocessing
 directives. You can use the !IF, !IFDEF, !IFNDEF, !ELSE, and !ENDIF
 directives to conditionally process the description file. With other
 preprocessing directives you can display error messages, include other
 files, undefine a macro, and turn certain options on or off. NMAKE reads and
 executes the preprocessing directives before processing the description file
 as a whole.

 Preprocessing directives (listed in Table 10.8) begin with an exclamation
 point (!), which must appear at the beginning of the line. You can place
 spaces between the exclamation point and the directive keyword. These
 directives are not case sensitive.

 Table 10.8  Preprocessing Directives

╓┌─────────────────────────┌─────────────────────────────────────────────────╖
 Directive                 Description
 ────────────────────────────────────────────────────────────────────────────
 !CMDSWITCHES              Turns on or off NMAKE options /D, /I, /N, and /S.
 {+| -}opt...              (See Section 10.4, "Command-Line Options.") Do
                           not specify the slash ( / ). If !CMDSWITCHES is
                           specified with no options, all options are reset
                           to the values they had when NMAKE was started.
                           This directive updates the MAKEFLAGS macro. Turn
                           an option on by preceding it with a plus sign (+
                           ), or turn it off by preceding it with a minus
                           sign (-).

 !ERROR text               Prints text, then stops execution.

 !IF constantexpression    Reads the statements between the !IF keyword and
                           the next !ELSE or !ENDIF keyword if
                           constantexpression evaluates to a nonzero value.

 Directive                 Description
 ────────────────────────────────────────────────────────────────────────────

 !IFDEF macroname          Reads the statements between the !IFDEF keyword
                           and the next !ELSE or !ENDIF keyword if
                           macroname is defined. NMAKE considers a macro
                           with a null value to be defined.

 !IFNDEF macroname         Reads the statements between the !IFNDEF keyword
                           and the next !ELSE or !ENDIF keyword if
                           macroname is not defined.

 !ELSE                     Reads the statements between the !ELSE and
                           !ENDIF keywords if the preceding !IF, !IFDEF, or
                           !IFNDEF statement evaluated to zero. Anything
                           following !ELSE on the same line is ignored.

 !ENDIF                    Marks the end of an !IF, !IFDEF, or !IFNDEF
                           block. Anything following !ENDIF on the same
                           line is ignored.

 Directive                 Description
 ────────────────────────────────────────────────────────────────────────────


 !INCLUDE filename         Reads and evaluates the description file
                           filename before continuing with the current
                           description file. If filename is enclosed by
                           angle brackets (< >), NMAKE searches for the
                           file first in the current directory and then in
                           the directories specified by the INCLUDE macro.
                           Otherwise, it looks only in the current
                           directory. The INCLUDE macro is initially set to
                           the value of the INCLUDE environment variable.

 !UNDEF macroname          Marks macroname as undefined in NMAKE's symbol
                           table.

 ────────────────────────────────────────────────────────────────────────────




 10.3.7.1  Expressions in Preprocessing

 The constantexpression used with the !IF directive can consist of integer
 constants, string constants, or program invocations. Integer constants can
 use the unary operators for numerical negation (-), one's complement (~),
 and logical negation (!). They can also use any binary operator listed in
 Table 10.9.

 Table 10.9  Preprocessing-Directive Binary Operators

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Operator                          Description
 ────────────────────────────────────────────────────────────────────────────
 +                                 Addition

 -                                 Subtraction

 *                                 Multiplication

 /                                 Division
 Operator                          Description
 ────────────────────────────────────────────────────────────────────────────
/                                 Division

 %                                 Modulus

 &                                 Bitwise AND

 |                                 Bitwise OR

 ^                                 Bitwise XOR

 &&                                Logical AND

 ||                                Logical OR

 <<                                Left shift

 >>                                Right shift

 ==                                Equality
 Operator                          Description
 ────────────────────────────────────────────────────────────────────────────
==                                Equality

 !=                                Inequality

 <                                 Less than

 >                                 Greater than

 <=                                Less than or equal to

 >=                                Greater than or equal to

 ────────────────────────────────────────────────────────────────────────────



 You can group expressions by enclosing them in parentheses. NMAKE treats
 numbers as decimal unless they start with 0 (octal) or 0x (hexadecimal). Use
 the equality (==) operator to compare two strings for equality, or the
 inequality (!=) operator to compare for inequality. Enclose strings in
 double quotation marks.


 Example

 The following example shows how preprocessing directives can be used to
 control whether the linker inserts CodeView information into the .EXE file:


   !INCLUDE <infrules.txt>
   !CMDSWITCHES +D
   winner.exe : winner.obj
   !IFDEF debug
   !   IF "$(debug)"=="y"
          LINK /CO winner.obj;
   !  ELSE
          LINK winner.obj;
   !   ENDIF
   !ELSE
   !   ERROR Macro named debug is not defined.
   !ENDIF

 In this example, the !INCLUDE directive inserts the INFRULES.TXT file into
 the description file. The !CMDSWITCHES directive sets the /D option, which
 displays the times of the files as they are checked. The !IFDEF directive
 checks to see if the macro  debug  is defined. If it is defined, the !IF
 directive checks to see if it is set to  y. If it is, NMAKE reads the LINK
 command with the /CO option; otherwise, NMAKE reads the LINK command without
 /CO. If the  debug  macro is not defined, the !ERROR directive prints the
 specified message and NMAKE stops.


 10.3.7.2  Executing a Program in Preprocessing

 NMAKE can invoke programs and check their status.

 You can invoke any program from within NMAKE by placing the program's name
 or path name within square brackets ( [ ] ). The program is executed during
 preprocessing, and its exit code replaces the program specification in the

 description file. A nonzero exit code usually indicates an error. You can
 use this value to control execution, as in the following example:

   !IF [c:\util\checkdsk] != 0
   !    ERROR Not enough disk space; NMAKE terminating.
   !ENDIF


 10.3.8  Extracting Filename Components

 "Special Macros," Section 10.3.4.3, showed how qualifiers could be added to
 macros that represented filenames in order to select components of the name
 or path. This feature is especially useful when creating a general-purpose
 description block that works with the name of any dependent.

 Besides these macro modifiers, NMAKE offers another feature that allows you
 to extract components of the name of the first dependent file as you have
 specified it in the description file or on the command line (not the full
 filename specification on disk). The components can then be recombined with
 specific paths, extensions, or directories to create the particular name or
 path you need, without having to specify the exact name or path when you
 write the description block.

 The first dependent file is the first file listed to the right of the colon
 on a dependency line. If a dependent is implied from an inference rule,
 NMAKE considers it to be the first dependent file. If more than one
 dependent is implied from inference rules, the .SUFFIXES list determines
 which dependent is first.

 You can use either of the following syntaxes:

   %s

   %|«parts»F

 where parts can be one or more of the following letters, or can be omitted:


 Letter                    Description
 ────────────────────────────────────────────────────────────────────────────
 No letter                 Complete name
 d                         Drive
 p                         Path
 f                         File base name
 e                         File extension

 You can specify more than one letter. The order of the letters is not
 significant; NMAKE constructs the filename that meets (or comes closest to
 meeting) all the specifications. The letters are case sensitive.

 The  %s  option substitutes the complete name; it is equivalent to both  %|F
  and  %|dpfeF.

 NMAKE interprets any percent symbol (%) within a command line (either in a
 description block or an inference rule) as the start of a file specifier
 using this syntax. Therefore, if you need to use a literal percent symbol
 within a command line, you must specify it as a double percent symbol (%%).



 Example

 The following example demonstrates this special syntax:

   sample.exe : c:\project\sample.obj
       LINK %|dpfF, a:%|pfF.exe;

 This example represents the following command:

   LINK c:\project\sample, a:\project\sample.exe;

 In this example, the sequence  %|dpfF  represents the same drive, path, and
 base name as the dependent on the dependency line, while the sequence  %|pfF
  represents only the path and base name of the dependent. The command tells
 the LINK utility to build the executable file on another drive in a
 directory of the same name.


 10.4  Command-Line Options

 NMAKE accepts a number of options, listed in Table 10.10. You can specify
 options in uppercase or lowercase and use either a slash or dash. For
 example, -A, /A, -a, and /a all represent the same option. This book uses a
 slash and uppercase letters.

 Table 10.10  NMAKE Options

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Option                            Action
 ────────────────────────────────────────────────────────────────────────────
 /A                                Forces execution of all commands in
                                   description blocks in the description
                                   file even if targets are not out-of-date
                                   with respect to their dependents. Does
                                   not affect the behavior of incremental
                                   commands such as ILINK; using /A does
                                   not force a full link.

 /C                                Suppresses nonfatal error or warning
                                   messages and the NMAKE copyright message.

 /D                                Displays the modification time of each
                                   file.

 /E                                Causes environment variables to override
                                   macro definitions in description files.
 Option                            Action
 ────────────────────────────────────────────────────────────────────────────
                                  macro definitions in description files.
                                   See Section 10.3.4, "Macros."

 /F filename                       Specifies filename as the name of the
                                   description file. If you supply a dash (
                                   -) instead of a filename, NMAKE gets
                                   description-file input from the standard
                                   input device. (Terminate keyboard input
                                   with either F6 or CTRL+Z.) If you omit
                                   /F, NMAKE searches the current directory
                                   for a file called
                                   MAKEFILE and uses it as the description
                                   file. If MAKEFILE doesn't exist, NMAKE
                                   uses inference rules for the
                                   command-line targets.

 /HELP                             Calls the QuickHelp utility. If NMAKE
                                   cannot locate the help file or QuickHelp,
                                   it displays a brief summary of NMAKE
 Option                            Action
 ────────────────────────────────────────────────────────────────────────────
                                  it displays a brief summary of NMAKE
                                   command-line syntax and exits to the
                                   operating system.

 /I                                Ignores exit codes from commands listed
                                   in the description file. NMAKE processes
                                   the whole description file even if
                                   errors occur.

 /N                                Displays but does not execute the
                                   description file's commands. This option
                                   is useful for debugging description
                                   files and checking which targets are
                                   out-of-date.

 /NOLOGO                           Suppresses the NMAKE copyright message.

 /P                                Displays all macro definitions,
                                   inference rules, target descriptions,
 Option                            Action
 ────────────────────────────────────────────────────────────────────────────
                                  inference rules, target descriptions,
                                   and the .SUFFIXES list on the standard
                                   output device.

 /Q                                Checks modification times for
                                   command-line targets (or first target in
                                   description file if no command-line
                                   targets are specified). NMAKE returns a
                                   zero exit code if all such targets are
                                   up-to-date and a nonzero exit code if
                                   any target is out-of-date. Only
                                   preprocessing commands in the
                                   description file are executed. This
                                   option is useful when running NMAKE from
                                   a batch file.

 /R                                Ignores inference rules and macros that
                                   are defined in the TOOLS.INI file or
                                   that are predefined.
 Option                            Action
 ────────────────────────────────────────────────────────────────────────────
                                  that are predefined.

 /S                                Suppresses the display of commands
                                   listed in the description file.

 /T                                Changes modification times for
                                   command-line targets (or first target in
                                   description file if no command-line
                                   targets are specified). Only
                                   preprocessing commands in the
                                   description file are executed. Contents
                                   of target files are not modified.

 /X filename                       Sends all error output to filename,
                                   which can be a file or a device. If you
                                   supply a dash (-) instead of a filename,
                                   error output is sent to the standard
                                   output device.

 Option                            Action
 ────────────────────────────────────────────────────────────────────────────

 /Z                                Used for internal communication between
                                   NMAKE (or NMK) and PWB.

 /?                                Displays a brief summary of NMAKE
                                   command-line syntax and exits to the
                                   operating system.

 ────────────────────────────────────────────────────────────────────────────




 Example

 The following command line specifies two NMAKE options:

   NMAKE /F sample.mak /C targ1 targ2

 The /F option tells NMAKE to read the description file SAMPLE.MAK. The /C
 option tells NMAKE not to display nonfatal error messages and warnings. The
 command specifies two targets (targ1  and  targ2) to update.

 In the following example, NMAKE updates the target  targ1:

   NMAKE /D /N targ1

 Since no description file is specified, NMAKE searches the current directory
 for a description file named MAKEFILE. The /D option displays the
 modification time of each file; the /N option displays the commands in
 MAKEFILE without executing them.


 10.5  NMAKE Command File

 If you find yourself repeatedly using the same sequence of command-line
 arguments, you can place them in a text file and pass the file's name as a
 command-line argument to NMAKE. NMAKE opens the command file and reads the
 arguments. This feature is especially useful if the argument list exceeds
 the maximum length of a command line (128 characters in DOS, 256 in OS/2).

 To provide input to NMAKE with a command file, type

 NMAKE @commandfile

 In the commandfile field, enter the name of a file containing the
 information NMAKE expects on the command line. You can split input between
 the command line and a command file. Use the name of the command file
 (preceded by @) in place of the input information on the command line.


 Example 1

 Assume you have created a filenamed UPDATE containing this line:

   /S "program = sample" sort.exe search.exe

 If you start NMAKE with the command

   NMAKE @update

 then NMAKE reads its command-line arguments from UPDATE. The at sign (@)
 tells NMAKE to read arguments from the file. The effect is the same as if
 you typed the arguments directly on the command line:

   NMAKE /S "program = sample" sort.exe search.exe

 NMAKE treats the file as if it were a single set of arguments and replaces
 each line break with a space. Macro definitions that contain spaces must be
 enclosed in quotation marks, just as if you had typed them on the command
 line.

 The quotation marks that delimit a macro force all characters between them
 to be interpreted literally. Therefore, if you split a macro between lines,
 an unwanted line break is inserted into the macro. Macros that span multiple
 lines must be continued by ending each line except the last with a backslash
 ( \ ):

   /S "program \
   = sample" sort.exe search.exe

 This file is equivalent to the first example. The backslash allows the macro
 definition ("program = sample") to span two lines.


 Example 2

 If the command-file UPDATE contains this line:

   /S "program = sample" sort.exe

 you can give NMAKE the same command-line input as in the example above by
 specifying the command

   NMAKE @update search.exe


 10.6  The TOOLS.INI File

 You can customize NMAKE by placing commonly used macros, inference rules,
 and description blocks in the TOOLS.INI initialization file. Settings for
 NMAKE must follow a line that begins with [NMAKE]. This section of the
 initialization file can contain macro definitions, .SUFFIXES lists, and
 inference rules. For example, if TOOLS.INI contains the following section:

   [NMAKE]
   CC=qcl
   CFLAGS=/Gc /Gs /W3 /Oat
   .c.obj:
       $(CC) /c $(CFLAGS) $*.c

 NMAKE reads and applies the lines following  [NMAKE]. The example redefines
 the macro CC to invoke the Microsoft QuickC (R) Compiler, defines the macro
 CFLAGS, and redefines the inference rule for making .OBJ files from .C
 sources. (Note that macros are case sensitive; a macro called cc is not
 substituted in a rule that uses $(CC).)

 NMAKE looks for TOOLS.INI in the current directory. If it isn't there, NMAKE
 searches the directory specified by the INIT environment variable.

 Macros and inference rules appearing in TOOLS.INI can be overridden. See
 Section 10.3.4.7, "Precedence among Macro Definitions," and Section
 10.3.5.5, "Precedence among Inference Rules."


 10.7  Inline Files

 NMAKE can create "inline files" which contain any text you specify. One use
 of inline files is to write a response file for another utility such as LINK
 or LIB. This eliminates the need to maintain a separate response file and
 removes the restraint on the maximum length of a command line.

 Use this syntax to create an inline file called filename:

 target : dependents  command << «filename» inlinetext <<«KEEP | NOKEEP»

 All inlinetext between the two sets of double angle brackets (<<) is placed
 in the inline file. The filename is optional. If you don't supply filename,
 NMAKE gives the inline file a unique name. NMAKE places the inline file in
 the directory specified by the TMP environment variable. If TMP is not
 defined, the inline file is placed in the current directory.

 Directives are not allowed in an inline file. NMAKE treats a directive in an
 inline file as literal text.

 The inline file can be temporary or permanent. If you don't specify the
 option, or if you specify NOKEEP, the file is temporary. Specify KEEP to
 retain the file after the build ends.


 Example

 The following description block creates a LIB response file named LIB.LRF:

   OBJECTS=add.obj sub.obj mul.obj div.obj
   math.lib : $(OBJECTS)
       LIB @<<lib.lrf
   $*.lib
   -+$(OBJECTS: = &^
   -+)
   listing;
   <<KEEP

 The resulting response file tells LIB which library to use, the commands to
 execute, and the name of the listing file to produce:

   math.lib
   -+add.obj &
   -+sub.obj &
   -+mul.obj &
   -+div.obj
   listing;

 The file MATH.LIB must exist beforehand for this example to work.


 Multiple Inline Files

 The inline file specification can create more than one inline file. For
 instance,

   target.abc : depend.xyz
       cat <<file1 <<file2
   I am the contents of file1.
   <<KEEP
   I am the contents of file2.
   <<KEEP

 The example creates the two inline files, FILE1 and FILE2. All inline text
 is written to the files sequentially. Therefore, the text

   I am the contents of file1.

 goes into FILE1, not FILE2, even though the text is nested between the angle
 brackets for FILE2 and the  <<KEEP  statement which follows. NMAKE then
 executes the command

   cat file1 file2

 The KEEP keywords tell NMAKE not to delete FILE1 and FILE2 when done.


 10.8  Sequence of NMAKE Operations

 When you are writing a complex description file, it can be helpful to know
 the sequence in which NMAKE performs operations. This section describes
 those operations and their order.

 NMAKE first looks for a description file.

 When you run NMAKE from the command line, NMAKE's first task is to find the
 description file:


   1.  If the /F option is used, NMAKE searches for the filename specified in
       the option. If NMAKE cannot find that file, it returns an error.

   2.  If the /F option is not used, NMAKE looks for a file named MAKEFILE in
       the current directory. If there are targets on the command line, NMAKE
       builds them according to the instructions in MAKEFILE. If there are no
       targets on the command line, NMAKE builds only the first target it
       finds in MAKEFILE.

   3.  If NMAKE cannot find MAKEFILE, NMAKE looks for target files on the
       command line and attempts to build them using inference rules (either
       defined by the user in TOOLS.INI or predefined by NMAKE). If no target
       is specified, NMAKE returns an error.


 Macro definitions follow a priority.

 NMAKE then assigns macro definitions with the following precedence (highest
 first):


   1.  Macros defined on the command line

   2.  Macros defined in a description file or include file

   3.  Inherited macros

   4.  Macros defined in the TOOLS.INI file

   5.  Predefined macros (such as CC and RFLAGS)


 Macro definitions are assigned in order of priority, not in the order in
 which NMAKE encounters them. For example, a macro defined in an include file
 overrides a macro with the same name from the TOOLS.INI file. Note that a
 macro within a description file can be redefined; the most recent definition
 in the description file is used.

 Inference rules also follow a priority.

 NMAKE also assigns inference rules, using the following precedence (highest
 first):


   1.  Inference rules defined in a description file or include file

   2.  Inference rules defined in the TOOLS.INI file

   3.  Predefined inference rules (such as .c.obj)


 You can use command-line options to change some of these precedences.


   ■   The /E option allows macros inherited from the environment to override
       macros defined in the description file.

   ■   The /R option tells NMAKE to ignore macros and inference rules that
       are defined in TOOLS.INI or are predefined.


 NMAKE preprocesses directives before running the description-file commands.


 Next, NMAKE evaluates any preprocessing directives. If an expression for
 conditional preprocessing contains a program in square brackets ( [ ] ), the
 program is invoked during preprocessing, and the program's exit code is used
 in the expression. If an !INCLUDE directive is specified for a file, NMAKE
 preprocesses the included file before continuing to preprocess the rest of
 the description file. Preprocessing determines the final description file
 that NMAKE reads.

 NMAKE updates targets in the description file.

 NMAKE is now ready to update the targets. If you specified targets on the
 command line, NMAKE updates only those targets. If you did not specify
 targets on the command line, NMAKE updates just the first target it finds in
 the description file. (This behavior differs from the MAKE utility's
 default; see Section 10.10, "Differences between NMAKE and MAKE.") If you
 specify a pseudotarget, NMAKE always updates the target. If you use the /A
 option, NMAKE always updates the target, even if the file is not
 out-of-date.

 If the dependents of the targets are themselves out-of-date or do not exist
 yet, NMAKE updates them first. If the target has no explicit dependent,
 NMAKE looks in the current directory for one or more files with the same
 base name as the target and whose extensions are in the .SUFFIXES list. (See
 Section 10.3.6, "Directives," for a description of the .SUFFIXES list.) If
 it finds such files, NMAKE treats them as dependents and updates the target
 according to the commands.

 Errors usually stop the build.

 NMAKE normally stops processing the description file when a command returns
 a nonzero exit code. In addition, if NMAKE cannot tell whether the target
 was built successfully, it deletes the target. If you use the /I
 command-line option, NMAKE ignores error codes and attempts to continue
 processing. The .IGNORE directive has the same effect as the /I option. To
 prevent NMAKE from deleting the partially created target if you interrupt
 the build with CTRL+C or CTRL+BREAK, specify the target name in the
 .PRECIOUS directive.

 Alternatively, you can use the dash (-) command modifier to ignore the error
 code for an individual command. An optional number after the dash tells
 NMAKE to continue if the command returns an exit code that is less than or
 equal to the number, and to stop if the exit code is greater than the
 number.

 You can document errors by using the !ERROR directive to print descriptive
 text. The directive causes NMAKE to print some text, then stop, even if you
 use /I, .IGNORE, or the dash (-) modifier.


 10.9  A Sample NMAKE Description File

 The following example illustrates many of NMAKE's features. The description
 file creates an executable file from C-language source files:

   #  This description file builds SAMPLE.EXE from SAMPLE.C,
   #  ONE.C, and TWO.C, then deletes intermediate files.

   CFLAGS   = /c /AL /Od $(CODEVIEW)  # controls compiler options
   LFLAGS   = /CO                     # controls linker options
   CODEVIEW = /Zi                     # controls CodeView data

   OBJS = sample.obj one.obj two.obj
   all : sample.exe

   sample.exe : $(OBJS)
       link $(LFLAGS) @<<sample.lrf
   $(OBJS: =+^
   )
   sample.exe
   sample.map;
   <<KEEP

   sample.obj : sample.c sample.h common.h
        CL $(CFLAGS) sample.c

   one.obj : one.c one.h common.h
        CL $(CFLAGS) one.c

   two.obj : two.c two.h common.h
        CL $(CFLAGS) two.c

   clean :
       -del *.obj
       -del *.map
       -del *.lrf

 Assume that this description file is named SAMPLE.MAK. To invoke it, enter

   NMAKE /F SAMPLE.MAK all clean

 NMAKE then builds SAMPLE.EXE and deletes intermediate files.

 Here is how the description file works. The CFLAGS, CODEVIEW, and LFLAGS
 macros define the default options for the compiler, linker, and inclusion of
 CodeView information. You can redefine these options from the command line
 to alter or delete them. For example,

   NMAKE /F SAMPLE.MAK CODEVIEW= CFLAGS= all clean

 creates an .EXE file that does not contain CodeView information.

 The OBJS macro specifies the object files that make up SAMPLE.EXE, so they
 can be reused without having to type them again. Their names are separated
 by exactly one space so that the space can be replaced with a plus sign (+)
 and a carriage return in the link response file. (This is illustrated in the
 second example in Section 10.3.4.4, "Substitution within Macros.")

 The  all  pseudotarget points to the real target,  SAMPLE.EXE. If you do not
 specify any target on the command line, NMAKE ignores the  clean
 pseudotarget but still builds  all, since  all  is the first target in the
 description file.

 The dependency line containing the target  sample.exe  makes the object
 files specified in OBJS the dependents of SAMPLE.EXE. The command section of
 the block contains only link instructions. No compilation instructions are
 given, since they are given explicitly later in the file. (You could also
 define an inference rule to specify how an object file is to be created from
 a C source file.)

 The link command is unusual in that the link parameters and options are not
 passed directly to LINK. Rather, an inline response file is created
 containing these elements. This eliminates the need to maintain a separate
 link response file. It also allows the LINK command line to exceed the
 normal limit on the length of a command line (128 characters in DOS, 256
 characters in OS/2).

 The next three dependencies define the relationship of the source code to
 the object files. The .H (header or include) files are also dependents,
 since any changes to them would require recompilation.

 The  clean  pseudotarget deletes unneeded files after a build. The dash
 modifier (-) tells NMAKE to ignore errors returned by the deletion commands.
 If you want to save any of these files, don't specify  clean  on the command
 line; NMAKE then ignores the  clean  pseudotarget.


 10.10  Differences between NMAKE and MAKE

 NMAKE replaces the Microsoft MAKE program. NMAKE differs from MAKE in the
 following ways:


   ■   NMAKE does not evaluate targets sequentially. Instead, NMAKE updates
       the targets you specify when you invoke it, regardless of their
       positions in the description file. If no targets are specified, NMAKE
       updates only the first target in the file.

   ■   NMAKE requires a special syntax when specifying a target in more than
       one dependency line. (See Section 10.3.1.8, "Specifying a Target in
       Multiple Description Blocks.")

   ■   NMAKE accepts command-line arguments from a file.

   ■   NMAKE provides more command-line options.

   ■   NMAKE provides more predefined macros.

   ■   NMAKE permits substitutions within macros.

   ■   NMAKE supports directives placed in the description file.

   ■   NMAKE allows you to specify include files in the description file.


 The first item in the list deserves special emphasis. While MAKE updates
 every target, working from beginning to end of the description file, NMAKE
 expects you to specify targets on the command line. If you do not, NMAKE
 builds only the first target in the description file.

 This difference is clear if you run NMAKE using a typical MAKE description
 file, which lists a series of subordinate targets followed by a higher-level
 target that depends on the following subordinates:

   pmapp.obj : pmapp.c
       CL /c /G2sw /W3 pmapp.c

   pmapp.exe : pmapp.obj pmapp.def
       LINK pmapp, /align:16, NUL, os2, pmapp

 MAKE builds both targets (PMAPP.OBJ and PMAPP.EXE), but NMAKE builds only
 the first target (PMAPP.OBJ).

 Because of these performance differences, you may want to convert MAKE files
 to NMAKE files. MAKE description files are easy to convert. One way is to
 create a new description block at the beginning of the file. Give this block
 a pseudotarget named  all  and list the top-level target as a dependent of
 all. To build  all, NMAKE must update every file upon which the target  all
 depends:

   all : pmapp.exe

   pmapp.obj : pmapp.c
       CL /c /G2sw /W3 pmapp.c

   pmapp.exe : pmapp.obj pmapp.def
       LINK pmapp, /align:16, NUL, os2, pmapp

 If the above file is named MAKEFILE, you can update the target PMAPP.EXE
 with the command

   NMAKE

 or the command

   NMAKE all

 It is not necessary to list PMAPP.OBJ as a dependent of  all. NMAKE builds a
 dependency tree for the entire description file and builds whatever files
 are needed to update PMAPP.EXE. If PMAPP.C has a later modification time
 than PMAPP.OBJ, NMAKE compiles PMAPP.C to create PMAPP.OBJ, then links
 PMAPP.OBJ to create PMAPP.EXE.

 The same technique is suitable for description files with more than one
 top-level target. List all the top-level targets as dependents of  all:

   all : pmapp.exe second.exe another.exe

 The example updates the targets PMAPP.EXE, SECOND.EXE, and ANOTHER.EXE.

 If the description file lists a single, top-level target, you can use an
 even simpler technique. Move the top-level block to the beginning of the
 file:

   pmapp.exe : pmapp.obj pmapp.def
       LINK pmapp, /align:16, NUL, os2, pmapp
   pmapp.obj : pmapp.c
       CL /c /G2sw /W3 pmapp.c

 NMAKE updates the second target (PMAPP.OBJ) whenever needed to keep the
 first target (PMAPP.EXE) current.


 10.11  Using NMK

 When you maintain a project under DOS or in a DOS session under OS/2, you
 will probably need to use the NMK utility. NMK uses only 5K of memory,
 leaving room for the programs called during the build. You run NMK the same
 way you run NMAKE, using the same command-line syntax and the same
 description-file syntax. NMK calls NMAKE to read the description file and
 perform the build.

 The behavior of NMK is slightly different from that of NMAKE. The
 fundamental difference is that NMAKE rechecks the update status of all files
 after each build step, whereas NMK checks file status only once, at the
 start of the build process. If your description file simply compiles a
 series of files and then links them, this difference never causes a problem.
 But consider the following example, which uses a pseudotarget to clean up
 old files during the build:

   all : clean example.exe

   example.exe : example.asm
       ML example

   clean :
       del example.obj
       del example.exe

 This description file erases EXAMPLE.OBJ and EXAMPLE.EXE, then recompiles.
 Under NMAKE, it works as intended; that is, it


   1.  Erases files

   2.  Checks the status of EXAMPLE.EXE

   3.  Rebuilds EXAMPLE.EXE because EXAMPLE.EXE is no longer present


 However, NMK checks the status of the environment only at the beginning of
 the build. Since EXAMPLE.EXE exists when the build starts, the preceding
 description file


   1.  Erases files

   2.  Stops execution, because EXAMPLE.EXE was present and up-to-date at the
       beginning of the process


 PWB never generates a description file that requires dynamic status checking
 to run correctly, so you can use PWB-created description files with either
 NMAKE or NMK.


 10.12  Using Exit Codes with NMAKE

 NMAKE stops execution if a program executed by one of the commands in the
 NMAKE description file encounters an error. The exit code returned by the
 program is displayed as part of the error message.

 Assume the NMAKE description file  TEST  contains the following lines:

   TEST.OBJ : TEST.FOR
       FL /c TEST.FOR

 If the source code in  TEST.FOR  causes an error (but not a warning), you
 would see the following message the first time you use NMAKE with the NMAKE
 description file  TEST:

   NMAKE : fatal error U1077: 'FL /c TEST.FOR' - return code '2'

 This error message indicates that the command  FL /c TEST.FOR  in the NMAKE
 description file returned exit code 2.

 You can cause NMAKE to ignore an exit code for a command by preceding the
 command with a dash modifier (-). If you specify a number after the dash
 modifier (-n), NMAKE stops only if the exit code is greater than the
 specified number. (See Table 10.1.) You disable this behavior for the entire
 description file by invoking NMAKE with the /I option.

 You can also test exit codes in NMAKE description files with the !IF
 preprocessing directive. See Section 10.3.7.2, "Executing a Program in
 Preprocessing."

 If you prefer to use DOS batch files instead of NMAKE description files, you
 can test the code returned with the IF command. See a DOS manual for more
 information.

 NMAKE returns an exit code to the operating system or the calling program. A
 value of 0 indicates execution of NMAKE with no errors. Warnings return exit
 code 0.

 Code               Meaning
 ────────────────────────────────────────────────────────────────────────────
 0                  No error
 2                  Program error
 4                  System error─out of memory


 10.13  Related Topics in Online Help

 In addition to information covered in this chapter, information on the
 following topics can be found in online help.

 Topics                                Access
 ────────────────────────────────────────────────────────────────────────────
 Syntax and procedural information on  From the list of Utilities on the
 NMAKE                                 "Microsoft Advisor Contents" screen,
                                       choose "NMAKE"

 Using TOOLS.INI                       From the "Microsoft Advisor Contents"
                                       screen, choose "Programmer's
                                       WorkBench"; then choose "Using
                                       TOOLS.INI" from the list of topics
                                       relating to customizing PWB







 Chapter 11  Creating Help Files with HELPMAKE
 ────────────────────────────────────────────────────────────────────────────

 If you've used the Programmer's WorkBench (PWB) or one of the Microsoft
 Quick languages, you already know the advantages of online help, or the
 Microsoft Advisor. The Microsoft Help File Maintenance utility (HELPMAKE)
 lets you extend these advantages by customizing the help files supplied with
 Microsoft language products, or by creating your own help files for them.

 HELPMAKE translates help text files into a help database accessible within
 these environments:


   ■   Microsoft Programmer's WorkBench (PWB)

   ■   Microsoft QuickHelp utility

   ■   Microsoft CodeView debugger

   ■   Microsoft Editor version 1.02

   ■   Microsoft QuickC compiler versions 2.0 and later

   ■   Microsoft QuickBasic(tm) versions 4.5 and later

   ■   Microsoft QuickPascal(tm) version 1.0

   ■   Microsoft Word version 5.5


 This chapter describes how to create and modify help files using the
 HELPMAKE utility.


 11.1  Structure and Contents of a Help Database

 HELPMAKE creates a help database from one or more input files that contain
 information formatted for the help system. This section defines some of the
 terms involved in formatting and outlines the formats that HELPMAKE can
 process.


 11.1.1  Contents of a Help File

 Each help input file consists of one or more help "topics." A topic is the
 fundamental unit of help information. It is usually a screenful of
 information about a particular subject. You identify the subject by one or
 more "context strings," which are the words and phrases for which you want
 to be able to request help. When help is requested on a context string, the
 topic is displayed.

 The .context command defines a context string for the topic that follows it.
 In the source file for C help, for example, this line introduces help for
 the #include directive:

   .context #include

 The .context command and other formatting elements are described in Section
 11.5, "Help Text Conventions."

 Whether a context string contains one word or several words depends on the
 application. For example, because Microsoft QuickBasic considers spaces to
 be delimiters, a context string in QuickBasic help files is limited to a
 single word. Other applications, such as PWB, can handle context strings
 that span several words. In either case, the application hands the context
 string to an internal "help engine" that searches the database for
 information.

 Often, especially with library routines, the same information applies to
 more than one subject. For example, the C-language string-to-number
 functions strtod, strtol, and strtoul share the same help text. The help
 file lists all three function names as contexts for one block of topic text.
 The converse, however, is not true. You cannot associate a single context
 string with several blocks of topic text located at different places in the
 help file.

 Cross-references help you navigate a help database.

 Cross-references make it possible to view information about related topics,
 including header files and code examples. The help for the C-language open
 function, for example, references the access function. Cross-references can
 point to other contexts in the same help database, to contexts in other help
 databases, or even to ASCII files outside the database.

 Help files can have two kinds of cross-references:


   ■   Implicit

   ■   Explicit, or hyperlinks


 Implicit cross-references are coded with an ordinary .context command.

 The word "open" is an implicit cross-reference throughout Microsoft C help,
 and introduces help for the open function. If you select the word "open"
 anywhere in C help, the help system displays information on the open
 function. The context for open begins with an ordinary .context command. As
 a result, anywhere that you select "open," the help system references this
 context.

 Hyperlinks are explicit cross-references marked by invisible text.

 A "hyperlink" is an explicit cross-reference tied to a word or phrase at a
 specific location in the help file. You create hyperlinks when you write the
 help text. The hyperlink consists of a word or phrase followed by invisible
 text that gives the context to which the hyperlink refers.

 For example, to cause an instance of the word "formatting" to display help
 on the printf function, you would create an explicit cross-reference from
 the word "formatting" to the context "printf." Elsewhere in the file,
 "formatting" has no special significance, but at that one position, it
 references the help for printf. For details on how to create hyperlinks, see
 Section 11.5.4.

 Formatting flags let you change the appearance of text.

 Help text can also include formatting attributes to control the appearance
 of the text on the screen. Using these attributes, you can make certain
 words appear in various colors, inverse video, and so forth, depending on
 the application displaying help and the graphics capabilities of your
 computer.


 11.1.2  Help File Formats

 You can create sources for help text files in any of three formats:


   ■   QuickHelp format

   ■   Rich Text Format (RTF)

   ■   Minimally formatted ASCII


 In addition, you can reference unformatted ASCII files, such as include
 files, from within a help database.

 An entire help system (such as the ones supplied with Microsoft C, FORTRAN,
 MASM, or QuickBasic) can use any combination of files formatted with
 different format types. With C, for example, the README.DOC information file
 is encoded as minimally formatted ASCII; the help files for the PWB, C
 language, and run-time library are written in QuickHelp format before being
 compressed by HELPMAKE. The database also cross-references the header
 (include) files, which are unformatted ASCII files stored outside the
 database.


 QuickHelp

 QuickHelp format is the default format into which HELPMAKE decodes help
 databases. Any text editor can create a QuickHelp-format help text file.
 QuickHelp format also lends itself to a relatively easy automated
 translation from other document formats.

 QuickHelp files can contain any kind of cross-reference or formatting
 attribute. Typically, you use QuickHelp format when modifying a
 Microsoft-supplied database.

 QuickHelp format makes use of dot commands (such as .context─see the
 description of QuickHelp dot commands in Section 11.6.1). To use dot
 commands other than .context and .comment, the / T option is required for
 encoding and decoding. For details, see Section 11.3, "Helpmake Options."


 Rich Text Format

 Rich Text Format (RTF) is a Microsoft word-processing format that several
 word processors support, including Microsoft Word version 5.0 and later, and
 Microsoft Word for Windows. You can use RTF as an intermediate format to
 simplify transferring help files from one format to another. Like QuickHelp
 files, RTF files can contain formatting attributes and cross-references.

 An RTF word processor provides the easiest way to create an RTF file, but
 you can manually insert RTF codes with an ordinary text editor. There are
 also utility programs that convert text files in other formats to RTF
 format.

 See Section 11.6.2, "Rich Text Format," for more information.


 Minimally Formatted ASCII

 Minimally formatted ASCII files define contexts and their topic text; they
 cannot contain screen-formatting commands or explicit cross-references.
 (Implicit cross-references work the same way they do in the other formats.)
 Minimally formatted ASCII files are often used to display text in a
 README.DOC or small help files that do not require compression. See Section
 11.6.3, "Minimally Formatted ASCII Format," for more information.


 Unformatted ASCII

 Unformatted ASCII files are exactly what their name implies: regular ASCII
 files with no formatting commands, context definitions, or special
 information. HELPMAKE does not process unformatted ASCII files in any
 special way. An unformatted ASCII file does not become part of the help
 database; only its name is used as the object of a cross-reference.
 Unformatted ASCII files are useful for storing program examples. Any word
 that is an implicit cross-reference in other help files is also an implicit
 cross-reference in unformatted ASCII files.


 11.2  Invoking HELPMAKE

 The HELPMAKE program can encode to create new help files or decode to modify
 existing ones. Encoding converts a text file to a compressed help database.
 HELPMAKE can encode text files written in QuickHelp, RTF, and minimally
 formatted ASCII format. Decoding converts a help database to a text file for
 editing. Regardless of the source format, HELPMAKE always decodes a help
 database into a QuickHelp-format text file.

 You invoke HELPMAKE with the following syntax:

   HELPMAKE {/E«n» | /D«c» |
   / H| /?} [[options]] sourcefiles

 The options modify the action of HELPMAKE; they are described in Section
 11.3, "HELPMAKE Options."

 You must supply either the /E (encode) or the /D (decode) option. When
 encoding, you must also use the /O option to specify the file name of the
 database.

 The sourcefiles field is required. It specifies the input file(s) for
 HELPMAKE. If you use the /D (decode) option, sourcefiles can be one or more
 help database files (such as PWB.HLP). HELPMAKE decodes the database files
 to the standard output device. If you use the /E (encode) option,
 sourcefiles can be one or more help text files (such as PWB.SRC). File names
 are separated with a space. You can use standard wild-card characters to
 specify a group of related files.

 The example below invokes HELPMAKE with the /V, /E, and /O options (see
 Section 11.3.1, "Options for Encoding"). HELPMAKE reads input from the text
 file  my.txt  and writes the compressed help database in the file  my.hlp.
 The /E option, without a compression specification, maximizes compression.
 Note that the DOS or OS/2 redirection symbol (>) sends a log of HELPMAKE
 activity to the file  my.log. You may want to redirect the log file because,
 in its verbose mode (given by /V), HELPMAKE can generate a lengthy log.

   HELPMAKE /V /E /Omy.hlp my.txt > my.log

 The example below invokes HELPMAKE to decode the help database  my.hlp into
 the text file  my.src, given with the /O option. Once again, the /V option
 results in verbose output, and the output is directed to the log file
 my.log. Section 11.3.2 describes additional options for decoding.

   HELPMAKE /V /D /Omy.src my.hlp > my.log


 11.3  HELPMAKE Options

 HELPMAKE accepts the command-line options described below. You can specify
 options in uppercase or lowercase letters and precede them with either a
 forward slash ( / ) or a dash ( - ). Most options apply only to encoding,
 others apply only to decoding, and a few apply to both. The /T option is
 required if you want to use dot commands with the QuickHelp format (which is
 the default format).


 11.3.1  Options for Encoding

 When you encode a file─that is, when you build a help database─you must
 specify the /E option. HELPMAKE also accepts other options to control
 encoding. The encoding options are listed below:

╓┌───────────┌───────────────────────────────┌───────────────────────────────╖
 Option      Action
 ────────────────────────────────────────────────────────────────────────────
 Option      Action
 ────────────────────────────────────────────────────────────────────────────
 /Ac         Specifies c as an
             application-specific control
             character for the help
             database file. The character
             marks a line that contains
             special information for
             internal use by the
             application. For example, the
             Microsoft Advisor uses the
             colon (:).

 /C          Makes context strings for this
             help file case sensitive.

 /E«n»       Creates (encodes) a help
             database from a specified text
             file. The n specifies the
             type(s) of compression. If n
             is omitted, HELPMAKE
 Option      Action
 ────────────────────────────────────────────────────────────────────────────
            is omitted, HELPMAKE
             compresses the file as much as
             possible (about 50%). The
             value of n is in the range 0
             -15. It is the sum of
             successive integral powers of
             2 representing various
             compression techniques:

             Value                           Technique

             0                               No compression

             1                               Run-length compression

             2                               Keyword compression

             4                               Extended keyword compression

 Option      Action
 ────────────────────────────────────────────────────────────────────────────

             8                               Huffman compression

             Add values to combine
             compression techniques. For
             example, use / E3 to get
             run-length and keyword compres-

             sion. Use / E0 in the testing
             stages of help database
             creation where you need to
             create the database quickly
             and are not yet concerned with
             size.

 /Kfilename  Optimizes keyword compression
             by supplying a list of
             characters that act as word
             separators. The filename is a
 Option      Action
 ────────────────────────────────────────────────────────────────────────────
            separators. The filename is a
             file containing your list of
             separator characters.


             The / E2 and / E3 options tell
             HELPMAKE to identify
             "keywords"─words occurring
             often enough to justify
             replacing them with shorter
             character sequences. A word is
             any series of characters that
             do not appear in the separator
             list. The default separator
             list includes all ASCII
             characters from 0 to 32, ASCII
             character 127, and the
             following characters:

 Option      Action
 ────────────────────────────────────────────────────────────────────────────

             ! " # & ` ' ( ) * + - , / : ;
             < = > ? @ [ \ ] ^ _ { | } ~

             You can improve keyword
             compression by designing a
             separator list tailored to a
             specific help file. If your
             help file contains #include
             directives, #include is
             encoded (by default) as
             include. To encode #include as
             a keyword, create a separator
             list that omits the #:

             !   " & ` ' ( ) * + - , / : ;
             < = > ? @ [ \ ] ^ _ { | } ~

             Characters in the range 0 -31
 Option      Action
 ────────────────────────────────────────────────────────────────────────────
            Characters in the range 0 -31
             are always separators, so you
             need not include them. A
             customized list must include
             all other separators, however,
             including the space (which
             follows ! in the list above).
             If you omit the space,
             HELPMAKE encodes sequences of
             words as keywords.

 /L          Locks the generated file so
             that it cannot later be
             decoded.

 /NOLOGO     Suppresses the HELPMAKE
             copyright message.

 /Ooutfile   Specifies outfile as the name
 Option      Action
 ────────────────────────────────────────────────────────────────────────────
/Ooutfile   Specifies outfile as the name
             of the help database.

 /Sn         Specifies the type of input
             file, according to the
             following n values:

             Option                          File Type

             /S1                             Rich Text Format (RTF)

             /S2                             QuickHelp (default)

             /S3                             Minimally formatted ASCII

 /T          Translates dot commands into
             internal format. If your help
             file contains dot commands
             other than .context and
 Option      Action
 ────────────────────────────────────────────────────────────────────────────
            other than .context and
             .comment, you must supply this
             option when encoding it. Dot
             commands are described in
             Section 11.6.1,"QuickHelp
             Format," and in later sections.
             The /T option causes the
             option /A: to be assumed.


 /V«n»       Controls verbosity of
             diagnostic and informational
             output. Larger values of n add
             more information. Omitting n
             produces a full listing. The
             values of n are listed below:

             Option                          Output

 Option      Action
 ────────────────────────────────────────────────────────────────────────────

             /V                              Maximum diagnostic output

             /V0                             No diagnostic output and no
                                             banner

             /V1                             HELPMAKE banner only

             /V2                             Pass names

             /V3                             Contexts on first pass

             /V4                             Contexts on each pass

             /V5                             Any intermediate steps within
                                             each pass

             /V6                             Statistics on help file and
                                             compression
 Option      Action
 ────────────────────────────────────────────────────────────────────────────
                                            compression

 /Wwidth     Indicates the fixed width of
             the resulting help text in
             number of characters. The
             value of width can range from
             11 to 255. If the /W option is
             omitted, the default is 76.
             When encoding an RTF source
             (/S1), HELPMAKE automatically
             formats the text to width.
             When encoding QuickHelp (/S2)
             or minimally formatted ASCII
             (/S3) files, HELPMAKE
             truncates lines to this width.




 11.3.2  Options for Decoding

 The /D option decodes a help database into QuickHelp files. HELPMAKE also
 accepts other options to control decoding. The decoding options are listed
 below:

╓┌────────────┌──────────────────────────────┌───────────────────────────────╖
 Option       Action
 ────────────────────────────────────────────────────────────────────────────
 /D«c»        Decodes the input file into
              its original text or
              component parts. If a
              destination file is not
              specified with the /O option,
              the help file is decoded to
              the standard output device.
              The form of decoding is
              controlled by the form of /D«
              c» specified:

              Form                           Effect
 Option       Action
 ────────────────────────────────────────────────────────────────────────────
             Form                           Effect

              /D                             Fully decodes the help
                                             database, leaving all
                                             cross-references and
                                             formatting information intact.

              /DS                            Splits a concatenated help
                                             database into its components
                                             using their original names. If
                                             the database was not created
                                             by concatenation, HELPMAKE
                                             copies it to a file with its
                                             original name. The database is
                                             not decompressed.

              /DU                            Decompresses the database and
                                             removes all screen formatting
                                             and cross-
 Option       Action
 ────────────────────────────────────────────────────────────────────────────
                                            and cross-
                                             references. The output can be
                                             used later for input and
                                             recompression, but all screen
                                             formatting and
                                             cross-references are lost.

              /NOLOGO                        Suppresses the HELPMAKE
                                             copyright message.

 /O«outfile»  Specifies outfile for the
              decoded output from HELPMAKE.
              If outfile is omitted, the
              help database is decoded to
              the standard output device.
              HELPMAKE always decodes help
              database files into QuickHelp
              format.

 Option       Action
 ────────────────────────────────────────────────────────────────────────────

 /T           Translates dot commands from
              internal format into
              dot-command format. You must
              always supply this option
              when decoding a help database
              that contains dot commands
              other than .context and
              .comment.




 /V«n»        Controls verbosity of
              diagnostic and informational
              output. Larger values of n
              add more information.
              Omitting n produces a full
              listing. The values of n are
 Option       Action
 ────────────────────────────────────────────────────────────────────────────
             listing. The values of n are
              listed below:

              Option                         Output

              /V                             Maximum diagnostic output

              /V0                            No diagnostic output and no
                                             banner

              /V1                            HELPMAKE banner only

              /V2                            Pass names

              /V3                            Contexts on first pass




 11.3.3  Options for Help

 The following are the options for help.

 Option                            Action
 ────────────────────────────────────────────────────────────────────────────
 / ?                               Displays a brief summary of HELPMAKE
                                   command-line syntax and exits without
                                   encoding or decoding any files. All
                                   other information on the command line is

                                   ignored.

 / «HELP»                          Calls the QuickHelp utility and displays
                                   help about HELPMAKE. If HELPMAKE cannot
                                   find QuickHelp or the help file, it
                                   displays the same information as with
                                   the /? option. No files are encoded or
                                   decoded. All other information on the
                                   command line is ignored.



 11.4  Creating a Help Database

 There are two ways to create a Microsoft-compatible help database.

 The first method is to decompress an existing help database, modify the
 resulting help text file, and recompress the help text file to form a new
 database.

 The second method is to append a new help database to an existing help
 database. This method involves the following steps:


   1.  Create a help text file in QuickHelp format, RTF, or minimally
       formatted ASCII.

   2.  Use HELPMAKE to create a help database file. The example below invokes
       HELPMAKE, using  yourhelp.txt  as the input file and producing a help
       database file named  yourhelp.hlp:

       HELPMAKE /V /E /Oyourhelp.hlp yourhelp.txt > yourhelp.log


   3.  Back up the existing database.

   4.  Append the new help database file to the existing database. The
       example below appends the new database  yourhelp.hlp  to the
       alang.hlp  database. (In the example, the / b modifier for the DOS
       COPY command combines the files as binary files.)

       COPY alang.hlp /b + yourhelp.hlp /b


   5.  Test the database. Assume  yourhelp.hlp  contains the context  sample.
       If you type  sample  in PWB and request help on it, the help window
       should display the text associated with the context  sample.


 ────────────────────────────────────────────────────────────────────────────
 WARNING

 The PWB editor truncates lines longer than about 250 characters. Some
 databases contain lines longer than this. To edit or create database files
 with extremely long lines, you must either use an editor (such as Microsoft
 Word) that does not restrict line length, or extend long lines using the
 backslash (\) line-continuation character.
 ────────────────────────────────────────────────────────────────────────────


 11.5  Help Text Conventions

 The source text that HELPMAKE uses to create Microsoft help databases must
 follow specific organizational conventions. The following sections explain
 these conventions.


 11.5.1  Structure of the Help Text File

 The Microsoft help system is simply a data-retrieval tool. It imposes no
 restrictions on the content or organization of help data. However, the
 HELPMAKE utility and the data-display routines in the help system expect a
 help file to follow a standard format. This section explains how to create
 correctly formatted help text files.

 In all three help text formats, the help text source file is a sequence of
 topics, each preceded by one or more context definitions. The following
 table lists the various formats and the corresponding context definition
 statements:

 Format                            Context Definition
 ────────────────────────────────────────────────────────────────────────────
 QuickHelp                         .context context

 RTF                               \ par >>context \ par

 Minimally formatted ASCII         >>context

 Unformatted ASCII                 None

 In QuickHelp format, each topic begins with one or more .context statements.
 These statements link the context string to its topic text. The topic text
 consists of all subsequent lines up to the next .context statement.

 In RTF format, each context definition must be in a paragraph of its own
 (denoted by \ par), beginning with the help delimiter (>>). As in QuickHelp,
 the topic text consists of all subsequent paragraphs up to the next context
 definition.

 In minimally formatted ASCII, each context definition must be on a separate
 line, and each must begin with the help delimiter (>>). As in RTF and
 QuickHelp files, all subsequent lines up to the next context definition
 constitute the topic text.

 See Section 11.6, "Using Help Database Formats," for detailed information
 about these three formats.

 ────────────────────────────────────────────────────────────────────────────
 WARNING

 HELPMAKE warns you if it encounters a duplicate context string definition
 within a given help source file. Each context string must be unique.
 ────────────────────────────────────────────────────────────────────────────


 11.5.2  Local Contexts

 Context strings beginning with the "at" sign (@) are "local." Making a
 context local saves file space and speeds access. However, local contexts
 cannot be cross-referenced with an implicit link, and they have no meaning
 outside the local file.

 When you use a local context, HELPMAKE does not generate a global context
 string (a context string that is known throughout the help system). Instead,
 it embeds an encoded cross-reference that has meaning only within the
 current context. For example,

   .context normal
   This is a normal topic, accessible by the context string "normal".
   [button\v@local\v] is a cross-reference to the following topic.

   .context @local

   This topic can be reached only by the explicit cross-reference
   in the previous topic (or by browsing the file sequentially).

 In the example above, the text  button\v@local\v  references  local  as a
 local context. If the user selects the text  button  or scrolls through the
 file, the help system displays the topic text that follows the context
 definition for  local. Because  local  is defined with the "at" sign @, it
 can be accessed only by a hyperlink within the same help file or by
 sequentially browsing the file.

 If you want a topic to be accessible in both local and global contexts, you
 simply mark the topic text with both global and local .context statements.
 For example, to make  topic  both global and local, add the following
 statements:

   .context topic
   .context @topic

 Naturally, both .context statements must appear immediately before the topic
 text to which they point.

 To create a context that begins with a literal @, precede it with a
 backslash ( \ ).


 11.5.3  Context Prefixes

 Microsoft help databases use several "context prefixes." A context prefix is
 a single letter followed by a period. It appears before a context string
 with a predefined meaning. These contexts may appear in the resulting text
 file when you decode a Microsoft help database.

 Context prefixes are used internally by Microsoft.

 Except for the h. prefix described below, the context prefixes are used by
 Microsoft to mark environment- or product-specific features. You would not
 normally add them to the help files you write.

 You can use the h. prefix to identify standard help-file contexts. For
 instance, h.default identifies the default help screen (the screen that
 normally appears when you select top-level help). Table 11.1 lists the
 standard h. contexts.

 Table 11.1  Standard h. Contexts

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Context                           Description
 ────────────────────────────────────────────────────────────────────────────
 h.contents                        The table of contents for the help file.
                                   You should also define the string
                                   "contents" for direct reference to this
                                   context.

 h.default                         The default help screen, typically
                                   displayed when the user presses SHIFT+F1
                                   at the "top level" in some applications.

 h.index                           The index for the help file. You can
                                   also define the string "index" for
                                   direct reference to this context.

 h.notfound                        The help text displayed by some
                                   applications when the help system cannot
                                   find information about the requested
                                   context. The text could be an index of
                                   contexts, a topical list, or general
 Context                           Description
 ────────────────────────────────────────────────────────────────────────────
                                  contexts, a topical list, or general
                                   information about using help.

 h.pg#                             A specific page within the help file.
                                   This is used in response to a "go to
                                   page #" request.

 h.pg$                             The help text that is logically last in
                                   the file. This is used by some
                                   applications in response to a "go to the
                                   end" request made within the help window.

 h.pg1                             The help text that is logically first in
                                   the file. This is used by some
                                   applications in response to a "go to the
                                   beginning" request made within the help
                                   window.

 h.title                           The title of the help database.
 Context                           Description
 ────────────────────────────────────────────────────────────────────────────
h.title                           The title of the help database.

 ────────────────────────────────────────────────────────────────────────────



 The context prefixes in Table 11.2 are internal to Microsoft products. They
 appear in decompressed databases, but you do not need to use them.

 Table 11.2  Microsoft Product Context Prefixes

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Prefix                            Purpose
 ────────────────────────────────────────────────────────────────────────────
 d.                                Dialog box. Each dialog box is assigned
                                   a number. Its help context string is d.
                                   followed by the number (for example,
                                   d.12).

 Prefix                            Purpose
 ────────────────────────────────────────────────────────────────────────────

 e.                                Error number. If a product supports the
                                   error-numbering scheme used by Microsoft
                                   languages, it displays help for each
                                   error using this prefix. For example,
                                   the context  e.P0105  refers to the
                                   Microsoft QuickPascal Compiler error
                                   message number P0105.

 h.                                Help item. Prefixes miscellaneous help
                                   context strings that may be constructed
                                   or otherwise hidden from the user. For
                                   example, most applications look for the
                                   context string h.contents when Contents
                                   is chosen from the Help menu.

 m.                                Menu item. Contexts that relate to
                                   product menu items are defined by their
                                   shortcut keys. For example, the Exit
 Prefix                            Purpose
 ────────────────────────────────────────────────────────────────────────────
                                  shortcut keys. For example, the Exit
                                   selection on the File menu item is
                                   accessed by ALT+F, X and is referenced
                                   in help by  m.f.x.

 n.                                Message number. Each message box is
                                   assigned a number. Its help context
                                   string is n. plus the number (for
                                   example,  n.5).

 ────────────────────────────────────────────────────────────────────────────




 11.5.4  Hyperlinks

 Explicit cross-references, or hyperlinks, are marked with invisible text in
 the help text file. A hyperlink is a word or phrase followed by invisible
 text that names the context to which the hyperlink refers.

 The keystroke that activates the hyperlink depends on the application.
 Consult the documentation for each product for the specific keystroke.

 When the user activates the hyperlink, the help system displays the topic
 referenced by the invisible text. The invisible cross-reference text is
 formatted as one of the following:

 Hidden Text                       Action
 ────────────────────────────────────────────────────────────────────────────
 contextstring                     Displays the topic associated with
                                   contextstring. For example,  exeformat
                                   displays the topic text for the context
                                   exeformat.

 filename!                         Treats filename as a single topic to be
                                   displayed. For example,
                                   $INCLUDE:stdio.h!  searches the
                                   directories in the INCLUDE environment
                                   variable for file  stdio.h  and displays
                                   it as a single help topic.

 filename!contextstring            Works the same as contextstring, except
                                   only the help file filename is searched
                                   for the context. If the file is not
                                   already open, the help system finds it
                                   (by searching either the current path or
                                   an explicit environment variable) and
                                   opens it. For example,
                                   $BIN:readme.doc!patches  searches for
                                   readme.doc  in the BIN environment
                                   variable and displays the topic
                                   associated with  patches.

 !command                          Executes the command specified after the
                                   exclamation point (!).

 In the following example, the word  Example  is a hyperlink. The \b,\p, and
 \v formatting flags mark hyperlinks in the help text. (The formatting flags
 are listed later in this chapter, in Table 11.4.)

   \bSee also:\p    Example\vopen.ex\v

 The hyperlink refers to  open.ex. If you select any of the letters of
 Example, the help system displays the topic whose context is  open.ex. On
 the screen, this line appears as follows:

   See also: Example

 An application might display  See also:  and  Example  in different colors
 or character types, depending on factors such as your default color
 selection and type of monitor.

 When a hyperlink needs to cross-reference more than one word, you must use
 an anchor, as in the following example:

   \bSee also:\p   \uExample\p\vprintf.ex\v, fprintf, scanf, sprintf,
   vfprintf, vprintf, vsprintf
               \aformatting table\vprintf.table\v

 This part of the example is an anchored hyperlink:

   \aformatting table\vprintf.table\v

 The anchor must fit on one line.

 The \ a flag creates an anchor for the cross-reference. In the example, the
 phrase following the \ a flag (formatting table) is the hyperlink. It refers
 to the context  printf.table. The first \v flag marks both the end of the
 hyperlink and the beginning of the invisible text. The name  printf.table
 is invisible; it does not appear on the screen when the help is displayed.
 The second \v flag ends the invisible text.


 11.6  Using Help Database Formats

 A database can be written in any of three text formats. The list below
 briefly describes these types. Sections 11.6.1-11.6.3 describe the
 formatting types in detail.

 An entire help system (such as the one supplied with PWB or QuickC) can
 handle any combination of formats. For example, the help files for Microsoft
 C are written in QuickHelp format, and the README.DOC file is unformatted
 ASCII.

 Type                              Characteristics
 ────────────────────────────────────────────────────────────────────────────
 QuickHelp                         Uses dot commands and embedded
                                   formatting characters (the default
                                   formatting type expected by HELPMAKE);
                                   supports highlighting, color, and
                                   cross-references. Files in this format
                                   must be compressed before use.

 RTF                               Uses a subset of standard RTF; supports
                                   highlighting, color, and
                                   cross-references; supports some dot
                                   commands. Files in this format must be
                                   compressed before use.

 Minimally formatted ASCII         Uses a help delimiter (>>) to define
                                   help contexts; does not support
                                   highlighting, color, or crossreferences.
                                   Files in this format can be compressed,
                                   but compression is not required.


 11.6.1  QuickHelp Format

 The QuickHelp format uses a dot command and embedded formatting flags to
 convey information to HELPMAKE.


 11.6.1.1  QuickHelp Dot Commands

 QuickHelp provides a number of dot commands that identify topics and convey
 other topic-related information to the help system. If your help file
 contains dot commands other than .context or .comment, you must supply the /
 T option when encoding and decoding with HELPMAKE.

 You can define more than one context for a single topic.

 The most important dot command is the .context command. Every topic in a
 QuickHelp file begins with one or more .context commands. Each .context
 command defines a context string for the topic text. You can define more
 than one context for a single topic, as long as you do not place any topic
 text between them.

 Typical .context commands are shown below. The first defines a context for
 the  #include C preprocessor directive. The second set illustrates multiple
 contexts for one block of topic text. In this case, the same topic text
 explains all of the string-to-number conversion routines in C.

   .context #include
          .
          . description of #include goes here
          .
   .context strtod
   .context strtol
   .context strtoul
          .
          . description of string-to-number functions goes here
          .

 The QuickHelp format includes several other dot commands. Table 11.3 lists
 the dot commands available in QuickHelp format.

 Table 11.3  QuickHelp Dot Commands

╓┌─────────────────────────────────────┌─────────────────────────────────────╖
 Command                               Action
 ────────────────────────────────────────────────────────────────────────────
 .category string                      Lists the category in which the
                                       current topic appears and its
                                       position in the list of topics. The
                                       category name is used by the
                                       QuickHelp Categories command, which
                                       displays the topics list. Supported
                                       only by QuickHelp.

 .command                              Indicates that the topic text is not
                                       a displayable help topic. Use this
                                       command to hide hyperlink topics and
                                       other internal information.

 .comment string                       The string is a comment that appears
 .. string                             only in the help source file.
 Command                               Action
 ────────────────────────────────────────────────────────────────────────────
.. string                             only in the help source file.
                                       Comments are not inserted in the
                                       help database, so they cannot be
                                       restored when you decompress a help
                                       file.

 .context string                       The string introduces a topic.

 .end                                  Ends a paste section. See the .paste
                                       command below. Supported only by
                                       QuickHelp.

 .freeze numlines                      Locks the first numlines lines at
                                       the top of the screen. This can be
                                       used to preserve a bar of
                                       cross-reference buttons for a help
                                       topic and prevent it from being
                                       scrolled.

 Command                               Action
 ────────────────────────────────────────────────────────────────────────────

 .length topiclength                   Indicates the default window size,
                                       in topiclength lines, of the topic
                                       about to be displayed.

 .line number                          Tells HELPMAKE to reset the line
                                       number to begin at number for
                                       subsequent lines of the input file.
                                       Line numbers appear in HELPMAKE
                                       error messages. HELPMAKE does not
                                       put the .line command into the help
                                       database, so it is not restored
                                       during decompression. See .source.

 .list                                 Indicates that the current topic
                                       contains a list of topics. QuickHelp
                                       displays a highlighted line; you can
                                       choose a topic by moving the
                                       highlighted line over the desired
 Command                               Action
 ────────────────────────────────────────────────────────────────────────────
                                      highlighted line over the desired
                                       topic and pressing ENTER. Help
                                       searches for the first word of the
                                       line. Supported only by QuickHelp.

 .mark name «column»                   Defines a mark immediately preceding
                                       the following line of text. The
                                       marked line shows a script command
                                       where the display of a topic begins.
                                       The name identifies the mark. The
                                       column is an integer value
                                       specifying a column location within
                                       the marked line. Supported only by
                                       QuickHelp.




 .next context                         Tells the help system to look up the
 Command                               Action
 ────────────────────────────────────────────────────────────────────────────
.next context                         Tells the help system to look up the
                                       next topic using
                                       context instead of the topic that
                                       physically follows it in the file.
                                       You can use this command to skip
                                       large blocks of .command or .popup
                                       topics.

 .paste pastename                      Begins a paste section. The
                                       pastename appears in the QuickHelp
                                       Paste menu. Supported only by
                                       QuickHelp.

 .popup                                Tells the help system to display the
                                       current topic as a popup instead of
                                       a normal, scrollable topic.
                                       Supported only by QuickHelp.

 .previous context                     Tells the help system to look up the
 Command                               Action
 ────────────────────────────────────────────────────────────────────────────
.previous context                     Tells the help system to look up the
                                       previous topic using context instead
                                       of the topic that physically
                                       precedes it in the file. You can use
                                       this command to skip large blocks of
                                       .command or .popup topics.

 .raw                                  Turns off special processing of
                                       certain characters by the
                                       application.

 .ref topic «, topic» ...              Tells the help system to display the
                                       topic in the Reference menu. You can
                                       list as many topics as needed;
                                       separate each additional topic with
                                       a comma. A .ref command is formatted
                                       without regard to the /W option.
                                       Supported only by QuickHelp.

 Command                               Action
 ────────────────────────────────────────────────────────────────────────────

                                       If no topic is specified, QuickHelp
                                       searches the line immediately
                                       following for a See: or See Also:
                                       reference; if present, the reference
                                       must be the first non-white-space
                                       characters on the line.

 .source filename                      Tells HELPMAKE that subsequent
                                       topics come from filename. By
                                       default, when an error occurs, the
                                       error message contains the name and
                                       line number of the input file. The
                                       .source command tells HELPMAKE to
                                       use filename in the error message
                                       instead of the name of the input
                                       file and to reset the line number to
                                       1. This is useful when you
                                       concatenate several sources to form
 Command                               Action
 ────────────────────────────────────────────────────────────────────────────
                                      concatenate several sources to form
                                       the input file. HELPMAKE does not
                                       put the .source command into the
                                       help database, so it is not restored
                                       during decompression. See .line.

 .topic text                           Defines text as the name or title to
                                       be displayed in place of the context
                                       string if the application help
                                       displays a title. This command is
                                       always the first line in the context
                                       unless you also use the .length or
                                       .freeze commands.

 ────────────────────────────────────────────────────────────────────────────




 11.6.1.2  QuickHelp Formatting Flags

 The QuickHelp format provides a number of formatting flags that are used to
 highlight parts of the help database and to mark hyperlinks in the help
 text.

 Each formatting flag consists of a backslash ( \ ) followed by a character.
 Table 11.4 lists the formatting flags.

 Table 11.4  QuickHelp Formatting Flags

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Formatting Flag                   Action
 ────────────────────────────────────────────────────────────────────────────
 \ a                               Anchors text for cross-references

 \ b, \ B                          Turns boldface on or off

 \ i, \ I                          Turns italics on or off

 \ p, \ P                          Turns off all attributes
 Formatting Flag                   Action
 ────────────────────────────────────────────────────────────────────────────
\ p, \ P                          Turns off all attributes

 \ u, \ U                          Turns underlining on or off

 \ v, \V                           Turns invisibility on or off
                                   (hides cross-references in text)

 \\                                Inserts a single backslash in text

 ────────────────────────────────────────────────────────────────────────────



 On monochrome monitors, text labeled with the bold, italic, and underline
 attributes appears in various ways, depending on the application (for
 example, high intensity and reverse video are commonly displayed). On color
 monitors, these attributes are translated by the application into suitable
 colors, depending on the user's default color selections.

 The \ b, \ i, \ u, and \v options are toggles, turning on and off their
 respective attributes. You can use several of these on the same text. Use
 the \ p attribute to turn off all attributes. Use the \v attribute to hide
 cross-references and hyperlinks in the text.

 HELPMAKE truncates the lines in QuickHelp files to the width specified with
 the / W option. Only visible characters count toward the character-width
 limit. Lines that begin with an application-specific control character are
 truncated to 255 characters regardless of the width specification. See
 Section 11.3.1, "Options for Encoding," for more information on truncation
 and application-specific control characters.

 In the example below, the \ b flag initiates boldface text for  Returns:,
 and the \ p flag changes the remaining text to plain text.

   \bReturns:\p    a handle if successful, or -1 if not.
               errno:  EACCES, EEXIST, EMFILE, ENOENT

 In the example below, the \ a flag anchors text for the hyperlink  Example.
 The \v flags define the cross-reference  sample_prog  and make the text
 between the \v flags invisible. Cross-references are described in the
 following section.

   \aExample \vsample_prog\v


 11.6.1.3  QuickHelp Cross-References

 Help databases contain two types of cross-references, implicit and explicit.
 They are described in Section 11.1.1, "Contents of a Help File."

 Any word that appears as a global context is implicitly cross-referenced.
 For example, any time you request help in PWB on close, the help window
 displays information about that function. You do not code implicit
 cross-references into your help text files.

 Insert formatting flags to mark explicit cross-references.

 Explicit cross-references (hyperlinks) are words or phrases on the screen
 that point to a context. For example, almost every "See:" and "See also:"
 reference in online help has a hyperlink pointing to the appropriate
 context. You can view the cross-referenced material immediately by
 activating the hyperlink, without having to search the help system's menus
 for the topic. You must insert formatting flags in your help text files to
 mark explicit cross-references.

 If the hyperlink consists of a single word, you can use invisible text to
 flag it in the source file. The \v formatting flag creates invisible text,
 as follows:

   hyperlink\vcontext\v

 Put the first \v flag immediately following the word you want to be the
 hyperlink. Following the flag, insert the context that the hyperlink points
 to. The second \v flag marks the end of the context; that is, the end of the
 invisible text. HELPMAKE generates a cross-reference whose context is the
 invisible text and whose hyperlink is the word.

 If the hyperlink consists of a phrase, rather than a single word, you must
 use anchored text to create explicit cross-references. Use the \ a and \v
 flags to create anchored text as follows:

   \ahyperlink-words\vcontext\v

 The \ a flag marks an anchor for the cross-reference. The text that follows
 the \ a flag is the hyperlink. The hyperlink must fit entirely on one line.
 The first \v flag marks both the end of the hyperlink and the beginning of
 the invisible text that contains the cross-reference context. The second \v
 flag marks the end of the invisible text.

 The C functions abs, cabs, and fabs in the following examples are implicit
 cross-references because they have a global context in the help system.

   See also: abs, cabs, fabs

 The next example shows the encoding for an explicit cross-reference to an
 example program and a function template from the help database for the
 Microsoft C run-time library:

   See also: Example\vopen.ex\v, Template\vopen.tm\v, close

 Here, the hyperlinks are  Example  and  Template, which reference the
 contexts  open.ex  and  open.tm. The example also contains an implicit
 cross-reference to the close function.

 The final example shows the encoding for an explicit cross-reference to an
 entire family of functions:

   See also: \ais... functions\vis_functions\v, atoi

 The cross-reference uses anchored text to associate a phrase, rather than
 just a word, with a context. In this example, the hyperlink is the anchored
 phrase  is... functions, and it cross-references the context  is_functions.
 In addition, the example contains an implicit cross-reference to the
 C-language atoi routine.


 11.6.1.4  QuickHelp Example

 The code below is an example in QuickHelp format that contains a single
 entry:

   .context open
   .length 13
   \bInclude:\p   <fcntl.h>, <io.h>, <sys\\types.h>, <sys\\stat.h>

   \bPrototype:\p  int open(char *path, int flag[, int mode]);
            oflag:  O_APPEND O_BINARY O_CREAT O_EXCL O_RDONLY
                    O_RDWR    O_TEXT    O_TRUNC  O_WRONLY
                    (can be joined by |)
            pmode:  S_IWRITE  S_IREAD   S_IREAD | S_IWRITE

   \bReturns:\p    a handle if successful, or -1 if not.
               errno:  EACCES, EEXIST, EMFILE, ENOENT

   \bSee also:\p  \uExample\p\vopen.ex\v, \uTemplate\p\vopen.tp\v,
                access, chmod, close, creat, dup, dup2, fopen, sopen,
   umask

 The .length command near the beginning of the example specifies the size of
 the initial window for the help text. Here, the initial window displays 13
 lines.

 The manifest constants (such as O_WRONLY and EEXIST), the C keywords (such
 as int and char), and the other functions (such as access and sopen) are
 implicit cross-references. The words  Example  and  Template  are explicit
 cross-references to the example  open.ex  and to the open template  open.tp,
 respectively. Note the use of double backslashes in the include file names.



 11.6.2  Rich Text Format

 Rich Text Format (RTF) is a Microsoft word-processing format supported by
 several word processors, including Microsoft Word 5.0 and Microsoft Word for
 Windows. RTF allows documents to be transferred between applications without
 loss of formatting. The HELPMAKE utility recognizes a subset of the full RTF
 syntax. If your file contains RTF codes that are not part of the subset,
 HELPMAKE discards them.

 To create an RTF-formatted file, enter the text and format it as you want it
 to appear: bold, underlined, hidden, italic, and so forth. (You can combine
 attributes.) You can also format paragraphs, selecting body and first-line
 indenting. The only items you need to insert into an RTF file manually are
 the help delimiter (>>) and the context string that start each entry.

 When you have entered and formatted the text, save it in RTF format. In
 Microsoft Word 5.0, for example, this means choosing Transfer Save, then
 highlighting RTF in the format: field.

 You do not see the RTF formatting codes when you load an RTF file into a
 compatible word processor; the word processor removes them and displays the
 text with the specified attribute(s). However, you can view these codes by
 loading an RTF file into a plain-text word processor.

 HELPMAKE recognizes the subset of RTF codes listed in Table 11.5.

 Table   11.5 RTF Formatting Codes

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 RTF Code
 ────────────────────────────────────────────────────────────────────────────
 \ b                               Boldface. The application decides how to
                                   display this; often it is
                                   intensified text.

 \ fin                             Paragraph first-line indent, n columns.

 RTF Code
 ────────────────────────────────────────────────────────────────────────────

 \ i                               Italic. The application decides how to
                                   display this; often it is reverse video.

 \ lin                             Paragraph indent from left margin, n
                                   columns.

 \ line                            New line (not new paragraph).

 \ par                             End of paragraph.

 \ pard                            Default paragraph formatting.

 \ plain                           Default attributes. On most screens,
                                   this is nonblinking normal
                                   intensity.

 \ tab                             Tab character.

 RTF Code
 ────────────────────────────────────────────────────────────────────────────

 \ ul                              Underline. The application decides how
                                   to display this; some adapters that do
                                   not support underlining display it as
                                   blue text.

 \ v                               Hidden text. Hidden text is used for
                                   cross-reference information and for some
                                   application-specific communications; it
                                   is not
                                   displayed.

 ────────────────────────────────────────────────────────────────────────────



 When HELPMAKE compresses the file, it formats the text to the width given
 with the / W option, ignoring the paragraph formats.

 As with the other text formats, each entry in the database source consists
 of one or more context strings, followed by topic text. An RTF file can
 contain QuickHelp dot commands.

 The help delimiter (>>) at the beginning of any paragraph marks the
 beginning of a new help entry. The text that follows on the same line is
 defined as a context for the topic. If the next paragraph also begins with
 the help delimiter, it also defines a context string for the same topic
 text. You can define any number of contexts for a block of topic text. The
 topic text comprises all subsequent paragraphs up to the next paragraph that
 begins with the help delimiter.

 The example below is a help database containing a single entry using subset
 RTF text. Note that RTF uses curly braces ( { } ) for nesting. Thus, the
 entire file is enclosed in curly braces, as is each specially formatted text
 item.

   {\rtf1
   \pard >>open\par
     {\b Include:}    <fcntl.h>, <io.h>, <sys\\types.h>, <sys\\stat.h>\par
   \par
     {\b Syntax:}     int open( char * filename, int oflag[, int pmode
   ] );\par
                oflag:  O_APPEND  O_BINARY  O_CREAT  O_EXCL  O_RDONLY\par

                        O_RDWR    O_TEXT    O_TRUNC  O_WRONLY\par
                        (may be joined by |)\par
                pmode:  S_IWRITE  S_IREAD   S_IREAD | S_IWRITE\par
   \par
     {\b Returns:}    a handle if successful, or -1 if not.\par
                errno:  EACCES, EEXIST, EMFILE, ENOENT\par
   \par
     {\b See also:}  Examples{\v open.ex}, access, chmod, close, creat,
   dup,\par
                dup2, fopen, sopen, umask\par
   >>open.ex\par
   To build this help file, use the following command:\par
   \par
   HELPMAKE /S1 /E15 /OOPEN.HLP OPEN.RTF\par
   \par

           < Back >{\v !B}
   }

 RTF files normally contain additional information that is not visible to the
 user; HELPMAKE ignores this extra information.


 11.6.3  Minimally Formatted ASCII Format

 A minimally formatted ASCII text file comprises a sequence of topics, each
 preceded by one or more unique context definitions. Each context definition
 must be on a separate line beginning with a help delimiter (>>). Subsequent
 lines up to the next context definition constitute the topic text.

 Minimally formatted ASCII files cannot contain highlighting.

 There are two ways to use a minimally formatted ASCII file. You can compress
 it with HELPMAKE, creating a help database, or an application can access the
 uncompressed file directly. Compressing minimally formatted ASCII files
 increases search speed. Uncompressed files are somewhat larger and slower to
 search. Minimally formatted ASCII files have a fixed width, and they cannot
 contain highlighting (or other nondefault attributes) or explicit
 cross-references.

 The following example, coded in minimally formatted ASCII, shows the same
 text as the QuickHelp example presented earlier in this section. The first
 line of the example defines  open  as a context string. The minimally
 formatted ASCII help file must begin with the help delimiter (>>), so that
 HELPMAKE or the application can verify that the file is indeed an ASCII help
 file.

   >>>>open

   Include:    <fcntl.h>, <io.h>, <sys\types.h>, <sys\stat.h>

   Prototype:  int open(char *path, int flag[, int mode]);
          oflag:  O_APPEND  O_BINARY  O_CREAT  O_EXCL  O_RDONLY
                  O_RDWR    O_TEXT    O_TRUNC  O_WRONLY
                  (can be joined by |)
          pmode:  S_IWRITE  S_IREAD   S_IREAD | S_IWRITE

   Returns:    a handle if successful, or -1 if not.
               errno:  EACCES, EEXIST, EMFILE, ENOENT

   See also:  access, chmod, close, creat, dup, dup2, fopen, sopen,
   umask

 When displayed, the help information appears exactly as it is typed into the
 file. Any formatting codes are treated as ASCII text.


 11.7  Related Topics in Online Help

 Information on the following related topics can be found in online help.

 Topic       Access
 ────────────────────────────────────────────────────────────────────────────
 HELPMAKE    Choose "HELPMAKE" from the "Microsoft Advisor Contents" screen
 QuickHelp   Choose "QH" from the "Microsoft Advisor Contents" screen






 Chapter 12  Linking Object Files with LINK
 ────────────────────────────────────────────────────────────────────────────

 This chapter describes the Microsoft Segmented-Executable Linker (LINK),
 which combines compiled or assembled object files into an executable file.
 It explains LINK's input syntax and fields and tells how to use options to
 control LINK. It discusses overlays in DOS programs and concludes with
 background information about LINK.


 12.1  Overview

 LINK combines 80x86 object files into either an executable file or a
 dynamic-link library (DLL). The object-file format is the Microsoft
 Relocatable Object-Module Format (OMF), based on the Intel 8086 OMF. LINK
 uses library files in Microsoft library format.

 LINK creates "relocatable" executable files and DLLs─that is, the operating
 system can load and execute these files in any unused section of memory.
 LINK can create DOS executable files with up to 1 megabyte of code and data
 (or up to 16 megabytes when using overlays), or OS/2 and Microsoft Windows
 programs with up to 16 megabytes.

 For more information on OMF, executable-file format, and the linking
 process, see the MS-DOS Encyclopedia.

 Use BIND to create an OS/2 program that also runs under DOS.

 The linker produces programs that run under DOS only or under OS/2 only, but
 not both. However, if an OS/2 program limits its OS/2 function calls to the
 Family API subset, you can use the Microsoft Bind Utility (BIND) to modify
 the OS/2 executable file so that it runs under both OS/2 and DOS. For more
 information, see online help.

 Use EXEHDR to examine the finished file.

 When the file (either executable or DLL) is created, you can examine the
 information that LINK puts in the file's header by using the Microsoft EXE
 File Header Utility (EXEHDR). For more information, see online help.

 Other programs can call LINK automatically.

 The Programmer's WorkBench (PWB) invokes LINK to create the final executable
 file or DLL. Therefore, if you develop your software with PWB, you might not
 need to read this chapter. However, the detailed explanations of LINK
 options might be helpful when you use the LINK Options dialog box in PWB.
 This information is also available in online help.

 The compiler or assembler supplied with your language (CL with C, FL with
 FORTRAN, ML with MASM) also invokes LINK. You can use most of the LINK
 options described in this chapter with this utility. Online help has more
 information about the compilers and assembler: select help for the
 appropriate language from the Compiler box of the help Contents screen.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 Unless otherwise noted, all references to "library" in this chapter refer to
 a static library, either a standard library created by the Microsoft Library
 Manager (LIB) or an import library created by the Microsoft Import Library
 Manager (IMPLIB), and not a DLL.
 ────────────────────────────────────────────────────────────────────────────


 12.2  LINK Output Files

 LINK is a bound application that runs under both DOS and OS/2 and can create
 executable files for DOS, OS/2, or Windows. You do not have to run LINK
 under OS/2 to create OS/2 applications, or under DOS to create DOS programs.
 The kind of file produced is determined by the way the source code is
 compiled and the information supplied to LINK, not the operating system LINK
 runs under.

 A program that runs under DOS is called an executable file or application. A
 program or DLL that runs under Windows or OS/2 is called a segmented
 executable file. LINK creates the appropriate file according to the
 following rules:


   ■   If a module-definition file or import library is not specified and the
       object files and libraries do not contain export definitions, LINK
       creates an application that runs under DOS.

   ■   If a module-definition file containing a LIBRARY statement is
       specified, LINK creates a DLL for Windows or OS/2.

   ■   If any other form of module-definition file is specified, or if any of
       the object files contains an exported definition, LINK creates an
       application to run under Windows or OS/2.


 LINK looks for the default run-time libraries named in the object files.
 Default libraries can be real or protected mode. (The mode is usually set
 when the language product is installed.) Protected-mode libraries contain
 export definitions. If LINK finds protected-mode default libraries, the
 output file will be a segmented executable file rather than a DOS file.

 The file OS2.LIB is an import library. Linking with OS2.LIB produces an OS/2
 application or DLL. When you use a Microsoft high-level language to compile
 for protected mode, the compiler automatically specifies OS2.LIB as a
 default library.

 LINK's output is either an executable file or a DLL. For simplicity, this
 chapter sometimes refers to this output as the "main file" or "main output."


 Map files list the segments and symbols in a program.

 LINK also creates a "map" file, which lists the segments in the executable
 file. The /MAP option adds public symbols to the map file, and the /LINE
 option adds line numbers.

 LINK produces other files when certain options are used.

 Other options tell LINK to create other kinds of output files. The /INCR
 option creates .ILK and .SYM files for incremental linking with ILINK. LINK
 produces a .COM file instead of an .EXE file when the /TINY option is
 specified. The combination of /CO and /TINY puts debugging information into
 a .DBG file. A Quick library results when the /Q option is specified. For
 more information on these and other options, see Section 12.5, "LINK
 Options."


 12.3  LINK Syntax and Input

 The LINK command has the following syntax:

   LINK objfiles«, «exefile» «,
   «mapfile»«, «libraries»«, deffile»
   » » »«;»

 The LINK fields perform the following functions:


   ■   The objfiles field is a list of the object files that are to be linked
       into an executable file or DLL. It is the only required field.

   ■   The exefile field lets you change the name of the output file from its
       default.

   ■   The mapfile field gives the map file a name other than its default
       name.

   ■   The libraries field specifies additional (or replacement) libraries to
       search for unresolved references.

   ■   The deffile field gives the name of a description file needed to
       create Windows and OS/2 applications and DLLs.


 Fields are separated by commas. You can specify all the fields or leave one
 or more fields (including objfiles) blank; LINK will then prompt you for the
 missing input. (For an explanation of how to use LINK prompts, see Section
 12.4, "Running LINK.") To leave a field blank, enter only the field's
 trailing comma.

 Options can be specified in any field. For descriptions of each of LINK's
 options, see Section 12.5, "LINK Options."

 The fields must be entered in the order shown, whether they contain input or
 are left blank. A semicolon (;) at the end of the LINK command line
 terminates the command and suppresses prompting for any missing fields. LINK
 then assumes the default values for the missing fields.

 If your file appears in or is to be created in another directory or device,
 you must supply the full pathname. Filenames are not case sensitive.

 The next five sections explain how to use each of the LINK fields.


 12.3.1  The objfiles Field

 The objfiles field specifies one or more object files to be linked. At least
 one filename must be entered. If you do not supply an extension, LINK
 assumes a default .OBJ extension. If the filename has no extension, add a
 period (.) at the end of its name.

 If you name more than one object file, separate the names with a plus sign
 (+) or a space. To extend objfiles to the following line, type a plus sign
 (+) as the last character on the current line, press ENTER, and continue. Do
 not split a name across lines.


 12.3.1.1  Load Libraries

 The objfiles field can also specify library files. A library specified this
 way becomes a "load library." You must specify the library's filename
 extension; otherwise, LINK assumes an .OBJ extension.

 LINK treats load libraries as any other object file: it puts every object
 module from a load library in the executable file, regardless of whether a
 module satisfies an unresolved external reference. The effect is the same as
 if you had specified all the library's object-module names in the objfiles
 field.

 Specifying a load library can therefore create an executable file or DLL
 that is larger than it needs to be. (A library named in the libraries field
 adds only those modules required to resolve external references.) However,
 loading an entire library can be useful when


   ■   Repeatedly specifying the same group of object files

   ■   Placing a library in an overlay

   ■   Debugging, so you can call library routines that would not be included
       in the release version of the program



 12.3.1.2  How LINK Searches for Object Files

 When searching for object (and load-library) files, LINK looks in the
 following locations in the order specified:


   1.  The directory specified for the file (if a path is included). If the
       file is not in that directory, the search terminates.

   2.  The current directory.

   3.  Any directories specified in the LIB environment variable.


 If LINK cannot find an object file, and a floppy drive is associated with
 that object file, LINK pauses and prompts you to insert a disk containing
 the object file.

 If you specify a library in the objfiles field, LINK treats it like any
 other object file. LINK therefore does not search for load libraries in
 directories named in the libraries field.


 12.3.1.3  Overlays

 A special syntax for the objfiles field lets you create DOS programs that
 use overlay modules. For more information about overlays, see Section 12.7,
 "Using Overlays under DOS."


 12.3.2  The exefile Field

 The exefile field is used to specify a name for the main output file. If you
 do not supply an extension, LINK assumes a default extension, either .EXE,
 .COM (when using the /TINY option), .DLL (when using a module-definition
 file containing a LIBRARY statement), or .QLB (when using the /Q option).

 If you do not specify an exefile, LINK gives the main output a default name.
 This name is the base name of the first file listed in the objfiles field,
 plus the extension appropriate for the type of executable file being
 created.

 LINK creates the main file in the current directory unless you specify an
 explicit path with the filename.


 12.3.3  The mapfile Field

 The mapfile field is used to specify a filename for the map file or to
 suppress creation of a map file. A map file lists the segments in the
 executable file or DLL.

 You can specify a path with the filename. The default extension is .MAP.
 Specify  NUL  to suppress the creation of a map file. The default for the
 mapfile field is one of the following:


   ■   If this field is left blank on the command line or in a response file,
       LINK creates a map file with the base name of the exefile (or the
       first object file if no exefile is specified) and the extension .MAP.

   ■   When using LINK prompts, LINK assumes either the default described
       above (if an empty mapfile field is specified) or  NUL.MAP, which
       suppresses creation of a map file.


 To add line numbers to the map file, use the /LINE option. To add public
 symbols, use the /MAP option. Both /LINE and /MAP force a map file to be
 created unless NULL is explicitly specified.


 12.3.4  The libraries Field

 You can specify one or more standard or import libraries (not DLLs) in the
 libraries field. If you name more than one library, separate the names with
 a plus sign (+) or a space. To extend libraries to the following line, type
 a plus sign (+) as the last character on the current line, press ENTER, and
 continue. Do not split a name across lines. If you specify the base name of
 a library without an extension, LINK assumes a default .LIB extension.

 If no library is specified, LINK searches only the default libraries named
 in the object files to resolve unresolved references. If one or more
 libraries are specified, LINK searches them in the order named before
 searching the default libraries.

 You can tell LINK to search additional directories for specified or default
 libraries by giving a drive name or path specification in the libraries
 field; end the specification with a backslash ( \ ). (If you don't include
 the backslash, LINK assumes the last element of the path is a library file.)
 LINK looks for files ending in .LIB in these directories.

 You can specify a total of 32 paths or libraries in the field. If you give
 more than 32 paths or libraries, LINK ignores the additional specifications
 without warning you.

 You might need to specify library names when you want to


   ■   Use a default library that has been renamed.

   ■   Specify a library other than the default named in the object file (for
       example, a library that handles floating-point arithmetic differently
       from the default library).

   ■   Search additional libraries.

   ■   Find a library not in the current directory and not in a directory
       specified by the LIB environment variable.



 12.3.4.1  Overriding Default-Library Searches

 Most compilers insert the names of the required language libraries in the
 object files. LINK searches for these default libraries automatically; you
 do not need to specify them in the libraries field. The libraries must
 already exist with the name expected by LINK. Default-library names usually
 refer to combined libraries built and named during setup; consult your
 compiler documentation for more information about default libraries.

 To make LINK ignore the default libraries, use the /NOD option. This leaves
 unresolved references in the object files, so you must use the libraries
 field to specify the alternative libraries that LINK is to search.


 12.3.4.2  Import Libraries

 You can specify import libraries created by the IMPLIB utility anywhere you
 can specify standard libraries. You can also use the LIB utility to combine
 import libraries and standard libraries. These combined libraries can then
 be specified in the libraries field.


 12.3.4.3  How LINK Resolves References

 LINK searches static libraries to resolve external references. A static
 library is either a standard library created by the LIB utility or an import
 library created by the IMPLIB utility. The linker searches first in the
 libraries and library directories you specify (in the order you specify
 them), then in the default libraries. If a default library is explicitly
 specified, it is searched in the order it is given.

 LINK uses only those library modules needed to resolve external references,
 not the entire library. However, if you enter a library as a load library in
 the objfiles field, all the modules of a load library are added to the main
 output.


 12.3.4.4  How LINK Searches for Library Files

 When searching for libraries, LINK looks in the following locations in this
 order:


   1.  The directory specified for the file (if a path is included). If the
       file is not in that directory, the search terminates. (The default
       libraries named in object files by Microsoft compilers do not include
       path specifications.)

   2.  The current directory.

   3.  Any directories in the libraries field.

   4.  Any directories specified in the LIB environment variable.


 If LINK cannot locate a library file, it prompts you to enter the location.
 The /BATCH option disables this prompting.


 Example

 The following is a specification in the libraries field:

   C:\TESTLIB\ NEWLIBV3 C:\MYLIBS\SPECIAL

 LINK searches NEWLIBV3.LIB first for unresolved references. Since no
 directory is specified for NEWLIBV3.LIB, LINK searches the following
 locations in this order:


   1.  The current directory

   2.  The C:\TESTLIB\ directory

   3.  The directories in the LIB environment variable


 If LINK still cannot find NEWLIBV3.LIB, it prompts you with the message

   Enter new file spec

 You can then enter either a path to the library or a full pathname for
 another library.

 If unresolved references remain after searching NEWLIBV3.LIB, LINK then
 searches the library C:\MYLIBS\SPECIAL.LIB. If LINK cannot find this
 library, it prompts you as described above for NEWLIBV3.LIB. If there are
 still unresolved references, LINK searches the default libraries.


 12.3.5  The deffile Field

 Use the deffile field to specify a module-definition file when you are
 linking a segmented executable file, which is an application or DLL for OS/2
 or Windows. A module-definition file is optional for an application but
 required for a DLL. If you specify a base name with no extension, LINK
 assumes a .DEF extension. If the filename has no extension, put a period (.)
 at the end of the name.

 By default, LINK assumes that no deffile needs to be specified. If you are
 linking for DOS, use a semicolon to terminate the command line before the
 deffile field (or accept the default NUL.DEF at the  Definitions File
 prompt).


 12.3.5.1  How LINK Searches for Module-Definition Files

 LINK searches for the module-definition file in the following order:


   1.  The directory specified for the file (if a path is included). If the
       file is not in that directory, the search terminates.

   2.  The current directory.


 For information on module-definition files, see Chapter 13.


 12.3.6  Examples

 The following examples illustrate various uses of the LINK command line.


 Example 1

   LINK FUN+TEXT+TABLE+CARE, , FUNLIST, XLIB.LIB;

 This command line links the object files FUN.OBJ, TEXT.OBJ, TABLE.OBJ, and
 CARE.OBJ. By default, the executable file is named FUN.EXE, because the base
 name of the first object file is  FUN, and no name is specified for the
 executable file. The map file is named FUNLIST.MAP. LINK searches for
 unresolved external references in the library XLIB.LIB before searching in
 the default libraries. LINK does not prompt for a .DEF file because a
 semicolon appears before the deffile field.


 Example 2

   LINK FUN, , ;

 This command produces a map file named FUN.MAP because a comma appears as a
 placeholder for the mapfile field on the command line.


 Example 3

   LINK FUN, ;
   LINK FUN;

 Neither of these commands produces a map file, because commas do not appear
 as placeholders for the mapfile field. The semicolon (;) terminates the
 command line and accepts all remaining defaults without prompting; the
 prompting default for the map file is not to create one.


 Example 4

   LINK MAIN+GETDATA+PRINTIT, , MAIN ;

 This command links the files MAIN.OBJ, GETDATA.OBJ, and PRINTIT.OBJ into a
 DOS executable file because no module-definition file is specified. The map
 file MAIN.MAP is created.


 Example 5

   LINK GETDATA+PRINTIT, , , , MODDEF

 This command links GETDATA.OBJ and PRINTIT.OBJ into a DLL if MODDEF.DEF
 contains a LIBRARY statement. Otherwise, it links them into a segmented
 executable file for OS/2 or Windows. LINK creates a map file named
 GETDATA.MAP.


 12.4  Running LINK

 The simplest use of LINK is to combine one or more object files with a
 run-time library to create an executable file. You type  LINK  at the
 command-line prompt, followed by the names of the object files and a
 semicolon (;). LINK combines the object files with language libraries
 specified in the object files to create an executable file. By default, the
 executable file takes the name of the first object file in the list.

 To interrupt LINK and return to the operating-system prompt, press CTRL+C at
 any time.

 LINK expects you to supply at least one input field (the objfiles field),
 and as many as five. There are several ways to supply the input fields LINK
 expects:


   ■   Enter all the required input directly on the command line.

   ■   Omit one or more of the input fields and respond when LINK prompts for
       the missing fields.

   ■   Put the input in a response file and enter the response-file name in
       place of the expected input.


 These methods can be used in combination. The LINK command line was
 discussed in Section 12.3. The following sections explain the other two
 methods.


 12.4.1  Specifying Input with LINK Prompts

 If any field is missing from the LINK command line and the line does not end
 with a semicolon, or if any of the supplied fields are invalid, LINK prompts
 you for the missing or incorrect information. LINK displays one prompt at a
 time and waits until you respond:

   Object Modules [.OBJ]:
   Run File [basename.EXE]:
   List File [NUL.MAP]:
   Libraries [.LIB]:
   Definitions File [NUL.DEF]:

 The LINK prompts correspond to the command-line fields described earlier in
 this chapter. If you want LINK to prompt you for every input field,
 including objfiles, type the command  LINK  by itself.

 Options can be entered anywhere in any field, before the semicolon if
 specified.


 12.4.1.1  Defaults

 The default values for each field are shown in brackets. Press ENTER to
 accept the default, or type in the filename(s) you want. The basename is the
 base name of the first object file you specified. To select the default
 responses for all the remaining prompts and terminate prompting, type a
 semicolon (;) and press ENTER.

 If you specify a filename without giving an extension, LINK adds the
 appropriate default extension. To specify a filename that does not have an
 extension, type a period (.) after the name.

 Use a space or plus sign (+) to separate multiple filenames in the objfiles
 and libraries fields. To extend a long objfiles or libraries response to a
 new line, type a plus sign (+) as the last character on the current line and
 press ENTER. You can continue entering your response when the same prompt
 appears on a new line. Do not split a filename or a pathname across lines.


 12.4.2  Specifying Input in a Response File

 You can supply input to LINK in a response file. A response file is a text
 file containing the input LINK expects on the command line or in response to
 prompts. Response files can be used to hold frequently used options or
 responses, or to overcome the 128-character limit on the length of a DOS
 command line.


 12.4.2.1  Usage

 Specify the name of the response file in place of the expected command-line
 input or in response to a prompt. Precede the name with an at sign (@), as
 in @responsefile. You must specify an extension if the response file has
 one; there is no default extension. You can specify a path with the
 filename.

 You can specify a response file in any field (either on the command line or
 when responding to prompts) to supply input for one or more consecutive
 fields or all remaining fields. Note that LINK assumes nothing about the
 contents of the response file; LINK simply reads the fields from the file
 and applies them, in order, to the fields for which it has no input. LINK
 ignores any fields in the response file or on the command line after the
 five expected fields are satisfied or a semicolon (;) appears.


 Example

 The following command invokes LINK and supplies all input in a response
 file, except the last input field:

   LINK @input.txt, mydefs


 12.4.2.2  Contents of the Response File

 Each input field must appear on a separate line or be separated from other
 fields on the same line by a comma. You can extend a field to the following
 line by adding a plus sign (+) at the end of the current line. A blank field
 can be represented by either a blank line or a comma.

 Options can be entered anywhere in any field, before the semicolon if
 specified.

 If a response file does not specify all the fields, LINK prompts you for the
 rest. Use a semicolon (;) to suppress prompting and accept the default
 responses for all remaining fields.


 Example

   FUN TEXT TABLE+
   CARE
   /MAP
   FUNLIST
   GRAF.LIB ;

 If the response file above is named  FUN.LNK, the command

   LINK @FUN.LNK

 causes LINK to


   ■   Link the four object files FUN.OBJ, TEXT.OBJ, TABLE.OBJ, and CARE.OBJ
       into an executable file named FUN.EXE.

   ■   Include public symbols and addresses in the map file.

   ■   Make the name of the map file FUNLIST.MAP.

   ■   Link any needed routines from the library file GRAF.LIB.

   ■   Assume no module-definition file.



 12.5  LINK Options

 This section explains how to use options to control LINK's behavior and
 modify LINK's output. It contains a description of each option following a
 brief introduction on how to specify options.


 12.5.1  Specifying Options

 The following paragraphs discuss rules for using options.


 12.5.1.1  Syntax

 All options begin with a slash ( / ). You can specify an option by using the
 shortest sequence of characters that uniquely identifies the option. The
 description for each option shows the minimum legal abbreviation with the
 optional part enclosed in double brackets. No gaps or transpositions of
 letters are allowed. For example,

   /B«ATCH»

 indicates that either /B or /BATCH can be used, as can /BA, /BAT, or /BATC.
 Option names are not case sensitive, so you can also specify /batch or
 /Batch. This chapter uses meaningful yet legal forms of the option names.


 12.5.1.2  Usage

 LINK options can appear on the command line, in response to a prompt, or as
 part of a field in a response file. They can also be specified in the LINK
 environment variable. (For more information, see Section 12.6, "Setting
 Options with the LINK Environment Variable.") Options can appear in any
 field before the last input, except as noted in the descriptions.

 If an option appears more than once (for example, on the command line and in
 the LINK variable), the effect is the same as if the option was given only
 once. If two options conflict, the most recently specified option takes
 effect. This means that a command-line option or one given in response to a
 prompt overrides one specified in the LINK environment variable. For
 example, the command-line option /SEG:512 cancels the effect of the
 environment-variable option /SEG:256.


 12.5.1.3  Numeric Arguments

 Some LINK options take numeric arguments. You can enter numbers either in
 decimal format or in standard C-language notation.


 12.5.2  The /ALIGN Option


 Option

   /A«LIGNMENT»:size

 The /ALIGN option aligns segments in a segmented executable file at the
 boundaries specified by size. The size argument must be an integer power of
 two. For example,

   /ALIGN:16

 indicates an alignment boundary of 16 bytes. The default alignment is 512
 bytes.

 This option reduces the size of the disk file by reducing the size of gaps
 between segments. It has no effect on the size of the file when loaded in
 memory.


 12.5.3  The /BATCH Option


 Option

   /B«ATCH»

 The /BATCH option suppresses prompting for libraries or object files that
 LINK cannot find. By default, the linker prompts for a new pathname whenever
 it cannot find a library that it has been directed to use. It also prompts
 you if it cannot find an object file that it expects to find on a floppy
 disk. When /BATCH is used, the linker generates an error or warning message
 (if appropriate). The /BATCH option also suppresses the LINK copyright
 message and echoed input from response files.

 Using this option can cause unresolved external references. It is intended
 primarily for users who use batch files or makefiles for linking many
 executable files with a single command and who wish to prevent linker
 operation from halting.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 This option does not suppress prompts for input fields. Use a semicolon (;)
 at the end of the LINK input to suppress input prompting.
 ────────────────────────────────────────────────────────────────────────────


 12.5.4  The /CO Option


 Option

   /CO«DEVIEW»

 The /CO option adds line numbers and symbolic data to the executable file
 for use with the Microsoft CodeView debugger. The /CO option has no effect
 if the object files do not contain CodeView debugging information.

 You can run the resulting executable file outside CodeView; the debugging
 data in the file is ignored. However, it increases file size and slows
 execution slightly. You should link a separate release version without the
 /CO option after the program has been debugged.

 When /CO is used with the /TINY option, debug information is put in a
 separate file with the same base name as the .COM file and with the .DBG
 extension.

 The /CO option is not compatible with the /EXEPACK option for DOS executable
 files.


 12.5.5  The /CPARM Option


 Option

   /CP«ARMAXALLOC»:number

 The /CPARM option sets the maximum number of 16-byte paragraphs needed by
 the program when it is loaded into memory. The operating system uses this
 value to allocate space for the program before loading it. This option is
 useful when you want to execute another program from within your program and
 you need to reserve memory for the program. The /CPARM option is valid only
 when linking DOS programs.

 LINK normally requests the operating system to set the maximum number of
 paragraphs to 65,535. Since this is more memory than DOS can supply, the
 operating system always denies the request and allocates the largest
 contiguous block of memory it can find. If the /CPARM option is used, the
 operating system allocates no more space than the option specified. Any
 memory in excess of that required for the program loaded is free for other
 programs.

 The number can be any integer value in the range 1 to 65,535. If number is
 less than the minimum number of paragraphs needed by the program, LINK
 ignores your request and sets the maximum value equal to whatever the
 minimum value happens to be. The minimum number of paragraphs needed by a
 program is never less than the number of paragraphs of code and data in the
 program. To free more memory for programs compiled in the medium and large
 models, link with /CPARM:1. This leaves no space for the near heap.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 You can change the maximum allocation after linking by using the EXEHDR
 utility, which modifies the executable-file header.
 ────────────────────────────────────────────────────────────────────────────


 12.5.6  The /DOSSEG Option


 Option

   /DO«SSEG»

 The /DOSSEG option forces segments to be ordered as follows:


   1.  All segments with a class name ending in CODE

   2.  All other segments outside DGROUP

   3.  DGROUP segments, in the following order:

       a.  Any segments of class BEGDATA. (This class name is reserved for
           Microsoft use.)

       b.  Any segments not of class BEGDATA, BSS, or STACK.

       c.  Segments of class BSS.

       d.  Segments of class STACK.



 In addition, /DOSSEG option defines the following two labels:

   _edata = DGROUP : BSS
   _end   = DGROUP : STACK

 The variables  _edata  and  _end  have special meanings for Microsoft
 compilers, so you should not define program variables with these names.
 Assembly-language programs can reference these variables but should not
 change them.

 The /DOSSEG option also inserts 16 null bytes at the beginning of the _TEXT
 segment (if this segment is defined). This behavior of the option is
 overridden by the /NONULLS option when both are used; use /NONULLS to
 override the DOSSEG comment record commonly found in standard Microsoft
 libraries.

 This option is principally for use with assembly-language programs. When you
 link high-level-language programs, a special object-module record in the
 Microsoft language libraries automatically enables the /DOSSEG option. This
 option is also enabled by assembly modules that use MASM directive .DOSSEG.



 12.5.7  The /DSALLOC Option


 Option

   /DS«ALLOCATE»

 The /DSALLOC option tells LINK to load all data starting at the high end of
 the data segment. At run time, the data segment (DS) register is set to the
 lowest data-segment address that contains program data.

 By default, LINK loads all data starting at the low end of the data segment.
 At run time, the DS register is set to the lowest possible address to allow
 the entire data segment to be used.

 The /DSALLOC option is most often used with the /HIGH option to take
 advantage of unused memory within the data segment. These options are valid
 only for assembly-language programs that create DOS .EXE files.


 12.5.8  The /EXEPACK Option


 Option

   /E«XEPACK»

 The /EXEPACK option directs LINK to remove sequences of repeated bytes
 (usually null characters) and to optimize the load-time relocation table
 before creating the executable file. (The load-time relocation table is a
 table of references relative to the start of the program, each of which
 changes when the executable image is loaded into memory and an actual
 address for the entry point is assigned.)

 The /EXEPACK option does not always produce a significant saving in disk
 space and may sometimes actually increase file size. Programs that have a
 large number of load-time relocations (about 500 or more) and long streams
 of repeated characters are usually shorter if packed. LINK notifies you if
 the packed file is larger than the unpacked file. The time required to
 expand a packed file may cause it to load more slowly than a file linked
 without this option.

 You cannot debug packed files with CodeView, because the /EXEPACK option
 removes symbolic information. A LINK warning message notifies you of this.

 The /EXEPACK option is not compatible with the /INCR option or with Windows
 programs.


 12.5.9  The /FARCALL Option


 Option

   /F«ARCALLTRANSLATION»

 The /FARCALL option directs the linker to optimize far calls to procedures
 that lie in the same segment as the caller. This can result in slightly
 faster code; the gain in speed is most apparent on 80286-based machines and
 later. The /PACKC option can be used with /FARCALL when linking for OS/2.
 /PACKC is not recommended when linking Windows applications with /FARCALL.

 The /FARCALL option is off by default. If an environment variable (such as
 LINK or FL) includes /FARCALL, you can use the /NOFARCALL option to override
 it.

 FARCALL optimizes by creating more efficient code.

 A program that has multiple code segments may make a far call to a procedure
 in the same segment. Since the segment address is the same (for both the
 code and the procedure it calls), only a near call is necessary. Far calls
 appear in the relocation table; a near call does not require a table entry.
 By converting far calls to near calls in the same segment, the /FARCALL
 option both reduces the size of the relocation table and increases execution
 speed, since only the offset needs to be loaded, not a new segment. The
 /FARCALL option has no effect on programs that make only near calls, since
 there are no far calls to convert.

 When /FARCALL is specified, the linker optimizes code by removing the
 instruction  call FAR label  and substituting the following sequence:

   nop
           push    cs
           call    NEAR label

 During execution, the called procedure still returns with a far-return
 instruction. However, because both the code segment and the near address are
 on the stack, the far return is executed correctly. The  nop  (no-op)
 instruction is added so that exactly five bytes replace the five-byte
 far-call instruction.

 In rare cases, /FARCALL should be used with caution.

 There is a small risk with the /FARCALL option. If LINK sees the far-call
 opcode (9A hexadecimal) followed by a far pointer to the current statement,
 and that segment has a class name ending in  CODE, it interprets that as a
 far call. This problem can occur when using  _based (segname  ("CODE")) in a
 C program. If a program linked with /FARCALL fails for no apparent reason,
 try using /NOFARCALL.

 Object modules produced by Microsoft high-level languages are safe from this
 problem because little immediate data is stored in code segments.
 Assemblylanguage programs are generally safe for use with the /FARCALL
 option if they do not involve advanced system-level code, such as might be
 found in operating systems or interrupt handlers.


 12.5.10  The /HELP Option


 Option

   /HE«LP»

 The /HELP option calls the QuickHelp utility. If LINK cannot find the help
 file or QuickHelp, it displays a brief summary of LINK command-line syntax
 and options. Do not give a filename when using the /HELP option.


 12.5.11  The /HIGH Option


 Option

   /HI«GH»

 At load time, the executable file can be placed either as low or as high in
 memory as possible. The /HIGH option causes DOS to place the executable file
 as high as possible in memory. Without the /HIGH option, DOS places the
 executable file as low as possible. This option is usually used with the
 /DSALLOC option. These options are valid only for assembly-language programs
 that create DOS .EXE files.


 12.5.12  The /INCR Option


 Option

   /INC«REMENTAL»

 The /INCR option must be used to prepare for subsequent linking with ILINK.
 This option produces a .SYM file and an .ILK file, each containing
 additional information needed by ILINK.

 When /INCR is specified, LINK creates the main output file as a segmented
 executable file. If the main output is a DOS application, LINK adds a stub
 loader so that the program can run under DOS. The file is slightly larger
 than it would be without /INCR.

 The /PADC and /PADD options are often used with the /INCR option to increase
 buffer size and thereby increase the likelihood that incremental linking
 will be successful. The /TINY and /EXEPACK options are not compatible with
 /INCR.

 You should not use /INCR or ILINK for the release version of a product.
 ILINK is intended to speed linking during development and debugging. In rare
 cases, linking with /INCR causes warning  L4001  to be generated. If this
 occurs, do not use this option or ILINK.


 12.5.13  The /INFO Option


 Option

   /INF«ORMATION»

 The /INFO option displays to the standard output information about the
 linking process, including the phase of linking and the names of the object
 files being linked. This option is a useful way to determine the locations
 of the object files being linked, the number of segments, and the order in
 which they are linked.


 12.5.14  The /LINE Option


 Option

   /LI«NENUMBERS»

 The /LINE option adds the line numbers and associated addresses from source
 files to the map file. The object file must contain line-number information
 for it to appear in the map file. If the object file has no line-number
 information, the /LINE option has no effect. (Use the /Zd or /Zi option with
 Microsoft compilers such as CL, FL, and ML to add line numbers to the object
 file.) If you also want to add public symbols to the map file, use the /MAP
 option.

 The /LINE option causes a map file to be created even if you did not
 explicitly tell the linker to do so. By default, the map file is given the
 same base name as the executable file with the extension .MAP. You can
 override the default name by specifying a new map filename in the mapfile
 field or in response to the  List File  prompt.


 12.5.15  The /MAP Option


 Option

   /M«AP»

 The /MAP option adds to the map file all public (global) symbols defined in
 object files. When /MAP is specified, the map file contains a list of all
 the symbols sorted by name and a list of all the symbols sorted by address.
 If you do not use this option, the map file contains only a list of
 segments. If you also want to add line numbers to the map file, use the
 /LINE option.

 The /MAP option causes a map file to be created even if you did not
 explicitly tell the linker to do so. By default, the map file is given the
 same base name as the executable file with the extension .MAP. You can
 override the default name by specifying a new map filename in the mapfile
 field or in response to the  List File  prompt.

 Under some circumstances, adding symbols slows the linking process. If this
 is a problem, do not use /MAP.


 12.5.16  The /NOD Option


 Option

   /NOD«EFAULTLIBRARYSEARCH»«:libraryname»

 The /NOD option tells LINK not to search default libraries named in object
 files. Specifying libraryname tells LINK to search all libraries named in
 the object files except libraryname. If you want LINK to ignore more than
 one library, specify /NOD once for each library. To tell LINK to ignore all
 default libraries, specify /NOD without a libraryname.

 High-level-language object files usually must be linked with a run-time
 library to produce an executable file. Therefore, if you use the /NOD
 option, you must also use the libraries field to specify an alternate
 library that resolves the external references in the object files.


 12.5.17  The /NOE Option


 Option

   /NOE«XTDICTIONARY»

 The /NOE option prevents the linker from searching extended dictionaries,
 which are lists of symbol locations in libraries created with LIB. The
 linker consults extended dictionaries to speed up library searches.

 Using /NOE slows the linker. Use this option when you are redefining a
 symbol or function defined in a library and you get the error

   L2044 symbol multiply defined, use /NOE


 12.5.18  The /NOFARCALL Option


 Option

   /NOF«ARCALLTRANSLATION»

 The /NOFARCALL option turns off far-call optimization (translation).
 Far-call optimization is off by default. However, if an environment variable
 (such as LINK or FL) includes the /FARCALL option, you can use /NOFARCALL to
 override /FARCALL.


 12.5.19  The /NOGROUP Option


 Option

   /NOG«ROUPASSOCIATION»

 The /NOGROUP option ignores group associations when assigning addresses to
 data and code items. It is provided primarily for compatibility with
 previous versions of the linker (2.02 and earlier) and early versions of
 Microsoft compilers. This option is valid only for assembly-language
 programs that create DOS .EXE files.>


 12.5.20  The /NOI Option


 Option

   /NOI«GNORECASE»

 This option preserves case in identifiers. By default, LINK treats uppercase
 and lowercase letters as equivalent. Thus  ABC,  Abc, and  abc  are
 considered the same name. When you use the /NOI option, the linker
 distinguishes between uppercase and lowercase, and considers these
 identifiers to be three different names.

 In most high-level languages, identifiers are not case sensitive, so this
 option has no effect. However, case is significant in C. It's a good idea to
 use this option with C programs to catch misnamed identifiers.


 12.5.21  The /NOLOGO Option


 Option

   /NOL«OGO»

 The /NOLOGO option suppresses the copyright message displayed when LINK
 starts. This option has no effect if not specified first on the command line
 or in the LINK environment variable.


 12.5.22  The /NONULLS Option


 Option

   /NON«ULLSDOSSEG»

 The /NONULLS option arranges segments in the same order they are arranged by
 the /DOSSEG option. The only difference is that the /DOSSEG option inserts
 16 null bytes at the beginning of the _TEXT segment (if it is defined), but
 /NONULLS does not insert the extra bytes.

 If both the /DOSSEG and /NONULLS options are given, the /NONULLS option
 takes precedence. You can therefore use /NONULLS to override the DOSSEG
 comment record found in run-time libraries. This option is for segmented
 executable files.


 12.5.23  The /NOPACKC Option


 Option

   /NOP«ACKCODE»

 This option turns off code-segment packing. Code-segment packing is normally
 off by default. However, if an environment variable (such as LINK or FL)
 includes the /PACKC option to turn on code-segment packing, you can use
 /NOPACKC to override /PACKC.


 12.5.24  The /OV Option


 Option

   /O«VERLAYINTERRUPT»:number

 This option sets an interrupt number for passing control to overlays. By
 default, the interrupt number used for passing control to overlays is 63 (3F
 hexadecimal). The /OV option allows you to select a different interrupt
 number. This option is valid only when linking DOS programs.

 The number can be any number from 0 to 255, specified in decimal format or
 in C-language notation. Numbers that conflict with DOS interrupts can be
 used; however, their use is not advised. You should use this option only
 when you want to use overlays with a program that already reserves interrupt
 63 for some other purpose.


 12.5.25  The /PACKC Option


 Option

   /PACKC«ODE»«:number»

 The /PACKC option turns on code-segment packing. The linker packs code
 segments by grouping neighboring code segments that have the same
 attributes. Segments in the same group are assigned the same segment
 address; offset addresses are adjusted accordingly. All items have the same
 physical address whether or not the /PACKC option is used. However, /PACKC
 changes the segment and offset addresses so that all items in a group share
 the same segment.

 The number specifies the maximum size of groups formed by /PACKC. The linker
 stops adding segments to a group when it cannot add another segment without
 exceeding number; then it starts a new group. The default segment size
 without /PACKC (or when /PACKC is specified without number) is 65,500 bytes
 (64K - 36 bytes).

 The /PACKC option produces slightly faster and more compact code. It affects
 only programs with multiple code segments. This option is off by default
 and, if specified in an environment variable, can be overridden with the
 /NOPACKC option.

 Code-segment packing provides more opportunities for far-call optimization
 (which is enabled with the /FARCALL option). The /FARCALL and /PACKC options
 together produce faster and more compact code. However, this combination is
 not recommended for Windows applications.

 Use caution when packing assembly-language programs.

 Object code created by Microsoft compilers can safely be linked with the
 /PACKC option. This option is unsafe only when used with assembly-language
 programs that make assumptions about the relative order of code segments.
 For example, the following assembly code attempts to calculate the distance
 between CSEG1  and  CSEG2. This code produces incorrect results when used
 with /PACKC, because /PACKC causes the two segments to share the same
 segment address. Therefore, the procedure would always return zero.

   CSEG1      SEGMENT PUBLIC 'CODE'
   .
   .
   .
   CSEG1      ENDS

   CSEG2      SEGMENT PARA PUBLIC 'CODE'
              ASSUME  cs:CSEG2

   ; Return the length of CSEG1 in AX

   codesize   PROC  NEAR
              mov   ax, CSEG2  ; Load para address of CSEG1
              sub   ax, CSEG1  ; Load para address of CSEG2
              mov   cx, 4      ; Load count
              shl   ax, cl     ; Convert distance from paragraphs
                               ;  to bytes
   codesize   ENDP

   CSEG2      ENDS


 12.5.26  The /PACKD Option


 Option

   /PACKD«ATA»«:number»

 The /PACKDoption turns on data-segment packing. The linker considers any
 segment definition with a class name that does not end in  CODE  as a data
 segment. Adjacent data-segment definitions are combined into the same
 physical segment. The linker stops adding segments to a group when it cannot
 add another segment without exceeding number bytes; then it starts a new
 group. The default segment size without /PACKD (or when /PACKD is specified
 without number) is 65,536 bytes (64K).

 The /PACKD option produces slightly faster and more compact code. It affects
 only programs with multiple data segments and is valid for OS/2 and Windows
 programs only. It might be necessary to use the /PACKD option to get around
 the limit of 255 physical data segments per executable file imposed by OS/2
 and Windows. Try using /PACKD if you get the following LINK error:

   L1073 file-segment limit exceeded

 This option may not be safe with other compilers that do not generate fixup
 records for all far data references.


 12.5.27  The /PADC Option


 Option

   /PADC«ODE»«:padsize»

 The /PADC option adds filler bytes to the end of each code segment for use
 when later linking with ILINK. If you use /PADC, you must also specify the
 /INCR option.

 The padsize is optional; the default is 0 bytes. If incremental linking
 fails, you can specify a padsize in decimal format or C-language notation.
 For example,  /PADC:256  adds an additional 256 bytes to each code segment.
 (You can also use  0400  or  0x100  to specify 256 bytes.)

 The linker recognizes code segments as segment definitions with class names
 that end in  CODE. Microsoft high-level languages automatically use this
 declaration for code segments. Code padding is not usually necessary for
 programs with multiple code segments but is recommended for mixed-model
 programs, programs with one code segment, and assembly-language programs in
 which code segments are grouped.


 12.5.28  The /PADD Option


 Option

   /PADD«ATA»«:padsize»

 The /PADD option adds filler bytes to the end of each data segment to permit
 subsequent linking with ILINK. If you use /PADD, you must also specify the
 /INCR option.

 The padsize is optional; the default is 16 bytes. The /INCR option itself
 adds 16 bytes. This default padding is usually sufficient for successful
 incremental linking. If incremental linking fails, you can specify a padsize
 in decimal format or C-language notation. (If you specify too large a
 padsize, you might exceed the 64K limitation on the size of the default data
 segment.) For example,  /PADD:32  adds an additional 32 bytes to each data
 segment. (You can also use  040  or  0x20  to specify 32 bytes.)


 12.5.29  The /PAUSE Option


 Option

   /PAU«SE»

 The /PAUSE option pauses the session before LINK writes the executable file
 or DLL to disk. This option is supplied for compatibility with machines that
 have two floppy drives but no hard disk. It allows you to swap floppy disks
 before LINK writes the executable file.

 If you specify the /PAUSE option, LINK displays the following message before
 it creates the main output:

   About to generate .EXE file
   Change diskette in drive letter and press <ENTER>

 The letter is the current drive. LINK resumes processing when you press
 ENTER.

 Do not remove a disk that contains either the map file or the temporary
 file. If LINK creates a temporary file on the disk you plan to remove,
 terminate the LINK session and rearrange your files so that the temporary
 file is on a disk that does not need to be removed. For more information on
 how LINK determines where to put the temporary file, see Section 12.9, "LINK
 Temporary Files."


 12.5.30  The /PM Option


 Option

   /PM«TYPE»:type

 This option specifies the type of Windows or OS/2 application being
 generated. The /PM option is equivalent to including a type specification in
 the NAME statement in a module-definition file.

 The type field can take one of the following values:

 Value                             Description
 ────────────────────────────────────────────────────────────────────────────
 PM                                Presentation Manager (PM) or Windows
                                   application. The application uses the
                                   API provided by PM or Windows and must
                                   be executed in the PM or Windows
                                   environment. This is equivalent to NAME
                                   WINDOWAPI.

 VIO                               Character-mode application to run in a
                                   text window in the
                                   PM or Windows session. This is
                                   equivalent to NAME
                                   WINDOWCOMPAT.

 NOVIO                             The default. Character-mode application
                                   that must run full screen and cannot run
                                   in a text window in PM or in Windows.
                                   This is equivalent to NAME
                                   NOTWINDOWCOMPAT.



 12.5.31  The /Q Option


 Option

   /Q«UICKLIBRARY»

 The /Q option directs the linker to produce a "Quick library" instead of an
 executable file. A Quick library is similar to a standard library in that
 both contain routines that can be called by a program. However, a standard
 library is linked with a program at link time; in contrast, a Quick library
 is linked with a program at run time.

 When /Q is specified, the exefile field refers to a Quick library instead of
 an application. The default extension for this field is then .QLB instead of
 .EXE.

 Quick libraries can be used only with programs created with Microsoft
 QuickBasic or early versions of Microsoft QuickC. These programs have the
 special code that loads a Quick library at run time.


 12.5.32  The /SEG Option


 Option

   /SE«GMENTS»«:number»

 The /SEG option sets the maximum number of program segments. The default
 without /SEG or number is 128. You can specify number as any value from 1 to
 16,384 in individual format or C-language notation. However, the number of
 segment definitions is constrained by available memory.

 LINK must allocate some memory to keep track of information for each
 segment; the larger the number you specify, the less free memory LINK has to
 run in. A relatively low segment limit (such as the 128 default) reduces the
 chance LINK will run out of memory. For programs with fewer than 128
 segments, you can minimize LINK's memory requirements by setting number to
 reflect the actual number of segments in the program. If a program has more
 than 128 segments, however, you must set a higher value.

 If the number of segments allocated is too high for the amount of memory
 available while linking, LINK displays the error message

   L1054 requested segment limit too high

 When this happens, try linking again after setting /SEG to a smaller number.



 12.5.33  The /STACK Option


 Option

   /ST«ACK»:number

 The /STACK option lets you change the stack size from its default value of
 2,048 bytes. The number is any positive value in decimal or C-language
 notation, up to 64K.

 Programs that pass large arrays or structures by value or with deeply nested
 subroutines may need additional stack space. In contrast, if your program
 uses the stack very little, you might be able to save space by decreasing
 the stack size. If a program fails with a stack-overflow message, try
 increasing the size of the stack.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 You can also use the EXEHDR utility to change the default stack size by
 modifying the executable-file header.
 ────────────────────────────────────────────────────────────────────────────


 12.5.34  The /TINY Option


 Option

   /T«INY»

 The /TINY option produces a .COM file instead of an .EXE file. The default
 extension of the output file is .COM. When the /CO option is used with
 /TINY, debug information is put in a separate file with the same base name
 as the .COM file and with the .DBG extension.

 Not every program can be linked in the .COM format. The following
 restrictions apply:


   ■   The program must consist of only one physical segment. You can declare
       more than one segment in assembly-language programs; however, the
       segments must be in the same group.

   ■   The code must not use far references.

   ■   Segment addresses cannot be used as immediate data for instructions.
       For example, you cannot use the following instruction:

       mov     ax, CODESEG


   ■   Windows and OS/2 programs cannot be converted to a .COM format.



 12.5.35  The /W Option


 Option

   /W«ARNFIXUP»

 The /W option issues the  L4000  warning when LINK uses a displacement from
 the beginning of a group in determining a fixup value. This option is
 provided because early versions of the Windows linker (LINK4) performed
 fixups without this displacement. This option is for linking segmented
 executable files.


 12.5.36  The /? Option


 Option

   /?

 The /? option displays a brief summary of LINK command-line syntax and
 options.


 12.6  Setting Options with the LINK Environment Variable

 You can use the LINK environment variable to set options that will be in
 effect each time you link. (Microsoft compilers such as CL, FL, and ML also
 use the options in the LINK environment variable.)


 12.6.1  Setting the LINK Environment Variable

 You set the LINK environment variable with the following operating-system
 command:

   SET LINK=options

 LINK expects to find options listed in the variable exactly as you would
 type them in fields on the command line, in response to a prompt, or in a
 response file. It does not accept input for other fields; filenames in the
 LINK variable cause an error.


 Example

   SET LINK=/NOI /SEG:256 /CO
   LINK TEST;
   LINK /NOD PROG;

 In the example above, the commands are specified at the system prompt. The
 file TEST.OBJ is linked using the options  /NOI,  /SEG:256, and  /CO. The
 file PROG.OBJ is then linked with the option  /NOD, in addition to  /NOI,
 /SEG:256, and  /CO.


 12.6.2  Behavior of the LINK Environment Variable

 You can specify options on the LINK command line or in a response file in
 addition to those in the LINK environment variable. If an option appears
 both in an input field and in the LINK variable, the input-field option
 overrides any environment-variable option it conflicts with. For example,
 the command-line option /SEG:512 overrides the environment-variable option
 /SEG:256.


 12.6.3  Clearing the LINK Environment Variable

 You must reset the LINK environment variable to prevent LINK from using its
 options. To clear the LINK variable, use the operating-system command

   SET LINK=

 To see the current setting of the LINK variable, type  SET  at the
 operatingsystem prompt.


 12.7  Using Overlays under DOS

 LINK can create DOS programs with "overlays." Overlays allow sections of a
 program to be loaded into memory only as needed. This permits running a
 program that would otherwise be too large to fit in available memory.
 Overlay programs execute more slowly, however, since the various program
 modules must be swapped into and out of memory.

 The CodeView debugger is compatible with overlaid modules. If you use
 CodeView to debug a program that has an overlay containing more than one
 code segment, you will see only the identifiers contained in the first
 segment of the overlay.


 12.7.1  Restrictions on Overlays

 Not all programs can use overlays. You will probably need to reorganize the
 code to accommodate the limitations explained in this section. Even after
 reorganization, some programs might not be convertible to overlay form or
 might not show a significant reduction in the amount of memory needed to
 execute them.

 Consider the following restrictions before trying to overlay a program:


   ■   You can use overlays only in programs with multiple code segments,
       because separate segment names are needed for overlays. Only code is
       overlaid, not data. The data becomes part of the "root" section of the
       program that is always in memory.

   ■   Only 255 overlays can be specified. The program can define only 255
       logical segments (segments with different names). This limits the
       total size of an overlaid program to 16 megabytes.

   ■   Only one overlay (in addition to the root) can be in memory at any one
       time. You must structure your program accordingly.

   ■   Duplicate names for different overlays are not supported; each module
       can appear only once in a program.

   ■   You must use far call/return instructions to transfer control between
       overlaid files. You cannot overlay files containing near routines if
       other overlays call those routines.

   ■   You cannot jump out of or into overlaid files using the longjmp
       C-library function. You can, however, use long jumps within an
       overlaid file.

   ■   You cannot use a function pointer to call a routine out of or into
       overlaid files. You can, however, use a function pointer to call a
       routine within an overlaid file.

   ■   You cannot use the same public name in different overlays.

   ■   The code required to manage overlays adds about 2K to 3K to the size
       of the root module.


 ────────────────────────────────────────────────────────────────────────────
 WARNING

 Never rename an executable program file containing overlays if it is to run
 under DOS 2.x and earlier. LINK records the .EXE filename in the program
 file. If you rename the file, the overlay manager may not be able to locate
 the proper file. You can rename an .EXE file that will run under DOS 3.x and
 later.
 ────────────────────────────────────────────────────────────────────────────


 12.7.2  Specifying Overlays

 Specify overlays by enclosing object-file (and possibly load-library) names
 in parentheses in the objfiles field. Each group of object files bracketed
 by parentheses represents one overlay. Overlays cannot be nested.

 The remaining modules (those not in parentheses), and any drawn from the
 run-time libraries, constitute the resident (or root) part of your program.
 The entry point to the program (for example,  main()  in a C program, or
 PROGRAM  in a FORTRAN program) must be in the root.


 Example

 The following list of files contains three overlays:

   a + (b+c) + (d+e) + f + (g)

 In this example, the groups  (b+c),  (d+e), and  (g)  are overlays. The
 remaining files  a  and  f  and any modules from libraries in the libraries
 field remain memory-resident throughout the execution of the program.

 It is important to remember that whichever object file first defines a
 segment gets all contributions to that segment. In the example above, if
 D.OBJ and F.OBJ both define the same segment, the contribution from F.OBJ to
 that segment goes into the  (d+e)  overlay rather than into the root.


 12.7.3  How Overlays Work

 Programs that use overlays require the overlay-manager code to handle module
 swapping. This code is included as part of the standard libraries for
 Microsoft high-level languages. If you specify overlays during linking, the
 code for the overlay manager is automatically linked with the rest of your
 program.

 LINK produces only one .EXE file. The overlay manager searches for this file
 whenever another overlay needs to be loaded. It first searches in the
 current directory. If the file is not there, the manager then searches the
 directories in the PATH environment variable. If the overlay manager still
 cannot find the file, it prompts for the pathname.


 Example

 Assume that an executable program called PAYROLL.EXE uses overlays and does
 not exist in either the current directory or the directories specified by
 PATH. If you run PAYROLL.EXE by entering a complete path specification, the
 overlay manager displays the following message when it attempts to load an
 overlay file:

   Cannot find PAYROLL.EXE
   Please enter new program spec:

 You can then enter the drive or directory, or both, where PAYROLL.EXE is
 located. For example, if the file is located in directory \EMPLOYEE\DATA\ on
 drive B, enter  B:\EMPLOYEE\DATA\; if the current drive is B, you can enter
 just  \EMPLOYEE\DATA\.

 If you later remove the disk in drive B and the overlay manager needs the
 overlay again, it does not find PAYROLL.EXE and displays the following
 message:

   Please insert diskette containing B:\EMPLOYEE\DATA\PAYROLL.EXE
   in drive B: and strike any key when ready.

 After the overlay file has been read from the disk, the overlay manager
 displays the following message:

   Please restore the original diskette.
   Strike any key when ready.


 12.7.4  Overlay Interrupts

 LINK replaces far calls to routines in overlays with interrupts (followed by
 the module identifier and offset). By default, the interrupt number is 63
 (3F hexadecimal). You can use the /OV option to change the interrupt number.



 12.8  Linker Operation under DOS

 LINK performs the following steps to produce a DOS executable file:


   1.  Reads the object modules submitted

   2.  Searches the given libraries, if necessary, to resolve external
       references

   3.  Assigns addresses to segments

   4.  Assigns addresses to public symbols

   5.  Reads code and data in the segments

   6.  Reads all relocation references in object modules

   7.  Performs fixups

   8.  Outputs an executable file (executable image and relocation
       information)


 Steps 5, 6, and 7 are performed iteratively─that is, LINK repeats these
 steps as many times as required before it progresses to step 8.

 The "executable image" contains the code and data that constitute the
 executable file. The "relocation information" is a list of references
 relative to the start of the program, each of which changes when the
 executable image is loaded into memory and an actual address for the entry
 point is assigned.

 The following sections explain the process LINK uses to concatenate segments
 and resolve references to items in memory.


 12.8.1  Segment Alignment

 LINK uses each segment's alignment type to set the starting address for the
 segment. The alignment types are BYTE, WORD, DWORD, PARA, and PAGE. These
 correspond to starting addresses at byte, word, doubleword, paragraph, and
 page boundaries, representing addresses that are multiples of 1, 2, 4, 16,
 and 256, respectively. The default alignment is PARA.

 When LINK encounters a segment, it checks the alignment type before copying
 the segment to the executable file. If the alignment is WORD, DWORD, PARA,
 or PAGE, LINK checks the executable image to see if the last byte copied
 ends at an appropriate boundary. If not, LINK pads the image with extra null
 bytes.


 12.8.2  Frame Number

 LINK computes a starting address for each segment in a program. The starting
 address is based on a segment's alignment and the sizes of the segments
 already copied to the executable file. The address consists of an offset and
 a "canonical frame number." The canonical frame number specifies the address
 of the first paragraph in memory containing one or more bytes of the
 segment. (A paragraph is 16 bytes of memory; therefore, to compute a
 physical location in memory, multiply the frame number by 16 and add the
 offset.) The offset is the number of bytes from the start of the paragraph
 to the first byte in the segment. For BYTE, WORD, and DWORD alignments, the
 offset may be nonzero. The offset is always zero for PARA and PAGE
 alignments. (An offset of zero means that the physical location is an exact
 multiple of 16.)

 The frame number of a segment can be obtained from the map file created by
 LINK. The first four digits of the start address give the frame number in
 hexadecimal. For example, a start address of  0C0A6  gives a frame number of
  0C0A.


 12.8.3  Segment Order

 LINK copies segments to the executable file in the same order that it
 encounters them in the object files. This order is maintained throughout the
 program unless LINK encounters two or more segments having the same class
 name. Segments having identical class names belong to the same class type
 and are copied as a contiguous block to the executable file.

 The /DOSSEG option might change the way in which segments are ordered.


 12.8.4  Combined Segments

 LINK uses combine types to determine whether two or more segments sharing
 the same segment name should be combined into one large segment. The valid
 combine types are PUBLIC, STACK, COMMON, and PRIVATE.

 If a segment has combine type PUBLIC, LINK automatically combines it with
 any other segments having the same name and belonging to the same class.
 When LINK combines segments, it ensures that the segments are contiguous and
 that all addresses in the segments can be accessed using an offset from the
 same frame address. The result is the same as if the segment were defined as
 a whole in one source file.

 LINK preserves each individual segment's alignment type. This means that
 even though the segments belong to a single large segment, the code and data
 in the segments do not lose their original alignment. If the combined
 segments exceed 64K, LINK displays an error message.

 If a segment has combine type STACK, LINK carries out the same combine
 operation as for PUBLIC segments. The only exception is that STACK segments
 cause LINK to copy an initial stack-pointer value to the executable file.
 This stack-pointer value is the offset to the end of the first stack segment
 (or combined stack segment) encountered.

 If a segment has combine type COMMON, LINK automatically combines it with
 any other segments having the same name and belonging to the same class.
 When LINK combines COMMON segments, however, it places the start of each
 segment at the same address, creating a series of overlapping segments. The
 result is a single segment no larger than the largest segment combined.

 A segment has combine type PRIVATE only if no explicit combine type is
 defined for it in the source file. LINK does not combine private segments.


 12.8.5  Groups

 Groups allow segments to be addressed relative to the same frame address.
 When LINK encounters a group, it adjusts all memory references to items in
 the group so that they are relative to the same frame address.

 Segments in a group do not have to be contiguous, belong to the same class,
 or have the same combine type. The only requirement is that all segments in
 the group fit within 64K.

 Groups do not affect the order in which the segments are loaded. Unless you
 use class names and enter object files in the right order, there is no
 guarantee the segments will be contiguous. In fact, LINK may place segments
 that do not belong to the group in the same 64K of memory. LINK does not
 explicitly check that all segments in a group fit within 64K of memory;
 however, LINK is likely to encounter a fixup-overflow error if this
 requirement is not met.


 12.8.6  Fixups

 Once the starting address of each segment in a program is known and all
 segment combinations and groups have been established, LINK can "fix up" any
 unresolved references to labels and variables. To fix up unresolved
 references, LINK computes an appropriate offset and segment address and
 replaces the temporary values generated by the assembler with the new
 values.

 LINK carries out fixups for the types of references shown in Table 12.1.

 The size of the value to be computed depends on the type of reference. If
 LINK discovers an error in the anticipated size of a reference, it displays
 a fixupoverflow message. This can happen, for example, if a program attempts
 to use a 16-bit offset to reach an instruction which is more than 64K away.
 It can also occur if all segments in a group do not fit within a single 64K
 block of memory.

 Table 12.1  LINK Fixups

 Type              Location of Reference    LINK Action
 ────────────────────────────────────────────────────────────────────────────
 Short             In JMP instructions      Computes a signed, eight-bit
                   that attempt to pass     number for the reference and
                   control to labeled       displays an error message if
                   instructions in the      the target instruction belongs
                   same segment or group.   to a different segment or group
                   The target instruction   (has a different frame address),
                   must be no more than     or if the target is more than
                   128 bytes from the       128 bytes away in either
                   point of reference.      direction.

 Near              In instructions that     Computes a 16-bit offset for
 self-relative     access data relative to  the reference and displays an
                   the same segment or      error if the data are not in
                   group.                   the same segment or group.

 Near              In instructions that     Computes a 16-bit offset for
 segment-relative  attempt to access data   the reference and displays an
                   in a specified segment   error message if the offset of
                   or group, or relative    the target within the specified
                   to a specified segment   frame is greater than 64K or
                                            less than 0, or if the
                   register.                beginning of the canonical
                                            frame of the target is not
                                            addressable.

 Long              In CALL instructions     Computes a 16-bit frame address
                   that attempt to access   and 16-bit offset for this
                   an instruction in        reference, and displays an
                   another segment or       error message if the computed
                   group.                   offset is greater than 64K or
                                            less than 0, or if the
                                            beginning of the canonical
                                            frame of the target is not
                                            addressable.

 ────────────────────────────────────────────────────────────────────────────



 12.9  LINK Temporary Files

 LINK uses available memory during the linking session. If LINK runs out of
 memory, it creates a disk file to hold intermediate files. LINK deletes this
 file when it finishes.

 When the linker creates a temporary disk file, you see the message

   Temporary file tempfile has been created.
   Do not change diskette in drive, letter.

 In the message displayed above, tempfile is the name of the temporary file
 and letter is the drive containing the temporary file. (The second line
 appears only for a floppy drive.)

 After this message appears, do not remove the disk from the drive specified
 by letter until the link session ends. If the disk is removed, the operation
 of LINK is unpredictable, and you might see the following message:

   Unexpected end-of-file on scratch file

 If this happens, run LINK again.


 Location of the Temporary File

 If the TMP environment variable defines a temporary directory, LINK creates
 temporary files there. If the TMP environment variable is undefined or the
 temporary directory doesn't exist, LINK creates temporary files in the
 current directory.


 Name of the Temporary File

 When running under OS/2 or DOS version 3.0 or later, LINK asks the operating
 system to create a temporary file with a unique name in the temporary-file
 directory.

 Under DOS versions earlier than 3.0, LINK creates a temporary file named
 VM.TMP. Do not use this name for your files. LINK generates an error message
 if it encounters an existing file with this name.


 12.10  LINK Exit Codes

 LINK returns an exit code (also called return code or error code) that you
 can use to control the operation of batch files or makefiles.

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Code                              Meaning
 ────────────────────────────────────────────────────────────────────────────
 0                                 No error.

 2                                 Program error. Commands or files given
                                   as input to the linker produced the
                                   error.

 4                                 System error. The linker

                                   Ran out of space on output files

                                   Was unable to reopen the temporary file

 Code                              Meaning
 ────────────────────────────────────────────────────────────────────────────

                                   Experienced an internal error

                                   Was interrupted by the user




 12.11  Related Topics in Online Help

 In addition to information covered in this chapter, information on the
 following topics can be found in online help.

 Topic                                 Access
 ────────────────────────────────────────────────────────────────────────────
 Syntax and procedural information on  Choose these topics from the
 LINK, BIND, and LIB                   "Microsoft Advisor Contents" screen

 Syntax and procedural information on  Choose "Miscellaneous" from the list
 EXEHDR                                of utilities on the "Microsoft
                                       Advisor Contents" screen







 Chapter 13  Module-Definition Files
 ────────────────────────────────────────────────────────────────────────────

 This chapter describes the contents of a module-definition file. It begins
 with a brief overview of the purpose of module-definition files. The rest of
 the chapter discusses each statement in a module-definition file and
 describes syntax rules, argument fields, attributes, and keywords for each
 statement.


 13.1  Overview

 A module-definition file is a text file that describes the name, attributes,
 exports, imports, system requirements, and other characteristics of an
 application or dynamic-link library (DLL) for OS/2 or Microsoft Windows.
 This file is required for DLLs and is optional (but desirable) for OS/2 and
 Windows applications.

 You use module-definition files in two situations:


   ■   You can specify a module-definition file in LINK's deffile field. The
       module-definition file gives LINK the information it needs to
       determine how to set up the application or DLL it creates.

   ■   You can provide LINK with the needed information when creating an
       application by using the Microsoft Import Library Manager utility
       (IMPLIB) to create an import library from a module-definition file (or
       from the DLL created by a module-definition file). You then specify
       the import library in LINK's libraries field.


 For more information about IMPLIB, see online help.


 13.2  Module Statements

 A module-definition file contains one or more "module statements." Each
 module statement defines an attribute of the executable file, such as its
 name, the attributes of program segments, and the number and names of
 exported and imported functions and data. Table 13.1 summarizes the purpose
 of the module statements and shows the order in which they are discussed in
 this chapter.

 Table 13.1  Module Statements

╓┌─────────────┌─────────────────────────────────────────────────────────────╖
 Statement     Purpose
 ────────────────────────────────────────────────────────────────────────────
 NAME          Names the application (no library created)
 LIBRARY       Names the DLL (no application created)
 DESCRIPTION   Embeds text in the application or DLL
 STUB          Adds a DOS executable file to the beginning of the file
 EXETYPE       Identifies the target operating system
 Statement     Purpose
 ────────────────────────────────────────────────────────────────────────────
EXETYPE       Identifies the target operating system
 PROTMODE      Specifies a protected-mode application or DLL
 REALMODE      Supported for compatibility
 STACKSIZE     Sets stack size in bytes
 HEAPSIZE      Sets local heap size in bytes
 CODE          Sets default attributes for all code segments
 DATA          Sets default attributes for all data segments
 SEGMENTS      Sets attributes for specific segments
 OLD           Preserves ordinals from a previous DLL
 EXPORTS       Defines exported functions
 IMPORTS       Defines imported functions
 ────────────────────────────────────────────────────────────────────────────



 13.2.1  Syntax Rules

 The syntax rules in this section apply to all statements in a
 module-definition file. Other rules specific to each statement are described
 in the sections that follow.


   ■   Statement and attribute keywords are not case sensitive. A statement
       keyword can be preceded by spaces and tabs.

   ■   A NAME or LIBRARY statement, if used, must precede all other
       statements.

   ■   Most statements appear at most once in a file and accept one
       specification of parameters and attributes. The specification follows
       the statement keyword on the same or subsequent line(s). If repeated
       with a different specification later in the file, the later statement
       overrides the earlier one.

   ■   The SEGMENTS, EXPORTS, and IMPORTS statements can appear more than
       once in the file and take multiple specifications, each on its own
       line. The statement keyword must appear once before the first
       specification and can be repeated before each additional
       specification.

   ■   Comments in the file are designated by a semicolon (;) at the
       beginning of each comment line. A comment cannot share a line with
       part or all of a statement but can appear between lines of a multiline
       statement.

   ■   Numeric arguments can be specified in decimal or in C-language
       notation.

   ■   Name arguments cannot match a reserved word.



 Example

 The sample module-definition file below gives a description for a DLL. This
 sample file includes one comment and five statements.

   ; Sample module-definition file

   LIBRARY

   DESCRIPTION 'Sample dynamic-link library'

   CODE       PRELOAD

   STACKSIZE  1024

   EXPORTS
       Init   @1
       Begin  @2
       Finish @3
       Load   @4
       Print  @5


 13.2.2  Reserved Words

 The following words are reserved by the linker for use in module-definition
 files. These names cannot be used as arguments in module-definition
 statements.

 (This figure may be found in the printed book.)

 * DOS4 and HUGE are obsolete but are still reserved by the linker.

 In addition to the words listed above, the following words are reserved for
 use by future or other versions of the linker and should be avoided.

 (This figure may be found in the printed book.)


 13.3  The NAME Statement

 The NAME statement identifies the executable file as an application (rather
 than a DLL). It can also specify the name and application type. The NAME or
 LIBRARY statement must precede all other statements. If NAME is specified,
 the LIBRARY statement cannot be used. If neither is used, the default is
 NAME and LINK creates an application.


 Syntax

   NAME «appname» «apptype» «NEWFILES»


 Remarks

 The fields can appear in any order.

 If appname is specified, it becomes the name of the application as it is
 known by OS/2 or Windows. This name can be any valid filename. If appname
 contains a space, begins with a nonalphabetic character, or is a reserved
 word, surround appname with double quotation marks. The name cannot exceed
 255 characters (not including surrounding quotation marks). If appname is
 not specified, the base name of the executable file becomes the name of the
 application.

 If apptype is specified, it defines the type of application. This
 information is kept in the executable-file header. The apptype field can
 take one of the following values:

 Value                             Description
 ────────────────────────────────────────────────────────────────────────────
 WINDOWAPI                         Presentation Manager (PM) or Windows
                                   application. The application uses the
                                   API provided by PM or Windows and must
                                   be executed in the PM or Windows
                                   environment. This is equivalent to the
                                   LINK option /PM:PM.

 WINDOWCOMPAT                      Character-mode application to run in a
                                   text window in the PM or Windows session.
                                   This is equivalent to the LINK option
                                   /PM:VIO.

 NOTWINDOWCOMPAT                   The default. Character-mode application
                                   that must run full screen and cannot run
                                   in a text window in PM or Windows. This
                                   is equivalent to the LINK option
                                   /PM:NOVIO.


 Specify NEWFILES to tell the operating system that the application supports
 long filenames and extended file attributes (available under OS/2 version
 1.2 and later). The synonym LONGNAMES is supported for compatibility.


 Example

 The example below assigns the name  calendar  to an application that can run
 in a text window in PM or Windows:

   NAME calendar WINDOWCOMPAT


 13.4  The LIBRARY Statement

 The LIBRARY statement identifies the executable file as a DLL. It can also
 specify the name of the library and the type of library-module
 initialization required. The NAME or LIBRARY statement must precede all
 other statements. If LIBRARY is specified, the NAME statement cannot be
 used. If neither is used, the default is NAME.


 Syntax

   LIBRARY «libraryname» «initialization»
   «PRIVATELIB»


 Remarks

 The fields can appear in any order.

 If libraryname is specified, it becomes the name of the library as it is
 known by OS/2 or Windows. This name can be any valid filename. If
 libraryname contains a space, begins with a nonalphabetic character, or is a
 reserved word, surround the name with double quotation marks. The name
 cannot exceed 255 characters. If libraryname is not given, the base name of
 the DLL file becomes the name of the library.

 If initialization is specified, it determines the type of initialization
 required. The initialization field can take one of the following values:

 Value                             Description
 ────────────────────────────────────────────────────────────────────────────
 INITGLOBAL                        The default. The library-initialization
                                   routine is called only when the library
                                   is initially loaded into memory.

 INITINSTANCE                      The library-initialization routine is
                                   called each time a new process gains
                                   access to the DLL. This keyword applies
                                   only to OS/2.


 If PRIVATELIB is specified, it tells Windows that only one application may
 use the DLL.


 Example

 The following example assigns the name  calendar  to the DLL being defined
 and specifies that library initialization is performed each time a new
 process gains access to  calendar:

   LIBRARY calendar INITINSTANCE


 13.5  The DESCRIPTION Statement

 The DESCRIPTION statement inserts specified text into the application or
 DLL. This statement is useful for embedding source-control or copyright
 information into a file.


 Syntax

   DESCRIPTION 'text'


 Remarks

 The text is a string of up to 255 characters enclosed in single or double
 quotation marks (' or "). To include a literal quotation mark in the text,
 either specify two consecutive quotation marks of the same type or enclose
 the text with the other type of quotation mark. If a DESCRIPTION statement
 is not specified, the default text is the name of the main output file as
 specified in LINK's exefile field. You can view this string by using the
 Microsoft EXE File Header Utility (EXEHDR).

 The DESCRIPTION statement is different from a comment. A comment is a line
 that begins with a semicolon (;). Comments are not placed in the application
 or library.


 Example

 The following example inserts the text  Tester's Version, Test "A",
 including a literal single quotation mark and a pair of literal double
 quotation marks, into the application or DLL being defined:

   DESCRIPTION "Tester's Version, Test ""A"""


 13.6  The STUB Statement

 The STUB statement adds a DOS executable file to the beginning of an OS/2 or
 Windows application or DLL. The stub is invoked whenever the file is
 executed under DOS. Usually, the stub displays a message and terminates
 execution. By default, LINK adds a standard stub for this purpose. Use the
 STUB statement when creating a dual-mode program.


 Syntax

   STUB {'filename' | NONE}


 Remarks

 The filename specifies the DOS executable file to be added. LINK searches
 for filename first in the current directory and then in directories
 specified with the PATH environment variable. The filename must be
 surrounded by single or double quotation marks (' or ").

 The alternate specification NONE prevents LINK from adding a default stub.
 This saves space in the application or DLL, but the resulting file will hang
 the system if loaded in DOS.


 Example

 The following example inserts the DOS executable file STOPIT.EXE at the
 beginning of the application or DLL:

   STUB 'STOPIT.EXE'

 The file STOPIT.EXE is executed when you attempt to run the application or
 DLL under DOS.


 13.7  The EXETYPE Statement

 The EXETYPE statement specifies under which operating system the application
 or DLL is to run. This statement is optional and provides an additional
 degree of protection against the program being run under an incorrect
 operating system.


 Syntax

   EXETYPE «OS2 | WINDOWS« version» |
   UNKNOWN»


 Remarks

 The EXETYPE keyword is followed by a descriptor of the operating system,
 either OS2 (for OS/2 applications and DLLs), WINDOWS (for WINDOWS
 applications and DLLs), or UNKNOWN (for other applications). The default
 without a descriptor or an EXETYPE statement is OS2.

 EXETYPE sets bits in the header which identify the operating system.
 Operating-system loaders can check these bits.


 Windows Programming

 The WINDOWS descriptor takes an optional version number. Windows reads this
 number to determine the minimum version of Windows needed to load the
 application or DLL. For example, if 3.0 is specified, the resulting
 application or DLL

 can run under Windows versions 3.0 and higher. If version is not specified,
 the default is 3.0. The syntax for version is

 number«.«number» »

 where each number is a decimal integer.

 In Windows programming, use the EXETYPE statement with a PROTMODE statement
 to specify an application or DLL that runs only under protected-mode
 Windows.


 13.8  The PROTMODE Statement

 The PROTMODE statement specifies that the application or DLL runs only under
 OS/2 or under Windows 3.0 standard mode and 386 enhanced mode. PROTMODE lets
 LINK optimize to reduce both the size of the file on disk and its loading
 time. However, an OS/2 program created with PROTMODE cannot be bound using
 BIND. Use PROTMODE in combination with an EXETYPE WINDOWS statement to
 define an application or DLL that runs only under protected-mode Windows.


 Syntax

   PROTMODE


 Example

 The following statement combination defines an application that runs only
 under protected-mode (standard or 386 enhanced) Windows version 3.0:

   EXETYPE WINDOWS 3.0
   PROTMODE





 13.9  The REALMODE Statement

 The REALMODE statement specifies that the application runs only in real
 mode. This statement is supported for compatibility with existing
 module-definition files. Use EXETYPE instead.


 Syntax

   REALMODE


 13.10  The STACKSIZE Statement

 The STACKSIZE statement specifies the size of the stack in bytes. It
 performs the same function as LINK's /STACK option. If both are specified,
 the STACKSIZE statement overrides the /STACK option.


 Syntax

   STACKSIZE number


 Remarks

 The number must be a positive integer, in decimal or C-language notation, up
 to 64K.


 Example

 The following example allocates 4,096 bytes of stack space:

   STACKSIZE 4096


 13.11  The HEAPSIZE Statement

 The HEAPSIZE statement defines the size of the application or DLL's local
 heap in bytes. This value affects the size of the default data segment
 (DGROUP). The default without HEAPSIZE is no local heap.


 Syntax

   HEAPSIZE {bytes | MAXVAL}


 Remarks

 The bytes field accepts a positive integer in decimal or C-language
 notation. The limit is MAXVAL; if bytes exceeds MAXVAL, the excess is not
 allocated.

 MAXVAL is a keyword that sets the heap size to 64K minus the size of DGROUP.
 This is useful in bound applications when you want to force a 64K
 requirement for DGROUP for the program in DOS. The bound program fails to
 load if 64K of memory is not available.


 Example

 The following example sets the local heap to 4,000 bytes:

   HEAPSIZE 4000


 13.12  The CODE Statement

 The CODE statement defines the default attributes for all code segments
 within the application or DLL. The SEGMENTS statement can override this
 default for one or more specific segments.


 Syntax

   CODE «attribute...»


 Remarks

 This statement accepts several optional attribute fields: conforming,
 discard, executeonly, iopl, load, movable, and shared. Each can appear once,
 in any order. These fields are described in Section 13.15, "CODE, DATA, and
 SEGMENTS Attributes."


 Example

 The following example sets defaults for the program's code segments. No code
 segments in the program are loaded until accessed, and all require I/O
 hardware privilege.

   CODE LOADONCALL IOPL


 13.13  The DATA Statement

 The DATA statement defines the default attributes for all data segments
 within the application or DLL. The SEGMENTS statement can override this
 default for one or more specific segments.


 Syntax

   DATA «attribute...»


 Remarks

 This statement accepts several optional attribute fields: instance, iopl,
 load, movable, readonly, and shared. Each can appear once, in any order.
 These fields are described in Section 13.15, "CODE, DATA, and SEGMENTS
 Attributes."


 Example

 The example below defines the application's data segment so that it cannot
 be shared by multiple copies of the program and cannot be written to. By
 default, the data segment can be read and written to and a new DGROUP is
 created for each instance of the application.

   DATA NONSHARED READONLY


 13.14  The SEGMENTS Statement

 The SEGMENTS statement defines the attributes of one or more individual
 segments in the application or DLL. The attributes specified for a specific
 segment override the defaults set in the CODE and DATA statements (except as
 noted below). The total number of segment definitions cannot exceed the
 number set using LINK's /SEG option. (The default without /SEG is 128.)

 The SEGMENTS keyword marks the beginning of the segment definitions, where
 each definition is on its own line. The SEGMENTS statement must appear once
 before the first specification (on the same or preceding line) and can be
 repeated before each additional specification. SEGMENTS statements can
 appear more than once in the file.


 Syntax

   SEGMENTS
             «'»segmentname«'» «CLASS 'classname'»
   «attribute...»


 Remarks

 Each segment definition begins with segmentname, optionally enclosed in
 single or double quotation marks (' or "). The quotation marks are required
 if segmentname is a reserved word.

 The CLASS keyword optionally specifies the class of the segment. Single or
 double quotation marks (' or ") are required around classname. If you do not
 use the CLASS argument, the linker assumes that the class is CODE.

 This statement accepts several optional attribute fields: conforming,
 discard, executeonly, iopl, load, movable, readonly, and shared. Each can
 appear once, in any order. These fields are described in the next section,
 "CODE, DATA, and SEGMENTS Attributes."


 Example

 The following example specifies segments named  cseg1,  cseg2, and  dseg.
 The first segment is assigned the class  mycode  and the second is assigned
 CODE by default. Each segment is given different attributes.

   SEGMENTS
       cseg1 CLASS 'mycode' IOPL
       cseg2    EXECUTEONLY PRELOAD CONFORMING
       dseg  CLASS 'data'   LOADONCALL READONLY


 13.15  CODE, DATA, and SEGMENTS Attributes

 The following attribute fields apply to the CODE, DATA, and SEGMENTS
 statements previously described. Refer to "Remarks" in each of the previous
 sections for the attribute fields that are used by each statement. Most
 fields are used by all three statements; others are used as noted. Each
 field can appear once, in any order.

 Listed with each attribute field below are keywords that are legal values
 for the field, along with descriptions of the field and values. The defaults
 are noted. If two segments with different attributes are combined into the
 same group, LINK makes decisions to resolve any conflicts and assumes a set
 of attributes.

╓┌───────────────────┌───────────────────────────────────────────────────────╖
 Attribute           Description
 ────────────────────────────────────────────────────────────────────────────
 conforming          {CONFORMING | NONCONFORMING}

                     For CODE and SEGMENTS statements only. Determines
                     whether a code segment is an 80286 "conforming"
                     segment for device drivers and system-level code. The
                     conforming attribute is for OS/2 only.

                     CONFORMING specifies that the segment executes at the
                     caller's privilege level. When IOPL=YES is specified
                     in CONFIG.SYS, no call gates are generated for calls
                     or jumps.

                     NONCONFORMING (the default) specifies that the segment
                     can be accessed from Ring 2. When IOPL=YES is
                     specified in CONFIG.SYS, call gates are generated.

                     For more information, refer to Intel documentation for
                     the 80286 processor and later.
 Attribute           Description
 ────────────────────────────────────────────────────────────────────────────
                    the 80286 processor and later.

 discard             {DISCARDABLE | NONDISCARDABLE}

                     For CODE and SEGMENTS statements only. Determines
                     whether a code segment can be discarded from memory
                     to fill a different memory request. If the discarded
                     segment is accessed later, it is reloaded from disk.
                     NONDISCARDABLE is the default. The discard attribute
                     is for Windows only.




 executeonly         {EXECUTEONLY | EXECUTEREAD}

                     For CODE and SEGMENTS statements only. Determines
                     whether a code segment can be read as well as executed.

 Attribute           Description
 ────────────────────────────────────────────────────────────────────────────

                     EXECUTEONLY specifies that the segment can only be
                     executed. The keyword EXECUTE-ONLY is an alternate
                     spelling.

                     EXECUTEREAD (the default) specifies that the segment
                     is both executable and readable. This attribute is
                     necessary for a program to run under the Microsoft
                     CodeView debugger.

 instance            {NONE | SINGLE | MULTIPLE}

                     For the DATA statement only. Affects the sharing
                     attributes of the default data segment (DGROUP). This
                     attribute interacts with the shared attribute.

                     NONE tells the loader not to allocate DGROUP. Use NONE
                     when a DLL has no data and uses an application's
                     DGROUP.
 Attribute           Description
 ────────────────────────────────────────────────────────────────────────────
                    DGROUP.

                     SINGLE (the default for DLLs) specifies that one
                     DGROUP is shared by all instances of the DLL or
                     application.

                     MULTIPLE (the default for applications) specifies that
                     DGROUP is copied for each instance of the DLL or
                     application.

 iopl                {IOPL | NOIOPL}

                     Determines whether a segment has I/O privilege. OS/2
                     only.

                     IOPL specifies that a code segment has I/O privilege
                     and that a data segment can be accessed only from an
                     IOPL code segment.

 Attribute           Description
 ────────────────────────────────────────────────────────────────────────────

                     NOIOPL (the default) specifies that there is no I/O
                     privilege for code and no protection for data.

 load                {PRELOAD | LOADONCALL}

                     Determines when a segment is loaded.



 (load,              PRELOAD specifies that the segment is loaded when the
 continued)          program starts.

                     LOADONCALL (the default) specifies that the segment is
                     not loaded until accessed and only if not already
                     loaded.

 movable             {MOVABLE | FIXED}

 Attribute           Description
 ────────────────────────────────────────────────────────────────────────────

                     Determines whether a segment can be moved in memory.
                     Windows only. FIXED is the default. An alternative
                     spelling for MOVABLE is MOVEABLE.

 readonly            {READONLY | READWRITE}

                     For DATA and SEGMENTS statements only. Determines
                     access rights to a data segment.

                     READONLY specifies that the segment can only be read.

                     READWRITE (the default) specifies that the segment is
                     both readable and writeable.

 shared              {SHARED | NONSHARED}

                     For real-mode Windows and for READWRITE data segments
                     under OS/2 only. Determines whether all instances of
 Attribute           Description
 ────────────────────────────────────────────────────────────────────────────
                    under OS/2 only. Determines whether all instances of
                     the program can share EXECUTEREAD and READWRITE
                     segments. (Under OS/2, all code segments and READONLY
                     data segments are shared.)

                     SHARED (the default for DLLs) specifies that one copy
                     of the segment is loaded and shared among all
                     processes accessing the application or DLL. This
                     attribute saves memory and can be used for code that
                     is not self-modifying. An alternate keyword is PURE.

                     NONSHARED (the default for applications) specifies
                     that the segment must be loaded separately for each
                     process. An alternate keyword is IMPURE.

                     This attribute and the instance attribute interact for
                     data segments. The instance attribute has the keywords
                     NONE, SINGLE, and MULTIPLE. If  DATA SINGLE  is
                     specified, LINK assumes SHARED; if  DATA MULTIPLE  is
 Attribute           Description
 ────────────────────────────────────────────────────────────────────────────
                    specified, LINK assumes SHARED; if  DATA MULTIPLE  is
                     specified, LINK assumes NONSHARED. Similarly,  DATA
                     SHARED  forces SINGLE, and  DATA NONSHARED  forces
                     MULTIPLE.




 13.16  The OLD Statement

 The OLD statement directs the linker to search another DLL for export
 ordinals. This statement preserves ordinal values used from older versions
 of a DLL. For more information on ordinals, see the sections below on the
 EXPORTS and IMPORTS statements.

 Exported names in the current DLL that match exported names in the old DLL
 are assigned ordinal values from the earlier DLL unless


   ■   The name in the old module has no ordinal value assigned, or

   ■   An ordinal value is explicitly assigned in the current DLL.


 Only one DLL can be specified; ordinals can be preserved from only one DLL.
 The OLD statement has no effect on applications.


 Syntax

   OLD 'filename'


 Remarks

 The filename specifies the DLL to be searched. It must be enclosed in single
 or double quotation marks (' or ").


 13.17  The EXPORTS Statement

 The EXPORTS statement defines the names and attributes of the functions and
 data made available to other applications and DLLs, and of the functions
 that run with I/O privilege. By default, functions and data are hidden from
 other programs at run time. A definition is required for each function or
 data item being exported.

 The EXPORTS keyword marks the beginning of the export definitions, each on
 its own line. The EXPORTS keyword must appear once before the first
 definition (on the same or preceding line) and can be repeated before each
 additional definition. EXPORTS statements can appear more than once in the
 file.

 Some languages offer a way to export without using an EXPORTS statement. For
 example, in C the _exports keyword makes a function available from a DLL.


 Syntax

   EXPORTS
   entryname«=internalname» «@ord«
   RESIDENTNAME» » «NODATA» «pwords»


 Remarks

 The entryname defines the function or data-item name as it is known to other
 programs. The optional internalname defines the actual name of the exported
 function or data item as it appears within the exporting program; by
 default, this name is the same as entryname.

 The optional ord field defines a function's ordinal position within the
 moduledefinition table as an integer from 1 to 65,535. If ord is specified,
 the function can be called by either entryname or ord. Use of ord is faster
 and can save space.

 The optional keyword RESIDENTNAME specifies that entryname be kept resident
 in memory at all times. This keyword is applicable only if ord is used. (If
 ord is not used, the name entryname is always kept in memory.)

 The optional keyword NODATA specifies that there is no static data in the
 function.

 The pwords field specifies the total size of the function's parameters in
 words. This field is required only if the function executes with I/O
 privilege. When a function with I/O privilege is called, OS/2 consults
 pwords to determine how many words to copy from the caller's stack to the
 I/O-privileged function's stack.


 Example

 The following EXPORTS statement defines the three exported functions
 SampleRead,  StringIn, and  CharTest. The first two functions can be called
 either by their exported names or by an ordinal number. In the application
 or DLL where they are defined, these functions are named  read2bin  and
 str1, respectively. The first and last functions run with I/O privilege and
 therefore are given with the total size of the parameters.

   EXPORTS
          SampleRead = read2bin @8              24
          StringIn   = str1     @4 RESIDENTNAME
          CharTest                              6


 13.18  The IMPORTS Statement

 The IMPORTS statement defines the names and locations of functions and data
 items to be imported (usually from a DLL) for use in the application or DLL.
 A definition is required for each function or data item being imported. This
 statement is an alternative to resolving references through an import
 library created by the IMPLIB utility; functions and data items listed in an
 import library do not require an IMPORTS definition.

 The IMPORTS keyword marks the beginning of the import definitions, each on
 its own line. The IMPORTS keyword must appear once before the first
 definition on the same or preceding line and can be repeated before each
 additional definition. IMPORTS statements can appear more than once in the
 file.


 Syntax

   IMPORTS
             «internalname=»modulename.entry


 Remarks

 The internalname specifies the function or data-item name as it is used in
 the importing application or DLL. Thus, internalname appears in the source
 code of the importing program, while the function may have a different name
 in the program where it is defined. By default, internalname is the same as
 the entry name. An internalname is required if entry is an ordinal value.

 The modulename is the filename of the exporting application or DLL that
 contains the function or data item.

 The entry field specifies the name or ordinal value of the function or data
 item as defined in the modulename application or DLL. If entry is an ordinal
 value, internalname must be specified. (Ordinal values are set in an EXPORTS
 statement.)

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 A given symbol (function or data item) has a name for each of three
 different contexts. The symbol has a name used by the exporting program
 (application or DLL) where it is defined, a name used as an entry point
 between programs, and a name used by the importing program where the symbol
 is used. If neither program uses the optional internalname field, the symbol
 has the same name in all three contexts. If either of the programs uses the
 internalname field, the symbol may have more than one distinct name.
 ────────────────────────────────────────────────────────────────────────────


 Example

 The following IMPORTS statement defines three functions to be imported:
 SampleRead,  SampleWrite, and a function that has been assigned an ordinal
 value of 1. The functions are found in the  Sample,  SampleA, and  Read
 applications or DLLs, respectively. The function from  Read  is referred to
 as  ReadChar  in the importing application or DLL. The original name of the
 function, as it is defined in  Read, may or may not be known and is not
 included in the IMPORTS statement.

   IMPORTS
                  Sample.SampleRead
                  SampleA.SampleWrite
       ReadChar = Read.1


 13.19  Related Topics in Online Help

 In addition to information covered in this chapter, information on the
 following topics can be found in online help.

 Topic                                 Access
 ────────────────────────────────────────────────────────────────────────────
 Syntax and procedural information on  Choose "LIB" from the list of
 LIB                                   utilities on the "Microsoft Advisor
                                       Contents" screen

 Module-definition files and IMPLIB    Choose "LINK" from the list of
                                       utilities on the "Microsoft Advisor
                                       Contents" screen









 Chapter 14  Customizing the Microsoft Programmer's WorkBench
 ────────────────────────────────────────────────────────────────────────────

 The Microsoft Programmer's WorkBench (PWB) is not just a text editor, but
 also a full-featured platform for program development. It is both flexible
 (you can customize it to match your working habits) and extensible (you can
 add your own functions and features).

 This chapter explains three ways to customize the Programmer's WorkBench:


   ■   Setting switches

   ■   Assigning keystrokes

   ■   Writing macros


 While this chapter explains customizing techniques, it does not document
 every customizable feature. Please consult online help for detailed
 information about these and other PWB features.

 This chapter assumes you are familiar with basic PWB operation and
 terminology. If not, please read "Using the Programmer's WorkBench" in
 Installing and Using the Microsoft Macro Assembler Professional Development
 System. The Programmer's WorkBench is supplied with both the Macro Assembler
 and Microsoft C so that you can customize one copy of PWB to work with these
 and other languages.


 14.1  Setting Switches

 The Programmer's WorkBench has a number of "switches," or user-configurable
 options, that control features such as how many lines the screen scrolls or
 whether you are prompted to save a file when you exit. Each switch has a
 name and can be assigned a value.

 There are two ways to set PWB switches. The easiest way is to choose Editor
 Settings from the Options menu. Saving the changes made to Editor Settings
 updates your TOOLS.INI initialization file. You can also directly edit
 TOOLS.INI. Either method can be used for more elaborate customizations, such
 as writing macros.


 14.1.1  Changing Current Assignments and Switch Settings

 You can change the current editor switches and key assignments. Choose
 Editor Settings or Key Assignments from the Options menu. PWB displays these
 settings in a new window labeled Current Assignments and Switch Settings.

 The <ASSIGN> pseudofile is associated with the Current Assignments and
 Switch Settings window. A pseudofile exists only in memory; it has no
 counterpart on disk until you explicitly save it. Saving the <ASSIGN>
 pseudofile automatically saves any changes you make in the Current
 Assignments and Switch Settings window.

 To change a switch, edit the line on which it appears. For instance, the
 vscroll switch controls how many lines PWB scrolls vertically; its default
 setting is 1. To change it, move to the corresponding line:

   vscroll:1

 Change the 1 to 3 and move the cursor to another line. PWB highlights the
 line to indicate that the change has been executed. (If you make an illegal
 change, PWB signals an error.) The change takes effect immediately: PWB now
 scrolls text three lines at a time.

 When you save <ASSIGN>, PWB updates your TOOLS.INI file.

 PWB discards all changes at the end of a session unless you explicitly save
 them. You save changes by saving <ASSIGN> as you would any other file.
 Select Save from the File menu, or press SHIFT+F2.

 You can also use this method for more elaborate customizations, such as
 writing macros (see Section 14.3, "Writing Macros"). Simply insert a few
 blank lines in the Current Assignments and Switch Settings window and enter
 the new information in them.

 If you add or modify a line of the Current Assignments and Switch Settings
 window, PWB immediately alters its behavior accordingly; the new or changed
 lines are saved in TOOLS.INI when you save the <ASSIGN> file. However,
 deleting a line has no effect, either on PWB's behavior or the contents of
 TOOLS.INI; you must edit TOOLS.INI to remove an assignment.


 14.1.2  Editing the TOOLS.INI Initialization File

 Another way to customize PWB is by editing TOOLS.INI, the initialization
 file used by PWB and other Microsoft language utilities. This is the most
 convenient way to perform extensive customizing.

 While the Current Assignments and Switch Settings window displays every
 customizable PWB item, the TOOLS.INI file contains lines only for items you
 have customized. PWB sets any items you omit from TOOLS.INI to a default
 value.

 TOOLS.INI is made up of sections that start with tags.

 Since TOOLS.INI can initialize a number of Microsoft tools, the file is
 divided into sections, one for each tool. Each section begins with a tag
 consisting of the tool's base name enclosed in square brackets:  [PWB]  for
 PWB.EXE,  [NMAKE]  for NMAKE.EXE, and so on.

 For example, assume you set the vscroll switch to 3 and saved the change,
 but you have not customized PWB in any other way. Your TOOLS.INI file will
 contain this section:

   [PWB]
   vscroll:3

 PWB reads TOOLS.INI at start-up and loads the settings from the  [PWB]
 section.

 You can also create sections of TOOLS.INI that configure PWB for specific
 programming languages or operating systems. For instance, your TOOLS.INI
 file could contain a section beginning with the tag

   [PWB-.C]

 for C source files, and

   [PWB-.ASM]

 TOOLS.INI sections contain customization information.

 for assembly-language (.ASM) source files. Each time you load a file with
 the designated extension, PWB reads the appropriate section of TOOLS.INI.
 You can have a different set of macros and other customizations for each
 file type.

 TOOLS.INI can also contain sections specific to an operating system. The
 following tag introduces a section specific to DOS version 3.31, for
 instance:

   [PWB-3.31]

 You can combine tags as needed. For example, the tag

   [PWB-3.0 PWB-10.10R]

 applies to DOS version 3.0 and OS/2 version 1.1 real mode.


 14.2  Assigning Functions to Keystrokes

 You can assign any PWB function to almost any keystroke. Keystroke
 assignments, like switches, are displayed in the Current Assignments and
 Switch Settings window (choose Key Assignments from the Options menu) and
 can be

 changed there. Suppose you want to assign the home cursor function to
 SHIFT+HOME. The default keystroke assignment for home is

   home:Goto

 If you change the assignment to

   home:Shift+Home

 SHIFT+HOME moves the cursor to the home (upper left) window position.

 You can assign the same function to more than one keystroke. For example,
 many keystrokes invoke the select function, which selects a text region. The
 preceding example adds a new keystroke (SHIFT+HOME) for the home function,
 but it does not remove the previous assignment (GOTO, the 5 key on the
 keypad).

 If you aren't sure whether a keystroke is already assigned, select the
 Current Assignments and Switch Settings window and press PGDN until you
 reach the Available Keys table. All unassigned keystrokes are displayed;
 once a keystroke is assigned, it no longer appears in this table.

 There are two limitations on keystroke assignments:


   ■   You should not reassign a keystroke that PWB assigns to a menu. For
       instance, ALT+F displays the File menu; PWB ignores any attempt to
       reassign ALT+F.

   ■   You should not reassign the ALT plus number keys 1- 6 (ALT+1, ALT+2,
       and so on). These keystrokes are reserved for the file history menu
       items.


 PWB uses the most recent duplicate key assignment.

 A keystroke can invoke only one function. If you accidentally assign a
 keystroke to more than one function, PWB uses the most recent assignment.
 For example,

   home:Ctrl+A
   setfile:Ctrl+A

 assigns the CTRL+A keystroke to two different functions, home and setfile.
 The second assignment overrides the first, assigning CTRL+A to setfile.

 You might occasionally want to "unassign," or disable, a keystroke. This is
 done by assigning the unassigned function to the keystroke. For example,

   unassigned:Ctrl+A

 disables CTRL+A. PWB signals an error when you press any unassigned key.

 As the list of assigned keystrokes shows, you can use SHIFT+CTRL as a
 prefix. For PWB to recognize this key combination, SHIFT must come first.
 For example, to use SHIFT+CTRL with M, you must type SHIFT+CTRL+M, not
 CTRL+SHIFT+M.


 14.3  Writing Macros

 If you need a feature or function that is not a part of PWB, the quickest
 way to create it is by writing a macro in the TOOLS.INI file. A macro can do
 something as simple as inserting a line of text, or it can perform complex
 operations by invoking PWB functions and other macros.


 14.3.1  Macro Syntax

 A macro can consist of any combination of PWB functions, literal text, and
 calls to previously defined macros. You can define up to 1,024 macros at one
 time.

 Anything inside quotation marks is literal text. Within literal text,
 quotation marks are represented by a backslash followed by quotation marks
 (\ ") and a backslash is represented by two consecutive backslashes (\ \).
 Only literal text is case sensitive; PWB ignores the case of everything
 else.

 The following macro "comments out" a line of MASM source code:

   comment:=begline "; "
   comment:alt+c

 The first line names the macro (comment); the macro commands follow the
 assignment operator ( := ). The begline editor function moves the cursor to
 the beginning of the current line. The text inside quotation marks (the MASM
 comment delimiter) is then inserted. The second line assigns a keystroke
 (ALT+C) to the macro.

 Macros can extend over one line.

 If a macro definition takes up more space than you have on one line (about
 250 characters in PWB), you can use the backslash ( \ ) to continue the
 definition on the next line. Consider, for instance, the following macro,
 which comments out a line of C source code:

   comment:=begline "/* " endline " */"

 It could be written as

   comment:=begline  \
   "/* " endline  \
   " */"

 Notice the extra space before each backslash. If you want a space between
 the end of one line and the beginning of the next, you must precede the
 backslash with two spaces.

 You can pass arguments to PWB macros.

 You can use the arg function to pass arguments to functions. For example,
 the following macro passes the argument  15  to the plines function (which
 scrolls text down):

   movedown:=arg "15" plines

 Because arg precedes the literal text, the text isn't written to the screen.
 Instead, it is passed as an argument to the next function, plines. The macro
 scrolls the current text down 15 lines.

 Arguments can also use regular-expression syntax (regular expressions are
 documented in online help):

   endword:=arg arg "([ .,;:()[\\»!$)" psearch cancel

 The arg arg sequence directs the psearch function to treat the text argument
 as a regular-expression search pattern. This search pattern tells PWB to
 search for the next space, period, comma, semicolon, colon, parentheses, and
 square brackets. (Note that a backslash must precede any character that has
 a special meaning in regular expressions─in this case, the right bracket.)

 A macro can invoke other macros:

   lcomment:= "/* "
   rcomment:= " */"
   commentout:=begline lcomment endline rcomment
   commentout:ctrl+o

 The  commentout  macro invokes the previously defined macros  lcomment  and
 rcomment.

 In addition to standard PWB functions, PWB macros can invoke user-defined
 macro functions. See Section 9.6, "Returning Values with Macro Functions."


 14.3.2  Macro Responses

 Some PWB functions ask you for confirmation. For example, the meta exit
 (quit without saving) function normally asks if you really want to exit.
 Such questions always take the answer "yes" (Y) or "no" (N).

 When you invoke such a function in a macro, the function assumes an answer
 of yes and does not ask for confirmation. For example, the macro definition


   quit:=meta exit
   quit:alt+x

 The meta prefix modifies the action of a function.

 invokes meta exit when you press ALT+X. Because the meta exit function is
 invoked from a macro, PWB exits without asking for confirmation.

 The following operators allow you to restore normal prompting or change the
 default responses:

 Operator                          Description
 ────────────────────────────────────────────────────────────────────────────
 <                                 Asks for confirmation; if not followed
                                   by another
                                   < operator, prompts for all further
                                   questions

 <y                                Assumes a response of "yes"

 <n                                Assumes a response of "no"


 A response operator applies to the function immediately preceding it. For
 example, you can add the < operator to the  quit  macro definition to
 restore the usual prompt:
 ────────────────────────────────────────────────────────────────────────────
   quit:=meta exit <
   quit:alt+x

 Now the macro prompts for a response before it exits.


 14.3.3  Macro Arguments

 If you enter an argument in PWB and then invoke a macro, the argument is
 passed to the first function in the macro that takes an argument:

   tripleit:=copy paste paste

 The  tripleit  macro invokes the copy and paste editing functions. When you
 highlight a text area and then invoke the macro, your highlighted argument
 is passed to the copy function, which copies the argument to the clipboard.
 The macro then invokes paste twice. The effect is to insert two copies of
 the highlighted text.

 You cannot pass more than one argument from PWB to a macro.

 You cannot pass more than one argument from PWB to a macro, even if the
 macro invokes more than one function that can accept an argument. The
 argument always goes to the first function in the macro that takes an
 argument.

 You can also prompt for input inside a macro and pass the input as an
 argument using the prompt function as shown below:

   newfile:=arg "Next file: " prompt setfile <
   newfile:alt+n

 The  newfile  macro prompts for a file name and then switches to the
 specified file. The sequence  arg "Next file: "  passes the text argument
 Next file:  to prompt, which prints it in the text-argument dialog box and
 waits for the user to respond. The response is passed as a text argument to
 the setfile function, which switches to that file.


 14.3.4  Macro Conditionals

 Macros can take different actions depending on certain conditions. Such
 macros take advantage of the fact that PWB editing functions return values─
 a TRUE (nonzero) value if successful or FALSE (zero) if unsuccessful.

 Macros can use four conditional operators:

 Operator    Description
 ────────────────────────────────────────────────────────────────────────────
 :>label     Defines a label that can be targeted by other operators
 =>label     Jumps to label
 +>label     Jumps to label if the previous function returns TRUE
 ->label     Jumps to label if the previous function returns FALSE

 For example, the  leftmarg  macro moves the cursor to the left margin of
 the editing window:
 ────────────────────────────────────────────────────────────────────────────
   leftmarg:=:>leftmore left +>leftmore

 The macro above invokes the left function repeatedly (jumping to the label
 leftmore) until it returns FALSE, indicating the cursor has reached the left
 margin.

 Macro execution depends on the status of conditionals.

 The label must appear immediately after the conditional operator, with no
 intervening spaces. A conditional operator without a label exits the macro
 immediately if the condition is satisfied. If the condition is not
 satisfied, the macro continues execution. The following example demonstrates
 this:

   turnon:=insertmode +> insertmode

 This macro turns on insert mode regardless of whether insert mode is
 currently on or off. If insert mode is off, the first invocation of
 insertmode toggles the mode on and returns TRUE, causing the +> operator to
 terminate the macro. If insert mode is currently on, the first invocation of
 insertmode turns insert mode off and returns FALSE. The macro then invokes
 insertmode a second time, turning insert mode back on.


 14.3.5  Recording Macros

 You can also create a macro by recording a procedure as you perform it. The
 keystroke sequence is saved and can be replayed, like any other macro. To
 record a macro:


   1.  Choose Set Record from the Edit menu. The Set Macro Record dialog box
       appears.

   2.  Type the name you want the macro to have in the Name text box.

   3.  Tab to the Key Assignment text box and press the key to which you are
       assigning the macro. (For example, press ALT+T to assign the macro to
       ALT+T. The name of the keystroke appears in the text box.) If the
       keystroke (such as ENTER, TAB, or ESC) would normally exit the dialog
       box or move to the next field, type in the keystroke's name.

   4.  Click the OK button.

   5.  Choose Record On from the Edit menu to start the recording.

   6.  Type the text or perform the actions you want to record. (You can
       select text or fields with the mouse as well as the keyboard. Mouse
       selections are automatically converted into equivalent keystrokes.)

   7.  Choose Record On again to end the recording.


 You have now created a named macro available through the assigned keystroke.
 Pressing this key replays the actions you recorded.

 ────────────────────────────────────────────────────────────────────────────
 WARNING

 If you do not select a name for your macro, it is assigned the default name
 recordvalue. Unless you plan to discard the macro when exiting, do not let a
 recorded macro's name default to recordvalue. Any subsequent macro recorded
 with the  recordvalue default name will overwrite the first recordvalue
 macro.
 ────────────────────────────────────────────────────────────────────────────

 A recorded macro is temporary; PWB discards it when you exit. To save a
 recorded macro:


   1.  Choose Edit Macro from the Edit menu. This opens the <RECORD>
       pseudofile and displays the macros you recorded.

   2.  Make any changes required. For example, you might want to change the
       macro's name or modify the keystroke sequence.

   3.  Save the macro using the Save command from the File menu.


 The macros defined in the <RECORD> pseudofile are added to your TOOLS.INI
 file when you save the <RECORD> file. PWB automatically reloads them at the
 next session.

 You can append functions to an existing macro without having to record the
 original steps again:


   1.  Choose Set Record from the Edit menu. The Set Macro Record dialog box
       appears.

   2.  Type the macro's name in the Name text box.

   3.  Tab to the Clear First check box and cancel selection. This causes any
       new actions to be appended to the original macro, rather than
       replacing (clearing) it.

   4.  Click the OK button.

   5.  Choose Record On from the Edit menu to start the recording.

   6.  Perform the actions you want added to the macro.

   7.  Choose Record On again to end the recording.


 Remember to save the modified macro before exiting, or the new version will
 be discarded.

 You can record a series of actions without executing them.

 You can make a "silent" recording, which records a series of actions without
 executing them. This allows you to create a macro without altering or
 damaging the file. Start the recording with a meta record command (press F9,
 SHIFT+CTRL+R). When the macro is complete, terminate recording with record
 (press SHIFT+CTRL+R).

 PWB gives no visual feedback during silent recording. If you need to see the
 macro being created, open the <RECORD> pseudofile in a second window as
 described above. This is an excellent way to get a better understanding of
 macros and editor functions.


 14.3.6  Temporary Macros

 You can use the assign function to create a macro that lasts only until the
 end of the current session. For example, the following steps create the
 comment  macro described above:


   ■   Press ALT+A

   ■   Type  comment:=begline "; "

   ■   Press ALT+=


 This key sequence tells PWB to open dialog boxes where the macro and key
 assignments are to be typed. To assign ALT+C to the macro,


   ■   Press ALT+A

   ■   Type  comment:alt+c

   ■   Press ALT+=


 The macro is available immediately and is discarded when you exit PWB.


 14.4  Related Topics in Online Help

 Information on the following related topics can be found in online help. All
 the topics listed below are found by choosing "Programmer's WorkBench" from
 the "Microsoft Advisor's Help System Contents" screen.

 Topic                             Access
 ────────────────────────────────────────────────────────────────────────────
 Writing macros                    Choose "Writing and Using Macros"

 TOOLS.INI                         Choose "Using TOOLS.INI"

 Regular expressions               Choose "Writing and Using Macros;" then
                                   choose "Regular Expressions" from under
                                   the "Building Macros" subhead

 The prompt and meta functions     Choose "Using PWB Functions," and from
                                   the next screen, choose "Alphabetical
                                   List"

 Assigning keystrokes              Choose "Setting PWB Switches" and then
                                   "Assign Function"






 Chapter 15  Debugging Assembly-Language Programs with CodeView
 ────────────────────────────────────────────────────────────────────────────

 You can diagnose software problems and locate programming errors quickly
 with the CodeView debugger. This chapter explains how to


   ■   Display and modify variables and memory

   ■   Control the flow of execution

   ■   Use advanced CodeView debugging techniques

   ■   Modify CodeView's behavior with command-line switches and the
       TOOLS.INI file


 CodeView supports the Microsoft mouse (or any fully compatible pointing
 device). This chapter first describes CodeView operations with the mouse,
 then with function keys. Command-window commands are not generally
 discussed, except when there is no comparable mouse or function-key command.
 Unless a specific mouse button is named, "clicking" means pressing and
 quickly releasing the left mouse button.


 15.1  Understanding Windows in CodeView

 CodeView divides the screen into logically separate sections called windows.
 Windows permit a large amount of information to be displayed in an organized
 and easy-to-read fashion.

 Each window displays a different type of data.

 Each CodeView window has a distinct function and operates independently of
 the others. The name of each window described below appears in the top of
 the window's frame:


   ■   The Source window displays the source code. You can open a second
       source window to view an include file, another source file, or the
       same source file at a different location. Any ASCII text file can be
       viewed in the Source window.

   ■   The Command window accepts debugging commands from the keyboard.

   ■   The Watch window displays the current values of selected variables.

   ■   The Local window lists the values of all variables local to the
       current procedure.

   ■   The Memory window shows the contents of memory. You can open a second
       Memory window to view a different section of memory.

   ■   The Register window displays the contents of the microprocessor's
       registers, as well as the processor flags.

   ■   The 8087 window displays the registers of the coprocessor or its
       software emulator.


 Figure 15.1 shows all CodeView windows.

 (This figure may be found in the printed book.)

 The first time you run CodeView, it displays three windows. The Local window
 is at the top, the Source window fills the middle of the screen, and the
 Command window is at the bottom. CodeView records which windows were open
 and how they were positioned at the time you exit. These settings become the
 default the next time you run CodeView.

 There are two ways to open windows. You can choose the desired window from
 the View menu or press its shortcut key. In addition, some operations (such
 as selecting a Watch variable) automatically open the appropriate window if
 it isn't already open.

 All displays are updated automatically.

 CodeView continually and automatically updates the contents of all windows.
 However, if you want to interact with a particular window (such as entering
 a command, setting a breakpoint, or modifying a variable), you must first
 select that window.

 The selected window is called the "active" window. The active window is
 marked in three ways:


   ■   The window's name is highlighted.

   ■   The text cursor appears in the window.

   ■   The vertical and horizontal scroll bars move into the window.


 Figure 15.2 shows the Source window as the active window.

 (This figure may be found in the printed book.)

 To select a new active window, click that window (position the mouse pointer
 in the window and press the left mouse button). You can also press F6 or
 SHIFT+F6 to move from one window to the next.

 Windows often contain more information than can be displayed in the area
 allotted to the window. There are several ways to view these additional
 contents.

 To view additional contents with the mouse:


   ■   Drag the scroll box on the horizontal or vertical scroll bars.
       (Position the mouse pointer on the scroll box and, while holding down
       the left mouse button, move the mouse in the appropriate direction.)

   ■   Click the arrows at the top and bottom of the scroll bars.

   ■   Click the gray area to either side of the scroll box in a scroll bar.


 To view additional contents with the keyboard:


   ■   Press the direction keys (LEFT, RIGHT, UP, DOWN) to move the cursor.

   ■   Press PGUP, PGDN, CTRL+PGUP (page left), and CTRL+PGDN (page right) to
       move the cursor to a different page of the window's contents.

   ■   Press CTRL+HOME to move the cursor to the beginning of the window's
       contents.

   ■   Press CTRL+END to move the cursor to the end of the window's contents.


 Typing commands when the Source window is active causes CodeView to
 temporarily shift its focus to the Command window. Whatever you type is
 appended to the last line in the Command window. If the Command window is
 closed, CodeView beeps in response to your entry and ignores the input.


 Adjusting the Windows

 Although you can't reposition the windows, you can change their size or
 close them. The Maximize, Size, and Close commands from the View menu
 perform these functions, or you can press CTRL+F10, CTRL+F8, and CTRL+F4,
 respectively. Window manipulation is especially easy with a mouse:


   ■   To maximize a window (enlarge it so it fills the screen), click the up
       arrow at the right end of the window's top border, or double-click the
       window's title. (Position the mouse pointer anywhere on the title and
       press the left mouse button twice, rapidly.) To restore the window to
       its original size, click the double arrow at the right end of the top
       border or press CTRL+F10.

   ■   To change the size of a window, position the mouse pointer anywhere
       along the line at the top of the window. Press and hold down the left
       mouse button, then drag the mouse to enlarge or reduce the window. The
       same action on a vertical border widens or narrows the window.

   ■   To close a window, click the dot at the left end of the top border.
       The adjacent windows automatically expand to recover the unused space.
       You can also close any window whose View menu name has a dot next to
       it: choose that window from the menu or press the window's shortcut
       key.


 CodeView remembers the last debugging session.

 CodeView stores session information in a file called CURRENT.STS, which is
 created in the directory pointed to by the INIT environment variable (or in
 the current directory, if there is no INIT variable). The session
 information includes such items as the name of the program being debugged,
 the CodeView windows that were open, breakpoint locations, and other status.
 This information becomes the default status the next time you run CodeView.



 15.2  Overview of Debugging Techniques

 There is no single best approach to debugging. CodeView offers a variety of
 debugging tools that let you select a method appropriate for the program or
 for your work habits. This section describes some approaches to solving
 debugging problems.

 Broadly speaking, two things can go wrong in a program:


   ■   The program doesn't manipulate the data the way you expected it to.

   ■   The flow of execution is incorrect.


 These problems usually overlap. Incorrect execution can corrupt the data,
 and bad data can cause execution to take an unexpected turn. Because
 CodeView allows you to trace program execution while simultaneously
 displaying whatever combination of variables you want, you don't have to
 know ahead of time whether the problem is bad data manipulation, a bad
 execution path, or some combination of both.

 CodeView has specific features that deal with the problems of bad data and
 incorrect execution:


   ■   You can view and modify any program variable, any section of memory,
       or any processor register. These features are explained in Section
       15.3, "Viewing and Modifying Program Data."

   ■   You can monitor the path of execution and precisely control where
       execution pauses. These features are explained in Section 15.4,
       "Controlling Execution."



 15.3  Viewing and Modifying Program Data

 CodeView offers a variety of ways to display the values of program
 variables, processor registers, and memory. You can also modify the values
 of all these items as the program executes. This section shows how to
 display and modify variables, registers, and memory.


 15.3.1  Displaying Variables in the Watch Window

 To add a variable to the Watch window, position the cursor on the variable's
 name, using the mouse or the direction keys (LEFT, RIGHT, UP, DOWN). Then
 choose the Add Watch command from the Watch menu, or press CTRL+W.

 A dialog box appears with the selected variable's name displayed in the
 Expression field. If you don't want to watch the variable shown, type in the
 name of another variable. Click the OK button or press ENTER to add this
 variable to the Watch window.

 The Watch window appears at the top of the screen. Selecting a Watch
 variable automatically opens the Watch window if the window isn't already
 open.

 A newly added variable may be followed by the message:

   <Watch Expression Not in Context>

 This message appears when execution has not yet reached the procedure where
 a local variable is defined. Global variables (those declared outside
 procedures) never cause CodeView to display this message; they can be
 watched from anywhere in the program.

 To remove a variable from the Watch window, choose the Delete Watch command
 from the Watch menu or press CTRL+U. Then select the variable to be removed
 from the list in the dialog box. You can also position the cursor on any
 line in the Watch window and press CTRL+Y to delete that line.

 You can watch an unlimited number of variables.

 You can place as many variables as you like in the Watch window; the
 quantity is limited only by available memory. You can scroll the Watch
 window to position it at those variables you want to view. CodeView
 automatically updates all Watch window variables as the program runs,
 including those not currently visible within the Watch window frame.

 A variable can be specified by its address as well as its name. You can give
 its address in segment:offset form, where either component can be a register
 name or a number. You can extract a variable's address by prefixing the &
 operator to its name. Prefixing a variable's address (or any address) with
 the BY, WO, or DW operator displays the byte, word, or doubleword value
 starting at that address.

 There are several ways to display a variable's value.

 By default, CodeView displays variables as decimal values. You can select
 the radix by typing  n8,  n10, or  n16  in the Command window for an octal,
 decimal, or hexadecimal display. CodeView remembers the current radix when
 you exit; it becomes the default radix the next time you run CodeView.


 15.3.2  Displaying Expressions in the Watch Window

 The Watch window is not limited to variables. You can enter an expression
 (that is, any valid combination of variables, constants, and operators) for
 CodeView to evaluate and display. You can also select the format in which
 CodeView displays the expression.

 MASM expressions are evaluated using C rules.

 CodeView does not include an expression evaluator specifically for MASM. It
 uses the C expression evaluator instead. This means you must enter MASM
 variables or expressions in a form the C evaluator recognizes, which is not
 always the way they appear in a MASM program. (Online help describes the
 operators and precedence order for C expressions. The last part of this
 section also gives examples of some of the more commonly used expression
 forms.)

 The Language command from the Options menu offers a choice of Auto, C,
 Basic, or FORTRAN expression evaluators. However, the Basic and FORTRAN
 expression evaluators do not support address evaluation, pointer
 conversions, type casting, or other operations needed when debugging
 assembly-language code.

 Besides arithmetic and memory-reference expressions, CodeView can also
 display Boolean expressions. For example, if a variable is never supposed to
 be larger than 100 or less than 25, the expression

   (var < 25 || var > 100)

 evaluates to one (TRUE) if  var  goes out of bounds.


 Changing Display Format

 By default, CodeView displays expression values in decimal form. You can
 change the display radix to octal or hexadecimal with the Radix (N) command
 described at the end of the previous section.

 Another way to change the display format is to append a comma and a
 single-digit format specifier to any watched variable, expression, or
 address. For example, to display  varname  in octal form, type  varname,o
 in the Watch expression box. (If  varname  is already in the Watch window,
 simply append a comma and the octal specifier  ,o  and then move the cursor
 off the line.) The following list describes the use of each specifier:

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Specifier                         Form Displayed
 ────────────────────────────────────────────────────────────────────────────
 c                                 Least-significant byte of the variable
                                   displayed as a single character

 d                                 Decimal value

 e or E                            Eight bytes displayed as a
                                   double-precision exponential
                                   number
 Specifier                         Form Displayed
 ────────────────────────────────────────────────────────────────────────────
                                  number

 f                                 Four bytes displayed as a
                                   single-precision floating-point
                                   number

 g or G                            Eight bytes displayed as a
                                   double-precision exponential
                                   number

 i                                 Signed integer value

 o                                 Unsigned octal value

 s                                 String; all following bytes displayed as
                                   ASCII characters, up to next null
                                   character (ASCII 0)

 u                                 Unsigned decimal value
 Specifier                         Form Displayed
 ────────────────────────────────────────────────────────────────────────────
u                                 Unsigned decimal value

 x or X                            Hexadecimal value, without leading  0x




 Displaying MASM Expressions

 Expressions using registers or indexes are more complex. The following
 sections show how to substitute CodeView expressions using the C expression
 evaluator for MASM expressions.

 Register Indirection - The C expression evaluator does not recognize
 brackets to indicate the memory location pointed to by a register. Instead,
 use the BY, WO, or DW operator to reference the corresponding byte, word, or
 doubleword value.

 MASM Expression              CodeView Equivalent
 ────────────────────────────────────────────────────────────────────────────
 BYTE PTR [bx]                BY bx
 WORD PTR [bp]                WO bp
 DWORD PTR [bp]               DW bp

 Register Indirection with Displacement - To perform based, indexed, or
 based-indexed indirection with a displacement, use the BY, WO, or DW
 operator combined with addition.

 MASM Expression                 CodeView Equivalent
 ────────────────────────────────────────────────────────────────────────────
 BYTE PTR [di+6]                 BY di+6
 BYTE PTR Test [bx]              BY &Test+bx
 WORD PTR [si] [bp+6]            WO si+bp+6
 DWORD PTR [bx] [si]             DW bx+si

 Address of a Variable - Use the address operator (&) instead of the OFFSET
 operator.

 MASM Expression              CodeView Equivalent
 ────────────────────────────────────────────────────────────────────────────
 OFFSET Var                   &Var

 PTR Operator - Use C type casts, or the BY, WO, and DW operators in
 conjunction with the address operator (&), to replace the PTR operator.

 MASM Expression                   CodeView Equivalents
 ────────────────────────────────────────────────────────────────────────────
 BYTE PTR Var                      BY &Var
                                   *(unsigned char*)&Var

 WORD PTR Var                      WO &Var
                                   *(unsigned *)&Var

 DWORD PTR Var                     DW &Var
                                   *(unsigned long*)&Var


 Strings - Add a comma and the string specifier  ,s  after the variable name.

 MASM Expression              CodeView Equivalent
 ────────────────────────────────────────────────────────────────────────────
 Stringvar                    Stringvar,s

 Because CodeView uses the C expression evaluator and C strings end with an
 ASCII null (zero), CodeView displays all characters up to the next null in
 memory when you request a string display. If you intend to debug a MASM
 program, you should terminate string variables with a null.

 Array and Structure Elements - The C expression evaluator equates an array
 name with the address of its first element. Therefore, you should prefix an
 array name with the address operator (&), then add the desired offset. The
 offset can be added directly, or it can appear within parentheses. It can be
 a number, a register name, or a variable.

 The following examples (using byte, word, and doubleword arrays) show how
 this is done:

 MASM Expression                   CodeView Equivalents
 ────────────────────────────────────────────────────────────────────────────
 String[12]                        BY &String+12
                                   *(&String+12)

 aWords[bx+di]                     WO &aWords+bx+di
                                   *(unsigned*)(&aWords+bx+di)

 aDWords[bx+4]                     DW &aDWords+bx+4
                                   *(unsigned long*)(&aDWords+bx+4)


 Pointers - MASM 6.0 lets you define pointer-type variables. Since these are
 the same as C pointers, the C expression evaluator works as it does with C
 programs.

 You dereference a pointer simply by typing its name in the Watch window. The
 pointer's address is displayed, followed by all the elements of the variable
 to which the pointer refers. Multiple levels of indirection (that is,
 pointers referencing other pointers) can be displayed simultaneously.


 15.3.3  Displaying Local Variables

 When your program is executing within the scope of a procedure, the Local
 window automatically displays the variables local to that procedure (stack
 variables). This includes arguments declared in PROC directives and
 variables explicitly declared as LOCAL within the procedure.

 Note that variables you create on the stack are not displayed in the Local
 window, since CodeView is aware only of the assembler-created stack. You can
 display user-defined stack variables in the Watch window by specifying their
 address in segment:offset form.


 15.3.4  Using Pointers to Display Arrays and Strings

 Unlike high-level-language compilers, MASM does not provide symbolic
 information for arrays. Consequently, CodeView cannot distinguish between a
 simple variable and an array, and therefore cannot directly display an
 assemblylanguage array in expanded form. (See Section 15.3.2, "Displaying
 Expressions in the Watch Window," to display individual array elements.)

 A user-defined pointer lets you view an expanded array.

 For debugging purposes, you can overcome MASM's lack of array information by
 using the TYPEDEF directive to define a pointer type, and from that a
 pointer variable for the array. (Place the directive and pointer definition
 within a conditional-assembly block, so the pointer won't be added to your
 release code.) You can then view the array from CodeView by placing the
 pointer in the Watch window. For example:

   array   BYTE    20 DUP (0)      ; array of 20 bytes

   IF debug
      PBYTE   TYPEDEF PTR BYTE     ; PBYTE type is pointer to bytes
      parray PBYTE array           ; parray points to array
   ENDIF

 If you declare multiple levels of pointers (pointers to pointers to
 pointers, and so on), multiple levels of indirection can be displayed
 simultaneously by expanding each subpointer.

 If it is inconvenient to view a character array in hexadecimal form, cast
 the variable's name to a character pointer by placing (char *) in front of
 the name. The character array is then displayed as a string delimited by
 apostrophes. You can also append the string-format specifier  ,s  to the
 expression.

 Note that the C expression evaluator expects a string to terminate with the
 ASCII null character (0). If you do not include a terminating null in the
 string's definition, the evaluator continues displaying memory as characters
 until it encounters a null. The Memory window is an effective way to view
 nonterminated strings.


 15.3.5  Displaying Structures

 MASM adds structure and union information to the debugging table. You can
 display MASM structures in expanded form, just as you would in C, Basic,
 Pascal, or FORTRAN.

 Structures contain multiple data values, often of different data types,
 arranged in one or more layers. Therefore, they are often referred to as
 "aggregate" data items. CodeView lets you control how much of a structure is
 shown; that is, whether all, part, or none of its components are displayed.


 The following example defines a structure and pointer types to implement a
 simple linked list:

   PTRLINKEDLIST TYPEDEF PTR LINKEDLIST
   PTRDATAWORD   TYPEDEF PTR WORD

   LINKEDLIST STRUCT
      ptrNext PTRLINKEDLIST 0
      ptrData PTRDATAWORD   0
   LINKEDLIST ENDS

   rootNode linkedList < >

 Once  rootNode  has been defined, the program calls the MALLOC function
 (which is available from the libraries of Microsoft high-level languages) to
 allocate memory for a structure pointer and a data pointer. The addresses of
 each are assigned to the corresponding pointers in  rootNode, readying the
 list for its first entry.

 The program stores a list item at the memory location specified by the
 preceding pointer, then calls MALLOC to allocate memory for the next list
 item. This process is repeated for each new list item, creating a linked
 list of data structures.

 To display the linked list of structures, add  rootNode  to the Watch
 window. It initially appears in the form:

   +rootnode = {...}

 The brackets indicate that this is an aggregate variable (since it's a
 structure). The plus sign (+) indicates that the structure has not yet been
 expanded to display its components.

 To expand  rootnode, double-click its display line. (Position the mouse
 pointer anywhere on the line and press the left mouse button twice,
 rapidly.) You can also move the cursor to the line and press ENTER. The
 Watch window display changes to

   -rootnode
      +ptrnext  = 0F00:1111
      ptrdata  = 0x0032  "2"

 The address and data values shown here are arbitrary. They depend on the
 data values stored and on the memory location from where MALLOC obtained
 free space. The minus sign (-) indicates that  rootnode  has been fully
 expanded; no further expansion is possible. The plus sign (+) indicates that
  ptrnext  points to another structure that has not been expanded.

 Any structure element can be independently expanded or contracted. To expand
 the next structure, double-click  ptrnext, or press ENTER when the cursor is
 on that line. The Watch window display changes to

   -rootnode
      -ptrnext  = 0F00:1111
         +ptrnext  = 0F00:2222
         ptrdata  = 0x0034  "4"
      ptrdata  = 0x0032  "2"

 Note that both the data value and its ASCII equivalent are displayed. To
 contract the structure, double-click its line a second time or position the
 cursor on the line and press ENTER.

 The process of expanding structures pointed to by  ptrnext  may be repeated
 indefinitely until you reach the last structure in the list. Its identifier
 will be prefixed with a minus sign, indicating that no more space for
 structures has been allocated.

 You can view individual elements instead of the entire structure.

 If you want to view only one or two elements of a large structure, indicate
 the specific structure elements in the Expression field of the Add Watch
 dialog box. Structure elements are separated by a dot (.), so you would type


   rootnode.ptrnext.ptrnext

 to view the pointer from the third structure in the list.


 15.3.6  Using Quick Watch

 Choose the Quick Watch command from the Watch menu (or press SHIFT+F9) to
 display the Quick Watch dialog box. If the cursor is in the Source, Local,
 or Watch window, the variable at the current cursor position appears in the
 dialog box. If it isn't the item you want to display, type in the desired
 expression or variable; then press ENTER. The Quick Watch window immediately
 displays the specified item.

 The Quick Watch display automatically expands structures and pointers to
 their first level. You can expand or contract an element just as you would
 in the Watch window: position the cursor on the appropriate line and press
 ENTER. If the array needs more lines than the Quick Watch window can
 display, drag the scroll box with the mouse, or press DOWN or PGDN to view
 the rest of the array.

 You can add Quick Watch variables to the Watch window.

 Choose the Add Watch button to add a Quick Watch item to the Watch window.
 Structures and pointers appear in the Watch window expanded as they were
 displayed in the Quick Watch dialog box.

 Quick Watch is a convenient way to take a quick look at a variable or
 expression. Since only one Quick Watch variable can be viewed at a time, you
 would not use Quick Watch for most of the variables you want to view.


 15.3.7  Displaying Memory

 Choosing the Memory command from the View menu opens a Memory window. Two
 Memory windows can be open at one time.

 By default, memory is displayed as hexadecimal byte values, with 16 bytes
 per line. At the end of each line is a second display of the same memory in
 ASCII form. Values that correspond to printable ASCII characters (decimal 32
 to 127) are displayed in that form. Values outside this range are shown as
 dots (.).

 You can display memory values in any form.

 Byte values are not always the most convenient way to view memory. If the
 area of memory you're examining contains character strings or floating-point
 values, you might prefer to view them in a directly readable form. Choosing
 the Memory Window command from the Options menu displays a dialog box with a
 variety of display options:


   ■   ASCII characters

   ■   Byte, word, or doubleword binary values

   ■   Signed or unsigned integer decimal values

   ■   Short (32-bit), long (64-bit), or ten-byte (80-bit) floating-point
       values


 Figures 15.3 and 15.4 show two of these different displays.

 (This figure may be found in the printed book.)

 (This figure may be found in the printed book.)

 Another way to choose a display format is to cycle through the formats by
 repeatedly pressing SHIFT+F3.

 Not every four-byte or eight-byte sequence represents a valid floating-point
 number. If a section of memory cannot be displayed in the floating-point
 format you select, the number displayed includes the characters NAN─"not a
 number."

 You can change the contents of the memory by simply overtyping new values in
 the Memory window. See Section 15.3.9 for more information on modifying
 values.


 Displaying Variables with a Live Expression

 Section 15.3.4 explained how to display a specific array element by adding
 the appropriate expression to the Watch window. You can also watch a
 particular array element or structure element in the Memory window. This
 CodeView display feature is called a "live expression." The term "live"
 means that CodeView dynamically displays memory starting at the current
 value of the address expression you specify.

 To create a live expression, choose the Memory Window command from the
 Options menu; then select the Live Expression check box. Type the element
 you want to view in the Address Expression field. For example, if  array  is
 a variable whose current value is being indexed by the value in the BI
 register and you wish to view it, type  array [bi]. Then choose the OK
 button or press ENTER.

 If no memory windows are open, a new Memory window opens. The first memory
 location in the window is the first memory location of the live expression.
 The section of memory displayed changes to the section the live expression
 currently references.

 You can use the Memory Window command from the Options menu to display the
 memory in a directly readable form. This is especially convenient when the
 live expression represents strings or floating-point values, which are
 difficult to interpret in hexadecimal form.

 It is usually more convenient to view an item in the Watch window than as a
 live expression. However, some items are more easily viewed as live
 expressions. For example, you can examine what is currently on top of the
 stack by entering SS:SP as the live expression. In fact, any legal
 combination of register values (such as ES:DI or DS:SI) can be entered in
 segment:offset form.


 15.3.8  Displaying the Processor Registers

 Choosing the Register command from the View menu (or pressing F2) opens a
 window on the right side of the screen. The microprocessor's current
 register values appear in this window. At the bottom of the window is a
 group of mnemonics representing the processor flags. Pressing F2 a second
 time closes the window.

 Video intensity shows changed values.

 When you first open the Register window, all register and flag values are
 shown in normal text. When you change a register or flag, the changed value
 is highlighted. For example, suppose the overflow flag is not set when the
 Register window is first opened. The corresponding mnemonic is NV and
 appears in light gray. If the overflow flag is subsequently set, the
 mnemonic changes to OV and appears in bright white. If your computer uses an
 80386/486 processor and you are running the real-mode version of CodeView
 choosing the 386 Instructions command from the Options menu displays the
 registers as 32-bit values. Choosing this command a second time returns to
 the 16-bit display.

 You can also display the registers of an 8087-80387 coprocessor (or the
 built-in coprocessor of the 80486) in a separate window by choosing the 8087
 command from the View menu. If your program uses the coprocessor emulator,
 the emulated registers are displayed instead.

 The Register values reveal program status.

 The Register window is a valuable debugging tool. Almost every assembly
 instruction alters a register or flag. As each line of code is executed, the
 register values and flags that change are highlighted, so you can see
 whether each instruction does what you intended it to.

 Also, when you execute an instruction whose operand has a memory location
 (such as a variable), the effective address of the operand, as well as the
 value stored at that address, is displayed at the bottom of the Register
 window.


 15.3.9  Modifying the Values of Variables, Memory, and Registers

 You can easily change the values of variables, memory locations, or
 registers displayed in the Watch, Local, Memory, Register, or 8087 windows.
 Simply position the cursor at the value you want to change and edit it to
 the appropriate value. In the Watch and Local windows, the change is
 accepted by CodeView when you move the cursor off the line. If you change
 your mind, press ALT+BKSP to undo the last change you made.

 You can also alter expressions in the Watch window by adding an operator or
 changing the variable displayed. When you have altered the expression and
 moved the cursor off the line, CodeView will immediately show the new value
 of the modified expression.

 The starting address of each line of memory displayed is shown at the left
 of the Memory window in segment:offset form. Altering the address
 automatically shifts the display to the corresponding section of memory.
 Under OS/2, if your program does not own that section of memory, memory
 values are displayed as double question marks (??).

 It's easy to change memory values...

 You can also change the values of memory locations by modifying the right
 side of the memory display (where memory values are shown in ASCII form).
 For example, to change a byte from decimal 75 to decimal 85, place the
 cursor over the letter K, which corresponds to the position where the memory
 value is 75 (K is ASCII 75), and type in U (ASCII 85).

 ...or flags.

 To toggle a processor flag, double-click its mnemonic. You can also position
 the cursor on a mnemonic, then press any key (except ENTER, TAB, or SPACE).
 Press ALT+BKSP (undo) to restore the flag to its previous setting.

 Be cautious when modifying memory or a register.

 The effect of changing a register, flag, or memory location can vary from no
 effect at all to crashing the operating system. Be cautious when altering
 these values.


 15.4  Controlling Execution

 There are two forms of program execution under CodeView:


   ■   Continuous; the program executes until either a previously specified
       breakpoint has been reached or the program terminates.

   ■   Single-step; the program pauses after each line of code has been
       executed.


 Sections 15.4.1 and 15.4.2 explain how each form of execution works and the
 most effective way to use each.

 As you are debugging, you can display the program in source-code form or
 assembly form. Section 15.4.3 explains the advantages of each.


 15.4.1  Continuous Execution

 Continuous execution lets you quickly execute the bug-free sections of code
 which would otherwise take a long time to execute one instruction at a time.


 The simplest form of continuous execution is to click the line of code you
 want to debug or examine in more detail with the right mouse button. The
 program executes up to the start of this line, then pauses. An alternative
 method is to position the cursor on this line, then press F7.

 You can also pause execution at a specific line of code with a "breakpoint."
 There are several types of breakpoints. Breakpoints are explained in the
 following section.


 Selecting Breakpoint Lines

 Breakpoints can be tied to lines of code.

 You can skip over those parts of the program that you don't want to examine
 by specifying one or more lines as breakpoints. The program executes up to
 the first breakpoint, then pauses. Pressing F5 continues program execution
 up to the next breakpoint, and so on. (You can halt execution at any time by
 pressing CTRL+C.)

 There is no limit to the number of breakpoints.

 You can set as many breakpoints as you like (limited only by available
 memory). There are several ways to set breakpoints:


   ■   Double-click anywhere on the desired breakpoint line. The selected
       line is highlighted to show that it is a breakpoint. To remove the
       breakpoint, double-click the line a second time.

   ■   Position the cursor anywhere on the line at which you want execution
       to pause. Press F9 to select the line as a breakpoint and highlight
       it. Press F9 a second time to remove the breakpoint and highlighting.

   ■   Display the Set Breakpoint dialog box by choosing Set Breakpoint from
       the Watch menu. Select one of the breakpoint options that permits a
       line ("location") to be specified. The line at the cursor is the
       default breakpoint line in the Location field. If this line is not the
       desired location, enter the line number desired. (You must place a
       period in front of the line number, or CodeView will interpret the
       number as an absolute address.) To remove the breakpoint, use F9 or
       choose Edit Breakpoints from the Watch menu to display the Edit
       Breakpoints dialog box.


 Not every line can be a breakpoint.

 A breakpoint line must be a program line that represents executable code.
 You cannot select a blank line, a comment, or a declaration (such as a
 variable declaration or a segment specifier) as a breakpoint.

 A breakpoint can also be set at an address. Type the address in
 segment:offset form in the Set Breakpoint dialog box. (Address breakpoints,
 unlike line breakpoints, are not saved in CodeView's status file, and
 therefore are not restored when you restart a debugging session.)

 A breakpoint can be set to the name of a procedure if the procedure was
 declared with the PROC directive. If not, the procedure must contain a
 labeled line. Type the procedure's name or the line's label in the Set
 Breakpoint dialog box.

 Once execution has paused, you can continue execution by clicking the F5=Go
 button in the display or by pressing F5. Execution continues to the next
 breakpoint. If there are no more breakpoints, execution continues to the end
 of the program, or until a fatal error occurs.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 The Set Breakpoint dialog box contains a Commands text box. You can type
 Command-window commands in this box, separated by semicolons. These commands
 are executed when the breakpoint is reached. See the Command Window section
 of CodeView online help for a full description of Command-window commands.
 ────────────────────────────────────────────────────────────────────────────


 Conditional Breakpoints

 Breakpoints are not limited to specific lines of code. CodeView can also
 pause when a variable reaches a particular value or just changes value. This
 is a "conditional breakpoint." In previous versions of CodeView, conditional
 breakpoints are called "watchpoints" and "tracepoints."

 You can associate a conditional breakpoint with a specific line of code, so
 that execution pauses at that line only if the variable has simultaneously
 reached a particular value or changed value. The check boxes in the Set
 Breakpoint dialog box select these other breakpoint types.

 To pause execution when a variable reaches a particular value, type an
 expression that is usually false in the Expression field of the Set
 Breakpoint dialog box. For example, if you want to pause when the variable
 looptest  equals 17, type  looptest == 17.

 To pause execution when a variable changes value, you need to type only the
 name of the variable in the Expression field. For large variables (such as
 arrays or character strings), you can specify the number of bytes you want
 checked (up to 32K) in the Length field. Execution pauses when any one of
 these values changes.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 CodeView checks every conditional breakpoint after executing each line of
 source code. Unless you have enabled the use of the debug registers with the
 CodeView /R command-line option, this computational overhead greatly slows
 execution. (Execution is even slower if you are executing in Mixed mode or
 Assembly mode, because conditional breakpoints are checked after each
 machine instruction.)

 For maximum speed when debugging, either associate conditional breakpoints
 with specific lines, or set conditional breakpoints only after you have
 reached the section of code that needs to be debugged. You can also use the
 Disable button in the Edit Breakpoints dialog box to temporarily suspend
 evaluation of a previously set conditional breakpoint.
 ────────────────────────────────────────────────────────────────────────────


 Using Breakpoints

 One of the most common bugs is a loop that executes too many or too few
 times. If you set a breakpoint on the statement that controls the loop
 statements, the program pauses after each iteration. With the loop variable
 or critical program variables in the Watch or Local windows, it should be
 easy to see what's going wrong in the loop.

 You can specify how many times a breakpoint is reached before stopping.

 You do not have to pause at a breakpoint the first time execution reaches
 it. CodeView lets you specify the number of times you want to ignore the
 breakpoint condition before pausing. Type the number in the Pass Count field
 of the Set Breakpoint dialog box. This feature can eliminate a lot of
 tedious singlestepping.

 Another programming error is erroneously assigning a value to a variable
 that should not change. Type the variable in the Expression field of the Set
 Breakpoint dialog box. Execution breaks whenever this variable changes─even
 unintentionally.

 You can assign new values to variables while execution is paused.

 Breakpoints are a convenient way to pause the program so you can assign new
 values to variables. For example, if a limit value is set by a variable, you
 can change the value to see whether program execution is affected.


 15.4.2  Single-Stepping

 In single-stepping, CodeView pauses after each line of code is executed. The
 next line to be executed is highlighted.

 There are two ways to single-step.

 You can single-step through a program with the Step and Trace commands. Step
 (executed by pressing F10) steps over procedure calls. All the code in the
 procedure is executed, but it appears to you as if the procedure executed in
 a single step. Trace (executed by pressing F8) traces through every step of
 all procedures. Each line of the procedure is executed as a separate step.

 You can alternate between Trace and Step as you like. The method you use
 depends only on whether you want to see what happens within a particular
 procedure. (Note that interrupt calls are always stepped over; you do not
 see individual steps of the execution.)

 If CodeView cannot locate the source code for a procedure in the current
 directory, it pauses and asks for the name of the file that contains the
 source. If you cannot supply a source file, CodeView disassembles the
 executable code and displays that instead. (If you are executing in Source
 mode, and the source code for a procedure is not available, CodeView steps
 over the procedure, even if you use the Trace command.)

 Note that breakpoints are active during both step and trace mode. If the
 procedure you step over contains a breakpoint, execution stops at the
 breakpoint.

 You can trace through the program continuously (without having to press F8
 at each step), using the Animate command from the Run menu. The speed of
 execution is controlled by the Trace Speed command from the Options menu.
 You can halt animated execution at any time by pressing any key.


 15.4.3  Changing the Program Display Mode

 The F3 function switches the display between Source mode, Mixed mode, and
 Assembly mode. You can also switch display modes by choosing the Source
 Window command from the Options menu and then selecting a display mode in
 the Source Window Options dialog box. (If the source-code text file cannot
 be located, CodeView automatically disassembles the executable file and
 displays it in assembly-language form.)

 The Source mode shows the program as you wrote it. The Mixed mode and
 Assembly mode each expand macros and code-generating directives (such as
 .STARTUP) into assembly-language instructions. You can execute these
 instructions one at a time (rather than as a single item), and verify that
 the assembler has created the correct instructions from the macro or the
 directive.

 Figures 15.5 and 15.6 show Mixed mode and Assembly mode, respectively, for
 the same code.

 (This figure may be found in the printed book.)

 (This figure may be found in the printed book.)


 15.5  Replaying a Debug Session

 CodeView can automatically create a "tape" (a disk file) with the debugging
 instructions and input data you entered when testing a program. The tape can
 then be "replayed" to repeat the debugging process. You initiate recording
 by choosing the History On command from the Run menu. Choosing History On a
 second time terminates recording. The recording is saved in the .CVH file in
 the current directory.

 Dynamic replay has several uses. The most obvious is repeating a debug
 session for the corrected version of a program. Dynamic replay usually works
 with slightly modified programs. However, the more you change the program,
 the less likely the new version will replay reliably.

 You can also use the recording as a bookmark. You can quit after a long
 debugging session, then pick up the session later in the same place.

 Dynamic replay makes it easy to correct a mistake.

 Most importantly, dynamic replay allows you to back up when you make an
 error or overshoot the section of code with the bug. This feature is
 important because not all bugs appear on the first path of execution you
 try.

 For example, you might have to manually execute a procedure many times
 before its bug appears. If you then enter a command that alters the
 machine's or program's status, thereby losing the information you need to
 find the cause of the bug, you would have to restart the program and
 manually repeat every debugging step to return to that point. Even worse, if
 you don't remember the exact sequence of events that exposed the bug, it
 could take hours to reproduce them.

 Dynamic replay of a recorded tape eliminates this problem. Choose the Undo
 command from the Run menu to automatically restart the program and
 continuously execute every command up to (but not including) the last one
 you entered. You can repeat this process as many times as you like until you
 return to the desired point in execution.

 You can add additional steps to an existing tape. Choose History On, then
 choose Replay. When replay has completed, perform whatever new debugging
 steps you want, then choose History On a second time to terminate recording.
 The new tape contains both the original and the added commands.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 CodeView records only those mouse commands that apply to CodeView. Mouse
 commands recognized by the application being debugged are not recorded.
 ────────────────────────────────────────────────────────────────────────────


 Replay Limitations under OS/2

 There are some limitations to dynamic replay when debugging under OS/2:


   ■   The program must not respond to asynchronous events. Replay under
       Presentation Manager is not currently supported because of this
       restriction.

   ■   Breakpoints must be specified at specific source lines or for specific
       symbols (rather than by absolute addresses), or replay may fail.

   ■   Single-thread programs behave normally during replay. However, one of
       the threads in a multithread program may cause an asynchronous event,
       violating the first restriction in this list. Multithread programs are
       therefore more likely to fail during replay.

   ■   Multiprocess replay will fail. Each new process invokes a new CodeView
       session. The existence of multiple sessions makes it impractical to
       record the sequence of events if you execute commands in a session
       other than the original session.



 15.6  Advanced CodeView Techniques

 Once you are comfortable displaying and changing variables, stepping through
 the program, and using dynamic replay, you might want to experiment with the
 advanced techniques explained below.


 Debugging OS/2 Programs

 You can debug protected-mode and bound programs under CodeView. See the
 Debug Multiple Processes and Debug Multiple Threads sections of CodeView
 online help for information about executing threads and multiple processes.



 Setting Command-Line Arguments

 If your program retrieves command-line arguments, you can specify them with
 the Set Runtime Arguments command from the Run menu. Type the arguments in
 the Command Line field before you begin execution. (Arguments entered after
 execution begins cause an automatic restart.)


 Opening Multiple Source Windows

 You can open two Source windows at the same time. The windows can display
 two different sections of the same program, or one window can show the
 calling program and the other a procedure file. You can move freely between
 the windows, executing lines of code as you like.


 Calling Procedures

 Any procedure in your program (whether user-written or from a library) can
 be called from the Command window or the Watch window. In the Command
 window, use the Display Expression command as follows:

   ?procname (arglist)

 The procedure procname is evaluated with the arglist arguments and the
 returned value is displayed in the Command window. (Note that CodeView
 cannot evaluate a function that returns an aggregate type.) In the Watch
 window, simply enter the procedure call. If the procedure does not return a
 value, the value displayed is the value of the AX register upon return from
 the procedure.

 You can evaluate any procedure, not just those called by your program. All
 object code specified to the linker is linked into the program. Any public
 functions in this code can be evaluated from the Command window.

 You can use this feature to call functions from within CodeView that you
 would not normally include in the final version of your program. For
 example, you could include the OS/2 API functions that control semaphores,
 then execute them from the Command window to manipulate the run-time
 environment at any point in the debugging process. (Remember that altering
 the environment during program execution may have unexpected side effects.)



 Executing Faster when Using Breakpoints

 Breakpoints can slow execution. You can increase CodeView's speed with the
 /R command-line option if you have an 80386/486-based computer and are
 running CodeView under DOS. This option enables the four debug registers,
 which support breakpoint-checking in hardware rather than in software. (The
 CodeView options are described in Section 15.7.)


 Printing Selected Items

 You can print all or part of the contents of any window with the Print
 command from the File menu. In the Print dialog box, a check box lets you
 print selected text from the window, the material currently displayed in the
 window, or the complete contents of the window. Select text by dragging the
 mouse across it, or by holding down the SHIFT key and pressing the direction
 keys (LEFT, RIGHT, UP, DOWN).

 By default, print output is to the file CODEVIEW.LST in the current
 directory. You can choose whether the new material is appended to an
 existing file or overwrites it, using the Append/Overwrite check box. If you
 want print output to go to a different file, type its name in the To File
 Name field. If you want the output to go to a printer, enter the appropriate
 device name such as LPT1 or COM2.


 Redirecting CodeView Input and Output

 The Command window accepts DOS-like commands that redirect input and output.
 These commands can also be included on the command line that invokes
 CodeView. Whatever items follow the /C option on the command line are
 treated as CodeView commands to be immediately executed at start-up.

   CV /c "<infile; t>outfile" myprog

 In the example above, input is redirected from  infile, which can contain
 start-up commands for CodeView. When CodeView exhausts all commands in the
 input file, focus automatically shifts to the Command window. Output is sent
 to  outfile  and echoed to the Command window. The  t  must precede the  >
 command for output to be sent to the Command window.

 Redirection is a useful way to automate CodeView start-up. It also lets you
 keep a viewable record of command-line input and output, a feature not
 available with dynamic replay. No record is kept of mouse operations. Some
 applications (particularly interactive ones) may need modification to allow
 for redirection of input to the application itself.


 Executing Faster with Additional Memory

 If you are running DOS and your computer uses expanded or extended memory,
 you can increase CodeView's execution speed by selecting the /X or /E
 option. CodeView moves as much as it can of itself and the symbolic CodeView
 information to higher memory (above the first megabyte).

 The /X option uses extended memory and gives the greatest speed increase.
 This option requires the HIMEM.SYS driver, which is included on your
 distribution disks. Add  DEVICE = HIMEM.SYS  to your CONFIG.SYS file to load
 HIMEM.SYS at boot time.

 The /E option uses expanded memory. The speed increase is not as great as
 that supplied by the /X option. The expanded memory manager (EMM) must be
 LIM 4.0, and no single module's debug information can exceed 48K. If the
 symbol table exceeds this limit, try reducing file-name information by not
 specifying full path names at compile time and by specifying CodeView
 information (/Zi) only with those program modules that need debugging.

 If you do not specify either /X or /E (or the /D disk-overlay option),
 CodeView automatically searches for the HIMEM.SYS driver and extended memory
 so it can implement the /X option. If it fails, CodeView searches for
 expanded memory to implement the /E option. If that search fails, CodeView
 uses a default disk overlay of 64K. (See the description of the /D option in
 the next section.)


 15.7  CodeView Command-Line Options

 The following options can be added to the command line that invokes
 CodeView. The Starting Up CodeView section of CodeView online help contains
 more information about these options.

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Option                            Description
 Option                            Description
 ────────────────────────────────────────────────────────────────────────────
 /2                                Two-monitor debugging. The display
                                   adapters must be configured for
                                   different addresses, such as Hercules (R)
                                   and VGA. The application is displayed on
                                   the primary monitor (the monitor the
                                   operating system normally directs output
                                   to), while CodeView's output appears on
                                   the secondary monitor.

 /25                               Display in 25-line mode.

 /43                               Display in 43-line mode.

 /50                               Display in 50-line mode.

 /B                                Display in black and white. This assures
                                   that the display is readable when a
                                   color display is not used. You should
                                   also specify this option along with the
 Option                            Description
 ────────────────────────────────────────────────────────────────────────────
                                  also specify this option along with the
                                   /2 option when the secondary monitor is
                                   black and white.

 /Ccommands                        Execute commands immediately on start-up.
                                   The commands must be separated with a
                                   semicolon. If any commands require a
                                   space, enclose the entire list in double
                                   quotation marks.

 /D«buffersize»                    Use disk overlays to increase the size
                                   of the program that can be debugged,
                                   where buffersize is the decimal size of
                                   the overlay buffer, in kilobytes.
                                   Smaller buffers leave more room for the
                                   program being debugged, while larger
                                   buffers increase the speed of execution.
                                   The acceptable range is 16K to 128K. The
                                   default size is 64K. (DOS only.)
 Option                            Description
 ────────────────────────────────────────────────────────────────────────────
                                  default size is 64K. (DOS only.)


 /E                                Use expanded memory for symbolic
                                   information and CodeView overlays. (DOS
                                   only.)

 /F                                Flip screen video pages (rather than
                                   swap). When your application does not
                                   use graphics, eight video screen pages
                                   are available. Switching from CodeView
                                   to the output screen is accomplished by
                                   directly selecting the appropriate video
                                   page. Cannot be used with /S. (DOS only.)

 /G                                Suppress "snow" on a CGA display. (DOS
                                   only.)

 /I«0 | 1»                         Control trapping of nonmaskable
 Option                            Description
 ────────────────────────────────────────────────────────────────────────────
/I«0 | 1»                         Control trapping of nonmaskable
                                   interrupts and 8259 interrupts. A value
                                   of 0 forces interrupt trapping on
                                   machines CodeView doesn't recognize as
                                   IBM-
                                   compatible. A value of 1 (the default)
                                   disables interrupt trapping. (DOS only.)

 /K                                Disable keyboard monitors (under OS/2)
                                   and keyboard interrupts (under DOS).
                                   This allows you to regain control of the
                                   computer under deadlock conditions, but
                                   prevents CodeView from recording
                                   keyboard entries when recording a debug
                                   session.

 /Ldll                             Load symbolic information for the
                                   specified dynamic-link libraries (DLL).
                                   (OS/2 only.) This option is required
 Option                            Description
 ────────────────────────────────────────────────────────────────────────────
                                  (OS/2 only.) This option is required
                                   only for DLLs loaded with DOSLOADMODULE.
                                   CodeView automatically loads debug
                                   information for statically linked DLLs.

 /M                                Disable CodeView's use of the mouse.
                                   This simplifies debugging programs that
                                   accept mouse commands.

 /N«0 | 1»                         Identical to /I, but applies only to
                                   nonmaskable interrupts. (DOS only.)

 /O                                Debug child processes ("offspring").
                                   (OS/2 only.)

 /R                                Use 80386/486 hardware debug registers
                                   to speed execution. (DOS only.)


 Option                            Description
 ────────────────────────────────────────────────────────────────────────────

 /S                                Swap screen in buffers (rather than
                                   flip). When your program uses graphics,
                                   all eight video pages must be used.
                                   Switching from CodeView to the output
                                   screen is accomplished by saving the
                                   previous screen in a buffer. Cannot be
                                   used with /F. (DOS only.)

 /TSF                              Toggle (invert) the sense of the
                                   Statefileread switch in TOOLS.INI. If
                                   Statefileread is set to no (do not read
                                   the status file), the status file is
                                   read, and vice-versa.

 /X                                Use extended memory for CodeView and
                                   symbolic information. (DOS only.)




 15.8  Customizing CodeView with the TOOLS.INI File

 The TOOLS.INI file customizes the behavior and user interface of several
 Microsoft products. The TOOLS.INI file is a plain ASCII text file. You
 should place it in a directory pointed to the INIT environment variable. (If
 you do not use the INIT environment variable, CodeView looks for TOOLS.INI
 only in the CodeView source directory.)

 The CodeView section of TOOLS.INI is preceded by the following line:

   [cv]

 If you run the protected-mode version of CodeView, use [cvp] instead. If you
 run both versions, include both: [cv cvp]. You can have separate sections
 for  cv  and  cvp  if you want different customizations.

 Most of the TOOLS.INI customizations for CodeView control screen colors, but
 you can also specify such things as start-up commands or the default name of
 the file that receives CodeView output. See the Configure CodeView section
 of CodeView online help for full information about all TOOLS.INI switches
 that control CodeView.


 15.9  Related Topics in Online Help

 In addition to information covered in this chapter, information on the
 following topics can be found in online help.

 Topic                             Access
 ────────────────────────────────────────────────────────────────────────────
 CodeView information              Choose "CodeView Debuggers" from the
                                   "Microsoft Advisor Contents" screen

 ML command-line options           Choose "Macro Assembler" from the
                                   "Command Line" section of the "Microsoft
                                   Advisor Contents" screen









 Chapter 16  Converting C Header Files to MASM Include Files
 ────────────────────────────────────────────────────────────────────────────

 The H2INC utility translates C header files into MASM-compatible include
 files. C header files normally have the extension .H; MASM include files
 normally have the extension .INC. This is the origin of the program's name:
 "H to INC."

 H2INC simplifies porting data structures from your C programs to MASM
 programs. This is especially useful when you have


   ■   A program that mixes C code and MASM code with globally accessible
       data structures

   ■   A program prototyped in C that you're translating to MASM for
       compactness and fast execution


 The H2INC program translates data declarations, function prototypes, and
 type definitions. H2INC does not convert C code into MASM code. When H2INC
 encounters a C statement that would compile into executable code, H2INC
 ignores the statement and issues a warning message to the standard output.

 H2INC accepts C source code compatible with Microsoft C 6.0 and creates
 include files suitable for MASM 6.0. These include files will not work with
 versions of MASM prior to 6.0.

 H2INC is designed to translate project header files that you have written
 specifically for translation to MASM 6.0 include files. It is not designed
 to translate header files such as PM.H and WINDOWS.H.

 This chapter explains how H2INC performs the C code translation and how the
 command-line options control the conversions.


 16.1  Basic H2INC Operation

 H2INC is designed to provide automatic translation of C declarations that
 you need to include in the MASM portions of an application. However, the set
 of C statements processed by H2INC must be those needed by and interpretable
 by MASM. H2INC converts only function prototypes, some preprocessor
 directives,

 and C declarations outside the scope of procedures. For example, H2INC
 translates the C statement

   #define MAX_EMPLOYEES 400

 into this MASM statement:

   MAX_EMPLOYEES EQU 400t

 The t specifies the decimal radix.

 H2INC does not translate C code into MASM code. Statements such as the
 following are ignored:

   printf( "This is an executable statement.\n" );

 H2INC translates declarations, not executable code.

 By default, H2INC creates a single .INC file. If the C header file includes
 other header files, the statements from the original and nested files are
 translated and combined into one .INC file. This behavior can be changed
 with the /Ni option (see Section 16.2).

 The program also preprocesses some statements, just as the C preprocessor
 would. For example, given the following statements, if  VERSION  is not
 defined, H2INC ignores the #ifdef block.

   #ifdef VERSION
   #define BOX_VALUE 4
   #endif

 If  VERSION  is defined, H2INC translates the statements inside the block
 from C syntax to MASM syntax.

 H2INC normally discards comments. If you use the /C option, C comments are
 passed to the output file. If the line starts with a  /*  or  // , the
 comment specifier is converted to a semicolon (;). If the line is part of a
 multiline comment, a semicolon is prefixed to each line.

 H2INC ignores anything that is not a comment or that cannot be translated.
 These items do not appear in the output file. If H2INC encounters an error,
 it stops translating and deletes the resulting .INC file.


 16.2  H2INC Syntax and Options

 To run H2INC, type  H2INC  at the command-line prompt, followed by the
 options desired and the names of the .H files you want to convert:

   H2INC [[options]] file.H ...

 You can specify more than one file.H. File names are separated by a space.
 The contents of each file.H are translated into a single file in the current
 directory with the name file.INC. The original file.H is not altered.

 The following lists describe the available options. You can specify more
 than one option. Note that the options are case sensitive except for /HELP.


 H2INC recognizes /? to display a summary of H2INC syntax, and /HELP to
 invoke QuickHelp for H2INC. If QuickHelp is not available, H2INC displays a
 short list of H2INC options. This option is not case sensitive.

 H2INC recognizes but ignores C 6.0 options that aren't specified in the
 following two lists.


 Options Directly Affecting H2INC Output

 This first list describes the options that directly affect the H2INC output:


 Option                            Action
 ────────────────────────────────────────────────────────────────────────────
 /C                                Passes comments in the .H file to the
                                   .INC file.

 /Fa «filename»                    Specifies that the output file contain
                                   only equivalent MASM statements. This is
                                   the default. If specified, the filename
                                   overrides the default, keeping the base
                                   name of the C header files and adding
                                   the .INC extension.

 /Fc «filename»                    Specifies that the output file contain
                                   equivalent MASM statements plus original
                                   C statements converted to comment lines.

 /Mn                               Assumes the .MODEL directive is not
                                   specified for the MASM source or the
                                   generated .INC files. Instructs H2INC to
                                   declare explicitly the distances for all
                                   pointers and functions.

 /Ni                               Suppresses the expansion of nested
                                   include files.

 /Zu                               Makes all structure and union tag names
                                   unique.


 Options Indirectly Affecting H2INC Output

 This second list describes the options that indirectly affect the H2INC
 output:

 Option                            Action
 ────────────────────────────────────────────────────────────────────────────
 /AT                               Specifies tiny memory model (.COM).

 /AS                               Specifies small memory model, the
                                   default.

 /AC                               Specifies compact memory model.

 /AM                               Specifies medium memory model.

 /AL                               Specifies large memory model.

 /AH                               Specifies huge memory model.

 /D«const«=value» »                Defines a constant or macro.

 /G0                               Enables 8086/8088 instructions (default).

 /G1                               Enables 80186/80188 instructions.

 /G2                               Enables 80286 instructions.

 /G3                               Enables 80386 instructions. Changes the
                                   default word size to DWORD.

 /G4                               Enables 80486 instructions. Changes the
                                   default word size to DWORD.

 /Gc                               Specifies Pascal as the default calling
                                   convention.

 /Gd                               Specifies C as the default calling
                                   convention for functions (default).

 /Gr                               Specifies the _fastcall calling
                                   convention for functions. Generates a
                                   warning since H2INC does not translate
                                   _fastcall functions and prototypes.

 /Ht                               Enables generation of text equates. By
                                   default, text items are not translated.

 /Ipaths                           Searches named paths for include files
                                   before searching the paths in the
                                   INCLUDE environment variable. Paths are
                                   separated with a semicolon (;).

 /J                                Changes default character type from
                                   signed char to unsigned char.

 /nologo                           Suppresses display of the sign-on banner.

 Option                            Action
 ────────────────────────────────────────────────────────────────────────────
 /Tc «filename»                    Enables the processing of files whose
                                   name does not end in .H.

 /uident                           "Undefines" one of the predefined
                                   identifiers. (See Section 16.3.1.)

 /U                                "Undefines" all predefined identifiers.
                                   (See Section 16.3.1.)

 /w                                Suppresses compiler warning messages;
                                   same as /W0.

 /W0                               Suppresses all warning messages.

 /W1                               Displays level 1 warning messages
                                   (default).

 /W2                               Displays level 1 and level 2 warning
                                   messages.

 /W3                               Displays level 1, 2, and 3 warning
                                   messages.

 /W4                               Displays all warning messages.

 /X                                Excludes search for include files in the
                                   standard places.

 /Za                               Disables language extensions (allows
                                   ANSI standard only).

 /Zc                               Causes functions declared as _pascal to
                                   be case insensitive.

 /Ze                               Enables language extensions (default).

 /Zn string                        Adds string to all names generated by
                                   H2INC. Used to eliminate name conflicts
                                   with other H2INC-generated include files.

 /Zp{1 | 2 | 4}                    Packs structure on a 1-, 2-, or 4-byte
                                   boundary, following C packing rules.
                                   Default is /Zp2.


 16.3  Converting Data and Data Structures

 The primary use of H2INC is to convert data automatically from C format into
 MASM format. This section shows how H2INC converts constants, variables,
 pointers, and other C data structures to definitions recognizable to MASM.

 Since the names of the items translated by H2INC may be distinguished only
 by the case of the names, you should specify OPTION CASEMAP:NONE in any MASM
 files that include .INC files generated with H2INC.


 16.3.1  User-Defined and Predefined Constants

 H2INC translates constants from C to MASM format. For example, C symbolic
 constants of the form

   #define CORNERS 4

 are translated to MASM constants of the form

   CORNERS EQU 4t

 in cases where  CORNERS  is an integer constant or is preprocessed to an
 integer constant. See Section 1.2.4, "Integer Constants and Constant
 Expressions," for more information on integer constants in MASM.

 TEXTEQU is new to MASM 6.0.

 When the defined expression evaluates to a noninteger value, such as a
 floating-point number or a string, H2INC defines the expression with TEXTEQU
 and adds angle brackets to create text macros. By default, however, these
 TEXTEQU expressions are not added to the include file. Set the /Ht option to
 tell H2INC to generate TEXTEQU expressions.

   /* #define PI 3.1415 */
   PI TEXTEQU <3.1415>

 H2INC uses this form when the expression is anything other than a constant
 integer expression. H2INC does not check the constant or string for
 validity. For example, although the following C definitions are valid, H2INC
 creates invalid string equates without generating an error.

 These C statements

   #define INT 6
   #define FOREVER for(;;)

 generate these MASM statements:

   INT EQU 6t
   FOREVER TEXTEQU <for(;;)>

 The first #define statement is invalid because INT is a MASM instruction; in
 MASM 6.0, instructions are reserved and cannot be used as identifiers. The
 for loop definition is invalid because MASM cannot assemble C code.

 Predefined constants control the contents of .INC files.

 You can make use of the following predefined constants in your C code to
 conditionally generate the code in .INC files. The predefined constants and
 the conditions under which they are defined are

 Predefined Constant               When Defined
 ────────────────────────────────────────────────────────────────────────────
 _H2INC                            Always defined

 M_I86                             Always defined

 MSDOS                             Always defined

 _MSC_VER                          Defined as 600 for this release

 M_I8086                           Defined if /G0 is specified

 M_I286                            Defined if /G0 is not specified

 NO_EXT_KEYS                       Defined if /Za is specified

 _CHAR_UNSIGNED                    Defined if /J is specified

 M_I86SM                           Defined if /AS is specified

 M_I86MM                           Defined if /AM is specified

 M_I86CM                           Defined if /AC is specified

 M_I86LM                           Defined if /AL is specified

 M_I86HM                           Defined if /AH is specified

 For example, if your C header file includes definitions which are specific
 to the C portion of the program or otherwise are not appropriate for
 translation by H2INC, you can bracket the C-specific code with

   #ifndef _H2INC
        /* C-specific code */
   #endif

 In this case, only the C compiler processes the bracketed code.

 The /u and /U options affect these predefined constants. The /uarg option
 undefines the constant specified as the argument. The /U option disables the
 definition of all predefined constants. Neither /u or /U affects constants
 defined by the /D option.

 H2INC places an OPTION EXPR32 directive in the .INC file so that MASM
 correctly handles long integers within expressions. This means that the .INC
 files as well as all the .ASM files which include .INC files created with
 H2INC will resolve integer expressions in 32 bits instead of 16 bits.


 16.3.2  Variables

 H2INC translates variables from C to MASM format. For example, this C
 declaration

   int my_var;

 is translated into the MASM declaration

   EXTERNDEF my_var:SWORD

 H2INC converts C variable types to MASM types as follows:

 C Type                            MASM Type
 ────────────────────────────────────────────────────────────────────────────
 char                              BYTE or SBYTE (controlled by /J option)

 signed char                       SBYTE

 unsigned char                     BYTE

 short                             SWORD

 unsigned short                    WORD

 int                               SWORD (SDWORD with /G3 or /G4 option)

 unsigned int                      WORD (DWORD with /G3 or /G4 option)

 long                              SDWORD

 unsigned long                     DWORD

 float                             REAL4

 double                            REAL8

 long double                       REAL10

 H2INC assumes that a variable is external unless the variable is explicitly
 declared as static. For example, the C declaration

   long big_data;

 is converted to this MASM declaration:

   EXTERNDEF big_data:SDWORD

 See Sections 1.2.6, "Data Types," and 4.1.1, "Allocating Memory for Integer
 Variables," for more information on MASM data types, and Section 8.2.2,
 "Declaring Symbols Public and External," for information on EXTERNDEF.

 H2INC does not allocate space for arrays since all variables are assumed to
 be external. For example, the C declaration

   int two_d[10][20];

 translates to

   EXTERNDEF two_d:SWORD

 H2INC does not translate static variables, since the scope of these
 variables extends only to the file where they are declared.


 16.3.3  Pointers

 H2INC translates C pointer variables into their MASM equivalents. The C
 declarations

   int *ptr_var;
   char NEAR *pCh;

 are translated into these MASM statements:

   EXTERNDEF ptr_var:PTR SWORD
   EXTERNDEF pCh:NEAR PTR SBYTE

 If you set the /Mn option, H2INC specifies all distances explicitly (for
 example, NEAR PTR instead of PTR). If /Mn is not set, the distances are
 generated only when they differ from the default values implied by the
 memory model specified by the /A command-line option.

 H2INC converts _segment and _based variables to type WORD in MASM.

 See Sections 1.2.6, "Data Types," and 3.3, "Accessing Data with Pointers and
 Addresses," for information about MASM pointers.


 16.3.4  Structures and Unions

 H2INC translates C structures and unions into their MASM equivalents. H2INC
 modifies the C structure or union definition to account for differences from
 MASM structure and union definitions. This list describes these
 modifications.


   ■   C allows a structure or union variable to have the same name as the
       type name, but MASM does not. The H2INC /Zu option prevents the
       structure name from matching a variable or instance by prefixing every
       MASM structure name with  @tag_.

   ■   If a C structure or union definition does not have a name, H2INC
       supplies one for the MASM conversion. These generated structure names
       take the form  @tag_n, where n is an integer that starts at zero and
       is incremented for each structure name H2INC generates.

   ■   If the /Zn option is specified, H2INC inserts the given string between
       the underscore and the number in the generated structure names. This
       eliminates name conflicts with other H2INC-generated include files.

   ■   H2INC adds the alignment value to the converted structure definition.


 The following examples show how these rules are applied when converting
 structures. (Union conversions are not shown; they are handled identically.)
 These examples assume that the C header file defines an alignment value of
 2. (See Section 5.2.1, "Declaring Structure and Union Types," for
 information on alignment values.)

 The following named C structure definition

   struct file_info
   {
      unsigned char  file_addr;
      unsigned int   file_size;
   };

 is converted to the following MASM form. Except for explicitly specifying
 the alignment value, the conversion is direct:

   file_info          STRUCT 2t
   file_addr          BYTE          ?
   file_size          WORD          ?
   file_info          ENDS

 If the same C structure definition is converted using the /Zu option, the
 @tag_  prefix is added to the structure's name so that the name does not
 duplicate the name of a structure component:

   @tag_file_info     STRUCT 2t
   file_addr          BYTE          ?
   file_size          WORD          ?
   @tag_file_info     ENDS

 If the original C structure definition is modified to be an unnamed-type
 declaration of a specific instance (myfile)

   struct
   {
      unsigned char  file_addr;
      unsigned int   file_size;
   } myfile ;

 its MASM conversion looks like the following example. (The specific integer
 added to the  @tag_  prefix is determined by the sequence in which H2INC
 creates tag names.)

   @tag_7          STRUCT 2t
   file_addr       BYTE ?
   file_size       WORD ?
   @tag_7          ENDS
   EXTERNDEF       C myfile:@tag_7

 Nested structures may have as many levels as desired; they are not limited
 to one level. Nested structures are "unnested" (expanded) in the correct
 hierarchical sequence, as shown with the C structure and H2INC-generated
 code in this example.

   /* C code: */
   struct phone
   {
       int  areacode;
       long number;
   };

    struct person
   {
       char   name[30];
       char   sex;
       int    age;
       int    weight;
       struct phone;
   } Jim;

   ; H2INC generated code:
   phone           STRUCT 2t
   areacode        SWORD          ?
   number          SDWORD         ?
   phone           ENDS

   person          STRUCT 2t
   name            SBYTE          30t DUP (?)
   sex             SBYTE          ?
   age             SWORD          ?
   weight          SWORD          ?
   STRUCT
     areacode      SWORD          ?
     number        SDWORD         ?
   ENDS
   person          ENDS

   EXTERNDEF       C Jim:person

 See Section 5.2 for information on MASM structures and unions.


 16.3.5  Bit Fields

 H2INC translates C bit fields into MASM records. H2INC looks at a structure
 definition; if it consists only of bit fields of the same type and if the
 total size of the bit fields does not exceed the type of the bit fields,
 then H2INC outputs a RECORD definition with the name of the structure. All
 bit-field names are modified to include the structure name for uniqueness,
 since record fields have global scope in MASM.

 For example,

   struct s
   {
      int i:4;
      int j:4;
      int k:4;
   }

 becomes:

   s       RECORD  @tag_0:4,
                   k@s:4,
                   j@s:4,
                   i@s:4

 The  @tag  variable pads out the record to the type size of the bit fields
 so alignment of the structures will be correct.

 If the bit fields are too large, are not of the same type, or are mixed with
 fields that are not bit fields, H2INC generates a RECORD definition inside
 the structure and then uses the definition.

 For example,

   struct t
   {
       int i;
       unsigned char a:4;
       int j:9;
       int k:9;
       long l;
   }  m;

 becomes:

   t         STRUCT 2t
   i         SWORD       ?
   rec@t_0   RECORD      @tag_1:4,
             a@t:4
   @bit_0    rec@t_0     <>
   rec@t_1   RECORD      @tag_2:7,
             j@t:9
   @bit_1    rec@t_1     <>
   rec@t_2   RECORD      @tag_3:7,
             k@t:9
   @bit_2    rec@t_2     <>
   l         SDWORD      ?
   t         ENDS

   EXTERNDEF C m:t

 Notice that  j  and  k  are not packed because their total size exceeds the
 16 bits of an integer in C.

 Since the  @bit  field names are local to the structure, these begin with  0
  for each structure type; the  @rec  variables have global scope and so
 their number always increases.

 The C bit-field declaration

   struct SCREENMODE
   {
      unsigned int disp_mode : 4;
      unsigned int fg_color  : 3;
      unsigned int bg_color  : 3;
   };

 is converted into the following MASM record:

   SCREENMODE      RECORD          disp_mode@SCREENMODE:4,
                                   fg_color@SCREENMODE:3,
                                   bg_color@SCREENMODE:3

 See Section 5.3 for information about MASM records.


 16.3.6  Enumerations

 H2INC converts C enumeration declarations into MASM EQU definitions that are
 treated as standard integer constants. If the C declaration is not assigned
 a value, the H2INC generates an EQU statement that supplies a value
 equivalent to its position in the list. For example, the C enumeration
 declaration

   enum tagName
   {
      id1,
      id2,
      id3 = 42,
      id4
   };

 is converted into the following EQU statements:

   id1     EQU     0t
   id2     EQU     1t
   id3     EQU     42t
   id4     EQU     43t

 See Section 1.2.4 for information on MASM integer constants.


 16.3.7  Type Definitions

 All type definitions using C base types are translated directly. For
 example, H2INC converts the C type definitions

   typedef int INTEGER;
   typedef float FLOAT;

 to these MASM forms:

   INTEGER TYPEDEF SWORD
   FLOAT   TYPEDEF REAL4

 Pointer types are converted in a similar fashion. The following declarations


   typedef int *PINT
   typedef int **PINT
   typedef int far *PINT

 become (respectively)

   PINT TYPEDEF PTR SWORD
   PINT TYPEDEF PTR PTR SWORD
   PINT TYPEDEF FAR PTR SWORD

 Addressing mode determines pointer size.

 The number of bytes allocated for the pointer is set by the addressing mode
 you have selected unless if is specifically overridden in the type
 definition.

 C statements using typedef which convert to a type with the same name as the
 type do not generate errors, but are not converted. For example, H2INC does
 not convert

   typedef int SWORD
   typedef unsigned char BYTE

 since these typedef statements would generate these MASM statements:

   SWORD   TYPEDEF SWORD
   BYTE    TYPEDEF BYTE

 See Section 3.3, "Accessing Data with Pointers and Addresses," for
 information on using TYPEDEF in MASM 6.0.


 16.4  Converting Function Prototypes

 When H2INC converts C function prototypes into MASM function prototypes, the
 elements of the C syntax are converted into the corresponding elements of
 the MASM syntax.

 The syntax of a C function prototype is

   [[storage]] [[distance]]
   [[ret_type]] [[langtype]]
   label ( [[parmlist»  )

 In C syntax, storage can be STATIC or EXTERN. H2INC does not translate
 static function prototypes because static functions are visible only within
 the current source module, and standard include files do not contain
 executable code.

 Procedures for returning values depend on the langtype specified.

 In C, the ret_type is the data type of the return value. Because the MASM
 PROTO directive does not specify how to handle return values, H2INC does not
 translate the return type. However, H2INC checks the langtype specified in
 the C prototype to determine how particular languages return the
 value─through the stack or through registers.

 For the Pascal, FORTRAN, or Basic langtype specifications, H2INC appends an
 additional parameter to the argument list if the return type is longer than
 four bytes. This parameter is always a near pointer with the type of the
 return value. If the value of the return value type is not supported, this
 parameter is an untyped near pointer.

 For the _cdecl langtype specification in the C prototype, all returned data
 is passed in registers (AX or AX plus DX). There is no restriction on the
 return type. Additional parameters are not necessary.

 The langtype represents the naming and passing conventions for a language
 type. H2INC accepts the following C language types and converts them to
 their corresponding MASM language types:

 C Language Type                   MASM Language Type
 ────────────────────────────────────────────────────────────────────────────
 _cdecl                            C

 _fortran                          FORTRAN

 _pascal                           PASCAL

 _stdcall                          STDCALL

 _syscall                          SYSCALL

 H2INC explicitly includes the langtype in every function prototype. If no
 language type is specified in the .H file prototype, the default language is
 _cdecl (unless the default is overridden by the /Gc command-line option).

 In the MASM prototype syntax, the label is the name of the function or
 procedure.

 If you select the /Mn option, H2INC specifies the distance of the function
 (near or far), whether or not the C prototype specifies the distance. If /Mn
 is not set, H2INC specifies the distance only when it is different from the
 default distance specified by the memory model.

 If the C prototype's parameter list ends with a comma plus an ellipsis (,
 ...), the function can accept a variable number of arguments. H2INC converts
 this to the MASM form: a comma followed by the :VARARG keyword (, :VARARG)
 appended to the last parameter.

 H2INC does not translate _fastcall functions. Functions explicitly declared
 _fastcall (or invoking H2INC with the /Gr option) generate a warning
 indicating that the function declaration has been ignored.

 The following examples show how the preceding rules control the conversion
 of C prototypes to MASM prototypes (when the memory model default is small).
 The example function is  my_func. The TYPEDEF generated by H2INC for the
 PROTO is given along with the PROTO statement.

   /* C prototype */
               my_func (float fNum, unsigned int x);
   ; MASM TYPEDEF
               @proto_0 TYPEDEF PROTO C :REAL4, :WORD
   ; MASM prototype
               my_func  PROTO   @proto_0

   /* C prototype */
               extern my_func1 (char *argv[]);
   ; MASM TYPEDEF
               @proto_1  TYPEDEF  PROTO C :PTR PTR SBYTE
   ; MASM prototype
               my_func1  PROTO    @proto_1

   /* C prototype */
               struct vconfig _far * _far pascal my_func2 (int, scri
   );
   ; MASM TYPEDEF
               @proto_2  TYPEDEF  PROTO FAR PASCAL :SWORD, :scri
   ; MASM prototype
               my_func2  PROTO    @proto_2

   /* C prototype */
               long pascal my_func3 (double y, struct vconfig vc);
   ; MASM TYPEDEF
               @proto_3  TYPEDEF  PROTO PASCAL :REAL8, :vconfig
   ; MASM prototype
               my_func3    PROTO    @proto_3

   /* C prototype */
               void _far _cdecl myfunc4 ( char _huge *, short);
   ; MASM TYPEDEF
               @proto_4  TYPEDEF  PROTO FAR C :FAR PTR SBYTE, :SWORD

   ; MASM prototype
               myfunc4  PROTO  @proto_4

   /* C prototype */
               short my_func5 (void *);
   ; MASM TYPEDEF
               @proto_5  TYPEDEF  PROTO C :PTR
   ; MASM prototype
               my_func5  PROTO  @proto_5

   /* C prototype */
               char my_func6 (int, ...);
   ; MASM TYPEDEF
               @proto_6  TYPEDEF  PROTO C :SWORD, :VARARG
   ; MASM prototype
               my_func6  PROTO  @proto_6

   /* C prototype */
               typedef char * ptrchar;
               ptrchar _cdecl my_func7 (char *);
   ; MASM TYPEDEF
               @proto_7  TYPEDEF  PROTO C :PTR SBYTE
   ; MASM prototype
               my_func7  PROTO  @proto_7

 See Section 7.3.6, "Declaring Procedure Prototypes," for more information on
 prototypes and Chapter 20, "Mixed-Language Programming," for information on
 calling conventions and mixed-language programs.


 16.5  Related Topics in Online Help

 In addition to information covered in this chapter, information on the
 following topics can be found in online help.

 Topic                             Access
 ────────────────────────────────────────────────────────────────────────────
 INCLUDE Directive                 From the "MASM 6.0 Contents" screen,
                                   choose "Directives" and then
                                   "Miscellaneous"

 Include files                     From the "MASM 6.0 Contents" screen,
                                   choose "Example Code"; then choose
                                   "INCLUDE Files" to see a list of the
                                   include files provided with MASM 6.0

 MASM data types (constants,       From the "MASM 6.0 Contents" screen,
 variables, structures, unions,    choose "Directives"; then choose "Data
 real numbers, records)            Allocation" or "Complex Data Types"

 TYPEDEF                           From the "MASM 6.0 Contents" screen,
                                   choose "Directives" and then "Complex
                                   Data Types"

 Procedures and prototypes         From the "MASM 6.0 Contents" screen,
                                   choose "Directives"; then choose
                                   "Procedure and Code Labels"






 Chapter 17  Writing OS/2 Applications
 ────────────────────────────────────────────────────────────────────────────

 Microsoft Operating System/2 (OS/2) takes full advantage of 80286 and later
 processors. It supports memory far beyond the DOS 640K limit and offers a
 rich set of multitasking system calls. Although OS/2 is much more powerful
 than DOS, you may ultimately find it easier to program for OS/2.

 This chapter shows how to develop an OS/2 application and how to write
 dual-mode programs to run under both OS/2 and DOS.

 To write OS/2 applications, you must learn OS/2 system calls. While this
 chapter mentions a few of these calls, you should consult the references
 listed in the book's introduction to learn more about OS/2 system functions.


 OS/2 supports two modes─real mode, which emulates the DOS environment, and
 protected mode, which supports all the advanced features. For simplicity's
 sake, the rest of this chapter equates OS/2 with protected mode.

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 Examples in this chapter support OS/2 1.x. Future versions of OS/2 may
 support different calling conventions.
 ────────────────────────────────────────────────────────────────────────────


 17.1  OS/2 Overview

 There are three steps in developing OS/2 or dual-mode applications:


   1.  Write the source code, using procedure calls rather than interrupts to
       call system functions.

   2.  Assemble and link the program with OS2.LIB.

   3.  Optionally, convert the program so that it can run under both OS/2 and
       DOS.


 This chapter explains each of these steps, first looking at specific
 differences in how you write DOS and OS/2 code. Then it illustrates the
 development of a simple OS/2 program. Finally, the chapter discusses
 register initialization and additional OS/2 utilities.


 17.2  Differences between DOS and OS/2

 Assembly language is assembly language. Most machine instructions you use in
 a DOS program are the same instructions you use in an OS/2 program. When you
 start making calls to the operating system, however, things change.

 You should understand the following differences between the two operating
 systems before attempting to write an OS/2 program.


 System Calls

 System calls control I/O and screen access.

 OS/2 is similar to DOS in that it offers a series of system calls that
 perform tasks such as opening or closing a disk file. The OS/2 system calls
 that handle keyboard input (KbdCharIn, for example) correspond to the
 interrupt 16h instructions in DOS. The OS/2 system calls for screen output
 (VioScrollDn, for example) correspond to DOS interrupt 10h calls. And the
 OS/2 disk and operating-system calls (DosGetDateTime, for example)
 correspond to DOS interrupt 21h calls.

 The effect is similar, but the way you actually make the calls is different.
 In DOS, you issue an interrupt. In OS/2, you make the system call with the
 INVOKE directive or the CALL instruction.


 New Instructions

 OS/2 is designed for advanced processors, and you may want to write programs
 that take advantage of the new instructions available on the 80286-80486. To
 use the new instructions and still target OS/2 1.x, place a .286 directive
 at the beginning of your source code.

 In general, you should avoid the directives that enable privileged
 instructions (.286P, .386P, and .486P), unless you are writing system-level
 code.

 Many OS/2 programs can be converted to run under DOS as well. To write
 programs to run on all DOS and OS/2 systems, use the default processor
 setting (.8086).


 The OS/2 Library

 MASM 6.0 provides OS2.INC and OS2.LIB.

 OS/2 programs must be linked to the system-call import library, OS2.LIB. The
 best way to perform this task is to use the INCLUDELIB directive, as shown
 in the example in the next section. In addition, you can include the OS2.INC
 file as an alternative to adding the prototypes for the OS/2 functions to
 your file.

 The OS2.LIB file makes system calls possible; it contains import definitions
 for all system calls. An import definition specifies the name of a procedure
 and the dynamic-link library (DLL) where the procedure resides. You can
 learn more about DLLs in Chapter 18, "Creating Dynamic-Link Libraries." To
 create an OS/2 application, however, you need to know only that OS2.LIB is
 required.


 Start-Up Code

 Unlike DOS, OS/2 automatically initializes all segment registers as required
 by the standard segment model. No special start-up sequence is required,
 although OS/2 places useful information in AX, BX, and CX (see Section 17.6,
 "Register and Memory Initialization") that you may want to save.


 Calling Conventions

 OS/2 1.x uses the Pascal calling convention.

 OS/2 system calls follow the Pascal calling and naming conventions. One way
 to enforce these conventions is to specify PASCAL in the .MODEL directive,
 then use the INVOKE directive to generate the correct code. Another is to
 include the OS2.INC file, which uses the PROTO directive to prototype the
 functions to follow the Pascal conventions. The prototypes specify Pascal as
 the calling convention. OS/2 functions return a value in AX. A nonzero value
 indicates an error. All registers except AX are preserved.

 The OS/2 2.x operating system uses different calling conventions. See the
 documentation provided with that product.


 Exit Code

 To exit an OS/2 program, call the OS/2 system function DosExit. If you use
 the .EXIT directive and the OS_OS2 attribute of the .MODEL statement, the
 assembler automatically generates the proper system call if you have a
 prototype for DosExit.


 Segment Restrictions

 Although OS/2 makes some operations easier, it does impose restrictions on
 the programmer. You cannot do segment arithmetic. That is, you cannot
 attempt to measure the distance between segments by subtracting one segment
 from another. In general, you also cannot add values to segment registers.
 Either operation may cause a protection violation, which would immediately
 terminate the program.

 Under OS/2, segment registers do not hold physical addresses; they hold
 "segment selectors." A segment selector is an index into the system's
 descriptor tables that hold the actual addresses. You can copy the segment
 selector or use it to access data, but you should not try to modify it.

 Huge pointer arithmetic is therefore different under OS/2. Under DOS, you
 can handle huge pointers easily by checking the OVERFLOW? flag after you
 increment or add to an offset address. If the result overflows (exceeds
 64K), then you increment the segment address. Under OS/2, manipulation of
 huge pointers requires special techniques. See your OS/2 documentation for
 more information.


 17.3  A Sample Program

 The following program prints  Hello, world. It runs under OS/2 protected
 mode.

   ; HELLO.ASM
   ;
               .MODEL  small, pascal, OS_OS2
               .286

               INCLUDELIB  os2.lib
               INCLUDE     os2.inc

               .STACK
               .DATA
   message     BYTE    "Hello, world.", 13, 10  ; Message to print
   bytecount   DWORD   ?                        ; Holds number of
                                                ;  bytes written
               .CODE

               .STARTUP
               push        1                  ; Select standard output
               push        ds                 ; Pass address of message
               push        OFFSET message
               push        LENGTHOF message   ; Pass length of message
               push        ds                 ; Pass address of count
               push        OFFSET bytecount   ;  returned by function
               call        DosWrite           ; Call system write
                                              ;  function
               .EXIT       0                  ; Exit with 0 return code
               END

 .STARTUP and .EXIT automatically generate code.

 The .STARTUP and .EXIT directives are very useful because they automatically
 produce correct code for the operating-system type specified with the .MODEL
 directive (see Section 2.2, "Using Simplified Segment Directives"). As
 described in Section 17.6, OS/2 initializes all segment registers;
 therefore, .STARTUP does nothing but indicate the starting point. To
 correctly exit an OS/2 program, you must call the DosExit function. The
 DosExit prototype is always available to MASM programs.

 In the example above, .EXIT automatically generates the following code under
 OS/2:

   .EXIT     0
   0011  6A 01             *   push    +000000001h ; Action 1 ends
   all threads
   0013  6A 00             *   push    +000000000h ; Pass 0 return code
   0015  9A ---- 0000 E    *   call    DosExit     ; Call system function
                                        END

 Between .STARTUP and .EXIT, the entire program consists of a single call to
 the DosWrite function. The program pushes the parameters on the stack and
 then makes the call. No POP or ADD instructions are needed to restore the
 stack after DosWrite returns; DosWrite observes the Pascal calling
 convention and restores the stack itself before returning.

 The .MODEL statement helps ensure that the assembler produces correct code
 for calling DosWrite:

   .MODEL  small, pascal, OS_OS2

 When you run HELLO.EXE, OS/2 looks at the import definitions in the
 executable-file header and makes sure that all needed DLLs are in memory. It
 then loads any needed DLLs not already in memory.

 The assembler must be informed that DosWrite and DosExit are far and observe
 the Pascal calling convention. This information is in the prototype.

 In the call to DosWrite, note that although  OFFSET message  is an immediate
 operand, the program pushes it directly onto the stack. This operation is
 legal on 80186-80486 processors but not on the 8086 or 8088:

   push    OFFSET message

 The processors you want to target determine the instructions you should use.


 Since OS/2 programs can execute only on the 80286 or later processors, it is
 reasonable to use extended operations not supported by the 8086. However, if
 you want to write a program that can be converted to run under both OS/2 and
 DOS (as shown in Section 17.5), then you should write code that can run on
 the 8086. For example,

   mov ax, OFFSET msg
           push    ax

 The following revision of the sample program illustrates the usefulness of
 the INVOKE directive. This version does everything the previous example did
 with far fewer statements:

   ; HELLO.ASM

               .MODEL  small, pascal, OS_OS2

               INCLUDE     os2.inc
               INCLUDELIB  os2.lib

               .STACK
               .DATA
   message     BYTE   "Hello, world.", 13, 10  ; Message to print
   bytecount   DWORD   ?                       ; Holds number of
                                               ;  bytes written
               .CODE
               .STARTUP

   INVOKE  DosWrite,
                       1,
                       ADDR message,
                       LENGTHOF message,
                       ADDR bytecount

               .EXIT 0                   ; Exit with return code 0
               END

 The INVOKE directive generates a call to the given procedure after first
 pushing all other arguments on the stack. Like a call statement in a
 high-level language, the INVOKE directive handles types in a sophisticated
 way.


 17.4  Building an OS/2 Application

 The easiest way to assemble and link the program is from the Programmer's
 WorkBench (PWB). From the Options Menu, select Link Options and choose OS/2
 Application. When you select Build from the Make menu, PWB calls ML and
 LINK, passing the proper options.

 From the command line, type

   ML hello.asm

 The next section discusses how to "bind" the program─that is, convert it so
 that it runs under either DOS or OS/2.


 17.5  Binding OS/2 MASM Programs

 You can convert many OS/2 programs to run under both OS/2 and DOS 3.x. This
 conversion is called "binding" because it binds system calls to the API.LIB
 file provided with MASM 6.0. This file simulates OS/2 functions under DOS.
 The program must use a restricted set of system calls or it cannot be bound.


 OS/2 function calls are known collectively as the applications program
 interface (API). If you restrict your system calls to a subset of these
 functions known as the Family API, the program can be bound. See the
 Microsoft Operating System/2 Programmer's Reference for a list of the Family
 API functions.

 Online help also provides information on these utilities.

 If you use PWB, binding is easy. Select Bound Application from the LINK
 Options command in the Options menu. PWB does the rest, calling the BIND.EXE
 utility.

 If you want to bind the program to run under either OS/2 or DOS, use this
 command line:

   ML /Fb hello.asm

 You can use system calls outside the Family API provided that you never use
 them when running under DOS. The program can check the operating system and,
 if running under OS/2, can execute system calls that do not belong to the
 Family API. To follow this strategy, list OS/2-only calls with the BIND's /N
 option. It is the program's responsibility to make sure these calls are
 never made under DOS; otherwise, execution is terminated.


 17.6  Register and Memory Initialization

 When you execute an OS/2 program, OS/2 stores information about the program
 directly in registers. With DOS programs, the information is kept in a
 separate program segment prefix (PSP). The registers hold these values when
 an OS/2 program begins:

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Register                          Contents at Program Start
 ────────────────────────────────────────────────────────────────────────────
 AX                                Segment address of program's environment

 BX                                Offset of command-line arguments within
                                   the
                                   environment

 CX                                Length of near data area (DGROUP)

 SP                                Offset of the top of the stack within
                                   the stack segment
 Register                          Contents at Program Start
 ────────────────────────────────────────────────────────────────────────────
                                  the stack segment

 CS:IP                             Program's entry point

 DS                                Segment address of near data area
                                   (DGROUP)

 SS                                Segment address of stack



 Note that OS/2 automatically initializes SS:SP correctly. If the .MODEL
 directive specifies FARSTACK, SS is initialized to its own segment address.
 If the model is NEARSTACK, OS/2 sets SS to DGROUP and SP to the top of the
 stack within DGROUP.

 You may want to save the AX, BX, and CX registers at startup.

 Upon start-up, AX, BX, and CX all contain information highly useful to some
 programs. If you want to access the program's command-line arguments or know
 the size of DGROUP, you must save the contents of these registers
 immediately:

   FPBYTE TYPEDEF FAR PTR BYTE

           .DATA

   args    FPBYTE  0
   cmds    FPBYTE  0

           .CODE

   mov WORD PTR args[0], ax ; Save segment of args
           mov WORD PTR args[2], 0  ; Offset is 0
           mov WORD PTR cmds[0], ax ; Save segment of cmds
           mov WORD PTR cmds[2], bx ; Save offset of cmds

 The AX register points to the segment value of the start of the program's
 environment. AX:BX points to the starting address of arguments within the
 environment, the first of which is the program name. This name is followed
 by a null (zero) byte and the command-line arguments exactly as typed at the
 command prompt. A second null marks the end of the arguments.

 If you use simplified segments, .DATA is equivalent to DGROUP.

 Under OS/2, the data segment register, DS, contains the segment of the near
 data area, DGROUP. If you use simplified segment directives, this is the
 .DATA segment. You must place one data segment in a group called DGROUP if
 you do not use the simplified directives:

   _DATA  SEGMENT WORD PUBLIC 'DATA'
          .
          .
          .
   _DATA  ENDS

   DGROUP GROUP _DATA
          ASSUME DS:DGROUP

 Calling the group anything other than DGROUP, or not having a DGROUP, causes
 an error. Only the memory required by the program is allocated by OS/2. This
 means that the system has space in reserve for later memory requests and for
 other programs.


 17.7  Other OS/2 Utilities

 In addition to LINK and BIND, MASM 6.0 provides other utilities useful for
 working with OS/2.


 EXEHDR

 The EXEHDR utility examines and can modify a DOS, Windows, or OS/2
 executable file header. In the case of OS/2 and Windows, EXEHDR reports a
 great deal more information: specifically, it displays the contents of
 segment tables and lists the attributes of the individual segments.


 IMPLIB

 The IMPLIB utility creates an import library that you can use when linking
 with a DLL or group of DLLs. Generally, there are three steps in using a
 DLL:


   1.  Copy the DLL to a directory listed in your CONFIG.SYS LIBPATH setting.

   2.  Run IMPLIB on the DLL to create an import library, or write a
       moduledefinition file.

   3.  Link the import library or module-definition file with any application
       that uses the DLL.


 An import library does not contain executable code but does contain the name
 and location of dynamic-link calls. These calls are resolved during run
 time.

 Chapter 18 goes into more detail about how to write DLLs.


 17.8  Module-Definition Files

 You can create a module-definition file for an application. A
 module-definition file is a text file that contains statements that give
 directions to the linker. These statements can alter the attributes of
 individual segments─for example, whether multiple instances of the program
 share data. Module-definition files are optional. If you use one, begin the
 file with the NAME statement. The following sample module-definition file
 specifies an application,  MYPROG, that shares the  CONSTDAT  segment:

   NAME MYPROG

   SEGMENTS CONSTDAT SHARED


 17.9  Related Topics in Online Help

 In addition to information covered in this chapter, information on the
 following topics can be found in online help:

╓┌─────────────────────────────────┌─────────────────────────────────────────╖
 Topic                             Access
 Topic                             Access
 ────────────────────────────────────────────────────────────────────────────
 BIND                              See the "Microsoft Advisor Contents"
                                   screen

 OS/2 Include files                Choose from the "MASM 6.0 Contents"
                                   screen

 PROTO, INVOKE                     From the "MASM 6.0 Contents" screen,
                                   choose "Directives" and then "Procedure
                                   and Code Labels"

 INCLUDE, INCLUDELIB               From the "MASM 6.0  Contents" screen,
                                   select "Directives" and then
                                   "Miscellaneous Language Directives"

 EXEHDR                            From the "Microsoft Advisor Contents"
                                   screen, select "Miscellaneous" under
                                   "Microsoft Utilities"

 INCL_NOCOMMON                     Select "OS/2 Include Files" from the
 Topic                             Access
 ────────────────────────────────────────────────────────────────────────────
INCL_NOCOMMON                     Select "OS/2 Include Files" from the
                                   "MASM 6.0 Contents" screen; from the
                                   next screen, select "Category Summary"

 CALL                              From the "MASM 6.0 Contents" screen,
                                   choose "Processor Instruction" and then
                                   "Control Flow"

 SHOW.EXE                          From the "MASM 6.0 Contents" screen,
                                   choose "Example Code" and then "SHOW
                                   (Text Viewer)"








 Chapter 18  Creating Dynamic-Link Libraries
 ────────────────────────────────────────────────────────────────────────────

 A "dynamic-link library" (DLL) links to the main program at run time (hence
 the term dynamic link). The program that calls the DLL is known as the
 "client program." One DLL can supply services for several clients
 simultaneously.

 The client program can choose to load the DLL into memory at the same time
 the main program loads, or it can choose to load the DLL only when it is
 needed.

 DLLs are available only in OS/2 and Windows. In non-Windows DOS programs,
 all object modules are statically linked to the program at link time. This
 chapter discusses DLL programming for OS/2 1.x only.

 After an overview of DLLs, this chapter describes the following stages in
 developing a DLL:


   ■   Understanding general DLL programming considerations

   ■   Writing an interface to the DLL's exported procedures and data

   ■   Writing initialization and termination code

   ■   Building the DLL


 The last step requires use of a module-definition file and an import
 library.


 18.1  DLL Overview

 Like a standard (object-code) library, a DLL contains procedures that one or
 more programs can call. Yet unlike standard-library procedures, DLL
 procedures are never copied into an application's executable file. They
 reside only on disk in the DLL file.

 DLLs have several advantages:


   ■   Dynamic link libraries save significant space since the DLL's code and
       data exist in only one place, no matter how many different programs
       call the DLL. Applications that need a particular DLL can share it.

       In contrast, a standard library routine (the printf function in C, for
       example) becomes part of the executable code for each application that
       uses it. For example, if three different programs use the statically
       linked printf function, three copies of the printf code are on disk.
       Furthermore, if all three programs run at once, the printf code occurs
       three times in memory. If the same function were part of a DLL, it
       would exist in only one location on disk and in memory.

   ■   Dynamic linking makes applications and libraries more independent, and
       therefore they are easier to maintain. You can update a DLL without
       having to relink any of the programs that use it.

   ■   Applications link faster because the executable code for a dynamic
       link function is not copied into the application's .EXE file. Instead,
       only an import definition is copied.


 The purpose of a DLL is to supply ("export") procedures and data to client
 programs at run time. Items not exported are visible only within the DLL.

 Exported procedures are visible to the client program.

 The concept of exporting is analogous to the action of the PUBLIC directive,
 but goes further. A public item is available only to other source modules
 within the same program or DLL. An exported item is available to all
 programs running on the system. In addition to global procedures and data, a
 DLL can contain other procedures and data definitions to support the
 operations of exported procedures.

 Finally, a DLL can contain initialization and termination code to allocate
 and release resources needed by the procedures. Resources are typically
 files or dynamic memory. System services for OS/2 and Windows are provided
 through DLLs.


 18.2  DLL Programming Requirements

 Four programming requirements arise from the nature of DLLs. These
 requirements apply to all code used in a dynamic-link call─both in an
 exported procedure and in any procedure it may call:


   ■   You cannot assume that the SS and DS registers hold the same value,
       unless you explicitly set SS equal to DS.

   ■   You should avoid using the math coprocessor or emulator routines
       unless you are certain a coprocessor or emulator library is available.

   ■   The DLL should be "re-entrant," because there is no guarantee that
       only one program will use the DLL. A re-entrant procedure is one that
       can be called by different programs concurrently. This creates
       problems for static data in the DLL, unless you declare data to be
       NONSHARED in the module-definitions file.

   ■   Be careful how you place data and code in segments. The location of
       data and code in different segments and the contents of the
       module-definition file also determine the content of the executable
       file.


 This section discusses these requirements.


 18.2.1  Separate Stack and Data Requirement

 The separate stack and data requirement involves both assembler assumptions
 and coding techniques. If you used the FARSTACK keyword as described in
 Section 18.3.1, "Choosing Module Attributes," the assembler makes correct
 assumptions about the contents of DS and SS.

 Do not assume that SS equals DS.

 In your own code, avoid any optimizing techniques that use SS to access
 items in the data segment or DS to access stack data. For example, the
 following code uses the ASSUME statement to be sure the correct stack is
 accessed:

   ASSUME  DS:DGROUP
           .
           .
           .
           push    ds
           lds     si, sourcead  ; Load DS for string ops
   ASSUME  DS:NOTHING
           .
           .
           .
   ASSUME  SS:STACK
           mov     bx, ss:thing  ; Access near data thing through SS
   ASSUME  SS:NOTHING

 Thread-specific variables can be stored on the stack, as shown in the
 example above.


 18.2.2  Floating-Point Math Requirement

 Don't assume the math coprocessor is available to the DLL.

 A stand-alone DLL─that is, a DLL created for general use by many programs─
 can make few assumptions about the calling program. Therefore, the safest
 way to perform floating-point calculations is to use alternate math
 routines. If you link to a Microsoft high-level language, you can access
 these routines through a language library. These routines give the fastest
 results possible without a coprocessor. See Section 6.3, "Using Emulator
 Libraries," for more information.

 Floating-point operations in DLLs can use a coprocessor or emulator routines
 if you are certain that a coprocessor or emulator libraries are available.


 18.2.3  Re-entrance Requirement

 A procedure may be called by any number of different programs concurrently.
 That is, program A may call a DLL procedure while program B is still
 executing the same procedure. The basic problem of re-entrance is how data
 is shared.

 Be aware that re-entering the DLL can modify its data.

 For example, suppose you have a DLL that contains an accounting package; one
 of the functions adds up an employee's salary for a whole year. First it
 initializes the total to zero; then it increments this total one week at a
 time. While program A is in the middle of this function, program B could
 enter the procedure; its first action would be to initialize the total to
 zero. Control could then pass back to program A, which would then have zero
 total for salary. The problem is that two instances of the DLL share the
 same variable for totals.

 A procedure in a DLL must therefore follow this rule: it can access static
 data items but must not alter them. Otherwise, one instance of a procedure
 could corrupt data relied on by another instance of the procedure.

 There are several exceptions to this rule. First, if data is declared
 NONSHARED in the module-definitions file, each instance has its own copy of
 the data segment, and there is no conflict. Second, you can use semaphores
 to allow mutually exclusive access to data items. Finally, there may be some
 items you deliberately want all instances to alter─such as a global counter
 to keep track of number of instances.

 Section 18.4.1, "Writing the Module-Definition File," explains how to
 declare some data items as SHARED while declaring others to be NONSHARED.


 18.2.4  Segment Strategy in a DLL

 Be careful how you place different kinds of data and code in different
 segments. When loading the DLL, OS/2 checks to see if the DLL is already in
 memory. If so, it loads only new copies of NONSHARED segments; it does not
 reload SHARED segments. Code segments are always SHARED.

 Control of DLL data and code works at the segment level. The DATA statement
 assigns default attributes for all data segments in the DLL, but the
 moduledefinition SEGMENTS statement overrides these attributes for any given
 segment.

 You may want to create a DLL that has some data shared between all programs
 that call the DLL and some data that is private to each instance. The
 following module-definition statement specifies that all data in  GLOBDAT
 is shared and all data in  PRIVDAT  is not:

   SEGMENTS
       GLOBDAT SHARED 'data'
       PRIVDAT NONSHARED 'data'

 The segments have class `code' unless you specifically define the class as
 shown in this example. See Section 18.4.1 for more information on
 module-definition files.


 18.3  Writing the DLL Code

 When you write the code for the DLL module, you need to select the correct
 module attributes, define the procedures and data in your DLL, and write the
 initialization and termination code. This section discusses these tasks.


 18.3.1  Choosing Module Attributes

 As noted in Chapter 2, there are four fields for the .MODEL directive:
 memory model, language type, operating system, and stack type. When you
 write a DLL, you can choose the attributes you would normally use for the
 first two fields. OS/2 system calls use the Pascal calling convention, so
 you may find it convenient to make all your modules use this convention as
 well.

 DLLs use the OS_OS2 and FARSTACK attributes.

 The operating system and stack fields should be OS_OS2 and FARSTACK,
 respectively. You should use the NEARSTACK attribute only if you switch
 execution to your own stack.

 A usable declaration is therefore

   .MODEL  large, pascal, os_os2, farstack

 If you are using full segment definitions, remember to generate an ASSUME
 directive for DS but not for SS.

   ASSUME  DS:DGROUP ; Necessary with full segment definitions


 18.3.2  Defining Procedures and Data

 Procedures and data in DLLs can be either global (available to the client
 process) or local (used only by the DLL). To create a global data item, make
 sure that it is public:

   EXTERNDEF   dllvar
               .DATA
   dllvar      WORD    0

 The variable must then be exported in a module-definition file, as shown in
 Section 18.4.1, "Writing the Module-Definition File." When executable files
 other than the DLL access the variable, they must treat it as far data, as
 in the following example:

   mov     ax, SEG dllvar
   mov     es, ax
   mov     bx, es:dllvar

 An exported procedure (often called a dynamic-link procedure) must follow
 these rules:


   ■   It must be declared far and public. The MASM keyword EXPORT does both
       of these.

   ■   The procedure should initialize DS upon entry (unless you are not
       going to be accessing any static near data).

   ■   Data pointers in the parameter list should be far.


 The easiest way to realize most of these requirements is to use the EXPORT
 keyword and LOADDS in the procedure's prologuearg list (see Section 7.3.8).
 LOADDS generates instructions to save DS and load it with the value of the
 DLL's data segment. The EXPORT keyword makes the procedure FAR and PUBLIC,
 overriding the memory model. You may also need to use FORCEFRAME, which
 instructs the assembler to generate a stack frame even if there are no
 parameters or locals.

 The example DLL used in the chapter, CSTR.DLL, illustrates how DLLs can be
 shared by several processes. The procedures in the DLL write a string and
 keep track of the number of times the string is written. When more than one
 process uses the DLL, they all increment the global variable  GCount, but
 each process increments its own private instance of the  PCount  variable.

 The only initialization code this DLL needs is code to set up the exit code.
 The next section shows how to write a module-definition file to create an
 import library and how to create a DLL from this code.

 The code for the CSTR.DLL example looks like this:

   .MODEL  small, pascal, os_os2, farstack
           .286

   INCL_NOCOMMON   EQU     1
   INCL_DOSPROCESS EQU     1
   INCL_VIO        EQU     1

           INCLUDE    OS2.INC
           INCLUDELIB OS2.LIB

           .DOSSEG

   VioWrtCStr PROTO FAR PASCAL, pchString:PCH, hv:HVIO
   GetGCount  PROTO     PASCAL
   GetPCount  PROTO     PASCAL
   CStrExit   PROTO FAR

           .STACK
           .DATA               ; Default segment is SHARED

   GCount  WORD    0           ; Count of all calls

   @CurSeg ENDS

   PRIVDAT SEGMENT WORD     ; Private segment is NONSHARED

   PCount  WORD    0           ; Count of all this process
                               ;  calls to VioWrtCStr
   PRIVDAT ENDS

           .CODE
           .STARTUP

       pusha

           ; Initialization goes here. In this case, the only
           ;  initialization is setting up the exit behavior.

           INVOKE  DosExitList, EXLST_ADD, CStrExit
           INVOKE  DosExitList, EXLST_EXIT,0

           popa
           retf

   VioWrtCStr PROC FAR PASCAL EXPORT <LOADDS> USES cx di si,
           pchString:PCH,
           hv:HVIO

           sub     al, al         ; Search for zero
           mov     cx, 0FFFFh     ; Set maximum length
           les     di, pchString  ; Load pointer
           mov     si, di         ; Copy it
           repne   scasb          ; Find null
           .IF     zero?          ; Continue if found
           sub     di, si         ; Calculate length
           xchg    di, si         ; Restore address and save length

           INVOKE  VioWrtTTy,     ; Let OS/2 do output
                   es:di,         ; Address of string
                   si,            ; Calculated length
                   hv             ; Video handle

           inc     GCount         ; Count as one of total calls

   ASSUME  DS:PRIVDAT
           mov     ax, PRIVDAT
           mov     ds, ax
           inc     PCount         ; Count as one of process calls
           ASSUME  DS:DGROUP
           sub     ax, ax         ; Success
           .ELSE
           mov     ax, 1          ; Error
           .ENDIF
           ret
   VioWrtCStr ENDP

   GetGCount PROC  FAR PASCAL EXPORT <LOADDS, FORCEFRAME>
           mov     ax, GCount
           ret
   GetGCount ENDP

   GetPCount PROC  FAR PASCAL EXPORT <LOADDS, FORCEFRAME> USES ds
           ASSUME  DS:PRIVDAT
           mov     ax, PRIVDAT
           mov     ds, ax
           mov     ax, PCount
           ASSUME  DS:NOTHING
           ret
   GetPCount ENDP

           .DATA
   szOut   BYTE    13, 10, "Exiting DLL...", 13, 10, 0
           .CODE

   CStrExit PROC   FAR  <LOADDS, FORCEFRAME>
           INVOKE  VioWrtCStr,
                   ADDR szOut,
                   0
           INVOKE  DosExitList, EXLST_EXIT, 0
   CStrExit ENDP

           END

 These generated code for the  VIOWrtCStr  procedure follows. The code marked
 with asterisks is generated by the assembler.

   VioWrtCStr PROC FAR PASCAL EXPORT <LOADDS> USES cx di si,
           pchString:PCH,
           hv:HVIO
    0000  55        *         push   bp
    0001  8B EC     *         mov    bp, sp
    0003  1E        *         push   ds
    0004  B8 ---- R *         mov    ax, DGROUP
    0007  8E D8     *         mov    ds, ax
    0009  51        *         push   cx
    000A  57        *         push   di
    000B  56        *         push   si
           .
           .                      ; Procedure code here
           .
                              ret
    000C  5E        *         pop    si
    000D  5F        *         pop    di
    000E  59        *         pop    cx
    000F  1F        *         pop    ds
    0010  C9        *         leave
    0011  CA 0006   *         ret    00006h
    0014               VioWrtCStr ENDP

 The DLL should establish its own data segment.

 The DLL should change DS in this manner because each client program has its
 own private version of DGROUP. When a program calls your dynamic-link
 procedure, DS points to the program's data area, not yours. The solution is
 to initialize DS so that it points to your own default data area.

 However, one side effect of this approach is that it alters DS so that it no
 longer is equal to SS. Consequently, all data pointers in the parameter list
 must be far pointers, even if the data was stack data or near data.


 18.3.3  Creating Initialization and Termination Code

 Begin initialization code with the .STARTUP directive.

 A DLL can contain procedures that require special resources, such as
 temporary files or dynamic memory blocks. Resources allocated during
 initialization exist for the lifetime of the client program and are removed
 when the client program exits. Usually the best method for managing these
 resources is to write initialization and termination code.

 A DLL can have a starting point just as an application does. In the case of
 a DLL, this starting point marks the beginning of the initialization code. A
 DLL does not need a starting point if it has no need for initialization. Do
 not use .EXIT, since .EXIT will terminate the client program.

 Attributes of the initialization code are defined in the module-definition
 file (see Section 18.4.1). Initialization code can have the INITGLOBAL or
 INITINSTANCE attribute.

 INITGLOBAL specifies that the initialization code executes only once─when
 the DLL is first loaded into memory. INITINSTANCE specifies that
 initialization code should execute once for each program that uses the DLL.
 INITGLOBAL is the default. You should use termination code only for DLLs
 that have been defined with INITINSTANCE unless you know that the first
 process to use the DLL is the last to terminate.

 To specify INITINSTANCE, place the LIBRARY statement in your
 moduledefinition file:

   LIBRARY CSTR INITINSTANCE

 In the statement above, CSTR is the name of the DLL.

 To include a termination procedure, invoke DosExitList in the initialization
 code. DosExitList is a system function that attaches a termination procedure
 to a program. When the program terminates, OS/2 executes the procedure as
 part of the program exit sequence. In the termination procedure itself,
 release any system resources (such as memory or files) allocated during
 initialization.

 This is the termination code for the CSTR.DLL module:

   CStrExit PROC   FAR  <LOADDS, FORCEFRAME>
           INVOKE  VioWrtCStr,
                   ADDR szOut,
                   0
           INVOKE  DosExitList, EXLST_EXIT, 0
   CStrExit ENDP

 The termination code in CSTR.DLL uses the INVOKE directive to set up a call
 to the DosExitList function. You can perform a similar operation by simply
 pushing arguments on the stack and observing the correct calling convention.


 The effect of DosExitList in the initialization code is to make OS/2 call
 the termination procedure when the current process exits. The "current
 process" in this case is the client program, not the DLL or the DLL
 initialization code.


 18.4  Building the DLL

 To create a DLL, you need to assemble the DLL code, write a
 module-definition file, use LINK to create the DLL, generate an import
 library, and then link the DLL to the client program.


 18.4.1  Writing the Module-Definition File

 A module-definition file is required for DLLs.

 The module-definition file is an ASCII text file that lists attributes of a
 library or application (in the case of an application, this file is
 optional). The moduledefinition file gives directions to the linker that
 supplement the information on the command line.

 This module-definition file tells the linker to create a DLL called CSTR.DLL
 with INITINSTANCE data. The library has exported procedure  VioWrtCStr,
 GetPCount, and  GetGCount, and the data segment  PRIVDAT  is not shared
 between programs:

   LIBRARY CSTR INITINSTANCE

   EXPORTS
       VioWrtCstr
       GetGCount
       GetPCount

   DATA SINGLE NONSHARED

 The LIBRARY statement need not specify a name. If the name is omitted, the
 linker gives the library the base filename of the module-definition file.
 The default file extension is .DLL. The INITINSTANCE attribute is optional
 and is significant only if you have initialization code. If you specify
 INITINSTANCE, then the library initialization is called each time a new
 process gains access to the library. Otherwise, it will be called once only.


  At least one procedure must be listed after EXPORTS.

 The EXPORTS statement lists identifiers (procedures and variables) that can
 be accessed directly by client programs. Note that if you give a procedure
 the EXPORTS attribute from within the source code, you do not need to list
 the procedure here. The EXPORTS keyword automatically exports the procedure
 by name, so putting the names of the procedures in the module-definition
 file is not required. However, exported variables must be listed in a
 module-definition file.

 The DATA statement lists attributes for data segments (DGROUP) in the DLL.
 The default for DLLs, SINGLE, specifies that one DGROUP is shared by all
 instances of the DLL. NONSHARED specifies that all other data segments are
 not to be shared. See Section 13.15, "CODE, DATA, and SEGMENTS Attributes."



 18.4.2  Generating an Import Library with IMPLIB

 The DLL exports a procedure; the client program imports it.

 Just as a procedure is exported by a DLL, it must be imported by an
 application. An application's EXE header must indicate what dynamic-link
 procedures are used and where they reside. The easiest way to specify this
 information is with an "import library," which is a .LIB file that contains
 the import information in object-record form. The IMPLIB utility automates
 this process for you.

 To create an import library, run the IMPLIB utility on the module-definition
 file:

   IMPLIB MYDYNLIB.LIB MYDYNLIB.DEF

 The result is the import library, MYDYNLIB.LIB, which you then link to any
 program that calls CSTR.DLL. You would then list MYDYNLIB.LIB in the
 libraries field (the fourth field) of the LINK command. Or, in
 assembly-language programs, you can link to this library automatically by
 just adding the following statement to the source code of your program:

   INCLUDELIB MYDYNLIB.LIB


 18.4.3  Creating and Using the DLL

 Now you can use LINK to create the DLL. The LINK utility uses the object
 module of the DLL code and the module definition to create the CSTR.DLL:

   LINK CSTR.OBJ , , , , MYDYNLIB.DEF

 If linking is successful, the linker creates a file with a .DLL extension.

 You can link several modules together to create a DLL. The following command
 line links several object modules and an object-code library (BIGLIB.LIB) to
 form a DLL. The module-definition file is MYDYNLIB.DEF:

   LINK  MOD1 MOD2 MOD3,,, BIGLIB, MYDYNLIB

 To use the DLL, copy the .DLL file to a directory listed in the LIBPATH
 setting in your CONFIG.SYS file.

 To create an executable file using the DLL, link the client program with the
 import library as shown:

   LINK CALLDLL.OBJ , , , MYDYNLIB.LIB

 By running CALLDLL.EXE in separate OS/2 windows, you can see that both
 client programs access the DDL at the same time. When the last process exits
 the DLL, the DLL is removed from memory.


 18.5  Related Topics in Online Help

 In addition to information covered in this chapter, information on the
 following topics can be found in online help.

 Topic                             Access
 ────────────────────────────────────────────────────────────────────────────
 LINK                              From the "Microsoft Advisor Contents"
                                   screen, select LINK

 Module-definition files           Select Module-Definition Files from the
                                   "LINK Contents" screen

 EXPORT                            Select from the MASM Language Index

 EXTERNDEF                         From the "MASM 6.0 Contents" screen,
                                   select "Directives"; then select "Scope
                                   and Visibility" from the next screen

 LOADDS,                           Choose "Proc" from the MASM Language
 FORCEFRAME                        Index

 IMPLIB                            Select "IMPLIB Summary" from the "LINK
                                   Contents" screen






 Chapter 19  Writing Memory-Resident Software
 ────────────────────────────────────────────────────────────────────────────

 Through its memory-management system, DOS allows a program to remain
 resident in memory after terminating. The resident program can later regain
 control of the processor to perform tasks such as background printing or
 "popping up" a calculator on the screen. Such a program is commonly called a
 TSR, from the Terminate-and-Stay-Resident function it uses to return to DOS.


 This chapter explains the techniques of writing memory-resident software.
 The first two sections present introductory material. Following sections
 describe important DOS and BIOS interrupts and focus on how to write safe,
 compatible, memory-resident software. Two example programs illustrate the
 techniques described in the chapter. These programs are also available as
 sample programs on the MASM 6.0 disks.


 19.1  Terminate-and-Stay-Resident Programs

 DOS maintains a pointer to the beginning of unused memory. Programs load
 into memory at this position. They terminate execution by returning control
 to DOS. Normally, the pointer remains unchanged, allowing DOS to reuse
 memory when loading other programs.

 A terminating program can, however, prevent other programs from loading on
 top of it. It does this by returning to DOS through the
 terminate-and-stay-resident function, which resets the free-memory pointer
 to a higher position. This leaves the program resident in a protected block
 of memory, even though it is no longer running.

 The terminate-and-stay-resident function (Function 31h) is one of the DOS
 services invoked through Interrupt 21h. The following fragment shows how a
 TSR program terminates using Function 31h and remains resident in a
 1000h-byte block of memory:

   mov     ah, 31h          ; Request DOS Function 31h
           mov     al, err          ; Set return code
           mov     dx, 100h         ; Reserve 100h paragraphs
                                    ;  (1000h bytes)
           int     21h              ; Terminate-and-stay-resident

 ────────────────────────────────────────────────────────────────────────────
 NOTE

 In current versions of DOS, Interrupt 27h also provides a
 terminate-and-stayresident service. However, Microsoft cannot guarantee
 future support for Interrupt 27h and does not recommend its use.
 ────────────────────────────────────────────────────────────────────────────


 19.1.1  Structure of a TSR

 TSRs consist of two distinct parts that execute at different times. The
 first part is the installation section, which executes only once, when DOS
 loads the program. The installation code performs any initialization tasks
 required by the TSR and then exits through the terminate-and-stay-resident
 function.

 A TSR consists of an installation section and a resident section.

 The second part of the TSR, called the resident section, consists of code
 and data left in memory after termination. Though often identified with the
 TSR itself, the resident section makes up only part of the entire program.

 The TSR's resident code must be able to regain control of the processor and
 execute after the program has terminated. Methods of executing a TSR are
 classified as either passive or active.


 19.1.2  Passive TSRs

 The simplest way to execute a TSR is to transfer control to it explicitly
 from another program. Because the TSR in this case does not solicit
 processor control, it is said to be passive. If the calling program can
 determine the TSR's memory address, it can grant control via a far jump or
 call. More commonly, a program activates a passive TSR through a software
 interrupt. The installation section of the TSR writes the address of its
 resident code to the proper position in the interrupt vector table (see
 Section 7.4, "DOS Interrupts"). Any subsequent program can then execute the
 TSR by calling the interrupt.

 Passive TSRs often replace existing software interrupts. For example, a
 passive TSR might replace Interrupt 10h, the BIOS video service. By
 intercepting calls that read or write to the screen, the TSR can access the
 video buffer directly, increasing display speed.

 Passive TSRs allow limited access since they can be invoked only from
 another program. They have the advantage of executing within the context of
 the calling program, and thus run no risk of interfering with another
 process, which could happen with active TSRs.


 19.1.3  Active TSRs

 The second method of executing a TSR involves signaling it through some
 hardware event, such as a predetermined sequence of keystrokes. This type of
 TSR is called active because it must continually search for its start-up
 signal. The advantage of active TSRs lies in their accessibility. They can
 take control from any running application, execute, and return, all on
 demand.

 An active TSR, however, must not seize processor control blindly. It must
 contain additional code that determines the proper moment at which to
 execute. The extra code consists of one or more routines called "interrupt
 handlers," described in the following section.


 19.2  Interrupt Handlers in Active TSRs

 The memory-resident portion of an active TSR consists of two parts. One part
 contains the body of the TSR─the code and data that perform the program's
 main tasks. The other part contains the TSR's interrupt handlers.

 An interrupt handler is a routine that takes control when a specific
 interrupt occurs. Although sometimes called an "interrupt service routine,"
 a TSR's handler usually does not service the interrupt. Instead, it passes
 control to the original interrupt routine, which does the actual interrupt
 servicing.

 Collectively, interrupt handlers ensure that a TSR operates compatibly with
 the rest of the system. Individually, each handler fulfills at least one of
 the following functions:


   ■   Auditing hardware events that may signal a request for the TSR

   ■   Monitoring system status

   ■   Determining whether a request for the TSR should be honored, based on
       current system status



 19.2.1  Auditing Hardware Events for TSR Requests

 Active TSRs commonly use a special keystroke sequence or the timer as a
 request signal. A TSR invoked through one of these channels must be equipped
 with handlers that audit keyboard or timer events.

 A keyboard handler receives control at every keystroke. It examines each
 key, searching for the proper signal or "hot key." Generally, a keyboard
 handler should not attempt to call the TSR directly when it detects the hot
 key. If the TSR cannot safely interrupt the current process at that moment,
 the keyboard handler is forced to exit to allow the process to continue.
 Since the handler cannot regain control until the next keystroke, the user
 has to press the hot key repeatedly until the handler can comply with the
 request.

 Instead, the handler should merely set a request flag when it detects a
 hot-key signal and then exit normally. Examples in the following paragraphs
 illustrate this technique.

 For computers other than the IBM PS/2 (R) series, an active TSR audits
 keystrokes through a handler for Interrupt 09, the keyboard interrupt:

   Keybrd  PROC    FAR
           sti                   ; Interrupts are okay
           push    ax            ; Save AX register
           in      al, 60h       ; AL = scan code of current key
           call    CheckHotKey   ; Check for hot key
           .IF     !carry?       ; If not hot key:
   ; Hot key pressed. Reset the keyboard to throw away keystroke.
           cli                   ; Disable interrupts while resetting
           in      al, 61h       ; Get current port 61h state
           or      al, 10000000y ; Turn on bit 7 to signal clear keybrd
           out     61h, al       ; Send to port
           and     al, 01111111y ; Turn off bit 7 to signal break
           out     61h, al       ; Send to port
           mov     al, 20h       ; Reset interrupt controller
           out     20h, al
           sti                   ; Reenable interrupts
           pop     ax            ; Recover AX
           mov     cs:TsrRequestFlag, TRUE ; Raise request flag
           iret                  ; Exit interrupt handler
           .ENDIF                ; End hot-key check

   ; No hot key was pressed, so let normal Int 09 service
   ; routine take over

           pop     ax            ; Recover AX and fall through
           cli                   ; Interrupts cleared for service
   KeybrdMonitor LABEL FAR       ; Installed as Int 09 handler for
                                 ;  PS/2 or for time-activated TSR
                                 ; Signal that interrupt is busy
           mov     cs:intKeybrd.Flag, TRUE
           pushf                 ; Simulate interrupt by pushing flags,
                                 ;  far-calling old Int 09 routine
           call    cs:intKeybrd.OldHand
           mov     cs:intKeybrd.Flag, FALSE
           iret
   Keybrd  ENDP

 A TSR running on a PS/2 computer cannot reliably read key-scan codes using
 the above method. Instead, the TSR must search for its hot key through a
 handler for Interrupt 15h (Miscellaneous System Services). The handler
 determines the current keypress from the AL register when AH equals 4Fh, as
 shown here:

   MiscServ PROC   FAR
           sti                   ; Interrupts okay
           .IF     ah == 4Fh     ; If Keyboard Intercept Service:
           push    ax            ; Preserve AX
           call    CheckHotKey   ; Check for hot key
           pop     ax
           .IF     !carry?       ; If hot key:
           mov     cs:TsrRequestFlag, TRUE ; Raise request flag
           clc                   ; Signal BIOS not to process the key
           ret     2             ; Simulate IRET without popping flags
           .ENDIF                ; End carry flag check
           .ENDIF                ; End Keyboard Intercept check
           cli                   ; Disable interrupts and fall through
   SkipMiscServ LABEL FAR        ; Interrupt 15h handler if PC/AT
           jmp     cs:intMisc.OldHand
   MiscServ ENDP

 The example program in Section 19.8 demonstrates how a TSR tests for a PS/2
 machine and then sets up a handler for either Interrupt 09 or Interrupt 15h
 to audit keystrokes.

 Setting a request flag in the keyboard handler allows other code, such as
 the timer handler (Interrupt 08), to recognize a request for the TSR. The
 timer handler gains control at every timer interrupt; the interrupts occur
 an average of 18.2 times per second. The following fragment shows how a
 timer handler tests the request flag and continually polls until it can
 safely execute the TSR.

   TestFlag PROC FAR
           .
           .
           .
           cmp     TsrRequestFlag, FALSE ; Has TSR been requested?
           je      exit                  ; If not, exit
           call    CheckSystem           ; Can system be interrupted
                                         ;  safely?
           jc      exit                  ; If not, exit
           call    ActivateTsr           ; If okay, call TSR

 Figure 19.1 illustrates the process. It shows a time line for a typical TSR
 signaled from the keyboard. When the keyboard handler detects the proper hot
 key, it sets a request flag called  TsrRequestFlag. Thereafter, the timer
 handler continually checks the system status until it can safely call the
 TSR.

 The timer itself can serve as the start-up signal if the TSR executes
 periodically. Screen clocks that continuously show seconds and minutes are
 examples of TSRs that use the timer this way. ALARM.ASM, a program described
 in the next section, shows another example of a timer-driven TSR.

 (This figure may be found in the printed book.)


 19.2.2  Monitoring System Status

 A TSR that uses a hardware device such as the video or disk must not
 interrupt while the device is active. A TSR monitors a device by handling
 the device's interrupt. Each interrupt handler need only set a flag to
 indicate that the device is in use and then clear the flag when the
 interrupt finishes.

 The following shows a typical monitor handler:

   NewHandler PROC    FAR
           mov     ActiveFlag, TRUE  ; Set active flag
           pushf                     ; Simulate interrupt by
                                     ;  pushing flags,
           call    OldHandler        ;  then calling original routine
           mov     ActiveFlag, FALSE ; Clear active flag
           iret                      ; Return from interrupt
   NewHandler ENDP

 Only hardware used by the TSR requires monitoring. For example, a TSR that
 performs disk input/output (I/O) must monitor disk use through Interrupt
 13h. The disk handler sets an active flag that prevents the TSR from
 executing during a read or write operation. Otherwise, the TSR's own I/O
 would move the disk head. This would cause the suspended disk operation to
 continue with the head incorrectly positioned when the TSR returned control
 to the interrupted program.

 In the same way, an active TSR that displays to the screen must monitor
 calls to Interrupt 10h. The Interrupt 10h BIOS routine does not protect
 critical sections of code that program the video controller. The TSR must
 therefore ensure that it does not interrupt such nonreentrant operations.

 The activities of the operating system also affect the system status. With
 few exceptions, DOS functions are not reentrant and must not be interrupted.
 However, monitoring DOS is somewhat more complicated than monitoring
 hardware. Discussion of this subject is deferred until Section 19.4.

 The following comments describe the chain of events depicted in Figure 19.1.
 Each comment refers to one of the numbered pointers in the figure.


   1.  At time = t, the timer handler activates. It finds the flag
       TsrRequestFlag  clear, indicating that the TSR has not been requested.
       The handler terminates without taking further action. Notice that
       Interrupt 13h is currently processing a disk I/O operation.

   2.  Before the next timer interrupt, the keyboard handler detects the hot
       key, signalling a request for the TSR. The handler sets
       TsrRequestFlag  and returns.

   3.  At time = t + 1/18 second, the timer handler again activates and finds
        TsrRequestFlag  set. The handler checks other active flags to
       determine if the TSR can safely execute. Since Interrupt 13h has not
       yet completed its disk operation, the timer handler finds
       DiskActiveFlag  set. The handler therefore terminates without invoking
       the TSR.

   4.  At time = t + 2/18 second, the timer handler again finds
       TsrRequestFlag set and repeats its scan of the active flags.
       DiskActiveFlag  is now clear, but in the interim, Interrupt 10h has
       activated as indicated by the flag  VideoActiveFlag. The timer handler
       accordingly terminates without invoking the TSR.

   5.  At time = t + 3/18 second, the timer handler repeats the process. This
       time it finds all active flags clear, indicating that the TSR may
       safely execute. The timer handler calls the TSR, which sets its own
       active flag to ensure that it will not interrupt itself if requested
       again.

   6.  The timer and other interrupts continue to function normally while the
       TSR executes.



 19.2.3  Determining Whether to Invoke the TSR

 Once a handler receives a request signal for the TSR, it checks the various
 active flags maintained by the handlers that monitor system status. If any
 of the flags are set, the handler ignores the request and exits. If the
 flags are clear, the handler invokes the TSR, usually through a near or far
 call. Figure 19.1 illustrates how a timer handler detects a request and then
 periodically scans various active flags until all the flags are clear.

 A TSR that changes stacks must not interrupt itself. Otherwise, the second
 execution would overwrite the stack data belonging to the first. A TSR
 prevents this by setting its own active flag before executing, as shown in
 Figure 19.1. A handler must check this flag along with the other active
 flags when determining whether the TSR can safely execute.


 19.3  Example of a Simple TSR: ALARM

 This section presents a simple alarm clock TSR that demonstrates some of the
 material covered so far. The program accepts an argument from the command
 line that specifies the alarm setting in military form, such as 1635 for
 4:35 P.M. For the sake of simplicity, the argument must consist of four
 digits, including leading zeros. To set the alarm at 7:45 A.M., for example,
 enter:

   ALARM 0745

 The installation section of the program begins with the  Install  procedure.
  Install  computes the number of five-second intervals that must elapse
 before the alarm sounds and stores this number in the word  CountDown. The
 procedure then obtains the vector for Interrupt 08 (timer) through Interrupt
 21h Function 35h and stores it in the far pointer  OldTimer. Interrupt 21h
 Function 25h replaces the vector with the far address of the new timer
 handler  NewTimer. Once installed, the new timer handler executes at every
 timer interrupt. These interrupts occur 18.2 times per second or 91 times
 every five seconds.

 Each time it executes, NewTimer  subtracts one from a secondary counter
 called  Tick91. By counting 91 timer ticks,  Tick91  accurately measures a
 period of five seconds. When  Tick91  reaches zero, it's reset to 91 and
 CountDown  is decremented by one. When  CountDown  reaches zero, the alarm