Towards an UNIX/POSIX to emx-OS/2 Porting Guide

Use the Imakefile!
or:
Some mean tricks to get the stuff compiled!

Copyright ⌐ for the compilation:
Arnd.H.Hanses@rz.ruhr-uni-bochum.de

"Good programmers know what to write.
Great ones know what to rewrite (and reuse)."
(ERIC S. RAYMOND, IN: "THE CATHEDRAL AND THE BAZAAR")


Alfieri: > E la fama? <
Gozzi: > E la fame? <
(COMMEDIA DELL'ARTE)

(Authorship is usually indicated. Please mail me any errors and suggested additions. I am looking forward to adding new chapters written by interested readers. If the author's name is omitted and not evident, I am citing usually an excerpt from the emx documentation. Those hundreds of trademarks are those of their respective owners.)

Contents

0.12

"I have found a cool X utility which I want to port. How should I proceed?
Normally, with X utilities, you get an Imakefile. Simply proceed as if you were under Unix. Run xmkmf -a to compile. Read the programming section on special issues."

("The XFree86/OS2 FAQ")

Preface
(What should be the aim of an emx-tutorial?)

Since I am not a professional programmer, I am advocating user friendly tools and user friendly documentation. I do not think such helpful things should aim at becoming large or comprehensive, but simple and down-to-earth. It should just enable the first simple steps and help to reduce the beginners problems, thus providing the necessary basis and some pointers so that you learn how to get access to the huge piles of helpful information on the Net and elsewhere. Well, it is time now to make the first step towards a (still epigonous and rudimentary) porting tutorial, compiling in the form of short articles some of the helpful information that's scattered in small bits and pieces everywhere (while trying to identify and name the author). Everybody may contribute such a small article for inclusion.

It is best used together with Alexander Mai's Unix to OS/2 Porting FAQ¹. If his concise notes are too short for your purposes, some explanation may be found here.

The first chapter 'Instead of an Introduction' is, what a novice should read. Those who want to use imake should read the respective chapter, etc.

Part 1
The Tutorial

Chapter 1
Instead of an Introduction

1.1  Why Are the Emx Development Tools Important for Writing Modern Portable Code ?

The number of new compilers for OS/2, which provide all (now standardized or still experimental) features of bleeding edge STL versions is very limited. 'Workframe' solutions fail, when a huge codebase with hundreds of megabytes of source is involved and a reasonable make process (i.e. based on Unix' 20 years of scientific and industrial experience with complex projects for heterogenous platforms and distributed networking, using quality software engineering strategies for code reliability) is required for any serious development. A prominent example is XFree86/OS2, where all make utilities and development tools except the GNU make utility and the GNU compiler gcc failed. Both are based on the emx libraries:

1.1.1  What's emx?

[excerpt from emx 0.9d INTRODUCTION 21-Dec-1998]

``Introduction

Welcome to emx 0.9d, an environment for creating 32-bit programs for OS/2 and DOS. You can use the GNU C compiler to compile programs for emx. The main design goal of emx is to simplify porting Unix software to OS/2 and DOS. Moreover, you can create `native' OS/2 programs, including Presentation Manager applications.

The emx package includes:

• emx.dll, emxio.dll, emxwrap.dll and emxlibc.dll dynamic link libraries for OS/2
• emx.exe DOS extender for running 32-bit programs under DOS
• emxfpemu, W. Metzenthen's FPU (387) emulator
• emxbind.exe for creating .exe files which work both under OS/2 and DOS
• emxomf.exe for converting .o object files (a.out) to .obj object files (OMF). With emxomf and LINK386 you can create OS/2 programs that don't need emx.dll
• emximp.exe for creating import libraries
• emxexp.exe for creating a list of export definitions from object files
• C header files and a complete C library
• a simple graphics library for 320x200 (256 colors) mode
• a library for screen output in windows
• header files and a library for using sockets of IBM TCP/IP for OS/2

Additionally, the following GNU programs are available compiled and with patches and sources:

• GCC 2.8.1, the GNU C compiler, supporting the C, C++, and Objective C languages
• GAS 2.6, the GNU assembler
• GDB 4.16, the GNU debugger
• ld, an ancient version of the GNU linker hacked for emx
• strip: a member of an ancient version of the GNU binary utilties
• ar, nm, size, objdump: a subset of the GNU binary utilities 2.6
• gprof, the BSD/GNU profiler
• texinfo, the GNU documentation system
• Patches for the GNU sources
• Patched source for GCC, GAS, GDB, ld, ar, nm, size, strip, objdump, info, makeinfo, texindex. You can compile all these programs with the files that come with emx (but you also need a make utility, such as dmake).

You can get the complete unpatched GNU sources by anonymous ftp from prep.ai.mit.edu and other archives such as ftp.leo.org.

Additionally, the following libraries are provided:

• some parts of the BSD C library
• the BSD data base library: dbopen()
• BSD curses library
• Henry Spencer's regexp library
• The GNU and BSD termcap libraries [...]

After unpacking the emx runtime and development system packages, you'll find further information in the following files:

• emxdocCOPYING.EMX emx & library license, license overview
• emxdocCOPYING GNU General Public License
• emxdocinstall.doc Installation guide
• emxdocemxrt.doc User's guide to the emx runtime package emxdocemxdev.doc Application developer's guide
• emxdocemxlib.doc C library reference
• emxdocfaq.doc Frequently asked questions (with answers)
• emxdocbuild.doc Compiling the sources
• emxdochistory.doc Change log [...]

After unpacking the GNU and BSD packages, you'll find further information in the following files: [...]

• emxdocemxgnu.doc GNU development tools information
• emxdocemxbsd.doc BSD library information [...]

emx is available for anonymous ftp on

• ftp.leo.org: /pub/comp/os/os2/leo/gnu/emx+gcc/ 
• ftp-os2.cdrom.com: /pub/os2/emx09d/ 
• ftp-os2.nmsu.edu: /pub/os2/dev/emx/v0.9d/ [...]

An emx-related mailing list has been created. The address for people to request to be added to or removed from the list is:

majordomo@iaehv.nl

To subscribe, send a message containing

subscribe emx

to majordomo@iaehv.nl. [...]

Highly optimized and bleeding edge C++-code can be compiled with the most recent experimental GNU Pentium Compiler Group's pgcc, which also contains tools updates, interesting for assembler (MMX) programmers.

1.1.2  What's That Hype About ``Code Reuse''?

Although not completely POSIX compliant, emx for OS/2 is nearly as close as can be for the pre-POSIX operating system OS/2. This makes it an ideal platform for CODE REUSE and OPEN PORTABILITY.

" X - a portable network transparent window system", as the authors call it, enables the reuse of Unix software for OS/2 and in general gives you the chance to write extremely portable gui code, running on a PC as well as on a super-computer by simply recompiling it, provided you have a POSIX-like programming interface.

"Why is writing portable code (and therefore emx) so important?"

"As for non-existing API, I would recommend to stick to well-known things like POSIX, BSD networking, X11 and avoid system-specific calls like DosWrite() etc. completely. If you're doing really multiplatform project. Really 'standard' calls are the same everywhere and rarely require re-implementation. There's some oddity here and there (I heard Ultrix(?) does not have strdup()?) but you can add it in transparent way, namely linking on Ultrix with your own implementation of strdup()."

(Dr. Sergey Ayukov, Sternberg Astronomical Institute Moscow, Russia http://www.ayukov.com, http://crydee.sai.msu.ru/index-asv.html)

"Good programmers know what to write. Great ones know what to rewrite (and reuse)."

• "What do I need to know?"
  Why? In short: Everything.
• "I don't have the necessary time?"
  Then just start the emx compilation. Learn and debug by and by. Look at Alexander Mai's emx faq to get started.

1.2  Where to Start?

"Heelpp, I'm stuck! Where to begin?"

(anonymous)

http://www.edm2.com/0408/introc1.html

"Use standard ANSI C (or C++) whereever possible!"

"ANSI C for Programmers on UNIX Systems",
by T. P. Love, CUED Cambridge University Engineering Department)
<tpl@eng.cam.ac.uk>,

ftp://svr-ftp.eng.cam.ac.uk/pub/misc/love_C.ps.Z"

In the following, to fill some gaps, I'll give a very short summary of porting-related concepts and specialties. For general information please refer to the online documentation that comes with recent Linux and FreeBSD distributions. It is also necessary that you have access to a good Unix/Linux tutorial, explaining those numerous concepts unfamiliar for OS/2, which cannot be covered here. For standards, try e.g.:

http://standards.ieee.org/announcements/opengroup.html 

This tutorial will concentrate on emx-specific questions that recently were discussed on the respective lists. Further sources of information are indexed by Timur Tabi:

http://www.edm2.com/links/index.html

http://hayai.freeshell.org/os2_xfree86.html

Chapter 2
Survive With Emx Tools (cc, make, gdb, emacs)

0.30section

"Those who do not understand Unix are condemned to reinvent it, poorly."

(Henry Spencer)

2.1  Let's contrast Emx With FreeBSD.

The following are some excerpts from a terse introduction into how to behave in a FreeBSD Unix development environment (``A User's Guide to FreeBSD Programming Tools``)

http://www.de.freebsd.org/tutorials.

``Simply proceed as if you were under Unix!''

2.1.1  Compiling With cc

``This section deals only with the GNU compiler for C and C++, since that comes with the base FreeBSD [and the emx] system. It can be invoked by either cc [write a cc.cmd file] or gcc. ... Once you've written your masterpiece, the next step is to convert it into something that will (hopefully!) run on FreeBSD [as well as on OS/2 after compiling your portable Posix source with emx gcc]. This usually involves several steps, each of which is done by a separate program:

1. Pre-process your source code to remove comments and do other tricks like expanding macros in C.

   1. Check the syntax of your code to see if you have obeyed the rules of the language. If you have not, it will complain! [Cf. the different warning options: Use at least -Wall.]
   2. Convert the source code into assembly language - this is very close to machine code, but still understandable by humans. Allegedly.
   3. Convert the assembly language into machine code - yep, we are talking bits and bytes, ones and zeros here.
   4. Check that you have used things like functions and global variables in a consistent way. For example, if you have called a non-existent function, it will complain. [Something like 'Unresolved reference foo in bar': Try to find out where you can steal the library, which contains this function or variable. Or maybe you forgot to specify -foolib on the command line?]
   5. If you are trying to produce an executable from several source code files, work out how to fit them all together. [This is the fixup of unresolved symbols, it is done by the linker ld or link386 (if you specified -Zomf), respectively.]
   6. Work out how to produce something that the system's run-time loader will be able to load into memory and run. [emxbind or link386]
   7. Finally, write the executable on the file system.

% cc foobar.c

% cc foo.c bar.c

There are lots and lots of options for cc, which are all in the man page. Here are a few of the most important ones, with examples of how to use them.

-o filename 

% cc -c foobar.c

% cc -g foobar.c 

% cc -O -o foobar foobar.c

Create an optimised version of the executable. The compiler performs various clevertricks to try and produce an executable that runs faster than normal. You can add a number after the -O to specify a higher level of optimisation, but this often exposes bugsin the compiler's optimiser. ... Optimisation is usually only turned on when compiling a release version. ...

The following three flags will force cc to check that your code complies to the relevant international standard, often referred to as the ANSI standard, though strictly speaking it is an ISO standard.

-Wall 

-ansi 

-pedantic 

Without these flags, cc will allow you to use some of its non-standard extensions to the standard. Some of these are very useful, but will not work with other compilers - in fact, one of the main aims of the standard is to allow people to write code that will work with any compiler on any system. This is known as portable code. Generally, you should try to make your code as portable as possible, as otherwise you may have to completely re-write the program later to get it to work somewhere else - and who knows what you may be using in a few years time?

% cc -Wall -ansi -pedantic -o foobar.exe foobar.c

% cc -o foobar foobar.c -lm

The most common example of this is when compiling a program that uses some of the mathematical functions in C. Unlike most other platforms [emx included], these are in a separate library from the standard C one and you have to tell the compiler to add it.

The rule is that if the library is called libsomething.a, you give cc the argument -lsomething. For example, the math library is libm.a, so you give cc the argument -lm. A common "gotcha" with the math library is that it has to be the last library on the command line. [Emx does not cut off the leading 'lib' part of library names and m.a or m.lib is just a dummy library which can be used for your own common extension modules.]

If you are compiling C++ code, you need to add -lg++, or -lstdc++ ... to the command line argument to link the C++ library functions. ...

Each of these will both produce an executable foobar from the C++ source file foobar.cc. Note that, on Unix systems, C++ source files traditionally end in .C, .cxx or .cc, rather than the MS-DOS&trade; style .cpp (which was already used for something else). gcc used to rely on this to work out what kind of compiler to use on the source file; however, this restriction no longer applies, so you may now call your C++ files .cpp with impunity!

2.1.2  Common cc Queries and Problems

Q: I am trying to write a program which uses the sin() function and I get an error like this. What does it mean?

/var/tmp/cc0143941.o: Undefined symbol `_sin' referenced from text segment

Q: All right, I wrote this simple program to practice using -lm. All it does is raise 2.1 to the power of 6.

#include <stdio.h>

     

int main() 

{

float f = pow(2.1, 6);

   printf("2.1 ^ 6 = %f", f);

 return 0;

}

% cc temp.c -lm

% ./a.out

2.1 ^ 6 = 1023.000000

A: When the compiler sees you call a function, it checks if it has already seen a prototype for it. If it has not, it assumes the function returns an int, which is definitely not what you want here.

Q: So how do I fix this?

A: The prototypes for the mathematical functions are in math.h. If you include this file, the compiler will be able to find the prototype and it will stop doing strange things to your calculation!

    #include <math.h>

    #include <stdio.h>

     

    int main() {

    ...

% ./a.out

2.1 ^ 6 = 85.766121

Q: I compiled my program and it seemed to run all right at first, then there was an error and it said something about core dumped. What does that mean?

A: The name core dump dates back to the very early days of Unix, when the machines used core memory for storing data. Basically, if the program failed under certain conditions, the system would write the contents of core memory to disk in a file called core, which the programmer could then pore over to find out what went wrong.

Q: Fascinating stuff, but what I am supposed to do now?

A: Use gdb to analyse the core (see Section 5).

Q: When my program dumped core, it said something about a segmentation fault. What's that?

A: This basically means that your program tried to perform some sort of illegal operation on memory; Unix is designed to protect the operating system and other programs from rogue programs.

Common causes for this are [Read this carefully, well meant advice]:

Trying to write to a NULL pointer, eg

char *foo = NULL;

strcpy(foo, "bang!");

 char *foo;

 strcpy(foo, "bang!");

int bar[20];

bar[27] = 6;

char *foo = "My string";

strcpy(foo, "bang!");

const char *foo = "My string"; /* constant! */

Doing naughty things with malloc() and free(), eg

char bar[80];

free(bar);

char *foo = malloc(27);

free(foo);

free(foo);

2.1.3  Make

2.1.3.1  What's make?

It reads in a file, called a makefile, that tells it how different files depend on each other, and works out which files need to be re-compiled and which ones don't. For example, a rule could say something like "if fromboz.o is older than fromboz.c, that means someone must have changed fromboz.c, so it needs to be re-compiled." The makefile also has rules telling make how to re-compile the source file, making it a much more powerful tool.

Makefiles are typically kept in the same directory as the source they apply to, and can be called makefile, Makefile or MAKEFILE. Most programmers use the name Makefile, as this puts it near the top of a directory listing, where it can easily be seen.

2.1.3.2  Example of using make

Here's a very simple make file:

    foo: foo.c

        cc -o foo foo.c

The creation line starts with a tab (press the tab key) and then the command you would type to create foo if you were doing it at a command prompt. If foo is out of date, or does not exist, make then executes this command to create it. In other words, this is the rule which tells make how to re-compile foo.c.

So, when you type make, it will make sure that foo is up to date with respect to your latest changes to foo.c. This principle can be extended to Makefiles with hundreds of targets-in fact, on FreeBSD, it is possible to compile the entire operating system just by typing make world in the appropriate directory! Another useful property of makefiles is that the targets don't have to be programs. For instance, we could have a make file that looks like this:

    foo: foo.c

        cc -o foo.exe foo.c

     

    install:

        cp foo /home/me

    % make target

If we just type make on its own, make will always look at the first target and then stop without looking at any others. So if we typed make here, it will just go to the foo target, re-compile foo if necessary, and then stop without going on to the install target.

Notice that the install target doesn't actually depend on anything! This means that the command on the following line is always executed when we try to make that target by typing make install. In this case, it will copy foo into the user's home directory. This is often used by application makefiles, so that the application can be installed in the correct directory when it has been correctly compiled.

This is a slightly confusing subject to try and explain. If you don't quite understand how make works, the best thing to do is to write a simple program like "hello world" and a make file like the one above and experiment. Then progress to using more than one source file, or having the source file include a header file. The touch command is very useful here-it changes the date on a file without you having to edit it. ...

2.1.3.3  More advanced uses of make

Make is a very powerful tool, and can do much more than the simple example above shows. Unfortunately, there are several different versions of make, and they all differ considerably. The best way to learn what they can do is probably to read the documentation-hopefully this introduction will have given you a base from which you can do this.

The version of make that comes with FreeBSD is the Berkeley make ... Many applications in the ports use GNU make, which has a very good set of "info" pages. If you have installed any of these ports, GNU make will automatically have been installed as gmake. It's also available as a port and package in its own right.

To view the info pages for GNU make, you will have to edit the dir file in the

/usr/local/info directory to add an entry for it. This involves adding a line like

 * Make: (make).                 The GNU Make utility.

2.2  Debugging

2.2.1  The Debugger

The debugger that comes with FreeBSD [and emx] is called gdb (GNU debugger). You start it up by typing

    % gdb progname

    M-x gdb RET progname RET

front-end for it xxgdb in the ports collection. [Emx calls it pmgdb.} ...

2.2.2  Running a program in the debugger

You'll need to have compiled the program with the -g option to get the most out of using gdb. It will work without, but you'll only see the name of the function you're in, instead of the source code. If you see a line like:

    ... (no debugging symbols found) ...

At the gdb prompt, type break main. This will tell the debugger to skip over the preliminary set-up code in the program and start at the beginning of your code. Now type run to start the program - it will start at the beginning of the set-up code and then get stopped by the debugger when it calls main(). (If you've ever wondered where main() gets called from, now you know!).

You can now step through the program, a line at a time, by pressing n. If you get to a function call, you can step into it by pressing s. Once you're in a function call, you can return from stepping into a function call by pressing f. You can also use up and down to take a quick look at the caller.

Here's a simple example of how to spot a mistake in a program with gdb. This is our program (with a deliberate mistake):

    #include <stdio.h>

     

    int bazz(int anint);

     

    main() {

        int i;

     

        printf("This is my programn");

        bazz(i);

        return 0;

    }

     

    int bazz(int anint) {

        printf("You gave me %dn", anint);

        return anint;

    }

When we compile and run the program we get

    % cc -g -o temp temp.c

    % ./temp

    This is my program

    anint = 4231

% gdb temp

GDB is free software and you are welcome to distribute copies of it

under certain conditions; type "show copying" to see the conditions.

There is absolutely no warranty for GDB; type "show warranty" for details.

GDB 4.13 (i386-unknown-freebsd), Copyright 1994 Free Software Foundation, Inc.

(gdb) break main                   Skip the set-up code

Breakpoint 1 at 0x160f: file temp.c, line 9. breakpoint at main()

(gdb) run                           Run as far as main()

Starting program: temp            Program starts running

     

Breakpoint 1, main () at temp.c:9     gdb stops at main()

(gdb) n                                   Go to next line

This is my program                     Program prints out

(gdb) s                                  step into bazz()

bazz (anint=4231) at temp.c:17   gdb displays stack frame

(gdb)

(gdb) up                                 Move up call stack

#1  0x1625 in main () at temp.c:11 gdb displays stack frame

(gdb) p i                            Show us the value of i

$1 = 4231                                 gdb displays 4231

    ...

    main() {

        int i;

     

        i = 5;

        printf("This is my programn");

    ...

in that area of memory when the program ran, which in this case happened to be 4231.

Note: gdb displays the stack frame every time we go into or out of a function, even if we're using up and down to move around the call stack. This shows the name of the function and the values of its arguments, which helps us keep track of where we are and what's going on. (The stack is a storage area where the program stores information about the arguments passed to functions and where to go when it returns from a function call).

2.2.3  Examining a core file

A core file is basically a file which contains the complete state of the process when it crashed. In "the good old days", programmers had to print out hex listings of core files and sweat over machine code manuals, but now life is a bit easier. Incidentally, under FreeBSD and other 4.4BSD systems, a core file is called progname.core instead of just core, to make it clearer which program a core file belongs to. [emx calls it just core.]

To examine a core file, start up gdb in the usual way. Instead of typing break or run, type

    (gdb) core progname.core [emx: gdb progname core]

You should see something like this:

    % gdb a.out

    GDB is free software and you are welcome to distribute copies of it

     under certain conditions; type "show copying" to see the conditions.

    There is absolutely no warranty for GDB; type "show warranty" for details.

    GDB 4.13 (i386-unknown-freebsd), Copyright 1994 Free Software Foundation, Inc.

    (gdb) core a.out.core

    Core was generated by `a.out'.

    Program terminated with signal 11, Segmentation fault.

    Cannot access memory at address 0x7020796d.

    #0  0x164a in bazz (anint=0x5) at temp.c:17

    (gdb)

Sometimes it's useful to be able to see how a function was called, as the problem could have occurred a long way up the call stack in a complex program. The bt command causes gdb to print out a back-trace of the call stack:

    (gdb) bt

    #0  0x164a in bazz (anint=0x5) at temp.c:17

    #1  0xefbfd888 in end()

    #2  0x162c in main() at temp.c:11

    (gdb)

2.2.4  Attaching to a running program

One of the neatest features about gdb is that it can attach to a program that's already running. Of course, that assumes you have sufficient permissions to do so. A common problem is when you are stepping through a program that forks, and you want to trace the child, but the debugger will only let you trace the parent.

What you do is start up another gdb, use ps to find the process ID for the child, and do

(gdb) attach pid

"That's all very well," you're probably thinking, "but by the time I've done that, the child process will be over the hill and far away". Fear not, gentle reader, here's how to do it (courtesy of the gdb info pages):

...

if ((pid = fork()) < 0)     /* _Always_ check this */

    error();

else if (pid == 0) {        /* child */

    int PauseMode = 1;

     

while (PauseMode)

    sleep(10);  /* Wait until someone attaches to us */

        ...

} else {            /* parent */

        ...

Using Emacs as a Development Environment

Emacs

[...] Emacs is basically a highly customisable editor-indeed, it has been customised to the point where it's more like an operating system than an editor! Many developers and sysadmins do in fact spend practically all their time working inside Emacs, leaving it only to log out. [I recommend also to test elvis, a vi clone.] It's impossible even to summarise everything Emacs can do here, but here are some of the features of interest to developers:

• Very powerful editor, allowing search-and-replace on both strings and regular expressions (patterns), jumping to start/end of block expression, etc, etc.
• Pull-down menus and online help.
• Language-dependent syntax highlighting and indentation.
• Completely customisable.
• You can compile and debug programs within Emacs.
• On a compilation error, you can jump to the offending line of source code.
• Friendly frontend to the info program used for reading GNU hypertext documentation, including the documentation on Emacs itself.
• Friendly front-end to gdb, allowing you to look at the source code as you step through your program.
• You can read Usenet news and mail while your program is compiling.
• And doubtless many more that I've overlooked.

Once it's installed, start it up and do C-h t to read an Emacs tutorial-that means hold down the control key, press h, let go of the control key, and then press t. (Alternatively, you can you use the mouse to select Emacs Tutorial from the Help menu).

Although Emacs does have menus, it's well worth learning the key bindings, as it's much quicker when you're editing something to press a couple of keys than to try and find the mouse and then click on the right place. And, when you're talking to seasoned Emacs users, you'll find they often casually throw around expressions like "M-x replace-s RET foo RET bar RET" so it's useful to know what they mean. And in any case, Emacs has far too many useful functions for them to all fit on the menu bars.

Fortunately, it's quite easy to pick up the key-bindings, as they're displayed next to the menu item. My advice is to use the menu item for, say, opening a file until you understand how it works and feel confident with it, then try doing C-x C-f. When you're happy with that, move on to another menu command.

If you can't remember what a particular combination of keys does, select Describe Key from the Help menu and type it in-Emacs will tell you what it does. You can also use the Command Apropos menu item to find out all the commands which contain a particular word in them, with the key binding next to it.

By the way, the expression above means hold down the Meta key, press x, release the Meta key, type replace-s (short for replace-string-another feature of Emacs is that you can abbreviate commands), press the return key, type foo (the string you want replaced), press the return key, type bar (the string you want to replace foo with) and press return again. Emacs will then do the search-and-replace operation you've just requested. If you're wondering what on earth the Meta key is, it's a special key that many Unix workstations have. Unfortunately, PC's don't have one, so it's usually the alt key (or if you're unlucky, the escape key).

Oh, and to get out of Emacs, do C-x C-c (that means hold down the control key, press x, press c and release the control key). If you have any unsaved files open, Emacs will ask you if you want to save them. (Ignore the bit in the documentation where it says C-z is the usual way to leave Emacs-that leaves Emacs hanging around in the background, and is only really useful if you're on a system which doesn't have virtual terminals).

2.3  Configuring Emacs

Emacs does many wonderful things; some of them are built in, some of them need to be configured.

Instead of using a proprietary macro language for configuration, Emacs uses a version of Lisp specially adapted for editors, known as Emacs Lisp. This can be quite useful if you want to go on and learn something like Common Lisp, as it's considerably smaller than Common Lisp (although still quite big!).

The best way to learn Emacs Lisp is to download the Emacs Tutorial. However, there's no need to actually know any Lisp to get started with configuring Emacs, as I've included a sample .emacs file, which should be enough to get you started. Just copy it into your home directory and restart Emacs if it's already running; it will read the commands from the file and (hopefully) give you a useful basic setup.

2.3.1  A sample .emacs file

Unfortunately, there's far too much here to explain it in detail; however there are one or two points worth mentioning.

• Everything beginning with a ; is a comment and is ignored by Emacs.
• In the first line, the -*-áEmacs-Lispá-*- is so that we can edit the .emacs file itself within Emacs and get all the fancy features for editing Emacs Lisp. Emacs usually tries to guess this based on the filename, and may not get it right for .emacs.

This file supports syntax highlighting for C, C++, Perl, Lisp and Scheme, by guessing the language from the filename.

Emacs already has a pre-defined function called next-error. In a compilation output window, this allows you to move from one compilation error to the next by doing M-n; we define a complementary function, previous-error, that allows you to go to a previous error by doing M-p. The nicest feature of all is that C-c C-c will open up the source file in which the error occurred and jump to the appropriate line.

We enable Emacs's ability to act as a server, so that if you're doing something outside Emacs and you want to edit a file, you can just type in

% emacsclient filename

2.3.2  Example A sample .emacs file

    ;; -*-Emacs-Lisp-*-

     

    ;; This file is designed to be re-evaled; use the variable first-time

    ;; to avoid any problems with this.

    (defvar first-time t  

      "Flag signifying this is the first time that .emacs has been evaled")

     

    ;; Meta

    (global-set-key "M- " 'set-mark-command)

    (global-set-key "M-C-h" 'backward-kill-word)

    (global-set-key "M-C-r" 'query-replace)

    (global-set-key "M-r" 'replace-string)

    (global-set-key "M-g" 'goto-line)

    (global-set-key "M-h" 'help-command)

     

    ;; Function keys

    (global-set-key [f1] 'manual-entry)

    (global-set-key [f2] 'info)

    (global-set-key [f3] 'repeat-complex-command)

    (global-set-key [f4] 'advertised-undo)

    (global-set-key [f5] 'eval-current-buffer)

    (global-set-key [f6] 'buffer-menu)

    (global-set-key [f7] 'other-window)

    (global-set-key [f8] 'find-file)

    (global-set-key [f9] 'save-buffer)

    (global-set-key [f10] 'next-error)

    (global-set-key [f11] 'compile)

    (global-set-key [f12] 'grep)

    (global-set-key [C-f1] 'compile)

    (global-set-key [C-f2] 'grep)

    (global-set-key [C-f3] 'next-error)

    (global-set-key [C-f4] 'previous-error)

    (global-set-key [C-f5] 'display-faces)

    (global-set-key [C-f8] 'dired)

    (global-set-key [C-f10] 'kill-compilation)

     

    ;; Keypad bindings

    (global-set-key [up] "C-p")

    (global-set-key [down] "C-n")

    (global-set-key [left] "C-b")

    (global-set-key [right] "C-f")

    (global-set-key [home] "C-a")

    (global-set-key [end] "C-e")

    (global-set-key [prior] "M-v")

    (global-set-key [next] "C-v")

    (global-set-key [C-up] "M-C-b")

    (global-set-key [C-down] "M-C-f")

    (global-set-key [C-left] "M-b")

    (global-set-key [C-right] "M-f")

    (global-set-key [C-home] "M-<")

    (global-set-key [C-end] "M->")

    (global-set-key [C-prior] "M-<")

    (global-set-key [C-next] "M->")

     

    ;; Mouse

    (global-set-key [mouse-3] 'imenu)

     

    ;; Misc

    (global-set-key [C-tab] "C-qt")   ; Control tab quotes a tab.

    (setq backup-by-copying-when-mismatch t)

     

    ;; Treat 'y' or <CR> as yes, 'n' as no.

    (fset 'yes-or-no-p 'y-or-n-p)

        (define-key query-replace-map [return] 'act)

        (define-key query-replace-map [?C-m] 'act)

     

    ;; Load packages

    (require 'desktop)

    (require 'tar-mode)

     

    ;; Pretty diff mode

    (autoload 'ediff-buffers "ediff" "Intelligent Emacs interface to diff" t)

    (autoload 'ediff-files "ediff" "Intelligent Emacs interface to diff" t)

    (autoload 'ediff-files-remote "ediff"

      "Intelligent Emacs interface to diff")

     

    (if first-time

        (setq auto-mode-alist

          (append '((".cpp$" . c++-mode)

                (".hpp$" . c++-mode)

                        (".lsp$" . lisp-mode)

                (".scm$" . scheme-mode)

                (".pl$" . perl-mode)

                ) auto-mode-alist)))

     

    ;; Auto font lock mode

    (defvar font-lock-auto-mode-list  

      (list 'c-mode 'c++-mode 'c++-c-mode 'emacs-lisp-mode 'lisp-mode 'perl-mode 'scheme-mode)

      "List of modes to always start in font-lock-mode")

     

    (defvar font-lock-mode-keyword-alist

      '((c++-c-mode . c-font-lock-keywords)

        (perl-mode . perl-font-lock-keywords))

      "Associations between modes and keywords")

     

    (defun font-lock-auto-mode-select ()

      "Automatically select font-lock-mode if the current major mode is

    in font-lock-auto-mode-list"

      (if (memq major-mode font-lock-auto-mode-list)  

          (progn

        (font-lock-mode t))

        )

      )

     

    (global-set-key [M-f1] 'font-lock-fontify-buffer)

     

    ;; New dabbrev stuff

    ;(require 'new-dabbrev)

    (setq dabbrev-always-check-other-buffers t)

    (setq dabbrev-abbrev-char-regexp "sw|s_")

    (add-hook 'emacs-lisp-mode-hook

          '(lambda ()  

             (set (make-local-variable 'dabbrev-case-fold-search) nil)

             (set (make-local-variable 'dabbrev-case-replace) nil)))

    (add-hook 'c-mode-hook

          '(lambda ()  

             (set (make-local-variable 'dabbrev-case-fold-search) nil)

             (set (make-local-variable 'dabbrev-case-replace) nil)))

    (add-hook 'text-mode-hook

          '(lambda ()  

             (set (make-local-variable 'dabbrev-case-fold-search) t)

             (set (make-local-variable 'dabbrev-case-replace) t)))

     

    ;; C++ and C mode...

    (defun my-c++-mode-hook ()

      (setq tab-width 4)

      (define-key c++-mode-map "C-m" 'reindent-then-newline-and-indent)

      (define-key c++-mode-map "C-ce" 'c-comment-edit)

      (setq c++-auto-hungry-initial-state 'none)

      (setq c++-delete-function 'backward-delete-char)

      (setq c++-tab-always-indent t)

      (setq c-indent-level 4)

      (setq c-continued-statement-offset 4)

      (setq c++-empty-arglist-indent 4))

     

    (defun my-c-mode-hook ()

      (setq tab-width 4)

      (define-key c-mode-map "C-m" 'reindent-then-newline-and-indent)

      (define-key c-mode-map "C-ce" 'c-comment-edit)

      (setq c-auto-hungry-initial-state 'none)

      (setq c-delete-function 'backward-delete-char)

      (setq c-tab-always-indent t)

    ;; BSD-ish indentation style

      (setq c-indent-level 4)

      (setq c-continued-statement-offset 4)

      (setq c-brace-offset -4)

      (setq c-argdecl-indent 0)

      (setq c-label-offset -4))

     

    ;; Perl mode

    (defun my-perl-mode-hook ()

      (setq tab-width 4)

      (define-key c++-mode-map "C-m" 'reindent-then-newline-and-indent)

      (setq perl-indent-level 4)

      (setq perl-continued-statement-offset 4))

     

    ;; Scheme mode...

    (defun my-scheme-mode-hook ()

      (define-key scheme-mode-map "C-m" 'reindent-then-newline-and-indent))

     

    ;; Emacs-Lisp mode...

    (defun my-lisp-mode-hook ()

      (define-key lisp-mode-map "C-m" 'reindent-then-newline-and-indent)

      (define-key lisp-mode-map "C-i" 'lisp-indent-line)

      (define-key lisp-mode-map "C-j" 'eval-print-last-sexp))

     

    ;; Add all of the hooks...

    (add-hook 'c++-mode-hook 'my-c++-mode-hook)

    (add-hook 'c-mode-hook 'my-c-mode-hook)

    (add-hook 'scheme-mode-hook 'my-scheme-mode-hook)

    (add-hook 'emacs-lisp-mode-hook 'my-lisp-mode-hook)

    (add-hook 'lisp-mode-hook 'my-lisp-mode-hook)

    (add-hook 'perl-mode-hook 'my-perl-mode-hook)

     

    ;; Complement to next-error

    (defun previous-error (n)

      "Visit previous compilation error message and corresponding source code."

      (interactive "p")

      (next-error (- n)))

     

    ;; Misc...

    (transient-mark-mode 1)

    (setq mark-even-if-inactive t)

    (setq visible-bell nil)

    (setq next-line-add-newlines nil)

    (setq compile-command "make")

    (setq suggest-key-bindings nil)

    (put 'eval-expression 'disabled nil)

    (put 'narrow-to-region 'disabled nil)

    (put 'set-goal-column 'disabled nil)

     

    ;; Elisp archive searching

    (autoload 'format-lisp-code-directory "lispdir" nil t)

    (autoload 'lisp-dir-apropos "lispdir" nil t)

    (autoload 'lisp-dir-retrieve "lispdir" nil t)

    (autoload 'lisp-dir-verify "lispdir" nil t)

     

    ;; Font lock mode

    (defun my-make-face (face colour &optional bold)

      "Create a face from a colour and optionally make it bold"

      (make-face face)

      (copy-face 'default face)

      (set-face-foreground face colour)

      (if bold (make-face-bold face))

      )

     

    (if (eq window-system 'x)

        (progn

          (my-make-face 'blue "blue")

          (my-make-face 'red "red")

          (my-make-face 'green "dark green")

          (setq font-lock-comment-face 'blue)

          (setq font-lock-string-face 'bold)

          (setq font-lock-type-face 'bold)

          (setq font-lock-keyword-face 'bold)

          (setq font-lock-function-name-face 'red)

          (setq font-lock-doc-string-face 'green)

          (add-hook 'find-file-hooks 'font-lock-auto-mode-select)

     

          (setq baud-rate 1000000)

          (global-set-key "C-cmm" 'menu-bar-mode)

          (global-set-key "C-cms" 'scroll-bar-mode)

          (global-set-key [backspace] 'backward-delete-char)

                        ;      (global-set-key [delete] 'delete-char)

          (standard-display-european t)

          (load-library "iso-transl")))

     

    ;; X11 or PC using direct screen writes

    (if window-system

        (progn

          ;;      (global-set-key [M-f1] 'hilit-repaint-command)

          ;;      (global-set-key [M-f2] [?C-u M-f1])

          (setq hilit-mode-enable-list   

            '(not text-mode c-mode c++-mode emacs-lisp-mode lisp-mode

              scheme-mode)

            hilit-auto-highlight nil

            hilit-auto-rehighlight 'visible

            hilit-inhibit-hooks nil

            hilit-inhibit-rebinding t)

          (require 'hilit19)

          (require 'paren))

      (setq baud-rate 2400)         ; For slow serial connections

      )

     

    ;; TTY type terminal

    (if (and (not window-system)  

         (not (equal system-type 'ms-dos)))

        (progn

          (if first-time

          (progn

            (keyboard-translate ?C-h ?C-?)

            (keyboard-translate ?C-? ?C-h)))))

     

    ;; Under UNIX

    (if (not (equal system-type 'ms-dos))

        (progn

          (if first-time

          (server-start))))

     

    ;; Add any face changes here

    (add-hook 'term-setup-hook 'my-term-setup-hook)

    (defun my-term-setup-hook ()

      (if (eq window-system 'pc)

          (progn

    ;;  (set-face-background 'default "red")

        )))

     

    ;; Restore the "desktop" - do this as late as possible

    (if first-time

        (progn

          (desktop-load-default)

          (desktop-read)))

     

    ;; Indicate that this file has been read at least once

    (setq first-time nil)

     

    ;; No need to debug anything now

    (setq debug-on-error nil)

     

    ;; All done

    (message "All done, %s%s" (user-login-name) ".")

2.3.3  Extending the Range of Languages Emacs Understands

Now, this is all very well if you only want to program in the languages already catered for in the .emacs file (C, C++, Perl, Lisp and Scheme), but what happens if a new language called "whizbang" comes out, full of exciting features?

The first thing to do is find out if whizbang comes with any files that tell Emacs about the language. These usually end in .el, short for "Emacs Lisp". For example, if whizbang is a FreeBSD port, we can locate these files by doing

% find /usr/ports/lang/whizbang -name "*.el" -print

So for example, if the output from the find command was

/usr/ports/lang/whizbang/work/misc/whizbang.el

# cp /usr/ports/lang/whizbang/work/misc/whizbang.el /usr/local/share/emacs/site-lisp

Find the auto-mode-alist entry in .emacs and add a line for whizbang, such as:

    ...

    (".lsp$" . lisp-mode)

    (".wiz$" . whizbang-mode)

    (".scm$" . scheme-mode)

    ...

ending in .wiz.

Just below this, you'll find the font-lock-auto-mode-list entry. Add whizbang-mode to it like so:

    ;; Auto font lock mode

    (defvar font-lock-auto-mode-list  

      (list 'c-mode 'c++-mode 'c++-c-mode 'emacs-lisp-mode 'whizbang-mode 'lisp-mode 'perl-mode 'scheme-mode)

      "List of modes to always start in font-lock-mode")

And that's all that's needed. If there's anything else you want done automatically when you open up a .wiz file, you can add a whizbang-mode hook (see my-scheme-mode-hook for a simple example that adds auto-indent).

2.4  Further Reading

• Brian W. Kernighan and Dennis M. Ritchie, The C Programming Language (2nd Edition), Prentice-Hall 1988, ISBN 0-13-110362-8
• J.F. Korsh and L.J. Garrett, Data Structures, Algorithms, and Program Style using C, PSW-Kent 1988
• Samuel P. Harbison and Guy L. Steele, C: A Reference Manual, Addison-Wesley 1987
• Andrew Koenig, C Traps and Pitfalls, Addison-Wesley 1989
• Mark Horton, Portable C Software, Prentice-Hall 1990
• William H. Press and Brian P. Flannery and Saul A. Teukolsky and William T. Vetterling, NUMERICAL RECIPES in C: The Art of Scientific Computing, Cambridge University Press 1988
• Brian W. Kernighan and Rob Pike, The Unix Programming Environment, Prentice-Hall 1984, ISBN 0-13-937681-X
• W. Richard Stevens, Advanced Programming in the Unix Environment, Addison-Wesley 1992, ISBN 0-201-56317-7
• W. Richard Stevens, Unix Network Programming, Prentice-Hall 1990, ISBN 0-13-949876-1
• Bjarne Stroustrup, The C++ Programming Language, Addison-Wesley 1991, ISBN 0-201-53992-6
• Winston and Horn, Lisp (3rd Edition), Addison-Wesley 1989, ISBN 0-201-08319-1

Part 2
Special Problems

Chapter 3
Frequently Used Concepts

0.40section

"Too much X around, I'm confused!"

(A novice to X )

3.1  What's ' X\protect '?

The Manual Page

Running 'man X' on Unix - and on your system if the necessary backends are installed ('man' or 'xman' is generally a good idea for terse descriptions of terms, apps and interfaces) - spits out something (cryptical) like:

NAME 

X - a portable, network-transparent window system 

SYNOPSIS 

The  X  Window System is a network transparent window system which 

runs on a wide range of computing and graphics machines. It should 

be relatively straightforward to build the  X  Consortium software 

distribution on most ANSI C and POSIX compliant systems. Commercial 

implementations are also available for a wide range of platforms.

The  X  Consortium requests that the following names be used when 

referring to this software:

X ;  X  Window System;  X  Version 11;  X  Window System, Version 11; 

X 11

X  Window System is a trademark of  X  Consortium, Inc. 

DESCRIPTION 

X  Window System servers run on computers with bitmap displays. The 

server distributes user input to and accepts output requests from 

various client programs through a variety of different interprocess 

communication channels. Although the most common case is for the 

client programs to be running on the same machine as the server, 

clients can be run transparently from other machines (including 

machines with different architectures and operating systems) as well.

X  supports overlapping hierarchical subwindows and text and graphics 

operations, on both monochrome and color dis plays. For a full 

explanation of the functions that are available, see the Xlib - 

C Language  X  Interface manual, the  X  Window System Protocol 

specification, the  X  Toolkit Intrinsics - C Language Interface 

manual, and various toolkit documents.

The number of programs that use  X  is quite large. Programs provided 

in the core  X  Consortium distribution include: a terminal emulator, 

xterm; a window manager, twm; [...] access control programs, xauth, 

xhost, and iceauth; user preference setting programs, xrdb, xcmsdb, 

xset, xsetroot, xstdcmap, and xmodmap; [...]; utilities for listing 

information about [...] displays, xdpyinfo, xlsatoms, and xprop; 

[...]; a display server and related utilities, Xserver [...]; a 

utility to terminate clients, xkill; [...] 

Many other utilities, window managers, games, toolkits, etc. are 

included as user-contributed software in the  X  Consortium distribution, 

or are available using anonymous ftp on the Internet. See your site 

administrator for details.

3.2  The 'official' port of XFree86

The newest OS/2-port is introduced in "The XFree86/OS2 FAQ" (by Holger Veit, Sebastien Marineau):

http://www-set.gmd.de/~veit/os2/x11os2faq.html 

3.2.1  General Information

*1.1 What is X11, X11R6.3, XFree86, XFree86/OS2?

X11, more precisely called "The X Window System" is a complete window system that usually runs as the de-facto standard in Unix environments. X11R6.3 is the name of the current release (precisely, the most recent one is X11R6.4, but this is not part of the 3.X server line. XFree86 is a port of the X Window System to x86-based systems. XFree86/OS2 is a particular port of XFree86 for OS/2 based systems.

1.2 Where can I find more information?

Books about X11 exist in any well-sorted CS bookstore.

Some URLs:

http://www.x11.org/ (this has a rather large bibliography) 

http://www.xfree86.org/

http://set.gmd.de/~veit/os2/xf86os2.html (XFree86/OS2) 

Newsgroups:

comp.windows.x.* (X11 specific things) 

comp.os.os2.programmer.* (OS/2 specific things) 

Mailing list (for XFree86/OS2): see Q 1.7 [...]

*1.7 Is there a mailing list for XFree86/OS2?

Yes, read

http://set.gmd.de/~veit/os2/xf86mail.html 

for details. [...] There is also a digest version available, read the above URL. [...]

1.16 Are there some X11 books?

Sure, ask in your book store. A user's guide is for instance:

Niall Mansfield, The X Window System - A User's Guide, Addison Wesley, or The Definite Guides to the X Windows System, Volume Three, X Window System User's Guide, O'Reilly&Associates

The latter one is part of an eight (nine,ten?) volume documentation set for the X11 programmer. From my biased point of view of a programmer, this is the most comprehensive must-have for the serious programmer (however, it does not belong to the cheaper booksets, unfortunately)."

3.2.2  The XFree86/OS2 Infrastructure

• The hardcoded XFree86/OS2 directory structure looks like this and must not be changed:

  $(X11ROOT)XFree86  
  $(X11ROOT)XFree86/bin (binaries in PATH) 
  $(X11ROOT)XFree86/lib (DLL's and development libraries) 
  $(X11ROOT)XFree86/lib/X11 (subdir's with config files)
  yourhome_directory - (config files for various applications)

  while $(X11ROOT) normally is a drive like 'D:' and $(HOME) is free.
• GNU File Utilities

  XFree86 and most of it's ports stem from Unix/POSIX; it is advisable that you download the Posix' standard utilities like rm, grep, mkdir, cp etc.

• Shells

  The required shell for Rexx is CMD.EXE. While this is familiar, porting will require a shell that acts like AT&T 'sh', like the xksh. I recommend a recent version of ash/bsh, the Berkeley Unix standard shell; other ported shells often have too small command-line buffers.

3.2.3  Tip: Everblue

• Look at

  "The OS/2 OpenSource Project: http://www.netlabs.org"

  and search for the brand new project Everblue.

  "Adrian Gschwend" <ktk@netlabs.org> introduces it: "Today Brian Smith released the first beta of EverBlue, the XLib-PM implementation for OS/2 which allows you to compile Xlib (Unix/Linux) applications as native PM applications.

  The beta is not really usable for endusers but developers can use it to compile PM versions of X-Applications. There is just a small demo included so maybe it does not make sense to download it now if you are not a developer.

  Sourcecode of this betadrop is also included in the package, just install it. If you want to support this project you can subscribe to the EverBlue mailinglist at Egroups, we still need a lot of support because the project is not yet complete. I will set up a CVS server at Netlabs as soon as we decided how to organise the structure of the source. Informations about CVS will be posted to the EverBlue mailinglist.

  Everblue mailinglist: http://www.egroups.com/group/everblue-dev

  Get the file from: (self installing exe-file)

  ftp://ftp.netlabs.org/pub/everblue/everblue.exe"

3.3  And Posix, etc.?

Run e.g. on Linux

man 2 intro 

NAME 

intro - Introduction to system calls 

DESCRIPTION 

This chapter describes system calls.

Calling Directly In most cases, it is unnecessary to invoke a system 

call directly, but there are times when the Standard C library does 

not implement a nice function call for you.

Synopsis #include <unistd.h> [...]  

CONFORMING TO 

Certain codes are used to indicate Unix variants and standards to 

which calls in the section conform. These are:

SVr4 

System V Release 4 Unix, as described in the "Programmer's Reference 

Manual: Operating System API (Intel processors)" (Prentice-Hall 1992, 

ISBN 0-13-951294-2)

SVID 

System V Interface Definition, as described in "The System V Interface 

Definition, Fourth Edition", available at ftp://ftp.fpk.novell.com/pub/unix- standards/svid in Postscript files.

POSIX.1 

IEEE 1003.1-1990 part 1, aka ISO/IEC 9945-1:1990s, aka "IEEE Portable 

Operating System Interface for Computing Environments", as elucidated 

in Donald Lewine's "POSIX Programmer's Guide" (O'Reilly & Associates, 

Inc., 1991, ISBN 0-937175-73-0.

POSIX.1b 

IEEE Std 1003.1b-1993 (POSIX.1b standard) describing real-time 

facilities for portable operating systems, aka ISO/IEC 9945-1:1996, as 

elucidated in "Programming for the real world - POSIX.4" by Bill O. 

Gallmeister (O'Reilly & Associates, Inc. ISBN 1-56592-074-0). 

[Note: No specific emx support!]

4.3BSD/4.4BSD 

The 4.3 and 4.4 distributions of Berkeley Unix. 4.4BSD was 

upward-compatible from 4.3.

V7 Version 7, the ancestral Unix from Bell Labs. 

FILES 

/usr/include/linux/_unistd.h"

emx/include/unistd.h  
emx/include/io.h 

System calls are documented in /emx/doc/system.doc.

All system calls are declared in <emx/syscalls.h>. Interface routines are in the emx.a and emx.lib libraries. System call emulation routines are in the sys.lib libraries. Please do not use system calls from application programs - always use C library functions. If you need a system call, put a function into the library which issues the system call. [...] The goal is to have Unix-like system calls.

Moreover if we go into details we should explain XPG, X/OPEN, Spec1170, etc. Please consult Unix tutorials here.

And to How Get Posix, etc. Defined?

Specify the desired environment (cf. emx GNU gcc manual 2.12/2.8):

"The preprocessor CPP defines the symbols __32BIT__ and 

__EMX__. If the -Zmt, -Zmts, or -Zmtd option is given on the 

GCC command line, the symbol __MT__ is defined." 

"The following assertions are predefined in the emx port of 

gcc: 

#system(unix), 

#system(emx), 

#cpu(i386) and 

#machine(i386)."

`-ansi'

switch to the GNU C compiler defines __STRICT_ANSI__. This is recommended, those definitions in the headers which contradict ANSI will not be included; most of the extensions are usually not necessary. Tip: Use

#define inline __inline__ 

to become standard conformant and to avoid error messages.

`-pedantic'

switch complains loudly about non_ANSI code,

`-posix'

defines _POSIX_SOURCE.

If more than one of these are defined, they accumulate. For example:

__STRICT_ANSI__, _POSIX_SOURCE and _POSIX_C_SOURCE 

together give you ISO C, 1003.1, and 1003.2, but nothing else:

__STRICT_ANSI__ ISO Standard C. 

_POSIX_SOURCE IEEE Std 1003.1. 

_POSIX_C_SOURCE 

If ==1, like _POSIX_SOURCE; 

if >=2 add IEEE Std 1003.2; 

if >=199309L, add IEEE Std 1003.1b-1993 

[Note: No specific emx support! Also cf. 

/XFREE86/INCLUDE/X11/Xos.h from XFree86/OS2]

'-v -Q'

gives you important debug info, including builtin symbols and assertions, as well as a complete list of enabled optimization switches. The latter is crucial for pgcc's experimental optimizing. I've experienced internal compiler problems ('virtual memory exhausted') or unexpected program behaviour with pgcc v. 1.1.3 (gcc v. 2.92.66) and the following optimizations (if in any doubt, use the switches below). I also recommend to disable the experimental C++ feature 'rtti' and 'exceptions':

-fno-gcse -fno-schedule-insns2 -fno-reg-reg-copy-opt -fno-omit-frame-pointer -fno-exceptions -fno-rtti

Chapter 4
A Silly Demo and Some Comments

Now time has come to compile some simple demo code. Here I will not add another one to the pile of introduction to ANSI C, several excellent tutorials exist. All I can express here is a kind of compliment to Eberhard Mattes for his emx development tools and libraries: There is not much to say about compiling standard C (and C++; provided you use a recent compiler version) with emx. Just start.

4.1  Read the Emx Documentation

This is just the right moment to re-read the README and INSTALL file of emx. Then compile the sample code provided by Eberhard Mattes and have a close look at the first chapters of the Emx Application Developers Guide, which introduce its philosophy. Doing both carefully might save you trouble later:

``There are three methods for creating executable files:

(E1) using ld and emxbind

(E2) using emxomf, emxomfld and LINK386, the program will use the emx.dll dynamic link library for performing system calls

(E3) using emxomf, emxomfld and LINK386, the program will be linked with a system call library (creating a stand-alone application or DLL)

The assembler [GNU as] creates a Unix-style a.out object file (.o file). When using method (E1), .o files created by the assembler are linked by a Unix-style linker [GNU ld] with Unix-style libraries (.a files) to create a Unix-style a.out file. Then, emxbind is used to turn this file into an .exe file that can be executed under both OS/2 and DOS. Using method (E1) enables core dumps and fork(). Moreover, programs created with method (E1) can be debugged using GDB, the GNU debugger. Programs created using method (E1) use emx (emx.dll under OS/2, emx.exe under DOS) for system calls.

When using method (E2), the .o files created by the assembler are converted to Object Module Format files (.obj files). These files are linked with the OS/2 linker LINK386. The libraries are .lib files. emxomfld is a front end to LINK386 which converts the ld command line to a LINK386 command line. Programs created with method (E2) cannot create core dumps, cannot call fork() and cannot be debugged with GDB. Method (E2) works only under OS/2 and creates programs that work only under OS/2. You can use IBM's IPMD and SD386 debuggers to debug programs created with methods (E2) and (E3). Files created with method (E2) are usually smaller. The emx.dll dynamic link library is used for system calls. The -Zomf option of GCC selects method (E2).

When using method (E3), the program won't call the emx.dll dynamic link library. A system call library (emx emulator) is linked to the program. The system call library maps system calls to OS/2 API calls. Only a subset of the emx system calls is available with method (E3). For instance, the general terminal interface is not available. Functions which are not available or are limited with method (E3) are marked [*] in the library reference. Use a module definition file and the STACKSIZE statement to set the stack size. Alternatively, you can use the -Zstack option of GCC.''

4.2  Source with Comments

Now a trivial example to illustrate some Posix compliant emx interfaces. Refer to the Emx C Library Reference for the details about open(), read(), write(), etc.

(For an example how to use the more efficient ANSI streams counterparts fopen(), fread(), fwrite(), etc., and how to change file attributes like date and time, please cf. below: cpfile.c in the Chapter ``Some Notes on Portability and Coding Style''. It is an example for a widely portable symlink() replacement ).

Please read the comments in the source; in fact the source has only been written to illustrate the comments and to introduce the compiler commandline frontend 'gcc.exe'.

-----snip: cut here-------

/* demonstrate_EMX_and_XFree86_os2_io.c: gcc -Zbin-files

 

Just try to compile this primitive example with any 

commercial OS/2 compiler

   */

 

#define TRUE 1

#define _POSIX_C_SOURCE 2

 

#include <unistd.h> /* Posix io and constants; contains  

                        #define _POSIX_SOURCE, 

                        #define _POSIX_VERSION          199009L 

                    */

#include <stdio.h>      /* ANSI conformant io, Posix extensions:

                         * fdopen(), fileno() 

                         */

#include <stdlib.h>     /* ANSI C: getenv(); Posix: environ */

#include <sys/types.h>  /* Posix: for open() you need 

                           sys/types.h, */ 

#include <sys/stat.h>   /* sys/stat.h and fcntl.h */

#include <fcntl.h>      /* man pages/Lib Ref. of the command say 

                 which include files need to be mentioned, for

                 portability follow ANSI and Posix,  X apps should 

                 conform to the similar MIT  X Open standard */

#include <string.h>

 

 

/* MIT  X Open/Posix and ANSI type globals */

 

ssize_t bytes_read;     /* ssize_t is in unistd.h, sys/types.h */

size_t fp_bytes_read;   /* size_t is in unistd.h, stdio.h */

int fd, fd2;            /* File descriptors - unbuffered */

FILE *fp, *fp2;         /* stdio.h - buffering */

char buffer[BUFSIZ] = ""; /* BUFSIZ is set up in stdio.h */

 

 

  /* Use File descriptors - unbuffered */

        static inline void

rawRead(const char *fileOrDevice)

{

#ifndef __EMX__ /* use only this for emx specific code, 

                   not: OS2, 32BIT, etc. */

  fd = open(fileOrDevice, O_RDONLY); /* include in low-level 

                                        fd-table */

#else

  fd = open(fileOrDevice, O_RDONLY|O_TEXT); /* cf: emx C                                           Library Reference */

#endif 

  if (fd == -1){

     printf("Error! Reading fileOrDevice: ");

     perror(fileOrDevice);

     exit(1);

  }

 

  while (TRUE){

     bytes_read = read(fd, buffer, BUFSIZ);

     if (bytes_read > 0)

        printf("%d bytes read from %s.n", bytes_read, 

                fileOrDevice);

     else {

        if (bytes_read == 0){

          printf("End of file %s reachedn", fileOrDevice);

          close(fd); /* Remove from the file-descriptor table! 

                        Never mix buffered/raw i/o on the same 

                        file or device */

          break;

        }

        else if (bytes_read == -1){

          printf("Error! Reading fileOrDevice: ");

          perror(fileOrDevice);

          exit(2);

        }

     }

}

 

 /* use buffering */

        static inline char const *

bufferedRead(void)

{

 char const *etc_password = 

             (char const*)malloc(sizeof(getenv("ETC")) + 

              sizeof("TCPOS2.INI")); 

 strcpy((char*)etc_password, getenv("ETC"));

              /* '/tcpos2.ini' is equivalent; cf. below

               * the section about Unix file systems.*/

 strcat((char*)etc_password, "TCPOS2.INI");

 

 fp = fopen(etc_password, "rt"); /* "t" is ANSI */

 if (fp == NULL){    /* read below about changing fopen(), etc. */

    printf("fopen failed to open %s.", etc_password);

    fputs("No secret file foundn", stderr);

    exit(1);

 }

 

 while(TRUE){

    fp_bytes_read = fread(buffer, 1, BUFSIZ, fp);

    if (fp_bytes_read == 0)

       break;

    printf("n%ld bytes read and copied from %s.n", 

                fp_bytes_read, etc_password);

 }

 rewind(fp); /* go back to the start of the file */

 

/* Find the descriptor associated with a stream */

  fd2 = fileno(fp);

  if (fd2 == -1)

    fputs("fileno() failedn", stderr);

 

/* Find the stream associated with a descriptor */

  fp2 = fdopen (fd2, "rt");

  if (fp2 == NULL)

    fputs("fdopen failedn", stderr);

  fclose(fp2);

  return (etc_password);

}

 

        int 

main(int argc, const char *argv[])

{

  const char *secret = bufferedRead();

 

  printf("Your secret file found in: %sn", secret); 

        if (argv[1] == NULL) 

                rawRead("/config.sys");

        else

                rawRead(argv[1]);

 

  return(0);

}

-----snip: cut here-----

4.3  Module Definition File - Executable

compile the example with

'gcc -Wall -O2 -Zomf -Zcrtdll -Zbin-files -o demo.exe 

demonstrate_EMX_and_XFree86_os2_io.c'

set LINK386=/PM:VIO/NOIGNORECASE/NOO

Note

#1:

In my opinion, too small stack sizes, usually chosen with the -Zomf switch and its far too small default size of 32k, are the #1 reason for unexpected program behaviour or seemingly random crashes. The effect often only becomes visible if large data objects are processed. So please remember ! Don't be too shy to preallocate several Mbytes, as OS/2 will efficiently manage physical memory.

-----snip------ ; 
;EXE.DEF: module definition file for an executable 

;running in an OS/2 window or under XFree86/OS2, 

; 

NAME WINDOWCOMPAT NEWFILES 

DESCRIPTION 

  'demo programs, emx 0.9d runtime required'

; cf. emx Reference: Equal to 'gcc -Zstack 512' 

STACKSIZE 

  524288 

; enlarge if necessary

HEAPSIZE 

  524288

; enlarge if necessary

; Note:  X  apps coming from Unix often need huge values, 

; 32 Mbytes won't hurt!

; end module definition file 

------snap------

'gcc -Wall -O2 -Zomf -Zcrtdll -Zbin-files -o demo.exe  
EXE.DEF demio.c'

Note

#2:

``The malloc() implementation of emx 0.9c and later can use multiple heap objects under OS/2; in consequence, the heap size is no longer limited under OS/2. However, if you use another malloc() implementation which does not support non-contiguous memory allocation or if the application is run under DOS, the maximum size of the heap available for malloc() is fixed. [...]

The default heap size is 32 MByte under OS/2. For the emx implementation of malloc(), this is the size of the initial heap object and the minimum size of additional heap objects. If ld and emxbind are used for linking, you can change the heap size with the -h<heap_size> option of emxbind.

If emxomf and LINK386 without -Zsys are used for linking, the heap size cannot be changed. If -Zsys is used, initialize the variable _sys_heap_size to the desired heap size:

unsigned _sys_heap_size = 0x4000000; /* 64 MByte */ 

This can be done in any module explicitely linked to your application. ``

Chapter 5
XFree86: Hardware Gets Involved

0.40section

In the old days, when men still were REAL MEN and wrote their own device drivers...

(Linus Thorvalds)

5.1  Devices, X\protect Servers, Clients

5.1.1  '/dev/fastio$', '/dev/console$', '/dev/ptyP0$', etc.

Many uncommon devices are being used for X.

A GUI through slip and a 9600 baud modem ain't no fun. And even an overcrowded Ethernet makes you spend your youth in front of a monitor waiting for response... So you will prefer fast local communication. For this you should compile your apps on OS/2!

This is what you will find in the current XFree86/OS2 implementation's device front:

"If you must do port I/O, use the functions from xf86sup.sys for this; this driver is there anyway, so why not take advantage of it? Common precautions for directly accessing hardware apply. Code that relies on Unix security, job control, tty/pty handling, needs major rewriting. In most cases, it is easier to make a separate file for OS/2 rather than clutter the original one with #ifdef clauses."

Q: What is the XF86SUP.SYS device driver good for?
"This device driver provides certain essential base functionality which is missing from standard OS/2." [...]

"When I use xconsole, there is always a warning message in the first line. What is that? You mean the text Warning: process PID=????? is accessing /dev/fastio$, right? This is an intentional notification that some process uses the fastio device to perform access to I/O ports. It is also intentional that this message cannot be disabled. As long as the PID is the process ID of the X server, everything is okay. If the PID belongs to another, unknown process, you should seriously ask the author of the program what he/she is doing with our hardware. [...]"

(The XFree86/2 faq)

Cf. http://set.gmd.de/~veit/os2/xf86sup.html

and the excerpts of the faq, reordered and abridged by me for a faster overview:

5.1.2  Concepts Related to Hardware (vio) Access and Kernel Level API 's

OS/2 API : "You may use all Dos* API functions, as long as they don't conflict with emx somehow (e.g. _beginthread vs. DosCreateThread). Stay away from Kbd*, Mou*, Vio* calls - none of those will do what you expect, but you can severely influence the operation of the X11 system. [...]"

Hint: Use instead ANSI C ('stdio.h') (or POSIX ('unistd.h')) interfaces!

"3.11 Can I use another shell like 'bash' in place of CMD.EXE in an xterm? Most shells are subtly incompatible with the PTY mechanism used in xterm. A symptom may be that they won't work either if you redirect console I/O to a COM terminal which should be possible. bash is such an incompatible shell. See also Q 3.12 and Q 3.13.

Meanwhile, there are special ports of tcsh and ksh which work in an xterm.

3.12 I use 4OS/2 (or some other shell) as my shell. Will it work in xterms? 4OS2 has a number of misfeatures, politely spoken, in an xterm. We attempt to fix this in the future. For now, don't use it in an an xterm. If 4OS2 or another shell is in your OS2_SHELL or SHELL variable, please add an environment variable named X11SHELL to point to a valid CMD.EXE path, e.g. SET X11SHELL=D:OS2CMD.EXE in your CONFIG.SYS. This will override the OS2_SHELL or SHELL setting which is used otherwise. 3.13 Why does OS2MORE.COM give a SYS0447 in an xterm? This is a program which silently assumes it has access to the keyboard and the screen: it uses functions from KBDCALLS, MOUCALLS, or VIOCALLS. There are more programs of this kind, for instance: most *.COM in os2, ATTRIB.EXE, BACKUP.EXE, CACHE.EXE, EAUTIL.EXE, FIND.EXE, HELPMSG.EXE, LINK.EXE, LINK386.EXE(!), PATCH.EXE, PSTAT.EXE, REPLACE.EXE, RESTORE.EXE, SETBOOT.EXE, SORT.EXE, SPOOL.EXE SYSLEVEL.EXE, TEDIT.EXE, TRACE.EXE, UNPACK.EXE, XCOPY.EXE, XDFCOPY.EXE, but also unfortunately TELNET and FTP, and some others; infact, almost any 16 bit application is a possible candidate. Note that CMD.EXE is an exception of the rule: it is largely still 16 bit, but is clean for XFree86/OS2 use.

You can use the EWS utility EXEMAP.EXE to check if these DLLs are linked in.

Note: the existance of these DLLs does not mean that the program does not work at all, actually EMX.DLL could call KBD API functions but usually does not in the X11 environment. However, certain unexpected effects may occur in an xterm. I'll try to find a workaround for this in the future, but don't count on this. 3.14 FTP seems to work in an xterm, but it does not hide the password? This is a side effect of what was described in Q 3.13. So actually FTP does NOT work. [...]

5.2  Inter Process (Session) Communication

Now some concepts and short hints related to the NAMED PIPE for local communication with the X server. Emx unfortunately does not yet have a user friendly Posix compatible mkfifo() interface, this is due to the still missing Posix compatible file system. You can use the native dos*npipe() interfaces of OS/2 to create /PIPE/* devices, then register file handles via _imphandle() to use them with emx library functions as well as with the native OS/2 kernel api. Note: Their behaviour is not Posix compatible; cf. for details OS/2 api reference manuals.

"2.52 What does the error message "_X11TransOs2OpenClient: Open server pipe /PIPE/Xxf86.0 failed" mean?
This is for the local named pipe communication the same problem as Q 2.31 is for the network connection. It basically means: the server crashed for some reason, and now a client, e.g. xterm cannot connect to the server. The reason is probably hidden somewhere else in the XF86Config file, or some other setup problem (e.g. network configuration). [...]

2.31 What does "SocketINETConnect() can't connect: errno = 65" mean? What does "SocketINETConnect() can't connect: errno = 61" mean? These messages *can* point out a network installation problem, in many cases they are secondary errors, though, and are really caused by a different problem. Nevertheless, you should ensure that your network is setup correctly. [...]

3.8 How can I enable authorization for host foo.edu without using xhost?
Create the file X0.hosts in XFree86libX11. To enable connections from a certain host, add in a line containing inet:hostname to the file. For example, your file may look like this: inet:foo.edu, inet:friendly.host.edu .. Note that the X0 refers to the zero in the display name os2systemname:0, i.e. in the case of Q 3.6 this would become a X1.hosts then. [...]

5.3  Emx and device driver programming (hardware devices)

Q: Can emx be used to generate 16 bit code for programming OS/2 device drivers?

No!. :

(Holger Veit): "as gas lacks the ability to produce arbitrary segment types beyond 32 bit .text/.data/.bss segments. It can generate 16 bit instructions in a 32 bit segment (prefixed with 0x66/0x67), though."

"But you can write device drivers in 32-bit using some (already existing) thunking mechanism

Check out

/pub/os2/system/drivers/filesys/32drv170.zip 

on hobbes."

http://rover.wiesbaden.netsurf.de/~meile/warpstock/wsp02/en/index.htmlfor the details...

Chapter 6
Building Libraries

6.1  Explaining Some Library Concepts

6.1.1  Dynamic Link Libraries (DLL's) Versus Static Libraries

Dynamic link libraries are great, but ain't easy.

6.1.1.1  Static Libraries (*.a or *.lib)

Historic pre-Posix systems of the Unix family, using the a.out binary format, allowed static linkage only:

Common functions could only be archived for faster linking with the librarian ar and the librarian indexer ranlib (or ar -s), thus creating a.out archives. Most prominent is the standard C library system call library (libc.a), which every program needs. It is linked in automatically. This is equivalent to c.a (a.out) and c.lib (omf) of emx, plus some additional system specific helper libraries. Conversion is possible, cf. emxomf docs.

A difference is the math library (libm.a): Traditional Unix uses a separate libm.a. It must be specified explicitly as last entry of the linker command line (-lm).

Emx automatically statically links those functions in, - in fact they are a part of libc.a -, whereas libm.a is only a dummy. If you use -Zcrtdll, the equivalent import library versions for implicit dynamic linkage are used instead; more details later.

6.1.1.2  Shared Objects (*.so)

AT&T Unix System V Release 4 (SYSVr4) introduced the new ELF binary format, which allowed for the first time an efficient and flexible use of shared global functions and variables (shared objects; *.so).

This is achieved by creating shared read-only code and special shared read-write data segments, which can be used by multiple processes and are now better supported by the new binary format. The global emx picture is here with a grain of salt nearly the same as the one for Unix; omf binary format is potentially somewhat more complex, though. A process (session) consists mainly of the following memory sections:

• Fixed .text: machine langage commands and constants,
• variable global .data: global variables of your program and libraries taken from the heap
• and last but not least .bss: a stack pile, where your processor temporarily jots down it's masses of 'telephone notices' so that it can get them back in the right order; the poor piece of silicium cannot keep everything constantly in mind, aka registers.

All are assembled to relocatable object code - which still contain unresolved symbols - or data segments and later, after symbol fixup, moved to their final memory location by the OS's linker and the executable loader, respectively. A.out segments need in addition a small a.out-loader, which is added by emxbind or they need to be converted by -Zomf and emxomf to native object module format before linking with link386.

6.1.2  Shared and Dynamic Libraries (TOG Spec1170)

Don't confuse the following, just because on OS/2 and on ELF systems they use the same file type, i.e.: *.so or *.DLL, respectively:

Shared libraries

are something implicitly linked against the main executable (i.e., more precisely forwarded to it by import libraries, for systems that use this fixup technique, respectively) and loaded at the start-up of the executable.

Dynamic libraries

are not linked to the main module, but loaded at the run-time, if need be. Any systems comforming to "The Single UNIX Specification version 1/2" from The Open Group, often also referred to as the former Spec1170 by X/Open Ltd., can use the:

dlopen()

call to open dynamic libraries at the run-time in a unified manner and can call

dlsym()

to actually resolve symbols, if needed.

Ports implemented by Alexander Mai, Andreas Kaiser, or an almost identical implementation copyrighted by someone else exist.

6.2  Exporting symbols from a DLL (OS/2 OMF)

Unlike most Posix systems, OS/2-emx does not export automatically any function or global variable. You must write a "module definition file", listing exported symbols. This gives you the chance to make sure that all entry points are identified by ordinal numbers, which have to be *identical* for new library revisions (binary compatibility). You can add new exports at the end of the list, though. If you don't make sure that your entry-points are identical among library revisions and that the called functions and exported global variables or constants do exactly what they should do, you will loose binary compatibility and you will have to relink all applications using the dll. After linking the dll this "module definition file" is then passed to the 'emximp' tool in order to create an "import library" (link library) that contains necessary fixups for the linker to resolve the exporty dll symbols when linking you application against this dll.

"EXPORTS Make functions and variables visible outside the .exe or .dll file. All entry points of a dynamic link library must be exported using EXPORTS. Exporting entry points of .exe files is less common. " (Emx docs).

The 'emxexp' tool as well as studying the library's public header file may help you to identify which symbols are worth to be exported for later linker fixup, when compiling your client application.

Hints:

Use short names (symbols) in a library, especially with C++. The omf linker (link386.exe) has a symbol length restrictions (512 characters). Else you'll probably see multiply defined symbols linker errors. Some more special tips at the end of this chapter.

6.3  'dynamic library' and 'shared object' (*.so) versus DLL

You'll encounter a 'dynamic library' that differs from a 'shared object' (*.so), if you try to port recent apps like 'dia', that come from Posix/SYSVr4-Unix:

"Dia plugins are all dynamic libraries, so the program stops working once you start using NONAME keyword."

Dynamic libraries are NOT linked to the main module but loaded at the run-time. Don't confuse them, just because on OS/2 and on ELF systems they are the same file. Any systems comforming to "The Single UNIX Specification version 1/2" from 'The Open Group' can use the dlopen() system call to load dynamic libraries at the run-time in a unified manner and call dlsym() to resolve symbols. Ports of this special mechanism are available separately; emx does not provide the dlopen() and dlsym() system calls.

6.4  Demo

Here a demo (the common libxpm library); cf. the port's diff by H. Veit. This xpm library revision is e.g. linked into xforms.dll rev. 0.88.

Warning:

Different xpm versions are incompatible. Apparently random crashes are often a sign that you mixed xpm versions and/or headers. You cannot mix versions! Use only the version your toolkit libraries have been tested and/or linked with. Headers are incompatible among versions, too.

Note:

In XFree86-4.0, libXpm will be now finally integrated and frozen.

;Xpmos2.def: sample module definition file for an  X  dll 

;under XFree86/OS2, linked with '-Zcrtdll'

; initialize shared objcts per instance

LIBRARY XPM INITINSTANCE TERMINSTANCE

DESCRIPTION "X11R6 XFree86 libXPM xpm.dll for OS/2 EMX09C VERSION=7 (xpm3.4g)"

;

;reduce memory fragmentation (for larger and less frequently used library 

;code save memory and use the 'CODE LOADONCALL' linker flag instead)

CODE

        PRELOAD

;

;.text objects are globally shared memory, .data is individual to any client 

;process

DATA

        MULTIPLE NONSHARED

;

;is this really needed here? Well, at least doesn't hurt.

STACKSIZE 65536

;

;exported client interface symbols for linker fixup, 'NONAME' reduces table

;size, helps that you link only with binary compatible libraries, else

;you'll experience immediate errors/crash (yes, we'll go the hard way ;-)

EXPORTS

        XpmCreatePixmapFromData @ 1 NONAME

        XpmCreateDataFromPixmap @ 2 NONAME

        XpmReadFileToPixmap @ 3 NONAME

        XpmWriteFileFromPixmap @ 4 NONAME

        XpmCreateImageFromData @ 5 NONAME

        XpmCreateDataFromImage @ 6 NONAME

        XpmReadFileToImage @ 7 NONAME

        XpmWriteFileFromImage @ 8 NONAME

        XpmCreateImageFromBuffer @ 9 NONAME

        XpmCreatePixmapFromBuffer @ 10 NONAME

        XpmCreateBufferFromImage @ 11 NONAME

        XpmCreateBufferFromPixmap @ 12 NONAME

        XpmReadFileToBuffer @ 13 NONAME

        XpmWriteFileFromBuffer @ 14 NONAME

        XpmReadFileToData @ 15 NONAME

        XpmWriteFileFromData @ 16 NONAME

        XpmAttributesSize @ 17 NONAME

        XpmFreeAttributes @ 18 NONAME

        XpmFreeExtensions @ 19 NONAME

        XpmFreeXpmImage @ 20 NONAME

        XpmFreeXpmInfo @ 21 NONAME

        XpmGetErrorString @ 22 NONAME

        XpmLibraryVersion @ 23 NONAME

        XpmReadFileToXpmImage @ 24 NONAME

        XpmWriteFileFromXpmImage @ 25 NONAME

        XpmCreatePixmapFromXpmImage @ 26 NONAME

        XpmCreateImageFromXpmImage @ 27 NONAME

        XpmCreateXpmImageFromImage @ 28 NONAME

        XpmCreateXpmImageFromPixmap @ 29 NONAME

        XpmCreateDataFromXpmImage @ 30 NONAME

        XpmCreateXpmImageFromData @ 31 NONAME

        XpmCreateXpmImageFromBuffer @ 32 NONAME

        XpmCreateBufferFromXpmImage @ 33 NONAME

; X11.def (created by emximp)

LIBRARY X11 INITINSTANCE TERMINSTANCE

 

DESCRIPTION

  'X11.DLL (MIT  X Standards Library for emx-port of XFree86).'

 

CODE 

  LOADONCALL

 

DATA

  LOADONCALL MULTIPLE NONSHARED

 

STACKSIZE   ;is this really useful?

   0x80000

 

HEAPSIZE

   0x160000

 

EXPORTS

  XSaveContext                     @1  NONAME

.... etc.

6.5  How to Link Your DLL

I recommend the following simple (unorthodox, but nearly 'foolproof') procedure:

Use the static archive provided on the net or use the Makefile (normally created by Imake, but today more and more by the OS/2 port of autoconf; details later) to make the static archive foo_s.a. Then:

1. 'emxomf -s -l foo_s.a' (convert to native omf format)
2. 'emximp -o foo.a foo.def' (create import library for later ld linker fixup)
3. 'emxomf -s -l foo.a' (convert to native omf format for link386)
4. 'gcc foo.def -o foo.dll -Zomf -Zdll -Zmtd -Zsysv-signals -Zbin-files -Zlinker /EXEPACK:2 -Zlinker /NOI -Zlinker /NOL -Zlinker /NOD -s -fstack-check -lfoo_s -lmore_foo_libs_here -lXmu -lXt -lSM -lICE -lX11 -lXext -lbsd -lsocket'

Linking Order

Linking order is crucial! The traditional C-style linker must find the fixup for still unresolved symbols in the same library or in another one which is _following_ on the command line. You might even need to link in rare cases like: -lfoo1 -lfoo2 -lfoo1

where foo1 needs symbols exported by foo2 and foo2 needs symbols exported by foo1.

6.6  Library Debugging

For a debuggable library compiled with '-g' you'd use the same line (except: '-Zomf -s' and the '-Zlinker blurb', which are specific to link386).

6.7  Building a C++ DLL

Cf. emx docs:

"You should not use the GNU C++ libraries (libg++ and libstdc++) if you create a DLL (unless you are building a stand-alone DLL or a private C runtime DLL) as there is not yet a DLL version of these libraries.

As creating the module definition file for the DLL by hand is quite boring, you should let your computer do it: see the description of the emxexp tool for details. The sign sample of emxample.zip uses emxexp to create the module definition file.

To call the constructors and destructors for static objects in the DLL, make sure that the _DLL_InitTerm() function of the DLL calls __ctordtorInit() and __ctordtorTerm(). You should call _CRT_init() before __ctordtorInit() to ensure that the C library has been initialized for use by the constructors, even if OS/2 happens to initialize the DLLs in a wrong sequence. " [The default function, being linked automatically with -Zdll, should suffice.]

6.8  Some Special Tricks for libraries

6.8.1  Instead of using preprocessor macros, you can use import libraries

Beware: Redefines of functions are debugging headaches: The actual code won't be explicitly in the sources; you will forget, that you redef'd!

(On Mon, 26 Apr 1999 08:59:44 +0900, Shigeru Miyata wrote:)

Instead of using preprocessor macros, you can use import libraries for the purpose in some cases. If the prototypes of the functions are declared somewhere, you can run emximp to create .a/.lib on the module definition file like this:

LIBRARY emxlibcm 

EXPORTS 

strcasecmp @219

strncasecmp @225 

;getcwd @779 

;chdir @764 

;lstat @561

EXPORTS entryname [=internalname] [@ordinal [RESIDENTNAME|NONAME]] 

[...] 

entryname is the name of the function as made visible outside of the .exe or .dll file. entryname is always converted to upper case. internalname is the name of the function as defined in your program. If =internalname is omitted, it is assumed to be identical to entryname. internalname is case sensitive. Exported functions not only have a name (entryname), they also have an ordinal number, the position within the name table. Using ordinal numbers when importing saves space and is supposed to be faster.

Example:

EXPORTS 

my_qsort=qsort1 @1 RESIDENTNAME 

Use NONAME to avoid putting entryname into the name tables.

6.8.2   Debugging huge DLL's

(On Fri, 15 Oct 99 08:59:15, Asbjoern Pettersen wrote:)

>I'm no popuplog expert. Where do I find more info about it ?

IBM service pac CDs contain a Redbook ("The OS/2 Debugging Handbook, #0.6"), if you really want to undergo the pain of low level (Intel architecture) debugging.

Can't you just compile a.out (no -Zomf and -g, this links also the dll's with ld.exe, but you might have to link the *.o files directly, linking against archives *.a sometimes miss symbols despite ar -s) and use pmgdb or gdb? Moreover you have then the luxury of nice core dumps, that poor non-emx developers can only envy!

So you ought to provide debuggable a.out binaries. What you can do (and what I've done to build LYX dll's) is to compile without -Zomf and with -s. After that you'll compile an unbound a.out symbol file (same procedure as before, just leave out the '.dll' suffix from -o foo.dll. If need be, you can always feed the debugger this symbol file, while the dll has a reasonable size.

>> Well surely I can create "-g" dll monsters ;) But there is a problem, ld.exe

>> refuses to build dll from my .def files

>> complaining about some symbols, etc.. while link386 is up and ok.

>>link386 has /NOI flag, while ld does not have an equivalent.

>I haven't tested but modifying the line 47 of emx/src/emxbind/export.c

>may help.

>Anyway if I were you, I would compile only the relevant object into

>the exe with debugging on. It seems you haven't created the map file

>for glib12.dll Try using exemap as follows. Since I don't have

>gtk-1.2.5 here, the addresses below are not exact.

>Suppose you have a SIGSEGV in GLIB12.DLL 0001:00000e08

>exemap glib12.dll|grep "1 00000"|sort +2

>Now you will have something like

> 6 1 00000038 03 Entry32 EXPORT GLOBAL (G_ARRAY_NEW)

> 4 1 000000F8 03 Entry32 EXPORT GLOBAL (G_ARRAY_FREE)

> 3 1 00000168 03 Entry32 EXPORT GLOBAL (G_ARRAY_APPEND_VALS)

> 7 1 000001A8 03 Entry32 EXPORT GLOBAL (G_ARRAY_PREPEND_VALS)

> 5 1 00000208 03 Entry32 EXPORT GLOBAL (G_ARRAY_INSERT_VALS)

>[...]

> 27 1 00000C74 03 Entry32 EXPORT GLOBAL (G_CACHE_NEW)

> 24 1 00000DD4 03 Entry32 EXPORT GLOBAL (G_CACHE_DESTROY)

> 25 1 00000E34 03 Entry32 EXPORT GLOBAL (G_CACHE_INSERT)

> 28 1 00000EF4 03 Entry32 EXPORT GLOBAL (G_CACHE_REMOVE)

> 26 1 00000FB4 03 Entry32 EXPORT GLOBAL (G_CACHE_KEY_FOREACH)

>(The table above is that of glib12.dll v1.2.2 for emx/OS2)

>Now you know that the crash occurs in somewhere between "1 00000DD4"

>and "1 00000E34". Hence it is in G_CACHE_DESTROY @24 which is in

>gcache.obj

>It is a good bet that the problem is in fact enclosed in gcache.c

>and a few more sources. Then you don't have to build whole the

>library with debugging symbols.

>Add gcache.c to the source of scriptfu.exe, build and run it under

>gdb.

6.8.3   Length limitation for symbol names in OMF-Format

(On Sun, 17 Oct 1999 14:34:21 +0200 (MET DST), Stefan Neis wrote):

>       Hi,

>> 1) When i compiling and linking programme with key -Zomf, i get in result link

>> error message: symbol defined more then once.

>> 2) When compiling and linking without omf libs (with ld) all ok.

>>

>AFAIR, there's some length limitation for symbol names in OMF-Format.

>gcc tends to generate really huge names.  Especially for template

>instantiations a very large part at the beginning of two different

>instantiations might be identical, so the automatic truncating will give

>you the same name for different symbols.

>'nm' might be helpful to verify, if this is your problem.

>If it is, you can't do very much about it, besides praying for

>either IBM removing this stupid length limit or

>somebody modifying gcc to generate shorter unique symbols :-(

Note: pgcc already contains an experimental symbol name compression extension.

6.8.4   Specify the name without the extension

(On Wed, 20 Oct 1999 17:28:36 +0200, Alexander Mai wrote):

>Potential problem might be the DLL name: IIRC one can either use

>  foo

>or

>  z:barfoo.dll

>to access foo.dll (if in LIBPATH).

'foo.dll' generally does not work.  It works if and only if it is already

loaded into memory by other processes (or the process itself).  If you

want to use DLL as a dynamic library, specify the name either without

the extension or by the fullpath with the extension.

6.8.5   C++-DLL's

(Compiling qt apps under EMX; S.Miyata)

" -Qt 1.x apps must be compiled either with

        emx09c without special flags

     or

        emx09d with the flags -fno-rtti -fno-exceptions

  -Qt 2.x apps must be compiled with emx09d without special flags.

  -Zomf is not recommended for Qt 2.x, the link may fail.

Now the problem with stdcpp.a turns out to be a bug in emx09d fixpack2

(missing exports in emxlibc?.dll). Using stdcpp.a from emx09c might work, but

it is the last resort. I cannot guarantee the result.

Also stdcpp.a is compiled with the default flags in gcc 2.8 with '-frtti

-fexceptions'.  Only fix is to recompile stdcpp.a yourself with '-fno-rtti

-fno-exceptions' if you want to use it with Qt 1.x"

On Mon, 1 Nov 1999 16:25:35 +0100, Alexander Mai wrote:

>As long as things are rather simple, small and free of 'exceptions'

>everything seems to be fine. e.g. fltk rarely causes problems.

>And problems aren't solved in current egcs/gcc. It fails for valid

>code and in addition it fails for broken code which used to compile

>for years. It doesn't matther whom to blame for this,

>but from a user's point of view, it's a bad situation.

>Also it makes clear that an EMX 09e/ egcs 2.95.x wouldn't

>solve all problems (if any).

I recommend not to use exceptions, as they are slow and often buggy with all

current compilers. So the idea to recompile 'stdcpp.a' might often be the only

safe solution. Emx docs recommend not to link 'stdcpp.a' into a dll, as binary

compatibility is hard to achieve then.

6.8.6   Multiply defined symbols error with C++-DLL's

I observed a wierd link386.exe bug wit C++-DLL's:

Linking against a static module/library (*.o/*.a) only works, if the system default libraries, which are passed automatically to the linker by the emx frontend, do not contain fixups for already resolved references. I.e.: Your mylib.lib contains your own version of foo::foo() and hence the symbol 'wierd-C++-symbol-name-for-foo::foo-function'. But foo::foo() is already defined in some standard library, which is automatically passed to the linker. Normally this is no problem, because the linker will only use the first reference and ignore the second one, so that you can safely override the standard version withyour private function.

But for some reason it behaves different for certain C++-symbols and tries to link against both contradicting function definitions. So you must temporarily disable the passing of the second symbol. To achieve this you may create a modified import library, which does no more contain the symbols you want to override.

6.8.7   Sys 0008

On Fri, 12 Nov 1999 16:44:44 +0100, Holger Veit wrote:

>> Hi!

>>

>> Today when I run xchat in my X with panel and GNOME icq applet I received approx.

>> such message:

>>

>> Sys 0008: ZVT Not enough memory to run program. I checked theseus I had 30Mb of

>> RAM free.

>

>I have seen this before with memory intensive programs, or debuggable X11 DLLs

>(very large). Another resource is probably exhausted, such as the space

>available for DLLs (64MB) or the process heap (about 300MB).

So, respect DLL resource limits. Process heap limits may be lifted in future versions

of OS/2, meanwhile one must live with the restriction of of 512 MB per process address

space.'

Chapter 7
Imakefile or configure, that's the question!

Traditionally X apps come with an Imakefile to use the old tool available for virtually all official X11R6 platforms (and some others). According to a set of rules and system-specific templates, imake generates a giant Makefile for Unix- style make. OS/2-emx and XFree86/OS2 then uses the GNU make utility for conditional compilation and linkage. Unfortunately this one differs from original AT&T-style make especially in the handling of libraries (archives). If you experience problems, please cf. the original documentation, available in GNU INFO-format.

The FSF GNU project uses a different suite of test scripts called auto* tools, written for Unix: autoconf, autoheader, automake, ... An experimental and useful, yet - due to the different file system, kernel and API design necessarily imperfect - port to OS/2 exists.

7.1  A short overview: Autoconf vs. Imakefile
(by Hung-Chi Chu <hcchu@r350.ee.ntu.edu.tw>)

I would like to share my experience about autoconf and Imakefile:

• autoconf/emx is not perfect, especially on Makefiles generated automatically. I don't like to modify any auto-generated files since they are "monsters" and the modification will be overwritten when the next configure runs.

  Tip #1

  (<miyata@kusm.kyoto-u.ac.jp>): "It also helps to edit config.status and run it, instead of running configure again."

• autoconf/emx is nearly perfect for generating config.h except that occasionally I need to add/modify a few lines in configure.in for searching some special functions in add-on libraries [Cf. the list of ported libraries in the faq].

  Tip #2:

  "Read below about some hackish #define's that simulate Unix API interfaces. Add necessary '#include' and '-lFOO' statements to the test scripts."

• for other auto-generated and non-intermediate files which have path definition in them, such as the script 'gtk-config', write a new file called *.os2; substitute the drive letter with ${X11ROOT} et al.

  Tip #3

  (http://set.gmd.de/~veit/os2/xf86os2.html, by Holger Veit <Holger.Veit@gmd.de>):
  [...] XFree86/OS2 stores all configuration files in a common directory. This is:
  X11ROOT:XFree86libX11, where X11ROOT is an environment variable which contains the drive letter the directory tree is located.
  The common places like /usr/X11, /usr/X11R6, /usr/XFree86, /etc which are used for XFree86 in various Unix flavors, are not supported [...]. It is also there where you should put your XF86Config file, the color database (rgb.txt), the host authorization files (X0.hosts), and various other files. Unix does not have drive letters, so a file (resources, fonts, config, etc.) path starts with '/' normally. [...]
  To prepend the X11ROOT part to a file path, a special function named __XOS2RedirRoot exists which you are supposed to use in these cases, rather than an insane own solution. The prototype of this function is:

  const char *__XOS2RedirRoot(const char *pathname);

- The effort of rewriting Imakefile/Makefile.emx is almost at finding all DEFINES for gcc. This can be easily done by searching -Dfoo in Makefiles generated by configure.

- Following are the main steps I apply to do port:

1. use autoconf/emx to generate config.h, Makefile et al.
2. for the first version:

   1. rewrite Imakefile/Makefile.emx from Makefile.am
   2. check some auto-generated and non-intermediate files. Modify and rename them as *.os2 if necessary
   3. for dynamic library, make a module definition file *.def and
   4. a response file *os2.rsp (necessary for Imakefile)
3. for the following versions:

   1. apply the patch of the previous version
   2. diff -u previous_version/Makefile.am this_version/Makefile.am, and if necessary modify Imakefile/Makefile.am
   3. diff -u of any other special *.in files and modify the corresponding *.os2
   4. for dynamic libraries, check *.def and modify it if necessary
4. for Imakefile: xmkmf; make Makefiles; make or xmkmf -a; make
   for Makefile.emx: gmake -f Makefile.emx all
5. make install and make a patch file by diff -ruN if all work well
6. make the zip package. :-)

7.2  Preliminary steps

The following is meant to overcome some simple, nevertheless annoying problems and to make things work faster: Quick and simple fixes, some dirty stolen tricks.

7.2.1  RTF docs

1. Read the README file! Or should it better be called DON'TREADME (psychology?)
2. Read the INSTALL file!
3. Read any additional info files from the program authors. Many problems ocurring under special UN*X flavours have a similar counterpart on OS/2 (others haven't, but this shall be a simple guide :-)
4. Read the /XFree86/lib/X11/config/README and /XFree86/lib/X11/config/host.def files, for a short introduction to adapting Imake templates. Excerpts introducing some important concepts:

`` The easiest way to write an Imakefile is to find another one that does something similar and copy/modify it! To change any of these variables, edit the site.def file.

Imake.tmpl provides defaults for the following variables: [...; cf. your current version of README]. Library.tmpl provides defaults for the following variables: [...; cf. README]. Server.tmpl provides defaults for the following variables: [...; cf. README]. Threads.tmpl provides defaults for the following variables: [...; cf. README]. An <os>Lib.rules file provides defaults for the following variables: [...; cf. README]. An <os>Lib.tmpl file provides defaults for the following variables: [...; cf. README]. The following variables are used by some part of the tree: [...; cf. README].

Make Variables

The following make variables are used by imake rules and may be set in an individual Imakefile.: [...; cf. README]. Many other make variables are set up by the imake config files and can be used in an Imakefile.  The easiest way to discover them is to look at the Makefile generated by an empty Imakefile.

Comments

Use C comment syntax in an Imakefile for comments that should not appear in the resulting Makefile. Use "XCOMM" at the start of each line to produce a comment that will appear in the Makefile. (The "XCOMM" will be translated into the Makefile comment character "#" by imake.) Do NOT use "#" as a comment character in Imakefiles; it confuses the C preprocessor used by imake on some systems.

Imake variables

Don't abuse the variables in X11.tmpl that describe particular pieces of X by using them to describe your own subsystems. Instead, create new variables that are defaulted using Imake.tmpl variables. [...]

Common Rules

Here are some of the common rules for building programs.  How to use them is described in Imake.rules and in the O'Reilly book "Software Portability with imake."

Basic program-building rules

All of these except NormalProgramTarget also generate rules to install the program and its manual page, and to generate dependencies.

SimpleProgramTarget

Use if there is only one program to be made and it has only one source file.

ComplexProgramTarget

Use if there is only one program to be made and it has multiple source files.  Set SRCS to the names of the source files, set OBJS to the names of the object files, and set DEPLIBS to the libraries that this program depends on.

ComplexProgramTarget_1

Like ComplexProgramTarget, but uses SRCS1, OBJS1, and DEPLIBS1 and can be used with ComplexProgramTarget_2 and ComplexProgramTarget_3 to build up to three programs in the same directory. Set PROGRAMS to the programs built by all of these rules.  For more than 3 programs, use NormalProgramTarget for each.

ComplexProgramTarget_2

Use after ComplexProgramTarget_1 for the second program in a directory.  Uses SRCS2, OBJS2, and DEPLIBS2.

ComplexProgramTarget_3

Use after ComplexProgramTarget_2 for the third program in a directory.  Uses SRCS3, OBJS3, and DEPLIBS3.

NormalProgramTarget

Build a program.  Can be used multiple times with different arguments in the same Imakefile.

[...]

Imakefile for directory with subdirectories

XCOMM sample Imakefile for subdirectories

#define IHaveSubdirs

#define PassCDebugFlags CDEBUGFLAGS="$(CDEBUGFLAGS)"

SUBDIRS = list of subdirs ...

MakeSubdirs($(SUBDIRS))

DependSubdirs($(SUBDIRS))

[...]''

7.2.2  Some additions to /XFree86/lib/X11/config/os2.cf and os2.rules

SMiyata <miyata@kusm.kyoto-u.ac.jp> pointed out in posting to the XFree86/OS2 mailing list:

As "far as possible, the changes must be confined to Imakefiles and definition files bundled in the application. The settings in os2.cf and os2.rules are shared by other applications, and hence may cause conflicts in them. Also, the changes to the template files are local to your setting, thus, although it may help to build binary on your system, it is not suitable for porting, unless they are reflected in the distribution of XFree86 itself.

There are indeed the cases that modifications must be made on template files coming with XFree86 rather than Imakefile and definition files coming with the applications [those often have uncommon names like local.cf or machinefile, etc.]:

Akira Hatakeyama <akira@sra.co.jp> reported elsewhere that, in order to build Xvnc on OS/2, the lines

#define BuildXKB YES  
#define BuildLBX YES

in os2.cf must be modified to

#ifndef BuildXKB  
#define BuildXKB YES  
#endif  
#ifndef BuildLBX  
#define BuildLBX YES  
#endif

since, else, the settings in vnc.def is overwritten by those of os2.cf"

#define OSName           OS/2

#define OSVendor         IBM

#define OSMajorVersion    3

#define OSMinorVersion    0

 

/*

 * C library features

 */

#ifdef __EMX__

 

/* set this to what you like (almost :-) 

 * OS/2 does a nonstandard bootstrap of imake 

 */

#define BootstrapCFlags   -DBSD43

 

/* X_WCHAR and X_LOCALE are set in Xosdefs.h */

#define StandardDefines   /**/

#define XawI18nDefines -DUSE_XWCHAR_STRING -DUSE_XMBTOWC

 

/* This will redirect everything to /XFree86

 * Care will be taken that programs that need a file 

 * from this tree will add the environment variable 

 * X11ROOT to the search path so you can redirect the 

 * stuff to a different drive

 */

#ifdef ProjectRoot

#undef ProjectRoot

#endif

/* This will be mangled with a drive letter in most OS/2 

 * rules ... 

 */

#define ProjectRoot     /XFree86

 

/* ... but not when a user tries to compile something 

 * from a different drive Please report any directories 

 * missed to Holger.Veit@gmd.de 

*/

#ifdef UseInstalled

#define IncRoot $(X11ROOT)/XFree86/include

#define LdPreLib -L$(X11ROOT)/XFree86/lib

#endif

 

/* more directories */

#define AdmDir    $(LIBDIR)/adm

#define ConfigDir $(X11ROOT)$(LIBDIR)/config

        DESTDIR = $(X11ROOT)

 

/* for the man pages: Make clear they are from X11 */

#define BookFormatManPages YES

#define ManDirectoryRoot /XFree86/man/man

#define ManSuffix       1X11

#define LibManSuffix    3X11

#define FileManSuffix   4X11

#define XmanSearchPath  /XFree86/man/man

#define ManSourcePath   $(MANPATH)

 

/* you must have installed groff! */

#define TblCmd          tbl

#define ColCmd          cat

#define NeqnCmd         eqn 

/* groff -Tascii -mandoc is for format portability */

#define NroffCmd        groff -Tascii -mandoc 

#define EqnCmd          eqn -Tascii

#define TroffCmd  groff -Tps

         NEQN = NeqnCmd

        NROFF = NroffCmd

 

/* A set of standard defines (warning: underscored emx 

 * extensions are now excluded from the headers unless 

 * -D_WITH_UNDERSCORE is defined): 

 */

#define OSDefines -D__EMX__ -D__i386__ -DOS2_ 

-Demxos2 -D_POSIX_SOURCE -D_BSD_SOURCE -D_GNU_SOURCE 

 

#ifndef BuildXKB

# define BuildXKB                YES

#endif

#ifndef BuildImplibs

# define BuildImplibs            YES

#endif

#ifndef BuildLBX

# define BuildLBX                YES

#endif

 

/* some more properties of the OS/2 implementation */

#define HasNdbm                 YES

#define HasPutenv               YES

#define HasSockets              YES

#define HasSnprintf             YES

#define HasBsearch              NO

#define HasLargeTmp             NO

#define HasPoll                 NO

#define HasBSD44Sockets         NO

#define HasSecureRPC            NO

#define HasSymLinks             NO

#define HasVFork                NO

#define HasVarDirectory         NO

/* #define HasStrcasecmp  NO 

   see below  */

#define HasStrcasecmp           YES

#define Malloc0ReturnsNull      YES

#define SetTtyGroup             NO

#ifndef HasLibCrypt

# define HasLibCrypt            NO

#endif

#ifndef HasShm

# define HasShm                 YES

#endif

#define ConnectionFlags -DTCPCONN -DOS2PIPECONN

#define StdIncDir       $(C_INCLUDE_PATH)

 

/* XF86Setup stuff - does not work yet! */

#define HasTcl                  NO

/*define TclLibDir $(X11ROOT)$(SHLIBDIR) */

/*define TclIncDir $(X11ROOT)$(INCROOT)/tcl */

/*define TclLibName xtcl */

 

#define HasTk                   NO

/*define TkLibDir $(X11ROOT)$(SHLIBDIR)  */

/*define TkIncDir $(X11ROOT)$(INCROOT)/tk */

/*define TkLibName xtk */

 

/*

 * Compiler Features

 */

#define HasGcc                  YES

#define HasGcc2                 YES

#define HasCplusplus            YES

#define HasGcc2ForCplusplus     YES

#define GccUsesGas              YES

#define UseGas                  YES

#define GnuCpp                  YES

#define DoRanlibCmd             YES

#define NeedConstPrototypes     YES

#define NeedFunctionPrototypes  YES

#define NeedNestedPrototypes    YES

#define NeedVarargsPrototypes   YES

#define NeedWidePrototypes      NO

 

#define CppCmd  cpp

#define StandardCppDefines      -traditional

#define PreProcessCmd           CppCmd

 

#define CcCmd gcc -Wall -mpentium -O2

#define DefaultCCOptions -D__ST_MT_ERRNO__ 

 -Dstrcasecmp=stricmp -Zmtd 

#define LibraryCCOptions -D__ST_MT_ERRNO__ 

 -Dstrcasecmp=stricmp -Zsysv-signals -Zmtd 

#define ServerCCOptions  -D__ST_MT_ERRNO__ 

 -Dstrcasecmp=stricmp -Zsysv-signals -Zmtd 

 

#define CplusplusCmd g++ -O3 -Wall -mpentium /* or gcc */

#define CplusplusOptions -Dstrcasecmp=stricmp -Zmtd  

 -Zsysv-signals 

 

#define MakeCmd x11make 

#define AsCmd   gcc -S

#define GccGasOption    -DGCCUSESGAS

#define AsmDefines      -DUSE_GAS

 

#define InstallCmd      install

 

#define LdCmd   ld

#define ExtraLoadFlags  -Zbin-files -Zmtd -Zsysv-signals 

#define ExtraLibraries  -lsocket -lbsd

 

/* quite a number of programs you need  */

#define ArCmd   ar cq

#define RanlibCmd       ar s

#define BourneShell     /**/

#define LexCmd  flex -l

#define LexLib  -lfl

#define YaccCmd bison

#define LintCmd /**/

#define MvCmd   mv

#define CompressCmd     compress

#define GzipCmd gzip

#define LnCmd cp

#define CpCmd cp

#define RmCmd   ImakeHelper 4 

 

#ifndef UseInstalled

#define ImakeCmd imake

#define MkdirHierCmd mkdirhier

#define DependCmd makedepend

#else

/* imake and mkdirhier come from Imake.tmpl, but DependCmd must be overriden */

#define DependCmd makedepend

#endif

 

/* "shell scripts" in OS/2 have this extension */

#define SHsuf cmd

 

#define InstPgmFlags /**/

#define InstBinFlags /**/

#define InstUidFlags /**/

#define InstLibFlags /**/

#define InstIncFlags /**/

#define InstManFlags /**/

#define InstDatFlags /**/

#define InstallFlags /**/

 

/*#define OptimizedCDebugFlags  DefaultGcc2i386Opt*/

#define OptimizedCDebugFlags 

#define ServerOSDefines XFree86ServerOSDefines -DDDXTIME

#define ServerExtraDefines GccGasOption XFree86ServerDefines

 

#if HasShm

# define ServerExtraSysLibs     -lshm

#endif

 

/*

 * Make & install Features

 */

 

#define AvoidNullMakeCommand    YES

_NULLCMD_ = @ rem

#define NullMakeCommand @ rem

#define StripInstalledPrograms  NO

#define CompressAllFonts        YES

#define CompressManPages        YES

#define GzipFontCompression     YES

#define DefaultUserPath  .;/os2;/emx/bin;/tcpip/bin;$(BINDIR)

#define DefaultSystemPath     /os2;/emx/bin;$(BINDIR)

 

#ifndef ExtraFilesToClean

# define ExtraFilesToClean *.~* *.exe *.dll *.obj *.lib 

   *.map

#endif

 

#if CompressManPages

#define CompressManCmd  gzip -n

   COMPRESSMANCMD = CompressManCmd

#endif

 

/* Use ProgramTargetName() macro: OS/2 has '.exe' suffix

 *   Change your Imakefile!

 */

#define ProgramTargetName(target) target.exe

 

/* we don't name libraries lib*.a 

 *   Change your Imakefile!

 */

#define LibraryTargetName(libname) libname.a

#define LibraryTargetNameSuffix(libname,suffix) Concat(libname,suffix.a)

 

/* ... and we even don't do it in rules that should have

 * used the above LibraryTargetName() macro

 */

#ifndef UnSharedLibReferences

#define UnsharedLibReferences(varname,libname,libsource)                @@

Concat3(DEP,varname,LIB) = _UseCat($(USRLIBDIR)/,$(BUILDLIBDIR)/,libname.a)     @@

Concat(varname,LIB) = LoaderLibPrefix Concat(-l,libname)                @@

LintLibReferences(varname,libname,libsource)

#endif

 

#ifndef SharedLibReferences

#define SharedLibReferences(varname,libname,libsource,revname,rev)      @@

Concat3(DEP,varname,LIB) = SharedLibDependencies(libname,libsource,revname) @@

Concat(varname,LIB) = LoaderLibPrefix Concat(-l,libname)                @@

LintLibReferences(varname,libname,libsource)

#endif

 

#include <os2.rules>

#include <os2Lib.rules>

 

# include <xfree86.cf>

 

#else

#error Edit os2.cf for your (non-EMX) OS/2 system

#endif

7.3  Common initial problems

Initially I thought imake was a labyrinthic beast, deserving the title of the Minotaurus among the make tools. But with small and simple fixes to work around some annoying bugs I was able to tame it a bit, so that 'xmkmf' at least does something useful:

7.3.1  'xmkmf -a' crashes because of 'sh' and 'Rexx' conflicts

Usually Imakefiles silently assume they are running on UNIX in a Bourne 'sh' environment, since authors intentionally do not support non UNIX platforms.

Just try running the following changed Rexx scripts first, to create a working environment (make.cmd):

/* REXX */

'@echo off' PARSE ARG a1 a2 a3 a4 a5 a6 a7 a8 a9 a10

/* x11make.exe 'MAKE=x11make.exe SHELL=d:/os2/cmd.exe' a1 a2 a3 a4 a5 a6 a7 a8 a9 a10 */ 

'SET SHELL=D:OS2CMD.EXE' 

'SET X11SHELL=D:OS2CMD.EXE' 

'set emxshell=D:OS2CMD.EXE' 

'set CONFIG_SHELL=D:OS2CMD.EXE' 

'set MAKE_SHELL=D:OS2CMD.EXE'

x11make.exe 'MAKE=x11make.exe SHELL= MAKE_SHELL=os2cmd.exe 

CONFIG_SHELL=d:os2cmd.exe' a1 a2 a3 a4 a5 a6 a7 a8 a9 a10

/* REXX */

'@echo off' 

PARSE ARG a1 a2 a3 a4 a5 a6 a7 a8 a9 a10 

'set SHELL=d:/bin/ash.exe' 

'SET X11SHELL=D:OS2CMD.EXE' 

'set emxshell=d:/bin/ash.exe' 

'set MAKE_SHELL=d:/bin/ash.exe'

/* x11make.exe 'MAKE=x11make.exe SHELL=' a1 a2 a3 a4 a5 a6 a7 a8 a9 a10 */

x11make.exe 'MAKE=x11make.exe SHELL=d:/bin/ash.exe MAKE_SHELL=d:/bin/ash.exe CONFIG_SHELL=d:/bin/ash.exe' a1 a2 a3 a4 a5 a6 a7 a8 a9 a10

Now lets assume xmkmf worked. Theoretically the steps are:

xmkmf; make Makefiles; make depend; make all 

make install; make install.man

7.3.2  'make' crashes because of 'sh' and 'Rexx' conflicts

Imake is misconfigured on most (commercial) UN*X platforms. Authors normally compensate this by including many lines of (normally highly complicated) Bourne shell code and calling several Bourne shell scripts in the Imakefile. In this case 'make' won't work. Nor does a sh.exe as make shell work: The current implementation depends on cmd.exe and Rexx. Finding and rewriting everything in those immense Imakefiles, helper scripts and Makefiles isn't really an alternative. Just try alternatively make and gmake (having installed e.g. /bin/ash.exe). Now make should simply resume where it has crashed. Just delete 'products' of those crashes (empty script output, truncated object files, core dumps, etc.) before resuming the process. This makes using Imake real fun!

Tips by John Williams <jwilliams@commerce.otago.ac.nz> (concerning 'configure' of GNU 'autoconf', fixed by the OS/2-port, but still valid for certain parallel Bourne sh-scripts very common in Imakefiles and/or config.local, etc.):

"3. Configure couldn't search the OS/2 path for programs because the shell construct:

for foo in bar do blah

apparently needs 'bar' to be a space or colon delimited list, whereas the OS/2 $PATH is semi-colon delimited.

4. Configure could not check to see if it's test programs were being compiled because it tests for success using test -s which means test that the file is not empty. The emx port of GCC creates foo.exe executables and touches foo, leaving an empty file. [...]

5. Overcome the above problems by setting an environment variable UNIX_PATH to be a colon delimited path. Next, search and replace using sed thus:

sed -f os2patch.sed configure > configure.os2

where os2patch.sed looks like:

s/PATH/UNIX_PATH/ s/test -s/test -f/ s/host=NONE/host=OS2/ s/"${IFS}:"$/"${IFS};"/

and then execute sh Configure.os2. Lines three and four are due to Ilya Zakharevich, who also pointed out the following issues:

1.if /tmp doesn't exist, make it

2. ensure CONFIG_SHELL=sh is set in shell environment you are running Configure from.

6. You can find what functions are contained in what object files and libraries by using nm (comes with emx). This also tells you which functions are referenced (called) by functions in that file."

7.4  Imakefile examples

config/README states: "The easiest way to write an Imakefile is to find another one that does something similar and copy/modify it!"

Here some strongly shortened examples that use Bourne shell code: sh -c 'for i in '"$(TEXT) $(SRCS) $(MISC) $(HDRS)"' ; do... etc., etc.

7.4.1  creating a static library (archive):
NormalLibraryTarget(rw, $(OBJS))

Excerpt from Imakefile for xpaint read/write directory:

#include "../version" 

#include "../Local.config" 

INCLUDES = -I.. $(XPM_INCLUDE) $(TIFF_INCLUDE) 

 $(JPEG_INCLUDE) $(PNG_INCLUDE)

 

# Nothing to change below here

TIFF_SRC = writeTIFF.c readTIFF.c 

#ifdef HaveTIFF 

TIFF_OBJ = writeTIFF.$O readTIFF.$O 

TIFF_DEF = -DHAVE_TIFF 

#endif 

[...] 

XPM_SRC = readWriteXPM.c 

XPM_OBJ = readWriteXPM.$O 

SGI_SRC = readWriteSGI.c 

#ifdef SGIArchitecture 

SGI_OBJ = readWriteSGI.$O 

SGI_DEF = -DHAVE_SGI 

#endif

 

DEFINES = $(XPM_DEF) $(TIFF_DEF) $(SGI_DEF) 

 $(JPEG_DEF) $(PNG_DEF) 

 

SRCS = rwTable.c  readWriteXBM.c readWritePNM.c   readWriteXWD.c writePS.c  readGIF.c writeGIF.c   $(XPM_SRC) $(TIFF_SRC) $(SGI_SRC) $(JPEG_SRC)

 $(PNG_SRC) libpnmrw.c 

 

OBJS = rwTable.$O  readWriteXBM.$O readWritePNM.$O 

 readWriteXWD.$O writePS.$O  readGIF.$O writeGIF.$O  

 $(XPM_OBJ) $(TIFF_OBJ) $(SGI_OBJ)  $(JPEG_OBJ) 

 $(PNG_OBJ) libpnmrw.$O 

 

HDRS = libpnmrw.h rwTable.h 

MISC = Imakefile

NormalLibraryTarget(rw, $(OBJS))

 

makelist: sh -c 'for i in '"$(TEXT) $(SRCS) $(MISC) $(HDRS)"' ; do  echo $(CURRENT_DIR)/$$i >> ../filelist ; done '

7.4.2  creating an executable: ComplexProgramTarget(xart)

#include "./version"

#include "./Local.config"

DEFINES = $(ARCH_DEFINES) $(EXTRA_DEFINES) 

$(XPM_INCLUDE) -DXPAINT_VERSION="$(VERSION)"

 

XPSRC = [...]

 

SRCS = $(XPSRC) $(OPSRC) $(XPWIDSRC)

OBJS = $(XPOBJ) $(OPOBJ) $(XPWIDOBJ)

 

HDRS =  [...]

 

SYS_LIBRARIES = XawClientLibs -lm

 

#ifdef SGIArchitecture

SGI_LIB = -limage

#endif

 

DEPLIBS = always xart.man xart.ad

LOCAL_LIBRARIES = LibraryTargetName(rw/librw.a) $(XPM_LIB) $(TIFF_LIB)

$(SGI_LIB) $(JPEG_LIB) $(PNG_LIB)

 

SUBDIRS = rw

 

ComplexProgramTarget(xart)

InstallAppDefaults(XArt)

 

NamedMakeSubdirs(always, $(SUBDIRS))

MakefileSubdirs($(SUBDIRS) bitmaps Doc)

CleanSubdirs($(SUBDIRS))

 

# Other targets...

 

clean::

    rm -f xart.ad.h DefaultRC.txt.h Help.txt.h PGP.*  

     xart.man xart.ad

 

includes:: xart.ad.h DefaultRC.txt.h Help.txt.h

 

xart.man:       xart.man.in version

        sed -e 'sßXPAINT_VERSIONß$(VERSION)ß' < xart.man.in > $@

 

Colormap.$O: ColormapP.h Colormap.h

[...]

makelist:: 

        -$(RM) filelist

        @touch filelist

 

makelist:: 

        sh -c 'for i in '"$(TEXT) $(SRCS) $(MISC) $(HDRS)"' ; 

        do echo $(CURRENT_DIR)/$$i >> $(TOP)/filelist ; done '

 

NamedTargetSubdirs(makelist, $(SUBDIRS) bitmaps Doc, , , makelist)

 

 

kit: makelist

        sh -c 'sum="`cat filelist`" ; makekit -oMANIFEST MANIFEST

$$sum'

 

tar: makelist

        sh -c 'cd .. ; rm -f xpaint-$(VERSION).tar.gz; 

        tar czf xpaint-$(VERSION).tar.gz 

        `sed -e "s:^./::" -e "s:^:xpaint/:" xpaint/filelist`'

 

TAGS:   $(SRCS) $(HDRS) $(RWSRC)

        etags -t -o ./TAGS $(XPSRC) $(OPSRC) $(XPWIDSRC) $(RWSRC)

$(HDRS)

 

certification:  makelist

        -$(RM) $(CERTIFICATION) 

        certify `cat filelist`

7.5  Some maybe useful defines

You may want to add parts of it to your Imakefile (or Local.config, etc.) in case of problems. Use comments, else you'll forget about these hackish def's!! Better: Include the necessary defines in config.h. You find them below being commented out (XCOMM XCOMM).

Beware: Redefines of functions are debugging headaches: The actual code won't be explicitly in the sources; you will forget, that you redef'd! Awful coding style!! Read the Portability chapter.

Giant command lines may require a good shell with a large input buffer. You can avoid this by adding necessary redefinitions to 'config.h' and not to the command line (-Dfoo).

XCOMM XCOMM /* OS/2 does not have everything */

#ifdef OS2Architecture 

XCOMM XCOMM remove 'XCOMM XCOMM' comments, if you receive linkage errors

XCOMM XCOMM  -DNEED_STRCASECMP 

#include <string.h>

#define StrcasecmpDefines -Dstrcasecmp=stricmp 

    -Dstrncasecmp=strnicmp -Dmemcasecmp=memicmp

 

ARCH_DEFINES = -DSTDC_HEADERS=1 -DHAVE_STRING_H=1 -DHAVE_STDLIB_H=1   

 -DHAVE_PARAM_H=1 -DHAVE_MEMORY_H=1 -DHAVE_ALLOCA_H=1 -DHAVE_ALLOCA=1 

 -DHAVE_STRERROR=1 -Dlstat=stat -DS_ISLNK=false -DS_ISBLK=false 

 -Dsymlink()=(0) -Dreadlink(s,t,l)=(strcpy(t,s),strlen(t)) 

 

XCOMM XCOMM  Try to include it in config.h

XCOMM XCOMM #define S_ISLNK() false

XCOMM XCOMM #define S_ISBLK() false

XCOMM XCOMM #define symlink() (0)

XCOMM XCOMM #define readlink(s,t,l) (strcpy(t,s),strlen(t))

 

XCOMM XCOMM /* Compare S1 and S2, ignoring case.  */

XCOMM XCOMM int memicmp(__const__ char *, __const__ char *, size_t);

 

XCOMM XCOMM #ifndef memcasecmp 

XCOMM XCOMM #define memcasecmp memicmp  // emx uses M$ fn name

XCOMM XCOMM #endif

 

XCOMM XCOMM /* Compare S1 and S2, ignoring case.  */

XCOMM XCOMM int stricmp(__const__ char *, __const__ char *);

 

XCOMM XCOMM #ifndef strcasecmp

XCOMM XCOMM #define strcasecmp stricmp

XCOMM XCOMM #define _strcasecmp stricmp 

XCOMM XCOMM #endif 

 

XCOMM XCOMM #if !defined(__STRICT_ANSI__) && !defined(_POSIX_SOURCE)

 

XCOMM XCOMM /* Compare no more than N chars of S1 and S2, ignoring case. */

XCOMM XCOMM int strnicmp(__const__ char *, __const__ char *, size_t);

 

XCOMM XCOMM #ifndef strncasecmp 

XCOMM XCOMM #define strncasecmp strnicmp

XCOMM XCOMM #define _strncasecmp strnicmp

XCOMM XCOMM #endif 

 

XCOMM XCOMM #endif /* !defined(__STRICT_ANSI__) */

 

XCOMM XCOMM /* drive letter support needs major rewriting:

XCOMM XCOMM  * #define getcwd _getcwd2

XCOMM XCOMM  * #define chdir _chdir2 

XCOMM XCOMM  */

 

XCOMM XCOMM  /* Adapt hardcoded path names to XFree86/OS2 structure */

 

XCOMM XCOMM  #if defined (__cplusplus)

XCOMM XCOMM  extern "C" {

XCOMM XCOMM  #endif

XCOMM XCOMM  const char* __XOS2RedirRoot(const char* pathname);

XCOMM XCOMM  #if defined (__cplusplus)

XCOMM XCOMM  }

XCOMM XCOMM  #endif

 

XCOMM XCOMM  #define "/usr/X11/foo" (__XOS2RedirRoot("/xfree86/lib/x11/foo"))

 

#endif

Chapter 8
XFree86/2 and Multi-threading

Most X applications are single-threaded, i.e. multi-threading effects are 'emulated' using the Unix fork() syscall.

The Port of XFree86 to OS/2 is not thread-safe, i.e. you can only make X11 calls from a single-thread. emx select() (including all X i/o that depends on select()) are not (!) thread-safe. This means that using multi-threading within X applications is for experts only. Calls to a second thread that interacts with X will crash your application. (unless you use some inter-thread communication to serialize the calls).

Compile and link with

-D__ST_MT_ERRNO__ -Zmtd

Notes:

Existing servers and clients which use Posix threads (PTHREADS) have not yet been ported to OS/2. There is ongoing work to achieve PTHREADS compatibility for OS/2-emx, though. XFree86/2 libraries (DLL's) and applications are prepared for later extensive use of multi-threading.

Cf. emx docs:

"Use _beginthread() to start a new thread. Do not use DosCreateThread unless the new thread doesn't call C library functions.

The C library functions in mt/c.a, mt/c.lib, and emxlibcm.dll are not yet completely thread-safe. [...]

If you are very careful, you can write multithread programs that use the single-thread libraries. Only the main thread is allowed to call library functions that have side effects, including functions that set errno. All other threads should use OS/2 API functions instead. " [Parallel precautions apply to multithreaded X applications.]

Chapter 9
Numerical Math

9.1  Types, Limits, Conversion.

For the following, please cf. the macros in <limits.h>:

The smallest integral type in C is named char, with the variants signed char and unsigned char. char differs from the other integral types in that it is not defined whether signed or unsigned. The next integral types are short int (often just referred to as short), int (including enum !) and long int (often just long.) All of them have unsigned versions.

To be on the safe side you must specify which you mean. However, conversions to other types require some precaution (see below). The maximum value for a char is CHAR_MAX, and the minimum CHAR_MIN, for signed char SCHAR_MAX and SCHAR_MIN and the maximum unsigned char is UCHAR_MAX (the minimum unsigned anything is always 0.)

There are three floating point types in C, float, double and long double. The floating point types are always signed. If you have operations with mixed types, the type with the largest value space decides the precision the operation will be done with. For char they're done at int precision. For float the arithmetics are done at double precision.

Integral Types:

The value space of an integral type is not rigorously defined; check CHAR_BIT, which on OS/2 is always 8. ANSI C only requires that the value space of short is a subset of that of int which is a subset of that of long, and that int is the type most natural for the processor. For 32-bit OS/2 this means that int and long is 32 bits. The Alpha uses 16-bits for short and 64 bit long. Signed integral types are usually not quite symmetric due to the representation used (although they actually can be, this depends on the processor.) Usually the negative range is one number larger than the positive range. For example the maximum signed 16-bit number is 32767 and the minimum signed 16-bit number is -32768 for Intel processors.

Floats and Doubles :

The smallest floating point type float is, just like the smallest integral type char a special case. Arithmetic is not done with them; they are just a way of storing data with a certain precision. Arithmetic is instead done with the precision of a larger type. K&R encouraged the interchangeable use of float and double since all expressions with such data types where always evaluated using the double representation - a real nightmare for those implementing efficient numerical algorithms in C.
According to the ANSI standard such programs will continue to exhibit the same behavior as long as one does not prototype the function. Therefore, when prototyping functions make sure the prototype is included when the function definition is compiled so the compiler can check if the arguments match.

• Keep in mind that the double representation does not necessarily increase the precision. Actually, in most implementations the worst-case precision decreases but the range increases.
• Do not use double unnecessarily since in most cases there is a large performance penalty. Furthermore, there is no point in using higher precision if the additional bits which will be computed are garbage anyway. The precision one needs depends mostly on the precision of the input data and the numerical method used.

Conversions

are, as far as possible, sign preserving. Overflow wraps around, according to the processor model. With VisualAge C++, where char is unsigned by default, the output of the following becomes:  
unsigned => 255, signed => -1, unknown => 255.
Emx gives:  
unsigned => 255, signed => -1, unknown => -1:

#include <stdio.h>

int main(void) { 

  unsigned char uc=255; 

  signed char sc = 255; 

  char c = 255;

  int iuc = uc, isc = sc, ic = c;

    printf(

"unsigned => %d, signed => %d, unknown => %dn", 

           iuc, isc, ic); 

    return 0; 

}

 

9.2  IEEE Maths.

Emx provides programming interfaces which comply to the recent standards of the ANSI mathematical amendments and IEEE, including infinity and NaN (not-a-number) values.

Infinity

is a value with an associated sign that is mathematically greater in magnitude than any binary floating-point number.

NaN

is a value in floating-point computations that is not interpreted as a mathematical value, and that contains a mask state and a sequence of binary digits.

The value of infinity can be computed from 1.0/0.0. The value of a NaN can be computed from 0.0/0.0. Depending on its bit pattern, a NaN can be either quiet (NaNQ) or signaling (NaNS), as defined in the ANSI/IEEE Standard for Binary Floating-Point Arithmetic (754-1982). A NaNQ is masked and never generates exceptions. A NaNS may be masked and may generate an exception, but does not necessarily do so. Emx currently supports only quiet NaN values; all NaN values discussed below refer to quiet NaNs.

NaN and infinity values are defined as macro constants in the <math.h> header file. You can get the corresponding float and long double values by a cast, you may obtain negative values by using the unary minus operator. The value of 0.0 can also be positive or negative, e.g. 1.0/(-0.0) results in -INFINITY.

Note:

Although positive and negative infinities are specific bit patterns, NaNs are not. A NaN value is not equal to itself or to any other value. For example, if you assign a NaN value to a variable x, you cannot check the value of x with the statement if (NAN == x). Instead, use the statement if (x != x). All relational and equality expressions involving NaN values always evaluate to FALSE or zero (0), with the exception of not equal (!=), which always evaluates to TRUE or one (1).

Testing for equality:

Real numbers are handled in ways that don't guarantee expressions to yield exact results. It's risky to test for equality. Better is to use something like

d = max(1.0, fabs(a), fabs(b))

and

then test fabs(a - b)/d against a relative error margin.
Useful constants in float.h are FLT_EPSILON, DBL_EPSILON, and LDBL_EPSILON, defined to be the smallest numbers such that

1.0f + FLT_EPSILON  != 1.0f 
1.0  + DBL_EPSILON  != 1.0  
1.0L + LDBL_EPSILON != 1.0L

respectively.

 

Avoiding over- and underflow:

You can test the operands before performing an operation in order to check whether the operation would work. You should always avoid dividing by zero. For other checks, split up the numbers into fractional and exponent part using the frexp() and ldexp() library functions and compare the resulting values against HUGE_VAL (all in <math.h>).

Infinity:

 

The IEEE standard for floating-point recommends a set of functions to be made available. Among these are functions to classify a value as NaN, Infinity, Zero, Denormalized, Normalized, and so on. Most implementations provide this functionality, although there are no standard names for the functions. Such implementations often provide predefined identifiers (such as _NAN, _INFINITY, etc) to allow you to generate these values.

If x is a floating point variable, then (x != x) will be TRUE if and only if x has the value NaN. Many Unix C implementations claim to be IEEE 748 conformant, but if you try the (x!=x) test above with x being a NaN, you'll find that they aren't.

In the mean time, you can write your own `standard' functions and macros, and provide versions of them for each system you use. If the system provides the functions you need, you #define your `standard' functions to be the system functions. Otherwise, you write your function as an interface to what the system provides, or write your own from scratch.

9.3  Useful Math Extensions

Some useful optimized extensions (functions and macros) are declared and defined in this header (simply include it instead of math.h):

/* gnumath.h (emx+gcc)

   Declarations and definitions for additional math functions.

   */

/*

   Parts: Cephes Math Library Release 2.3:  March, 1995

       Copyright 1984, 1995 by Stephen L. Moshier

   Documentation is included on the distribution media as

   Unix-style manual pages that describe the functions and their

   invocation.  The primary documentation for the library functions

   is the book by 

        Moshier, Methods and Programs for Mathematical Functions, 

        Prentice-Hall, 1989.

   Parts: Copyright (C) 1991, 92, 93, 95, 96, 97, 98 Free Software 

   Foundation, Inc. This file is part of the GNU C Library.

   See the GNU Library General Public License for more details.

 *

 *      Extensions to ISO C Standard: 4.5 MATHEMATICS   <math.h>

 */

#ifndef _GNUMATH_H__

#define _GNUMATH_H__

 

#if ( !defined (__STRICT_ANSI__) && !defined (_POSIX_SOURCE) && 

 !defined (__NO_C9X) ) || defined (_WITH_UNDERSCORE)

#undef _WITH_UNDERSCORE 

#define _WITH_UNDERSCORE 1

 

/* Machine-dependent inline versions */

#if (!__GNUC__ || __GNUC__ < 2 ||   

 (__GNUC__ == 2 && __GNUC_MINOR__ <= 7))

/* The gcc, version 2.7 or below, has problems with all this inlining

   code.  So disable it for this version of the compiler.  */

# undef __NO_MATH_INLINES

# define __NO_MATH_INLINES

#else

# define MATH_INL static __inline__

MATH_INL double expm1(double);

MATH_INL double log1p(double);

MATH_INL long double log1pl(long double);

MATH_INL double asinh(double);

MATH_INL long double asinhl(long double);

MATH_INL double acosh(double);

MATH_INL long double acoshl(long double);

MATH_INL double atanh(double);

MATH_INL long double atanhl(long double);

MATH_INL double coshm1(double);

MATH_INL double acosh1p(double);

MATH_INL double logb(double);

MATH_INL double drem(double, double);

MATH_INL void sincos(double, double *, double *);

MATH_INL double pow2(double);

MATH_INL double hypot(double x, double y);

#endif /* __NO_MATH_INLINES */

#endif /* ANSI */

 

/* Get machine-dependent HUGE_VAL value (returned on overflow).

        On all IEEE754 machines, this is +Infinity.  

   SVID wants HUGE instead of infinity.

   Get machine-dependent NAN value (returned for some domain errors).  */

#include <math.h>

 

/* X/Open wants another strange constant. */

#include <ieee.h>  /* MAXFLOAT */

#include <float.h> /* Some useful constants. */

#include <limits.h>

 

#if !defined (__STRICT_ANSI__) && !defined (_POSIX_SOURCE) && 

 !defined (__NO_C9X)

 

/* The above constants are not adequate for computation using `long double's.

   Therefore we provide as an extension constants with similar names as a

   GNU extension. (emx gcc version)

 */

#define M_El             2.71828182845904523536L        /* e */

#define M_LOG2El         1.44269504088896340736L        /* log 2e */

#define M_LOG10El        0.434294481903251827651L       /* log 10e */

#define M_LN2l           0.693147180559945309422L       /* log e2 */

#define M_LN10l          2.302585092994045684018L       /* log e10 */

 

#define M_PIl            3.14159265358979323846L

#define M_PI_2l          1.57079632679489661923L        /* pi/2 */

#define M_PI_4l          0.785398163397448309616L       /* pi/4 */

#define M_1_PIl          0.318309886183790671538L       /* 1/pi */

#define M_2_PIl          0.636619772367581343076L       /* 2/pi */

#define M_2_SQRTPIl      1.12837916709551257390L        /* 2/sqrt(pi) */

#define M_SQRT2l         1.41421356237309504880L        /* sqrt(2) */

#define M_SQRT1_2l       0.707106781186547524401L       /* 1/sqrt(2) */

 

/* Some Useful Constants From Cephes Math Library */

#define MAXNUML   LDBL_MAX

#define MACHEPL   LDBL_EPSILON    

#define MAXLOGL   1.1356523406294143949700E+04L  /* logl(MAXNUML) */

#define MINLOGL (-1.1355137111933024058900E+04L) /* logl(LDBL_MIN) */

#define LOGE2     6.9314718055994530941723E-1

#define LOGE2L    6.9314718055994530941723E-1L

#define LOG2EL    1.4426950408889634073599E0L

#define PIL       3.1415926535897932384626L

#define PIO2L     1.5707963267948966192313L

#define PIO4L     7.8539816339744830961566E-1L

#define NEGZEROL  (-0.0L)

#define NANL      (long double)NAN

#define INFINITYL (long double)INFINITY

#define NANF      (float)NAN

#define INFINITYF (float)INFINITY  

#include <errno.h>

 

#if !defined __NO_MATH_INLINES || defined _MATH_INLINES

 

#define _sgn(x)  (x == 0.0  ? 0.0  : (x > 0.0  ? 1.0  : -1.0 ))

#define _sgnl(x) (x == 0.0L ? 0.0L : (x > 0.0L ? 1.0L : -1.0L))

#define _sgn1(x)  (x >= 0.0  ? 1.0  : -1.0 )

#define _sgn1l(x) (x >= 0.0L ? 1.0L : -1.0L)

 

/* BSD Functions */

        MATH_INL double

drem(double x, double y) {

  register double value;

  __asm__ __volatile__("1: fprem1nt" "fstsw %%axnt" "sahfnt" "jp 1b"

     : "=t" (value) : "0" (x), "u" (y) : "ax", "cc");

  return value;

}

        MATH_INL double

expm1(double x) {

 register double value, exponent, temp;

  __asm__ __volatile__

    ("fldl2e                    # e^x - 1 = 2^(x * log2(e)) - 1nt"

     "fmul      %%st(1)         # x * log2(e)nt"

     "fstl      %%st(1)nt"

     "frndint                   # int(x * log2(e))nt"

     "fxchnt"

     "fsub      %%st(1)         # fract(x * log2(e))nt"

     "f2xm1                     # 2^(fract(x * log2(e))) - 1nt"

     "fscale                    # 2^(x * log2(e)) - 2^(int(x * log2(e)))nt"

     : "=t" (value), "=u" (exponent) : "0" (x));

  __asm__ __volatile__

    ("fscale                    # 2^int(x * log2(e))nt"

     : "=t" (temp) : "0" (1.0), "u" (exponent));

  temp -= 1.0;

  return temp + value;

}

 

/* Optimized versions for some non-standardized functions.  */

        MATH_INL double

hypot(double x, double y) {

  return sqrt(x * x + y * y);

}

        MATH_INL double

log1p(double x) {

 register double value;

  if (fabs (x) >= 1.0 - 0.5 * M_SQRT2)

    value = log(1.0 + x);

  else

    __asm__ __volatile__

      ("fldln2nt"

       "fxchnt"

       "fyl2xp1"

       : "=t" (value) : "0" (x));

  return value;

}

        MATH_INL long double

log1pl(long double x){

 register long double value;

  if (fabsl(x) >= 1.0L - 0.5L * M_SQRT2l)

    value = logl(1.0 + x);

  else

    __asm__ __volatile__

      ("fldln2nt"

       "fxchnt"

       "fyl2xp1"

       : "=t" (value) : "0" (x));

  return value;

}

        MATH_INL double

asinh(double x) {

 register __const__ double y = fabs(x);

        if (__isnan(x)) return x;

        if (x == 0.0) return x;

        if (!(__isfinite(x))) return x;

        if (y > 1.0e10) return log(x) + LOGE2;

#ifdef GNUFNS  /* This is the GNU original and it is wrong!!! */

/*   return log1p ((y * y / (sqrt (y * y + 1.0) + 1.0) + y)

                * _sgn1l(x));

 */

        return log1p(y +(y * y/(sqrt(y * y + 1.0)+ 1.0)))* _sgn1(x);

#else

        return   log(x +        sqrt(x * x + 1.0)       );

#endif

}

        MATH_INL double

acosh(double x) {

        if (__isnan(x)) return x;

        if (x < 1.0) {errno = EDOM; return NAN;}

        if (x > 1.0e10) {

                if (x == INFINITY) return INFINITY;

                return(log(x) + LOGE2); }

        return log(x + sqrt(x - 1.0) * sqrt(x + 1.0));

}

        MATH_INL double

atanh(double x) {

  register __const__ double y = fabs(x);

        if (x == 0.0) return x;

        if (y >= 1.0) {

                if (x == 1.0) return INFINITY;

                if (x == -1.0) return -INFINITY;

                errno = EDOM; return NAN; }

        if (y < 1.0e-8) return x;

#ifdef GNUFNS

        return -0.5 * log1p(-(y + y)/(1.0 + y)) * _sgn1(x);

#else

        return  0.5 * log  ( (1.0 + x)/(1.0 - x) );

#endif

}

                MATH_INL long double

asinhl(long double x) {

 register __const__ long double y = fabsl(x);

        if (__isnanl(x)) return x;

        if (x == 0.0L) return x;

        if (!(__isfinitel(x))) return x;

        if (y > 1.0e10L) return logl(x) + LOGE2L;

#ifdef GNUFNS  

        return log1pl( (y * y /(sqrtl(y * y + 1.0L) + 1.0L) + y) )

                        * _sgn1l(x);    /* This is corrected */

#else

        return logl( x + sqrtl(x * x + 1.0L));

#endif

}

        MATH_INL long double

acoshl(long double x) {

        if (__isnanl(x)) return x;

        if (x < 1.0L) {errno = EDOM; return NANL;}

        if (x > 1.0e10L) {

                if (x == INFINITYL) return INFINITYL; 

                return logl(x) + LOGE2L; }

        return logl(x + sqrtl(x - 1.0L) * sqrtl(x + 1.0L));

}

        MATH_INL long double

atanhl(long double x) {

  register __const__ long double y = fabsl(x);

        if (x == 0.0L) return x;

        if (y >= 1.0L) {

                if (x == 1.0L) return INFINITYL;

                if (x == -1.0L) return -INFINITYL;

                errno = EDOM; return NANL; }

        if (y < 1.0e-8L) return x;

#ifdef GNUFNS

        return -0.5L * log1pl(-(y + y)/(1.0L + y)) * _sgn1(x);

#else

        return  0.5L *   logl( (1.0L + x)/(1.0L - x) );

#endif

}

MATH_INL double

coshm1(double x) {

  register __const__ double __exm1 = expm1 (fabs (x));

  return 0.5 * (__exm1 / (__exm1 + 1.0)) * __exm1;

}

        MATH_INL double

acosh1p(double x) {

  return log1p(x + sqrt (x) * sqrt (x + 2.0));

}

        MATH_INL double

logb(double x) {

 register double value, __junk;

  __asm__ __volatile__

    ("fxtractnt"

     : "=t" (__junk), "=u" (value) : "0" (x));

  return value;

}

        MATH_INL void

sincos(double x, double *__sinx, double *__cosx) {

 register double __cosr, __sinr;

  __asm__ __volatile__ ("fsincos"

     : "=t" (__cosr), "=u" (__sinr) : "0" (x));

  *__sinx = __sinr;

  *__cosx = __cosr;

}

        MATH_INL double

pow2(double x) {

  register double value, exponent;

  int __p = (int) x;

 

  if (x ==(double) __p)

    return ldexp (1.0, __p);

 

  __asm__ __volatile__

    ("fldl      %%st(0)nt"

     "frndint                   # int(x)nt"

     "fxchnt"

     "fsub      %%st(1)         # fract(x)nt"

     "f2xm1                     # 2^(fract(x)) - 1nt"

     : "=t" (value), "=u" (exponent) : "0" (x));

  value += 1.0;

  __asm__ __volatile__

    ("fscale"

     : "=t" (value) : "0" (value), "u" (exponent));

 

  return value;

}

 

#endif /* _POSIX_SOURCE */

#endif /* __NO_MATH_INLINES */

 

#endif /* GNUMATH_H__ */

Chapter 10
What Shall I Better Change in the Sources?

Last but not least some well meant recommendations from Chapter 5 of the "emx APPLICATION DEVELOPER'S GUIDE", with some additional comments. You might use them as a short checklist.

10.1  Globbing

" If you want Unix-like wildcard expansion built into the program, use

int main (int argc, char *argv[]) { _wildcard (&argc, &argv); /* ... the program ... */ }

This should be done at the very beginning of main(), before ARGC and ARGV are used. See _wildcard() and _response()."

10.2  DOS "text" mode, binary reading, etc.

"Famous problems are related to the binary/text mode mess on OS/2, especially when using stdin/stdout." (Alexander Mai)

" Programs using stdin, stdout or stderr for binary data should call _fsetmode() to switch the stream to binary mode."

" Change all open(), fopen(), fdopen() and freopen() calls to use O_BINARY or "b", respectively, for binary files. If a file contains both binary and textual data, read the file in binary mode and do the conversion yourself."

Note: XFree86/OS2 uses '-Zbin-files'. So "do the conversion yourself" "and ignore carriage returns" and <CTRL Z>.

" Though fseek() and ftell() now work on text files, the offsets are different from what Unix programs expect. You may have to open the files in binary mode and ignore carriage returns (this has been done in GDB)."

" Programs reading a.out files should be changed to call _seek_hdr() or _fseek_hdr() before reading the header to support .exe files. More changes are usually required."

10.3  Process management and fork()

"Replace fork() and exec*() with spawn*(). Under OS/2, fork() is inefficient."
(emx docs)

Note:

This is usually not difficult; it is necessary to use '-Zomf'. Cf. the example below.

10.3.1  What is the difference between fork() and spawnvp()?

10.3.1.1  fork() and emx

The Posix syscall fork() is for traditional Unix systems the only possiblity to create a new process, i.e. a nearly identical clone of an existing one:

"This is achieved by copying the pointers in the PROCESS INFORMATION BLOCK (PIB) of the parent to the newly allocated block in the kernel. On the other hand, on DOS (Windows included) and on OS/2 (Windows NT included) there is no "fork" system call but there is a "spawn" instead. This system call just allocates a new process information block. Only file handles and the environment variable space are inherited and selectors to the address space of the parent process are not copied. So the child cannot access the contents of the memory of the parent. In fact, for OS/2 and Windows NT, it is impossible to implement "fork" in the same way as on the UNIX, since even root has only read access to the PIB table in the kernel. [In fact, there is no such security leak as 'root' in OS/2 (AH)]

The emx library function fork() is actually spawn() system call with a little bit of IPC to copy memory contents. It is inefficient because all the memory contents rather than just pointers are copied.

Well, the difference I was talking was that a spawned child has a clean address space, while a forked one might have a corrupted one." (SMiyata).

10.3.1.2  making it work

"fork() works only in programs linked by ld. It does not work in programs linked by LINK386." [-Zomf]

"fork() doesn't work correctly in multithread programs."

"If the parent process uses the general terminal interface for the keyboard, the child process cannot read from the keyboard using _read_kbd(), the general terminal interface, or the KBD OS/2 API functions.

If the process has a non-contiguous heap (that is, multiple heap objects), fork() will fail. Increase the initial heap size to work around this problem. If any DLL used by the program shares the heap with the program and uses the heap in _DLL_InitTerm() by calling sbrk(), malloc(), etc., fork() will fail. In both cases, fork() will return -1 and set errno to ENOMEM.

See also: alarm(), exec*(), _rmutex_create(), sbrk(), sigaction(), signal(), spawn*(), _uflags(), wait(), waitpid()"

10.3.2  replacing fork() and execvp(), etc.

In Posix the parent process clones itself with fork(), which generates a new process environment, whereas exec() overwrites its own present environment with a new one, keeping the PID. Usually this is combined: exec() is used in the forked child process, overwriting the parent's process status with a new one. Meanwhile the parent usually calls wait() or a home-brew function and waits for child termination, communicated by SIGCHLD.

ps (pstat.exe) gives you the details.

A common problem here is, that the child after termination does not find this wait() etc. point of the parent. Thus a 'zombie' remains. This may lead to a process table overflow, so that the system has to be shutdown and rebooted. All this hassle can be simplified by using spawn*().

Excerpt from exec*() documentation:

" When the new process ends, the parent process will be notified by SIGCHLD and wait() will return the original process ID (PID). [...]

Restrictions:

Native DOS programs cannot be run. The new process gets a new process ID. Unless the new process is an emx program (using emx.dll or emx.exe), the actions for all signals are reset to SIG_DFL in the new process"

"Replace exec*() with spawn*() and exit() if the parent process waits for the termination of the new process (by calling wait() or by waiting for SIGCLD). This is required to keep the process ID of the child process. In a forked process, however, you don't have to do this because emx.dll does it for you."

"Do not use the PTRACE_TRACEME request of ptrace(): use P_DEBUG instead when starting the process with spawn*()."

10.3.3  fork()/exec() and dual way pipe communication

It's no simple replacement when porting UNIX fork() and exec*() pairs to OS/2's spawn*(). The following two programs can be used as an example:

The first using fork() and execv() should be linked with ld but I have no *.a libraries at hand so cannot test it. But I think it is correct.

The second is linked with LINK386 and works fine. Both programs create a child process and then write a message through a pipe to the standard input of the child process. You can test it with cat.exe, which displays what has been read from stdin. This should give you a hint how to port fork() and exec*(). Look at the source of BASH 2.0 for OS/2 for another way to tackle this problem.

0.20subsubsection Vincent Kuo

/* pipe1.c, using fork() and exec*() */ 

#include <unistd.h> 

#include <io.h> 

static const char message[] = "Hello, kid!"; 

 

int

main(int argc, char**argv)

{   /* use the name of the program as argument */ 

 int two_fd[2], two_other_fd[2], pid;  

  if (pipe(two_fd)) /* the reading end is [0] */

    exit(1); /* error creating pipe ;-( */ 

  pid = fork();     /* now I am two! */

  if (pid) {        /* we are the parent! */ 

    if (pid == -1) 

      exit(2);      /* error fork()ing! ;-( */ 

    else { 

      close(two_fd[0]);  /* close unneeded reading end */ 

      write(two_fd[1], message, sizeof(message)); 

      close(two_fd[1]); 

    } 

  } else {              /* we are the forked child */ 

    dup2(two_fd[0], STDIN_FILENO);  /* connect reading 

          end to the new process' stdin, 'close' stdin */ 

    /* dup2(two_other_fd[0], STDOUT_FILENO); */  

    close(two_fd[0]); /* close the old file descriptor */ 

    close(two_fd[1]); /* close unneeded one */ 

    /* close(two_other_fd[0]); close(two_other_fd[1]); */  

    execv("cat", NULL); 

    exit(3);           /* if exec*() returns: Error ;-( */ 

  } 

  return 0; 

}

-------

/* pipe2.c, using spawn*() */ 

#include <process.h> 

#include <fcntl.h> 

#include <io.h> 
 
static const char message[] = "Hello, kid!"; 

 

/* supply the name of the program you want to run as argument */ 

int

main(int argc, char**argv) 
{ 

 int fd[2], ftmp; 

  if (pipe(fd))  

    exit(1);     /* error creating pipe */ 

  ftmp = dup(0);   /* save the stdin handle */ 

  dup2(fd[0],0); /* assign the read end of the pipe to stdin */   close(fd[0]); 

  fcntl(ftmp, F_SETFD, FD_CLOEXEC); 

  fcntl(fd[1], F_SETFD, FD_CLOEXEC);           spawnv(P_NOWAIT,argv[1],argv+1);

  /* restore the original stdin file, necessary if 

     more input from stdin is needed */ 

  dup2(ftmp,0); 

  close(ftmp);

  write(fd[1], message, sizeof(message)); 

  close(fd[1]); 

  return(0); 

}

-------

/*

 *  Routine: popenRW

 *  Returns: PID of child process

 *  Action : Exec program connected via pipe, connect int fd's to 

 *           both the stdin and stdout of the process.

 *  Params : Command - Program to run

 *           Pipes   - Array of 2 ints to store the pipe descriptors

 *                     Pipe[0] writes to child's stdin,

 *                     Pipe[1] reads from child's stdout/stderr

 */

#include <process.h> 

#include <fcntl.h> 

#include <io.h>

#define INCLUDE_DOS

#include <os2.h> 

int

popenRW(char *const argv[], int *pipes)

{

    HFILE Out1, Out2, In1, In2;

    HFILE Old0 = -1, Old1 = -1, Old2 = -1, Tmp;

    int pid;

    if (DosCreatePipe(&Out2, &Out1, 4096))

        return FALSE;

    if (DosCreatePipe(&In1, &In2, 4096))

    {

        DosClose (Out1);

        DosClose (Out2);

        return FALSE;

    }

    /* Save std{in,out,err} */

    DosDupHandle (STDIN, &Old0);

    DosSetFHState (Old1, OPEN_FLAGS_NOINHERIT);

    DosDupHandle (STDOUT, &Old1);

    DosSetFHState (Old2, OPEN_FLAGS_NOINHERIT);

    DosDupHandle (STDERR, &Old2);

    DosSetFHState (Old2, OPEN_FLAGS_NOINHERIT);

    /* Redirect std{in,out,err} */

    Tmp = STDIN;

    DosDupHandle (In1, &Tmp);

    Tmp = STDOUT;

    DosDupHandle (Out1, &Tmp);

    Tmp = STDERR;

    DosDupHandle (Out1, &Tmp);

    /* Close file handles not needed in child */

    

    DosClose (In1);

    DosClose (Out1);

    DosSetFHState (In2, OPEN_FLAGS_NOINHERIT);

    DosSetFHState (Out2, OPEN_FLAGS_NOINHERIT);

    /* Spawn we now our hoary brood. */

    pid = spawnvp (P_NOWAIT, argv[0], argv);

    /* Restore std{in,out,err} */

    Tmp = STDIN;

    DosDupHandle (Old0, &Tmp);

    DosClose (Old0);

    Tmp = STDOUT;

    DosDupHandle (Old1, &Tmp);

    DosClose (Old1);

    Tmp = STDERR;

    DosDupHandle (Old2, &Tmp);

    DosClose (Old2);

    if(pid < 0) 

      {

        DosClose (In2);

        DosClose (Out2);

        return -1;

      }

    

    pipes[0] = In2;

    _setmode (pipes[0], O_BINARY);

    pipes[1] = Out2;

    _setmode (pipes[1], O_BINARY);

    /* Save ID of write-to-child pipe for pclose() */

    ll_insert ((LL_KEY) In2, (LL_VAL) pid);

    

    return pid;

}

/*

 *  Routine: pclose

 *  Returns: TRUE on success

 *  Action : Close a pipe opened with popen();

 *  Params : Pipe - pipe to close

 */

int

pclose (FILE *Pipe) 

{

    RESULTCODES rc;

    PID pid, pid1;

    int Handle = fileno (Pipe);

    fclose (Pipe);

    rc.codeTerminate = -1;

    pid1 = (PID) ll_lookup ((LL_KEY) Handle);

    /* if pid1 is zero, something's seriously wrong */

    if (pid1 != 0)

      {

        DosWaitChild (DCWA_PROCESSTREE, DCWW_WAIT, &rc, &pid, pid1);

        ll_delete ((LL_KEY) Handle);

      }

    return rc.codeTerminate == 0 ? rc.codeResult : -1;

}

#if DIAGNOSTIC

void

main ()

{

  FILE *fp1, *fp2, *fp3;

  int c;

  ll_print ();

  fp1 = popen ("gcc --version", "r");

  ll_print ();

  fp2 = popen ("link386 /?", "r");

  ll_print ();

  fp3 = popen ("dir", "r");

  ll_print ();

  while ((c = getc (fp1)) != EOF)

    printf ("%c", c);

  while ((c = getc (fp2)) != EOF)

    printf ("%c", c);

  while ((c = getc (fp3)) != EOF)

    printf ("%c", c);

  pclose (fp1);

  ll_print ();

  pclose (fp2);

  ll_print ();

  pclose (fp3);

  ll_print ();

  return;

}

#endif /* DIAGNOSTIC */

10.3.4  Example (C++-Code from modified LYX 1.0.4 sources):

The following example demonstrates how the emx spawnvp() interface brings about a considerable simplification of the initial fork(), execvp(), waitForChild() implementation. The frequent usage of '#ifdef __EMX__' reduces readability, though.

interface header

// -*- C++ -*-

#include <sys/types.h>

#include <LString.h>

 

#ifdef __GNUG__

#pragma interface

#endif

 

 

/// Class to controll system-calls according to OS-specific interface

/**@Doc:

 Class, which controlls a system-call according to the specific interface of

 your operating system. 

  

  Instance starts and represents child processes.

  You should use this class if you need to start an external program in LyX.

  You can start a child in the background and have a callback function

  executed when the child finishes by using the DontWait starttype.

  */

class Systemcalls {

public:

        ///

        enum Starttype {

                System,

                Wait,

                DontWait

        };

        

        /// Callback function gets commandline and returnvalue from child

        typedef void (*Callbackfct)(LString cmd, int retval);

        

        ///

        Systemcalls();

        

        /** Generate instance and start childprocess 

          The string "what" contains a commandline with arguments separated 

          by spaces.

          When the requested program finishes, the callback-function is 

          called with the commandline and the returnvalue from the program.

          The instance is automatically added to a timercheck if starttype is

          DontWait (i.e. background execution). When a background child

          finishes, the timercheck will automatically call the callback

          function.

          */

        Systemcalls(Starttype how, LString what, Callbackfct call = 0);

        

        

        ///

        ~Systemcalls();

        

        /** Start childprocess. what contains a command on systemlevel. 

         */

        int Startscript(Starttype how, LString what, Callbackfct call = 0); // for reuse

        

        /** gets PID of childprocess. Used by timer */

        inline pid_t Getpid() { return pid; }

        

        /// Start callback

        inline void Callback() { if (cbk) cbk(command, retval); }

        

        /** Set return value. Used by timer */

        inline void setRetValue(int r) { retval = r; }

private:

        /// Type of execution: system, wait for child or background

        Starttype    start;

        /// Callback function

        Callbackfct  cbk;

        /// Commmand line

        LString      command;

        /// Process ID of child

        pid_t        pid;

        /// Return value from child

        int retval;

        

        ///

        int Startscript();

        

        

        /// Spawn child according to the interface of your OS (AHanses)

        pid_t Fork(int spawnFlag);

        

        /// Wait for child process to finish. Updates returncode from child.

#ifndef __EMX__

        /**

        Waitforchild not necessary with spawnvp() of emx.

        Note (concerning OS/2 specifics): Emx interface implementation of 

        spawnvp(P_WAIT, ...) returns returncode from child (AHanses).

           */

        void waitForChild();

#endif

};

implementation module

 

#include <config.h>

 

#ifdef __GNUG__

#pragma implementation

#endif

 

#include <sys/types.h>

#include <sys/wait.h>

#include <stdlib.h>

#include <stdio.h>

#include <unistd.h>

#include <process.h>

#include "syscall.h"

#include "syscontr.h"

 

//--------------------------------------------------------

// Class, which controlls a system-call

//--------------------------------------------------------

 

// constructor

Systemcalls::Systemcalls()

{

        pid = (pid_t) 0; // yet no child

}

 

// constructor

// 

// starts child

Systemcalls::Systemcalls(Starttype how, LString what, Callbackfct cback)

{

        start   = how;

        command = what;

        cbk     = cback;

        pid     = (pid_t) 0; // no child yet

        retval  = 0;

        Startscript();

}

 

// destructor

// not yet implemented (?)

Systemcalls::~Systemcalls()

{

}

 

// Start a childprocess

// 

// if child runs in background, add information to global controller.

 

int Systemcalls::Startscript()

{

        retval = 0;

        switch (start) {

        case System: 

                retval = system(command.c_str());

                Callback();

                break;

        case Wait:   

                pid = Fork(P_WAIT);

                if (pid>0) { // Fork succesful. Wait for child

#ifndef __EMX__

                        waitForChild();

#endif

                        Callback();

                } else

                        retval = 1;

                break;

        case DontWait:

                pid=Fork(P_NOWAIT);

                if (pid>0) {

                        // Now integrate into Controller

                        SystemcallsSingletoncontroller::Startcontroller starter;

                        SystemcallsSingletoncontroller *contr=

                                starter.GetController();

                        // Add this to contr

                        contr->AddCall(*this);

                } else

                        retval = 1;

                break;

                //default:  // error();

                //break;

        }

        return retval;

}

 

 

// Wait for child process to finish. Returns returncode from child.

#ifndef __EMX__

void Systemcalls::waitForChild()

{

        // We'll pretend that the child returns -1 on all errorconditions.

        retval = 1;

        int status;

        bool wait = true;

        while (wait) {

                pid_t waitrpid = waitpid(pid, &status, WUNTRACED);

                if (waitrpid == -1) {

                        perror("LyX: Error waiting for child");

                        wait = false;

                } else if (WIFEXITED(status)) {

                        // Child exited normally. Update return value.

                        retval = WEXITSTATUS(status);

                        wait = false;

                } else if (WIFSIGNALED(status)) {

                        fprintf(stderr,"LyX: Child didn't catch signal %d "

                                "and died. Too bad.n", WTERMSIG(status));

                        wait = false;

                } else if (WIFSTOPPED(status)) {

                        fprintf(stderr,"LyX: Child (pid: %ld) stopped on "

                                "signal %d. Waiting for child to finish.n", 

                                (long) pid, WSTOPSIG(status));

                } else {

                        fprintf(stderr,"LyX: Something rotten happened while "

                                "waiting for child %ldn", (long) pid);

                        wait = false;

                }

        }

}

#endif

 

// generate child in background

 

pid_t Systemcalls::Fork(int spawnFlag)

{

#ifndef __EMX__

        pid_t cpid=fork();

        if (cpid == 0) { // child

#endif

                LString childcommand(command); // copy

                LString rest = command.split(childcommand, ' ');

                const int MAX_ARGV = 255;

                char *syscmd = NULL; 

                char *argv[MAX_ARGV];

                int  index = 0;

                bool Abbruch;

                do {

                        if (syscmd == NULL) {

                                syscmd = childcommand.copy();

                        }

                        argv[index++] = childcommand.copy();

                        // reinit

                        Abbruch = !rest.empty();

                        if (Abbruch) 

                                rest = rest.split(childcommand, ' ');

                } while (Abbruch);

                argv[index] = NULL;

                // replace by command. Expand using PATH-environment-var.

#ifndef __EMX__

                execvp(syscmd, argv);

                // If something goes wrong, we end up here:

                perror("LyX: execvp failed");

        } else {

#else // spawn child in another process; fork() is not portable

                pid_t cpid = spawnvp(spawnFlag, syscmd, argv);

#endif

        if (cpid < 0) // error

                perror("LyX: Could not fork");

#ifndef __EMX__

        } else // parent

#endif

                return cpid;

        return 0;

}

 

 

// Reuse of instance

 

int Systemcalls::Startscript(Starttype how, LString what, Callbackfct cback)

{

        start   = how;

        command = what;

        cbk     = cback;

        pid     = (pid_t) 0; // yet no child

        retval  = 0;

        return Startscript();

}

 

 

 

//

// Mini-Test-environment for script-classes

//

#ifdef TEST_MAIN

#include <stdio.h>

 

 

int SimulateTimer;

void back(LString cmd, int retval)

{

        printf("Done: %s gave %dn", cmd.c_str(), retval);

        SimulateTimer = 0;

}

 

 

int main(int, char**)

{

        

        SystemcallsSingletoncontroller::Startcontroller starter; 

        SystemcallsSingletoncontroller *contr=starter.GetController();

        

        Systemcalls one(Systemcalls::System, "ls -ltag", back);

        Systemcalls two(Systemcalls::Wait, "ls -ltag", back);

        SimulateTimer = 1;

        Systemcalls three(Systemcalls::DontWait , "ls -ltag", back);

        // Simulation of timer

        while (SimulateTimer)

                {

                        sleep(1);

                        contr->Timer();

                }

}

#endif

10.4  Multithreading instead of fork()/vfork()

[to be written]

10.5  Unix file system issues

The separator of path components is a slash '/' as opposed to OS/2's ''. Note that many APIs understand the '/' on OS/2, but cmd.exe and most apps don't. Multiple paths (e.g. in environment variables like PATH) are separated by a colon ':' The root dir on every un*x system is '/'. Often code checks whether a given path is an absolute one by comparing the beginning with '/'. Use grep for all standard paths and all commands which are mentioned in this introduction. (_fullpath() and _abspath() can also be useful.)

Using

#define getcwd _getcwd2 

#define chdir _chdir2

"Watch out for Unix file system hacks: Unix allows deleting and renaming an open file (the file will be deleted after being closed). [unlink(), remove(), rename()] 

- Watch out for Unix file names (Unix is case sensitive, long file names and multiple dots are allowed). On OS/2's HPFS multiple dots are also allowed; however, trailing dots are not significant (except for the special file names `.' and `..').

- The null device is called /dev/null under Unix. The __open() system call translates the filenames "/dev/null" and "/dev/tty" (lower case, with slashes) to "nul" and "con", respectively. However,

system ("whatever >/dev/null");

won't work as the standard OS/2 and DOS command interpreters don't recognize /dev/null.

- If you want to use for separating directories, changes may be necessary. These changes are optional because / also works. "

"Unix does not have drive letters, so a file (resources, fonts, config, etc.) path starts with '/' normally. XFree86 is located in a tree below %X11ROOT%XFree86... To prepend the X11ROOT part to a file path, a special function named __XOS2RedirRoot exists which you are supposed to use in these cases rather than an insane own solution. The prototype of this function is:

const char *__XOS2RedirRoot(const char *pathname);'

[...] XFree86/OS2 stores all configuration files in a common directory. This is X11ROOT:XFree86libX11, where X11ROOT is an environment variable which contains the drive letter the directory tree is located. The common places like /usr/X11, /usr/X11R6, /usr/XFree86, /etc which are used for XFree86 in various Unix flavors, are not supported [...]. It is also there where you should put your XF86Config file, the color database (rgb.txt), the host authorization files (X0.hosts), and various other files." (FAQ)

"Potential problem might be the DLL name: IIRC one can either use

'foo' or 'x:barfoo.dll'

to access foo.dll (if in LIBPATH)." (Alexander Mai)

"Note that ///abc is a valid Unix filename. It's equivalent to /abc.

- Note that chdir ("..") is a no-op under Unix if the current working directory is the root directory. Under emx, chdir ("..") fails in the root directory.

- Use termio or termios or read the keyboard with _read_kbd() if you don't want to get input line by line.

- Under Unix, directories in environment variables (PATH, for instance) are separated by colons; use semicolons instead.

- The shell isn't called /bin/sh. Use system(). system() and popen() don't expand wildcards (unless COMSPEC points to a shell which expands wildcards)."

So, in short:

"grep for '/dev', '/bin', '/tmp', '/usr', '/proc', ... also check out all commands starting up another process:

exec*, popen, system,...

Remember that cmd.exe doesn't like slashes!" (Alexander Mai)

10.6  Sockets

"emx includes support for sockets of IBM TCP/IP for OS/2 (including the Internet Access Kit of OS/2 3.0). The headers files are derived from NetBSD and are packaged in bsddev.zip. Some manual pages are packaged in bsddoc.zip."

gcc mymain.c mysub.c -lmylib -lsocket -lm

``emx tries to provide seamless support for sockets, like Unix systems do. This approach is quite different from the one taken by IBM's TCP/IP for OS/2 Toolkit.

Sockets are not supported with the system call library sys.lib (-Zsys).

The header <sys/so_ioctl.h>, which includes TCP/IP-specific definitions for ioctl(), is included by <sys/ioctl.h> only if a TCP/IP-related header such as <sys/socket.h> has been included before <sys/ioctl.h>. If you don't want to reorder the #include directives, define the symbol _EMX_TCPIP before including <sys/ioctl.h>, or include <sys/so_ioctl.h> directly (however, the last alternative is not recommended).

When porting Unix applications, please note the following restrictions:

Socket handles are not inherited across exec*() and spawn*(). (Note that a process created by fork() inherits the socket handles of its parent process.) Inheriting socket handles is planned for a future release of emx.

The size of messages is restricted to 32767 bytes (this is a limitation of IBM TCP/IP for OS/2).

Initially, sockets are in binary mode. Use setmode() to switch to text mode. Text mode applies to read() and write(), only. recv() and send() always use binary mode.

The functions recvmsg(), sendmsg(), and socketpair() are not implemented."

#define _PATH_HEQUIV "/tcpip/etc/hosts.equiv" 

#define _PATH_HOSTS "/tcpip/etc/hosts" 

#define _PATH_NETWORKS "/tcpip/etc/networks" 

#define _PATH_PROTOCOLS "/tcpip/etc/protocols" 

#define _PATH_SERVICES "/tcpip/etc/services"

"The difficulty here is that one still needs to annotate each of these paths with a drive letter. XFree86 has to do this with almost any internal path by using a routine named __XOS2RedirRoot which prepends the %X11ROOT% variable. Here is the drawback of DOS like systems to behave like VMS ("everything is a device, and some of them, denoted by drive letters can contain file systems") in contrast to Unix philosophy ("everything is a file, and some of them may have special properties, like being a pipe, or a socket, or a device")." (Holger Veit <Holger.Veit@gmd.de>)

10.7  Environment (getenv())

EDITOR

Your favourite editor. In case you don't want to end up inside vi [a famous un*x editor] and don't know how to get out :q!

HOME

points to a directory which contains user specific data. Always non-NULL on un*x

PAGER

standard tool to page output. Something like "more" or "less". Use xless in XFree.

PATH

same as on OS/2. Note to use ';' instead of ':'.

PRINTER

default printer. Might be used in conjunction with printing tools like "lp", "lpr".

SHELL

standard shell to be used. Often used by Makefiles, ...

TERM

Name of the terminal currently used. (see termcap, terminfo). Make sure your software runs in an OS/2 window as well as in an xterm.

TMP

Usually not necessary, since all systems have a directory for temporary stuff in /tmp. If you don't have 'tmp', create it! Many software depends on this.

WINDOWID

Id of current X11 window.

10.8  Miscellanea

Unfortunately many OS/2 tools, including built-in cmd.exe commands have their counterparts on un*x with exactly the same name but different semantics/syntax. An incomplete list: echo, find, patch, sort.

" By default, signal processing is different when using signal(): SIG_ACK should be used instead of the signal handler address to re-enable a signal by calling signal() when the signal handler has been called. This behavior can be changed with the -Zbsd-signals and -Zsysv-signals options of GCC. If you use POSIX.1 functions for signal handling, SIG_ACK is not required.

- Printing single characters is inefficient. A solution is to use

setvbuf (stdout, NULL, _IOLBF, BUFSIZ)

and to use fflush (stdout) if you need the output immediately (flushing is required only after displaying prompting texts before reading input or displaying progress reports that don't end with a newline character). GDB output has been made much faster by using line buffering.

- Note that VEOF != VMIN and VEOL != VTIME. Programs which use VEOF and VEOL to access VMIN and VTIME, respectively, should be changed to use VMIN and VTIME. emx uses separate fields for VEOF, VEOL, VMIN and VTIME.

- To use termio, you have to reset the IDEFAULT bit of c_lflag. This does not apply to termios."

Part 3
Appendices

Chapter 11
Some Notes on Portability and Coding Style

SMiyata in a contribution to the XFree86/OS2 mailing list pointed out:

Having as many softwares as possible really helps even if their qualities are sometimes questionable. On the other hand, for example, I am using LYX for a practical purpose so that if the reliability of it is lower than that on some other platform, say, LINUX, I will eventually be forced to switch the operating system I use daily. And (more or less) functional ports do not appear over one night. I myself cannot grasp the contents of the source code at a glance. I have to test semi-working versions and investigate the relevant parts of the source again and again to have something working. [...] I was really helped that most of the non-working features wererather evident. I could pin-point the relevent parts of the source easily.

The problem of automating dirty workaround stuffs is that unless the one who ports a software awares that s/he is doing something dirty, it is really likely for him/her to let the semi-working workarounds unnoticed.

There are at least two remedies for this problem.You can let the dummy stub functions to print warning messages [...], although this may alarm also the users of the port. Another possiblity is to make the warning very clear in the doc.

Or do it like other projects: Use an environment variable to switch on/off those warnings. Might even include multiple levels, one printing statements once and another which reminds everytime of some missing functionality.

Write Proper Implementations.  
Example: cpfile.c, a widely portable replacement for link()/symlink().

Alexander Mai wrote:

I'm not sure at all what people think here, but I make some proposals ... :

a) use a file_copy routine for symlink (and don't clean up later perhaps)

b) let getdtablesize return some 'guessed' integer (w/o checking any docs, just get it working in most cases ...)

c) #define lstat stat (neglecting something one could imagine around TVFS)

/**** cpfile.c: cpfile() **** 

 

 From: "mithra" <mithra@bga.com>

 To: emx@queue.iaehv.nl

 Subject: Re: COPY/MOVE system function?

 

 Since I needed a copy I could use everywhere without invoking 

system(), here is cpfile.c:

 

 ****  

 * Copy file from 'in_name' to 'out_name'

 *

 * return:

 * 0 if all's well

 * 1 if can't open in_name

 * 2 if can't open out_name for writing

 * 3 if if stat() in_name returns an error

 * 4 if copy failed (read error)

 * 5 if copy failed (write error)

 * 6 if out_name close() failed (file system error!)

 *   Check to prevent silent data loss, e.g. with network data.

 * 7 if utime() failed (time set error)

 * 8 if chmod() of output file failed 

 * 9 if in_name close() failed (file system error!)

 *

 * in future may want access status of output file...

 * Curtis W. Rendon 12/20/1999 initial

 *  modified 04/01/00 A. Hanses 

 *

 * hereby released to the public domain, please keep my comments,

 * no warranty inferred, implied, or specified.

 *  Curtis W. Rendon mithra@earthling.net  12/29/1999

 *

 ***** Prototype:  */

 

int cpfile (const char *in_name, const char *out_name);

 

#include <stdio.h>

 

#ifdef VIOLATE_POSIX_USE_M$_EXTENSION

# include <io.h>          /* PC extension, not (Posix) portable */

#else

# include <unistd.h>        /* Emx has portable (Posix) headers */

#endif

 

#include <utime.h>

#include <sys/types.h>

#include <sys/stat.h>

#include <errno.h> /* ANSI/ISO C forbids explicit declaration ! *

 It is a thread local modifiable (int) lvalue (usually a macro) */

 

#ifndef lstat               /* This is a file system dependency */

#define lstat(x,y) stat(x,y)

#endif

 

static FILE *out_p;  /* This is the pointer to the new or trun- *

  cated old output file. Input file attributes are also copied. */

 

static struct stat file_status_buf;  /* File attributes buffers */

static struct utimbuf file_utime_buf;

 

static char buf[BUFSIZ];    /*  r/w no more than BUFSIZ at once */

static size_t count; /* read/write at least count bytes at once */

 

static int status = 0;                   /* default return code */

 

 

        int 

cpfile(const char *in_name, const char *out_name)

{

 FILE *const in_p = fopen(in_name, "rb");   /* point to in file */

 

  if (status)   /* reset and try your luck again, if a previous */

      status = 0;     /* call to cpfile() has returned an error */

 

  if (in_p == NULL)                    /* error opening in file */

      return (status = 1);

 

  if ((access(out_name, W_OK)) == 0)    /* overwriting out file */

      fprintf(stderr, "Warning: Overwriting '%s'...n", out_name);

  else if ((access(out_name, F_OK)) == 0) {

      if (fclose (in_p) == EOF) 

          perror("cpfile.c: Error closing input file");

      fprintf(stderr, "Error: Cannot write to '%s'n", out_name);

      return (status = 2);       /* error out file not writable */

  }         /* Note: Posix' link()/symlink() don't overwrite! To 

             * achieve an equivalent behaviour, you can test for 

             * W_OK and then return()

             */

  if ((out_p = fopen(out_name, "wb")) == NULL)

      status = 2;                     /* error opening out file */

  else {

      if (lstat(in_name, &file_status_buf) == -1) {

          status = 3; perror("cpfile.c: Input file stat() error");

      }

      do {                            /*  endless file i/o loop */

          count = fread( (void*) buf, 1, sizeof(buf), in_p);

 

          if ( (count < sizeof(buf)) && ferror(in_p) ) { 

              status = 4; perror("cpfile.c: Internal read error");

              break;

           }

           if (fwrite( (void*) buf, 1, count, out_p) < count) { 

              status = 5;

              perror("cpfile.c: Internal write error");

              break;

           }

      } while ( ! feof(in_p) );            /* end file i/o loop */

 

      if (fclose (out_p) == EOF)

          return (status = 6);  /* Now in danger of data loss ! */

 

      file_utime_buf.actime  = file_status_buf.st_atime;

      file_utime_buf.modtime = file_status_buf.st_mtime;

 

      if (utime(out_name, &file_utime_buf) == -1) {

          status = 7; perror("cpfile.c: utime() error");

      }

      if (chmod(out_name, file_status_buf.st_mode) == -1) {

          status = 8; perror("cpfile.c: chmod() error");

      }

  }

  if (fclose (in_p) == EOF)

       status = 9;

 

  return (status);

}

 

#ifdef DEBUG

int main(const int argc, const char *const argv[])

{

        if (argc != 3) {

                puts("Usage: cpfile INFILE OUTFILE");

                exit(-1);

        }

  errno = 0;

        if (cpfile(argv[1], argv[2])) {

                fprintf(stderr, "%s %s %s: Errorlevel: %dn", 

                   argv[0], argv[1], argv[2], status);

                if (errno) perror("cpfile.c");

        }

  return (status);

}

#endif

 

Macros can either be used to hide obscure hacks, or making the code more easily to understand. This depends on the situation. #ifdef clauses, namely nested ones are ill, ill, ill. Consider the following fragment in Xos.h:

#ifdef USG

#ifndef __TYPES__

#ifdef CRAY

#define word word_t

#endif /* CRAY */

#include <sys/types.h>     /* forgot to protect it... */

#define __TYPES__

#endif /* __TYPES__ */

#else /* USG */

#if defined(_POSIX_SOURCE) && (defined(MOTOROLA) || defined(AMOEBA))

#undef _POSIX_SOURCE

#include <sys/types.h>

#define _POSIX_SOURCE

#else

#include <sys/types.h>

#endif

#endif /* USG */

The X11 headers have many more of these 'pearls', and the GNU glibc include headers are even worse. The only way to find out what really goes into the code is to run gcc -E on the file that includes this file.

> >  b) a quick hack to satisfy compiler/linker

>

> using a function stub foo for a missing function:

>

> foo() 

  {

>       fprintf (stderr, "sorry, foo() not implemented! I'm just too lazy!

> You may consider this as a  bug; but I won't care at all ;-)" );

>       return -1;

> };

>

> >  and

> >  c) a proper implementation

>

> Providing all this ifdef'd in the sources (no stubs) and doing it

> according to the books in a way that will always work and is easy to

> debug.

You can't prevent laziness in user code; I myself have routines marked with

#define TBD(x) fprintf(stderr, #x ": not yet implementedn")

The point is rather: missing functions, namely prominent things like 'link', 'readlink', 'chown', 'setuid' and others should be moved into a separate OS-dependent file, rather than handled in the following way:

(config.h)

#define _HAS_CHOWN 0

(foo.c)

#if _HAS_CHOWN

        chown(blurb...);

#else

        /*do nothing*/

#endif

scattered all over the place. It is just the problem that people don't learn at all (even not in CS classes at universities) how to practically write portable code. Code is still written in a plumbing manner: Add some pipes in some arbitrary house so that the water just flows through it, and if you need another kitchen sink somewhere, just add a flange in a place whereever your editor goto-line command accidentally points you to. Not even talking about Booch's OOA/OOD/OOP or other techniques of software analysis and design, but the drawback of bazaar programming is distributed hacking and patching.

>

> Well, here I'm too inexperienced.

     II. Provide a LIBERTY.DLL

         -------

We'd consider making some libiberty.a (adding at least stubs for all

routines that are missing in EMX or are implemented differently in Unix

dialects) a definite standard. The XFree86-4.0 server now for instance

also needs ndbm and regex, for some reasons. I had to take this from

elsewhere." (Holger Veit)

"Not #ifdef but implement! Or at least provide stubs, to go without the

need to #ifdef. #ifdef's when not maintained tend to rust much faster than

code without them. And with such core-level thing as XFree86 the time to

live is essential." (Sergey Ayukov)

     III. Provide a BSD-like Paths.h Mechanism

          ------------

>       #define _PATH_HEQUIV    "/tcpip/etc/hosts.equiv"

>       #define _PATH_HOSTS     "/tcpip/etc/hosts"

>       #define _PATH_NETWORKS  "/tcpip/etc/networks"

>       #define _PATH_PROTOCOLS "/tcpip/etc/protocols"

>       #define _PATH_SERVICES  "/tcpip/etc/services"

>

> Well, how do we get things mapped to the current values here?

>

> E.g. there are lots of sources assuming an existing '/tmp'.

> Should we have an EMX internal variable/function as

> a replacement?

That's what a "Common Filesystem Standard" was supposed to be good for.

EMX or any other redirection software (e.g. TVFS) is then responsible to

map such paths to really existing locations. The above paths.h snippet

implies that there is some directory, from now on known as "/tcpip/etc"

which contains two files "protocols" and "services". TVFS might redirect

*this directory* to e.g. "/mptn/etc" but should not tear the files apart

resulting in obscure to find bugs (probably not in this case, but situations

are possible where the definition of such well-known paths is important)."

On Tue, 04 May 1999 21:36:19 -0100, Martin Krischik wrote:

how about:

typedef enum pathId

        {

        pathHEQUIV      ,

        pathHOSTS       ,

        pathNETWORKS    ,

        pathPROTOCOLS   ,

        pathSERVICES    ,

        } pathId;

extern void

pathCalc(unsigned char aPath[],

         enum PATH aID);

Would that not be a lot cleaner version. The Function would calculate the correct

filepath from the correct environment variable and filename.

        IV. Beware of Global Variables

            ---------

On Thu, 21 Oct 99 14:44:43, Asbjoern Pettersen wrote:

>When i started GIMP porting a long time ago 0.9.18 i run into this problems.

>The main problem is use of global variables and forking() of other processes (DLL

>loading).

>They use global variables as input !!  That have never been a good design on any

>OS'es.

>So the problem is how to set (and find) those global variables ! 

>This input is hidden and a new programmer will not know the real impact on changes

>he do. It's terrible! The program will get buggy and out of control.

        V. Try Hard to Respect Posix (and ANSI/ISO C/C++)

           ----------------

Remember: Emx is very close to ANSI/ISO and Posix.

* Avoid common GNU extensions.

* Don't cling to vendor-specific so-called 'standards'.

* Follow closely internationally accepted independent portability standards.

This means that emx programmers are sometimes reminded of the fact

that 'Posix does not buy you much...'. This is often a headache, but it

also shows you which code is portable, aka standard conformant, and

which code isn't, therefore has to add some quirks to gain portability:

On 06 Sep 1999 20:22:17 +0200, Lars Gullik Bjnnes wrote:

>Duncan Simpson <dps@io.stargate.co.uk> writes:

>| According to my man page strcasecmp is conforming to BSD 4.3 (and

>| all unicies I have seen so far, inclduign SunOS). AFAIK GNU libc 2.1 is

>| conforming to ANSI and Posix but also includes common extensions, like mmap,

>| lstat, setrlimit, snprintf, select, fileno, *BSD sockets and whole bunch of

>| other stuff that is  endemic in un*x source code and supported everywhere with

>| very few exceptions. 

>| I can use fdopen in Posix programs. There are also some things that only exist 

>| in glibc, and linux libc 5, like strfry(3) and memfrob(3), which are presumably

>| included for hack value.

>|

>| The average punter, and Un*x programmer, is not aware of how little Posix

>| gives him. As soon as you prevent a remote root exploit by using snprintf you 

>| have violated Posix

>

>And you don't know if your prog will compile on a system not running gnu libs.

>

>|-and of course by doing any networking and using

>| select(2) you have violated Posix anyway. All internationalisation is beyond

>| the Posix standard too.

>

>Not quite true, check man setlocale:

>CONFORMING TO

>       ANSI C, POSIX.1

>We do not want to use strcasecmp, and in LYX 1.1.x we are getting rid

>of it. (using more native C++ constructs instead.)

        "As for non-existing API, I would recommend to stick to well-known things

        like Posix, BSD networking, X11 and avoid system-specific calls like

        DosWrite() etc. completely. If you're doing really multiplatform project.

        Really 'standard' calls are the same everywhere and rarely require

        re-implementation. There's some oddity here and there (I heard Ultrix(?)

        does not have strdup()?) but you can add it in transparent way, namely

        linking on Ultrix with your own implementation of strdup()."

        (Dr. Sergey Ayukov, Sternberg Astronomical Institute Moscow, Russia

        http://www.ayukov.com, http://crydee.sai.msu.ru/index-asv.html)

"Ok, can someone explain (together with an example each) what is

a) a dirty trick, b) a quick hack to satisfy compiler/linker c) a proper implementation"

On Tue, 4 May 1999 10:44:58 +0200, Holger Veit wrote:

> >Ok, can someone explain (together with an example each) what is > > > > a) a dirty trick, > > using a preprocessor macro instead of #ifdef in the sources, so that > you can watch in place what is going on (if things go wrong; remember: > Everything that can go wrong will go wrong!)

Macros can either be used to hide obscure hacks, or making the code more easily to understand. This depends on the situation. #ifdef clauses, namely nested ones are ill, ill, ill. Consider the following fragment in Xos.h:

#ifdef USG #ifndef __TYPES__ #ifdef CRAY #define word word_t #endif /* CRAY */ #include <sys/types.h> /* forgot to protect it... */ #define __TYPES__ #endif /* __TYPES__ */ #else /* USG */ #if defined(_POSIX_SOURCE) && (defined(MOTOROLA) || defined(AMOEBA)) #undef _POSIX_SOURCE #include <sys/types.h> #define _POSIX_SOURCE #else #include <sys/types.h> #endif #endif /* USG */

The X11 headers have many more of these 'pearls', and the GNU glibc include headers are even worse.

The only way to find out what really goes into the code is to run gcc -E on the file that includes this file.

> > b) a quick hack to satisfy compiler/linker > > using a function stub foo for a missing function: > > foo() { >  fputs(stderr, "sorry, foo() not implemented! I'm just too lazy! > You may consider this as a bug; but I won't care at all ;-)" ); >  return -1; > }; > > > and

> > c) a proper implementation > > Providing all this ifdef'd in the sources (no stubs) and doing it > according to the books in a way that will always work and is easy to > debug.

You can't prevent laziness in user code; I myself have routines marked with

#define TBD(x) fprintf(stderr, #x ": not yet implementedn")

The point is rather: missing functions, namely prominent things like 'link', 'readlink', 'chown', 'setuid' and others should be moved into a separate OS-dependent file, rather than handled in the following way:

(config.h) #define _HAS_CHOWN 0

(foo.c) #if _HAS_CHOWN chown(blurb...); #else /*do nothing*/ #endif

scattered all over the place.

> But for great projects this is simply not feasible even if investing a > reasonable amount of time.

It is. It is just the problem that people don't learn at all (even not in CS classes at universities) how to practically write portable code. Code is still written in a plumbing manner: add some pipes in some arbitrary house so that the water just flows through it, and if you need another kitchen sink somewhere, just add a flange in a place whereever your editor goto-line command accidentally points you to. Not even talking about Booch's OOA/OOD/OOP or other techniques of software analysis and design, but the drawback of bazaar programming is distributed hacking and patching.

> >I'm not sure at all what people think here, but I make some proposals ... : > > >a) use a file_copy routine for symlink (and don't clean up later perhaps) > > yes this is missing > > >b) let getdtablesize return some 'guessed' integer (w/o checking any > > docs, just get it working in most cases ...) > >c) #define lstat stat (neglecting something one could imagine around TVFS) > > Well, here I'm too inexperienced.

II. Provide a LIBERTY.DLL -------

We'd consider making some libiberty.a (adding at least stubs for all routines that are missing in emx or are implemented differently in Unix dialects) a definite standard. The XFree86-4.0 server now for instance also needs ndbm and regex, for some reasons. I had to take this from elsewhere." (Holger Veit)

"Not #ifdef but implement! Or at least provide stubs, to go without the need to #ifdef. #ifdef's when not maintained tend to rust much faster than code without them. And with such core-level thing as XFree86 the time to live is essential." (Sergey Ayukov)

III. Provide a BSD-like Paths.h Mechanism ------------

> #define _PATH_HEQUIV "/tcpip/etc/hosts.equiv" > #define _PATH_HOSTS "/tcpip/etc/hosts" > #define _PATH_NETWORKS "/tcpip/etc/networks" > #define _PATH_PROTOCOLS "/tcpip/etc/protocols" > #define _PATH_SERVICES "/tcpip/etc/services" > > Well, how do we get things mapped to the current values here? > > E.g. there are lots of sources assuming an existing '/tmp'. > Should we have an emx internal variable/function as > a replacement?

That's what a "Common Filesystem Standard" was supposed to be good for. emx or any other redirection software (e.g. TVFS) is then responsible to map such paths to really existing locations. The above paths.h snippet implies that there is some directory, from now on known as "/tcpip/etc" which contains two files "protocols" and "services". TVFS might redirect *this directory* to e.g. "/mptn/etc" but should not tear the files apart resulting in obscure to find bugs (probably not in this case, but situations are possible where the definition of such well-known paths is important)."

On Tue, 04 May 1999 21:36:19 -0100, Martin Krischik wrote:

how about:

typedef enum pathId { pathHEQUIV , pathHOSTS , pathNETWORKS , pathPROTOCOLS , pathSERVICES , } pathId;

extern void pathCalc(unsigned char aPath[], enum PATH aID);

Would that not be a lot cleaner version. The Function would calculate the correct filepath from the correct environment variable and filename.

IV. Beware of Global Variables ---------

On Thu, 21 Oct 99 14:44:43, Asbjoern Pettersen wrote:

>When i started GIMP porting a long time ago 0.9.18 i run into this problems. >The main problem is use of global variables and forking() of other processes (DLL >loading). >They use global variables as input !! That have never been a good design on any >OS'es. >So the problem is how to set (and find) those global variables ! >This input is hidden and a new programmer will not know the real impact on changes >he do. It's terrible! The program will get buggy and out of control.

V. Try Hard to Respect Posix (and ANSI/ISO C/C++) ----------------

Remember: Emx is very close to ANSI/ISO and Posix.

* Avoid common GNU extensions.

* Don't cling to vendor-specific so-called 'standards'.

* Follow closely internationally accepted independent portability standards.

This means that emx programmers are sometimes reminded of the fact that 'Posix does not buy you much...'. This is often a headache, but it also shows you which code is portable, aka standard conformant, and which code isn't, therefore has to add some quirks to gain portability:

On 06 Sep 1999 20:22:17 +0200, Lars Gullik Bjnnes wrote:

>Duncan Simpson <dps@io.stargate.co.uk> writes:

>| According to my man page strcasecmp is conforming to BSD 4.3 (and >| all unicies I have seen so far, inclduign SunOS). AFAIK GNU libc 2.1 is >| conforming to ANSI and Posix but also includes common extensions, like mmap, >| lstat, setrlimit, snprintf, select, fileno, *BSD sockets and whole bunch of >| other stuff that is endemic in un*x source code and supported everywhere with >| very few exceptions. >| I can use fdopen in Posix programs. There are also some things that only exist >| in glibc, and linux libc 5, like strfry(3) and memfrob(3), which are presumably >| included for hack value. >| >| The average punter, and Un*x programmer, is not aware of how little Posix >| gives him. As soon as you prevent a remote root exploit by using snprintf you >| have violated Posix > >And you don't know if your prog will compile on a system not running gnu libs. > >|-and of course by doing any networking and using >| select(2) you have violated Posix anyway. All internationalisation is beyond >| the Posix standard too. > >Not quite true, check man setlocale: >CONFORMING TO > ANSI C, POSIX.1

>We do not want to use strcasecmp, and in LYX 1.1.x we are getting rid >of it. (using more native C++ constructs instead.)

"As for non-existing API, I would recommend to stick to well-known things like Posix, BSD networking, X11 and avoid system-specific calls like DosWrite() etc. completely. If you're doing really multiplatform project. Really 'standard' calls are the same everywhere and rarely require re-implementation. There's some oddity here and there (I heard Ultrix(?) does not have strdup()?) but you can add it in transparent way, namely linking on Ultrix with your own implementation of strdup()."

(Dr. Sergey Ayukov, Sternberg Astronomical Institute Moscow, Russia http://www.ayukov.com, http://crydee.sai.msu.ru/index-asv.html)

Preface

A. Instead of an Introduction

  I. Where to Start?

  II. Some Frequently Used yet Uncommon Concepts

    1. Another Interesting Name: What's 'X'?

    2. And Posix, etc.?

    3. And to How Get Posix, etc. Defined?

B. A Silly Demo and Some Comments

  I. Source with Comments

  II. Module Definition File

C. Notes on Libraries

  I. Dynamic link libraries (DLL)

  III. Shared libraries and dynamic libraries (TOG UNIX V.1/2)

  IV. Exporting symbols from a DLL (OS/2 OMF)

   V. 'dynamic library' and 'shared object' (*.so) versus DLL

     1. dynamic link library

     2. 'dynamic library' and 'shared object' (*.so) versus DLL

     3. demo

   VI. How to Link

  VII. Some Special Tricks for libraries

D. Now the Hardware Gets Involved

  I. Devices, X Servers, Clients

     1) '/dev/fastio$', '/dev/console$', '/dev/ptyP0$', etc.

     2) concepts related to hardware (vio) access and kernel level API 's

  II. Inter Process (Session) Communication

  III. Emx and device driver programming (hardware devices)

E. 'Imakefile' or 'configure', that's here the question!

  I. A short overview: Autoconf vs. Imakefile

  II. Preliminary steps

    1) RTF docs

    2) Some additions to /XFree86/lib/X11/config/os2.cf and os2.rules

  III. Common initial problems

    1) 'xmkmf -a' crashes because of 'sh' and 'Rexx' (path name) conflicts

    2) 'make' crashes because of 'sh' and 'Rexx' conflicts

  IV. Imakefile examples

    1) creating a static library (archive)

    2) creating an executable

    3) Some maybe useful defines

E. What Shall I Better Change in the Sources?

  I. Globbing

  II. DOS "text" mode problems and binary reading

  III. Process management and fork()

     1) What is the difference between fork() and spawnvp()?

       a) fork() and emx

       b) making it work

       c) replacing fork() and execvp(), etc.

     2) Example

  IV. Unix file system issues

  V. Sockets

  VI. Miscellanea

F. Some Notes on Portability and Coding Style

  I. Write Proper Implementations

  II. Provide a LIBERTY.DLL

  III. Provide a BSD-like Paths.h Mechanism

  IV. Beware of Global Variables

  V. Try Hard to Respect Posix (and ANSI/ISO C/C++)

     3) How to Link

        ----

I recommend the following simple (unorthodox, but nearly 'foolproof') procedure:

Use the static archive provided on the net or use the Makefile (normally created

by Imake, but today more and more by the OS/2 port of autoconf; details later)

to make the static archive foo_s.a. Then:

1. 'emxomf -s -l foo_s.a' (convert to native omf format)

2. 'emximp -o foo.a foo.def' (create import library for later ld linker fixup)

3. 'emxomf -s -l foo.a'  (convert to native omf format for link386)

4. 'gcc foo.def -o foo.dll -Zomf -Zdll -Zmtd -Zsysv-signals -Zbin-files

 -Zlinker /EXEPACK:2 -Zlinker /NOI -Zlinker /NOL -Zlinker /NOD -s -fstack-check

 -Zlinker /PACKDATA -lfoo_s -lmore_foo_libs_here -lXmu -lXt -lSM -lICE -lX11

 -lXext -lbsd -lsocket'

This produces a packed dll. While some linker flags aren't really necessary, they

won't hurt, unless you use OS/2 2.x (use '/EXEPACK' if in doubt). Isn't the

commandline nice?

Note #1 (linking order):

Linking order is crucial! The traditional C-style linker must find the

fixup for still unresolved symbols in the same library or in another one which

is _following_ on the command line. You might even need to link in rare cases

like:

        -lfoo1 -lfoo2 -lfoo1

where foo1 needs symbols exported by foo2 and foo2 needs symbols exported by foo1.

Note #2 (library debugging):

For a debuggable library compiled with '-g' you'd use the line, except '-Zomf -s' and

all the '-Zlinker blurb', which are specific to link386.

Note #3 (XFree86/2 and multi-threading):

Note #4 (recommended flags for speed):

(emx0.9c)

CXXFLAGS = -m486 -O3 -Zmtd -Zsysv-signals -D__ST_MT_ERRNO__ -Wall -malign-loops=2

-malign-jumps=2 -malign-functions=2 -ffast-math

LDFLAGS = -s -Zmtd -Zsysv-signals -Zbin-files

[your mileage may vary; add your testing results here!]

Trick#3: (On Sun, 17 Oct 1999 14:34:21 +0200 (MET DST), Stefan Neis wrote):

Hint: Use instead ANSI C ('stdio.h') (or Posix ('unistd.h')) interfaces!

"3.11

       Can I use another shell like 'bash' in place of CMD.EXE in an xterm?

       Most shells are subtly incompatible with the PTY mechanism used in

       xterm. A symptom may be that they won't work either if you redirect

       console I/O to a COM terminal which should be possible. bash is

       such an incompatible shell. See also Q 3.12 and Q 3.13.

       Meanwhile, there are special ports of tcsh and ksh which work in

       an xterm.

3.12

       I use 4OS/2 (or some other shell) as my shell. Will it work in xterms?

       4OS2 has a number of misfeatures, politely spoken, in an xterm. We

       attempt to fix this in the future. For now, don't use it in an an xterm.

       If 4OS2 or another shell is in your OS2_SHELL or SHELL variable,

       please add an environment variable named X11SHELL to point to a

       valid CMD.EXE path, e.g.

               SET X11SHELL=D:OS2CMD.EXE

       in your CONFIG.SYS. This will override the OS2_SHELL or SHELL

       setting which is used otherwise.

 3.13

       Why does OS2MORE.COM give a SYS0447 in an xterm?

       This is a program which silently assumes it has access to the

       keyboard and the screen: it uses functions from KBDCALLS,

       MOUCALLS, or VIOCALLS. There are more programs of this kind,

       for instance: most *.COM in os2, ATTRIB.EXE, BACKUP.EXE,

       CACHE.EXE, EAUTIL.EXE, FIND.EXE, HELPMSG.EXE, LINK.EXE,

       LINK386.EXE(!), PATCH.EXE, PSTAT.EXE, REPLACE.EXE, RESTORE.EXE,

       SETBOOT.EXE, SORT.EXE, SPOOL.EXE SYSLEVEL.EXE, TEDIT.EXE, TRACE.EXE,

       UNPACK.EXE, XCOPY.EXE, XDFCOPY.EXE, but also

       unfortunately TELNET and FTP, and some others; infact, almost any

       16 bit application is a possible candidate. Note that CMD.EXE is an

       exception of the rule: it is largely still 16 bit, but is clean for

       XFree86/OS2 use.

       You can use the EWS utility EXEMAP.EXE to check if these DLLs are

       linked in.

       Note: the existance of these DLLs does not mean that the program

       does not work at all, actually EMX.DLL could call KBD API functions

       but usually does not in the X11 environment. However, certain

       unexpected effects may occur in an xterm. I'll try to find a

       workaround for this in the future, but don't count on this.

 3.14

       FTP seems to work in an xterm, but it does not hide the password?

       This is a side effect of what was described in Q 3.13. So actually FTP

       does NOT work. [...]

     II. Inter Process (Session) Communication

         ------------

" 2.46

  I found that killing the X server or the window manager won't work well to end

X11. How can I shutdown X11?

Yes, we know it is a problem, mainly related to signal handling that is not fully

compatible to Unix signaling as well as some differences in session semantics.

XFree86 is not really intended to be started up and shutdown all the time [...]

2.52

  What does the error message "_X11TransOs2OpenClient: Open server pipe

PIPEXxf86.0 failed" mean?

This is for the local named pipe communication the same problem as Q 2.31 is

for the network connection. It basically means: the server crashed for some

reason, and now a client, e.g. xterm cannot connect to the server. The reason

is probably hidden somewhere else in the XF86Config file, or some other setup

problem (e.g. network configuration). [...]

2.31

  What does "SocketINETConnect() can't connect: errno = 65" mean?

What does "SocketINETConnect() can't connect: errno = 61" mean?

These messages *can* point out a network installation problem, in many cases

they are secondary errors, though, and are really caused by a different problem.

Nevertheless, you should ensure that your network is setup correctly. [...]

3.8

  How can I enable authorization for host foo.edu without using xhost?

Create the file X0.hosts in XFree86libX11. To enable connections from a

certain host, add in a line containing

       inet:hostname to the file. For example, your file may look

       like this:

       inet:foo.edu

       inet:friendly.host.edu

       ..

       Note that the X0 refers to the zero in the display name

       os2systemname:0, i.e. in the case of Q 3.6 this would

       become a X1.hosts then. [...]

      III. Emx and device driver programming (hardware devices)

           ------------------

Q: Can EMX be used to generate 16 bit code for programming OS/2 device drivers?

No!.

:

>(Holger Veit): as gas lacks the ability to produce arbitrary segment

>types beyond 32 bit .text/.data/.bss segments. It can generate 16 bit

>instructions in a 32 bit segment (prefixed with 0x66/0x67), though.

On Wed, 13 Oct 1999 17:58:35 +0100, Csaba Raduly wrote:

>But you can write device drivers in 32-bit using some (already existing)

>      thunking mechanism

>

>Check out /pub/os2/system/drivers/filesys/32drv170.zip on hobbes.

IBM distributes the device driver devel package (including compiler)

for free! You may want to look at

http://rover.wiesbaden.netsurf.de/~meile/warpstock/wsp02/en/index.html

for the details...

    II. Preliminary steps

        ------

The following is meant to overcome some simple, nevertheless annoying

problems and to make things work faster: Quick and simple fixes, some dirty

stolen tricks.

  1) RTF docs

     ---

- Read the README file! Or should it better be called DON'TREADME (psychology?)

- Read the INSTALL file!

- Read any additional info files from the program authors. Many

problems ocurring under special UN*X flavours have a similar counterpart on

OS/2 (others haven't, but this shall be a simple guide :-)

Read the /XFree86/lib/X11/config/README file, for a short introduction to

adapting Imake templates. The following may be useful, too.

  2) Some additions to /XFree86/lib/X11/config/os2.cf and os2.rules

     ---------------------

I tried to make life easier, so I've added some defines to the vendor specific

imake template file: /XFree86/lib/X11/config/os2.cf.

Reasons for that:

- Never edit generated Makefiles by hand: Far too much work for larger

  projects and errors are to be expected. (But it is sometimes unavoidable).

- If things work and don't break other projects, they should be mailed

to the respective lists in order to finally become standard:

But certainly, as SMiyata points out: As "far as possible, the changes must be

confined to Imakefiles and definition files bundled in the application.

The settings in os2.cf and os2.rules are shared by other applications, and hence

may cause conflicts in them.  Also, the changes to the template files are local to

your setting, thus, although it may help to build binary on your system, it is not

suitable for porting, unless they are reflected in the distribution of XFree86

itself.

There are indeed the cases that modifications must be made on template files coming

with XFree86 rather than Imakefile and definition files coming with the applications:

[those often have uncommon names like 'local.cf' or 'machinefile', etc.]

Akira Hatakeyama <akira@sra.co.jp> reported elsewhere that, in

order to build Xvnc on OS/2, the lines

  #define BuildXKB                YES

  #define BuildLBX                YES

        in os2.cf must be modified to

  #ifndef BuildXKB

  #define BuildXKB                YES

  #endif

  #ifndef BuildLBX

  #define BuildLBX                YES

  #endif

        since, else, the settings in vnc.def is overwritten by those of os2.cf"

My proposed changes of /XFree86/lib/X11/config/os2.cf:

  #define ManSuffix     1X11

  #define LibManSuffix  3X11

  #define FileManSuffix 4X11

        instead of

  #define ManSuffix     1 [... etc.]

(This helps a lot, if you already installed or will install several man pages from

bsd, Linux, etc. for the same function or tool.

I'm sure this will happen!! Also this is for good reasons now standard

on most recent UN*X versions.)

        Add the following (after: #define

HasStrcasecmp           NO):

#define StrcasecmpDefines -Dstrcasecmp=stricmp

-Dstrncasecmp=strnicmp [and probably: -DNEED_STRCASECMP]

(Just a hack, until emx has a working 'strcasecmp()' function interface).

        Change

#define CplusplusCmd g++

        to

#define CplusplusCmd gcc

#define CplusplusOptions -D__ST_MT_ERRNO__ -Zmtd -Zsysv-signals -O2

(I can't find any reason to call emx gcc by a different name for C++

or to use different options here.)

b) EMX docs recommend using object module format (-Zomf; link386.exe)

for better stability, if neither gdb.exe nor fork() shall be used. The

following additions to /XFree86/lib/X11/config/os2.rules are meant

to facilitate the switch in this case; nevertheless you'll have to

check dependency lines and rules, if a.out format is still used (and some

rules for libraries will become obsolete, but not harmful):

XCOMM $XConsortium: os2.rules /main/1 1996/10/31 14:47:27 kaleb $

XCOMM platform:  $XFree86: xc/config/cf/os2.rules,v 3.15 1997/01/05

11:49:39 dawes Exp $

.SUFFIXES: .a .lib .o .obj .c .C .cc .cpp

.c.$O:

        $(CC) $(CFLAGS) -c $*.c

#if HasCplusplus

.$C.$O:

        $(CXX) $(CFLAGS) -c $*.$C

#endif

XCOMM Uncomment if using OMF (emxomf called automatically by gcc)!

XCOMM CC  = gcc -Zomf

XCOMM O   = obj

XCOMM LD  = emxomfld

C   = C XCOMM or .cpp

O   = o

A   = .a

       III. Common initial problems

            --------

Initially I thought imake was a labyrinthic beast, deserving the title

of the Minotaurus among the make tools. But with small and simple fixes

to work around some annoying bugs I was able to tame it a bit, so that

'xmkmf' at least does something useful:

1) 'xmkmf -a' crashes because of 'sh' and 'Rexx' conflicts

   -------------------

Usually Imakefiles silently assume they are running on UNIX in a Bourne

'sh' environment, since authors intentionally do not support non UNIX

platforms.

Just try running the following changed Rexx scripts first, to create a

working environment:

                make.cmd

/* REXX */

'@echo off'

PARSE ARG a1 a2 a3 a4 a5 a6 a7 a8 a9 a10

/* x11make.exe 'MAKE=x11make.exe SHELL=d:/os2/cmd.exe' a1 a2 a3 a4 a5

a6 a7 a8 a9 a10 */

'SET SHELL=D:OS2CMD.EXE'

'SET X11SHELL=D:OS2CMD.EXE'

'set emxshell=D:OS2CMD.EXE'

'set CONFIG_SHELL=D:OS2CMD.EXE'

'set MAKE_SHELL=D:OS2CMD.EXE'

x11make.exe 'MAKE=x11make.exe SHELL= MAKE_SHELL=d:os2cmd.exe

CONFIG_SHELL=d:os2cmd.exe' a1 a2 a3 a4 a5 a6 a7 a8 a9 a10

        or

                gmake.cmd

/* REXX */

'@echo off'

PARSE ARG a1 a2 a3 a4 a5 a6 a7 a8 a9 a10

'set SHELL=d:/bin/ash.exe'

'SET X11SHELL=D:OS2CMD.EXE'

'set emxshell=d:/bin/ash.exe'

'set MAKE_SHELL=d:/bin/ash.exe'

/* x11make.exe 'MAKE=x11make.exe SHELL=' a1 a2 a3 a4 a5 a6 a7 a8 a9 a10

*/

x11make.exe 'MAKE=x11make.exe SHELL=d:/bin/ash.exe

MAKE_SHELL=d:/bin/ash.exe CONFIG_SHELL=d:/bin/ash.exe' a1 a2 a3 a4 a5

a6 a7 a8 a9 a10

==================================

Normally, running make.cmd should be sufficient, since the current OS/2

implementation by H. Veit and others makes extensive use of Rexx, needs

cmd.exe as command processor and is incompatible with Bourne shells.

Here a number of problems arise with many common Imakefiles.

Now lets assume xmkmf worked. Theoretically the steps are:

        xmkmf

        make Makefiles

        make depend

        make all (or simply: make)

        .programname.exe for testing

        make install

        make install.man

2) 'make' crashes because of 'sh' and 'Rexx' conflicts

   -----------------

Imake is misconfigured on most (commercial) UN*X platforms. Authors

normally compensate this by including many lines of (normally highly

complicated) Bourne shell code and calling several Bourne shell scripts in the

Imakefile. In this case 'make' won't work. Nor does a sh.exe as make shell work:

The current implementation depends on cmd.exe and Rexx. Finding and

rewriting everything in those immense Imakefiles, helper scripts and Makefiles

isn't really an alternative. Just try alternatively make and gmake (having installed

e.g. /bin/ash.exe). Now make should simply resume where it has crashed. Just

delete 'products' of those crashes (empty script output, truncated object

files, core dumps, etc.) before resuming the process. This makes using Imake real

fun!

Tips by John Williams <jwilliams@commerce.otago.ac.nz> (concerning 'configure' of

GNU 'autoconf', fixed by the OS/2-port, but still valid for certain parallel Bourne

sh-scripts very common in Imakefiles and/or config.local, etc.):

"3. Configure couldn't search the OS/2 path for programs because the shell construct:

        for foo in bar do blah

    apparently needs 'bar' to be a space or colon delimited list, whereas the OS/2

    $PATH is semi-colon delimited.

4. Configure could not check to see if it's test programs were being compiled because

   it tests for success using test -s which means test that the file is not empty.

   The emx port of GCC creates foo.exe executables and touches foo, leaving an empty

   file. Why does autoconf test for success with a non-empty file (-s) and not a

   `file exists' (-f) test?

5. Overcome the above problems by setting an environment variable UNIX_PATH to be

   a colon delimited path. Next, search and replace using sed thus:

        sed -f os2patch.sed configure > configure.os2

   where os2patch.sed looks like:

             s/PATH/UNIX_PATH/

             s/test -s/test -f/

             s/host=NONE/host=OS2/

             s/"${IFS}:"$/"${IFS};"/

  and then execute sh Configure.os2. Lines three and four are due to Ilya Zakharevich,

  who also pointed out the following issues:

       1.if /tmp doesn't exist, make it

       2.ensure CONFIG_SHELL=sh set in shell environment you are running Configure

         from.

6. You can find what functions are contained in what object files and libraries by

   using 'nm' (comes with emx). This also tells you which functions are referenced

   (called) by functions in that file."

     IV. Imakefile examples

         ------

$XFree86: config/README states:

  "The easiest way to write an Imakefile is to find another one that

does

  something similar and copy/modify it!"

Here some strongly shortened examples that use Bourne shell code:

sh -c 'for i in '"$(TEXT) $(SRCS) $(MISC) $(HDRS)"' ; do... etc., etc.

1) creating a static library (archive): NormalLibraryTarget(rw, $(OBJS))

----snip-----

# Imakefile for xpaint read/write directory

#include "../version"

#include "../Local.config"

INCLUDES = -I.. $(XPM_INCLUDE) $(TIFF_INCLUDE) $(JPEG_INCLUDE)

$(PNG_INCLUDE)

# Nothing to change below here

TIFF_SRC = writeTIFF.c readTIFF.c

#ifdef HaveTIFF

TIFF_OBJ = writeTIFF.$O readTIFF.$O

TIFF_DEF = -DHAVE_TIFF

#endif

[...]

XPM_SRC = readWriteXPM.c

XPM_OBJ = readWriteXPM.$O

SGI_SRC = readWriteSGI.c

#ifdef SGIArchitecture

SGI_OBJ = readWriteSGI.$O

SGI_DEF = -DHAVE_SGI

#endif

DEFINES = $(XPM_DEF) $(TIFF_DEF) $(SGI_DEF) $(JPEG_DEF) $(PNG_DEF)

SRCS = rwTable.c

        readWriteXBM.c readWritePNM.c readWriteXWD.c writePS.c

        readGIF.c writeGIF.c $(XPM_SRC) $(TIFF_SRC) $(SGI_SRC)

        $(JPEG_SRC) $(PNG_SRC) libpnmrw.c

OBJS = rwTable.$O

        readWriteXBM.$O readWritePNM.$O readWriteXWD.$O writePS.$O

        readGIF.$O writeGIF.$O $(XPM_OBJ) $(TIFF_OBJ) $(SGI_OBJ)

        $(JPEG_OBJ) $(PNG_OBJ) libpnmrw.$O

HDRS = libpnmrw.h rwTable.h

MISC = Imakefile

NormalLibraryTarget(rw, $(OBJS))

makelist:

        sh -c 'for i in '"$(TEXT) $(SRCS) $(MISC) $(HDRS)"' ; do

                echo $(CURRENT_DIR)/$$i >> ../filelist ; done '

rwTable.$O: ../Local.config

----snap------

        D. What Shall I Better Change in the Sources?

           --------------

Last but not least some well meant recommendations from Chapter 5 of

the "emx APPLICATION DEVELOPER'S GUIDE", with some additional comments.

You might use them as a short checklist.

I. Globbing

   ---

" If you want Unix-like wildcard expansion built into the program, use

      int main (int argc, char *argv[])

      {

        _wildcard (&argc, &argv);

        /* ... the program ... */

      }

  This should be done at the very beginning of main(), before ARGC and

  ARGV are used.  See _wildcard() and _response()."

II. DOS "text" mode, binary reading, etc.

    -------------

                                        "Famous problems are related to the

                                        binary/text mode mess on OS/2,

                                        especially when using stdin/stdout."

                                                                (Alexander Mai)

" Programs using stdin, stdout or stderr for binary data should call

  _fsetmode() to switch the stream to binary mode."

" Change all open(), fopen(), fdopen() and freopen() calls to use

  O_BINARY or "b", respectively, for binary files.  If a file contains

  both binary and textual data, read the file in binary mode and do

  the conversion yourself."

Note: XFree86/OS2 uses '-Zbin-files'. So "do the conversion yourself" "and ignore

carriage returns" and <CTRL Z>.

" Though fseek() and ftell() now work on text files, the offsets are

  different from what Unix programs expect.  You may have to open the

  files in binary mode and ignore carriage returns (this has been done

  in GDB)."

" Programs reading a.out files should be changed to call _seek_hdr()

  or _fseek_hdr() before reading the header to support .exe files.

  More changes are usually required."

III. Process management and fork()

     ----------

                        "Replace fork() and exec*() with spawn*(). 

                         Under OS/2, fork() is inefficient."

                                                                (emx docs)

Note: This is usually not difficult; it is necessary to use '-Zomf'. Cf. the

example below.

Replacing vfork()-type lightweight processes with threads is complicated.

  1) What is the difference between fork() and spawnvp()?

     ------------------

 a) fork() and emx

    -----

The Posix syscall fork() is for traditional Unix systems the only possiblity to

create a new process i.e. a nearly identical clone of an existing one:

"This is achieved by copying the pointers in the process information block (PIB)

of the parent to the newly allocated block in the kernel. On the other hand, on DOS

(Windows included) and on OS/2 (Windows NT included) there is no "fork" system call

but there is a "spawn" instead. This system call just allocates a new process

information block. Only file handles and the environment variable space are

inherited and selectors to the address space of the parent process are not copied.

So the child cannot access the contents of the memory of the parent. In fact, for

OS/2 and Windows NT, it is impossible to implement "fork" in the same way as on the

UNIX, since even root has only read access to the PIB table in the kernel.

[In fact, there is no such security leak as 'root' in OS/2 (AH)]

EMX library function fork() is actually spawn system call with a little bit of

IPC to copy memory contents. It is inefficient because all the memory contents

rather than just pointers are copied.

Well, the difference I was talking was that a *spawned* child has a clean address

space, while a *forked* one might have a corrupted one." (SMiyata).

 b) making it work

    -----

 "fork() works only in programs linked by ld. It does not work in programs linked

  by LINK386." [-Zomf]

 "fork() doesn't work correctly in multithread programs."

 "If the parent process uses the general terminal interface for the keyboard, the

  child process cannot read from the keyboard using _read_kbd(), the general

  terminal interface, or the Kbd OS/2 API functions.

  If the process has a non-contiguous heap (that is, multiple heap objects), fork()

  will fail. Increase the initial heap size to work around this problem. If any DLL

  used by the program shares the heap with the program and uses the heap in

  _DLL_InitTerm() by calling sbrk(), malloc(), etc., fork() will fail. 

  In both cases, fork() will return -1 and set errno to ENOMEM.

  See also: alarm(), exec*(), _rmutex_create(), sbrk(), sigaction(), signal(),

  spawn*(), _uflags(), wait(), waitpid()"

See the emx reference for more details about caveats. Use huge stack and heap size.

 c) replacing fork() and execvp(), etc.

    ------------

Excerpt from exec*() documentation:

" When the new process ends, the parent process will be notified by SIGCHLD

  and wait() will return the original process ID (PID). [...]

  Restrictions:

  Native DOS programs cannot be run.  The new process gets a new process ID. 

  Unless the new process is an emx program (using emx.dll or emx.exe),

  the actions for all signals are reset to SIG_DFL in the new process"

 "Replace exec*() with spawn*() and exit() if the parent process waits

  for the termination of the new process (by calling wait() or by

  waiting for SIGCLD). This is required to keep the process ID of the

  child process. In a forked process, however, you don't have to do this

  because emx.dll does it for you."

What does this mean? Well, in a Posix OS traditionally the parent process clones

itself with fork(), which generates a new process environment, whereas exec()

overwrites its own present environment with a new one, keeping the PID.

Usually this is combined:

exec() is used in the forked child process, overwriting the parent's process

status with a new one. Meanwhile the parent usually calls wait() or a home-brew

function and waits for child termination, communicated by SIGCHLD. ps (pstat.exe)

gives you the details.

        A common problem here is, that the child after termination does

not find this wait() etc. point of the parent. Thus a 'zombie' remains. This

may lead to a process table overflow, so that the system has to be shutdown and

rebooted. All this hassle can be simplified by using spawn*().

But for tracing and debugging:

 "Do not use the PTRACE_TRACEME request of ptrace(): use P_DEBUG

  instead when starting the process with spawn*()."

IV. Unix file system issues

    --------

The separator of path components is a slash '/' as opposed to OS/2's ''.

Note that many APIs understand the '/' on OS/2, but cmd.exe and most apps don't.

Multiple paths (e.g. in environment variables like PATH) are separated by a colon ':'

The root dir on every un*x system is '/'. Often code checks whether a given path is an

absolute one by comparing the beginning with '/'. Use grep for all standard paths and

all commands which are mentioned in this introduction. (_fullpath() and _abspath()

can also be useful.)

Using

     #define getcwd _getcwd2

     #define chdir _chdir2

may help to support drive letters. Read the EMX docs:

  "Watch out for Unix file system hacks: Unix allows deleting and

  renaming an open file (the file will be deleted after being closed).

  [unlink(), remove(), rename()]

- Watch out for Unix file names (Unix is case sensitive, long file

  names and multiple dots are allowed).  On OS/2's HPFS multiple dots

  are also allowed; however, trailing dots are not significant (except

  for the special file names `.' and `..').

- The null device is called /dev/null under Unix.  The __open() system

  call translates the filenames "/dev/null" and "/dev/tty" (lower

  case, with slashes) to "nul" and "con", respectively.  However,

      system ("whatever >/dev/null");

  won't work as the standard OS/2 and DOS command interpreters don't

  recognize /dev/null.

- If you want to use for separating directories, changes may be

  necessary.  These changes are optional because / also works.

- Implement support for drive names.  This can be done by using

      #define getcwd _getcwd2

      #define chdir _chdir2

  In addition, some changes will be necessary.  For instance, you have

  to change code which checks whether a filename is an absolute path

  name.  _fullpath() and _abspath() can also be useful."

        "Unix does not have drive letters, so a file (resources,

        fonts, config, etc.) path starts with '/' normally. XFree86 is

        located in a tree below %X11ROOT%XFree86...

        To prepend the X11ROOT part to a file path, a special function

        named __XOS2RedirRoot exists which you are supposed to use in

        these cases rather than an insane own solution. The prototype of

        this function is:

                const char *__XOS2RedirRoot(const char *pathname);'

        [...] XFree86/OS2 stores all configuration files in a common

        directory. This is X11ROOT:XFree86libX11, where X11ROOT is

        an environment variable which contains the drive letter the

        directory tree is located. The common places like /usr/X11,

        /usr/X11R6, /usr/XFree86, /etc which are used for XFree86 in

        various Unix flavors, are not supported [...].

        It is also there where you should put your XF86Config file,

        the color database (rgb.txt), the host authorization files

        (X0.hosts), and various other files." (FAQ)

"Potential problem might be the DLL name: IIRC one can either use

        'foo' or 'x:barfoo.dll'

to access foo.dll (if in LIBPATH)." (Alexander Mai)

 "Note that ///abc is a valid Unix filename.  It's equivalent to /abc.

- Note that chdir ("..") is a no-op under Unix if the current working

  directory is the root directory.  Under emx, chdir ("..") fails in

  the root directory.

- Use termio or termios or read the keyboard with _read_kbd() if you

  don't want to get input line by line.

- Under Unix, directories in environment variables (PATH, for

  instance) are separated by colons; use semicolons instead.

- The shell isn't called /bin/sh.  Use system(). system() and popen()

  don't expand wildcards (unless COMSPEC points to a shell which

  expands wildcards)."

So, in short:

  "grep for '/dev', '/bin', '/tmp', '/usr', '/proc', ...

   also check out all commands starting up another process:

        exec*, popen, system,...

   Remember that cmd.exe doesn't like slashes!" (Alexander Mai)

Footnotes:

¹Cf. the references below.