OS/2 Shareware BBS: 10 Tools/10-Tools.zip/bullet25.zip/common/bullet2.inf

OS/2 Help File | 1997-06-25 | 277KB | 11,413 lines

ΓòÉΓòÉΓòÉ 1. Copyright ΓòÉΓòÉΓòÉ

25-Jun-1997

          BULLET is Copyright 1992-97, and is owned by the author, Cornel Huth,
          and is protected by United States copyright laws and international
          treaty provisions.  License restrictions apply.


ΓòÉΓòÉΓòÉ 2. License Agreement ΓòÉΓòÉΓòÉ

Before using this software, BULLET you must agree to the following:

   1. A BULLET license grants you the right to use the BULLET library code on a
      royalty-free basis according to the terms of this License Agreement.

   2. You are required to have one license per developer that is programming
      with Bullet.

   3. There is no restriction on the number of users you may support, and no
      restriction on the number of different end-user programs you may
      distribute that use BULLET.  You may allow any number of simultaneous
      users to use your end-user program.

   4. The dynamic link library, BULLET*.DLL, may be distributed with your
      end-user program.  No other BULLET product may be distributed without
      permission.

   5. Evaluation use is limited to 300 records, and for the sole purpose of
      evaluating the software.  The BULLET library code may not be distributed
      in any form without a registered BULLET license.  A BULLET license is
      obtained only with purchase of a BULLET package, purchased from an
      authorized BULLET distributor.

   6. BULLET is owned by the author, Cornel Huth, and is protected by United
      States copyright laws and international treaty provisions.  You are not
      permitted to make copies of this software except for archival purposes.

   7. You may not rent or lease BULLET. You may not transfer this license
      without the written permission of the author.  If this software is an
      update or upgrade, you may not sell or give away previous versions.

   8. You may not reverse engineer, decompile, or disassemble this software if
      the intent or result is to alter the software.

   9. There are no expressed or implied warranties with this software.

  10. All liabilities in the use of this software rest with the user.

  11. U.S. Government Restricted Rights.  This software is provided with
      restricted rights.  Use, duplication, or disclosure by the Government is
      subject to restrictions as set forth in subparagraph (c)(1)(ii) of the
      Rights in Technical Data and Computer Software clause at 52.227-7013. The
      software is owned by Cornel Huth/6402 Ingram Rd/San Antonio Texas
      78238/USA.  This agreement is governed by the laws of the Great State of
      Texas, the United States of America, and all other countries of Earth.

 Any questions concerning this License Agreement should be directed to Product
 Support.

 Failure to comply with any part of this License Agreement may result in
 license revocation.


ΓòÉΓòÉΓòÉ 3. Installation ΓòÉΓòÉΓòÉ

Installation instructions are located in the README text file included with
your package.


ΓòÉΓòÉΓòÉ 4. Product Support ΓòÉΓòÉΓòÉ

Technical support in the use of Bullet is available for licensed users at the
40th Floor BBS or by way of the Internet.

40th Floor BBS:
[phone]  N-8-1

Internet:
[e-mail]
http://www.40th.com

Response time is usually within 24 hours by internet e-mail or if you leave a
message at the support BBS.  Alternatively, you may post a letter to Cornel
Huth/6402 Ingram Rd/San Antonio TX 78238/USA.

The latest in-version (2.x) release of BULLET 2 is available for download from
the BBS or from the internet at the www.40th.com site (alternate is
www.txdirect.net/~cornel/), or $5 by postal mail.  Re-createable bugs are fixed
immediately.  Contact support if you have any questions or requests.

For everyone else, or for general information, send e-mail to:
[e-mail]


ΓòÉΓòÉΓòÉ 4.1. Bug Report Form ΓòÉΓòÉΓòÉ

When requesting support for possible bug(s) use the following as a guideline:

   1. Include a complete problem description.
   2. Include sample source of the problem, if necessary (small is best).
   3. Include necessary data files, include files, etc.
   4. Include step-by-step procedure to follow in order to recreate the
      problem.

 Once done, send it to support by way of the BBS, e-mail, or postal mail to
 addresses listed in Product Support.


ΓòÉΓòÉΓòÉ 5. Ordering Information ΓòÉΓòÉΓòÉ

To order Bullet by check, bank check, money-order, or cash, use form ORDER.FRM.
For credit card order, use form ORDER.CC.

Payment Options

Γûá Check, money-order, cash

Your funds have to be in US Dollars and, if by check, have to be drawn on or
payable through a US bank.  If sending currency, use Registered AirMail.
Personal checks may require 10 working days to clear.  Most major non-American
banks have branch banks in the US.  Contact your bank for details.  Direct
wire-transfer is available: enquire to [e-mail].

To order, send payment and the order form ORDER.FRM to:

   Cornel Huth
   6402 Ingram Rd
   San Antonio, Texas 78238-3915
   USA

Γûá Credit card, Eurochecks in DM, other

For credit card orders or for orders that cannot use the Direct-To-Author order
form, use the special Bullet/BMT Micro order form, or see the file !ORDER.CC in
the distribution package.

Delivery Options

Standard AirMail to all destinations is available at no extra charge.  If a
printed manual is ordered, it will be AirMailed first-class.

Updates

Updates are available for $5 (to cover disk and shipping costs) for any 2.x
version update at your option level.  Updates may be downloaded from the Bullet
support BBS or intenet free of charge

Note:  All shipments outside of the US may go through your country's Customs.
Each package is valued at the total price of the order, less any shipping
costs, if any, unless pre-arranged otherwise.  Your Customs Office may delay
mail delivery if duty is required.

Click here [Γûá] for the check, money-order, cash order form.

Click here [Γûá] for the credit card order from.


ΓòÉΓòÉΓòÉ 5.1. Order Form for Check, Money-order, or Cash ΓòÉΓòÉΓòÉ

Printing this section from here may not fit on a single page.  Use the file
ORDER.FRM instead, or select the 'Services' menu item from here (upper-left),
and select 'Copy to file' to print this to .\TEXT.TMP.

Direct-To-Author Order Form (also see Credit Card Order Form)
Bullet 2.5 for OS/2 Order Form for Check, Money-order, or Cash


                                        Cost Per    Number of
       ORDERING:  BULLET 2.5            License     Licenses    Extended
     -----------------------------------            --------    --------

     Bullet 2 (includes all platforms)    $89.00  x          =
     -----------------------------------           --------     ---------

                                                             =
     Bullet 2 printed manual (OPTIONAL)   $20.00  x -------     ---------

                                                 SUB-TOTAL  = $
 When ordering direct-to-author you                             ---------
 save 15% on handling fees.  Allow at
 least 10 business days for your order       15% DISCOUNT   =  -
 to be processed when using this form                           ---------
 once it has been received.
                                           BULLET 2  TOTAL  = $
                                                                =========


Send TOTAL payment on check (US Dollars/US bank), money-order, or cash to

        CORNEL HUTH                       Always endorse your letter
        6402 INGRAM RD                             AirMail
        SAN ANTONIO TX 78238-3915          if you are not in the USA
        USA

   PLEASE PRINT - For comments add a separate page and mark here [  ]

Your Name>
           -------------------------------------------------- (required)

  Company>
           -------------------------------------------------- (optional)

  Address>
           -------------------------------------------------------------


           -------------------------------------------------------------


           -------------------------------------------------------------

    E-mail>                                               Disk
           ---------------------------------------------- Size> --------

 Telephone                             Fax
   Number> --------------------------- No.> ----------------------------

           This form is not for credit cards  Today's Date>
           ----------------------------------               ---/----/---
           Check must be Payable from US Bank


ΓòÉΓòÉΓòÉ 5.2. Order Form for Credit Card ΓòÉΓòÉΓòÉ

Printing this section from here will not fit on a single page.  Use the file
ORDER.CC instead, or select the 'Services' menu item from here (upper-left),
and select 'Copy to file' to print this to .\TEXT.TMP.  Cut-and-paste as
needed.

   To order Bullet by credit card (or Eurochecks in DM) use this form,
   payable to BMT Micro.  All other orders should use the Direct-To-
   Author order form.  All shipping is done by the author direct to
   you, and is always the current release of Bullet.  All Bullet 2
   versions include a printed manual and include free shipping to all
   destinations (AirMail).

                  Mail Orders To: BMT Micro
                                  PO Box 15016
                                  WILMINGTON NC 28408
                                  USA

                    Voice Orders: 800AM - 700PM EST (-5 GMT)
                                  [phone] (orders only)
                                  [phone]

                      Fax Orders: [phone]  24 hours / 7 Days
                                  [phone]  24 hours / 7 Days

         Online Orders via modem: [phone]  10 lines, all 14.4K
                                  [phone]  Direct 28.8K line

     Ordering and general ordering questions:
                         Via AOL: bmtmicro
                         via MSN: bmtmicro
                     Via Prodigy: HNGP66D
                  via Compuserve: 74031,307
                    via Internet: [e-mail]
                                  [e-mail]
                                  www.bmtmicro.com


     We accept Visa, Mastercard, Discover, American Express, Diners
     Club, Carte Blanche, Cashiers Check, Personal Check.   Personal
     checks are subject to clearance.  Eurochecks in DM are welcome.
     DM, Sterling, and US Currency is welcome but send only by
     registered mail, return reciept requested.   We cannot be liable
     for lost cash sent through the mail.

     Purchase orders are welcome, subject to approval.   The minimum
     amount is $250.00.

     Information for our German customers is explained in the last
     paragraph of this order form.

     -------------------------------cut here-----------------------------
     Name:
          ------------------------------------------------------(required)
     Company:
             ---------------------------------------------------(optional)
     Address:
             -------------------------------------------------------------

             -------------------------------------------------------------
     City:                                State/Province:
          --------------------------------               -----------------
     Postal/ZIP Code:                     Country:
                     ---------------------        ------------------------
     Phone:
           ---------------------------------------------------------------
     Fax:
         -----------------------------------------------------------------
     E-Mail #1
              ------------------------------------------------------------
     E-Mail #2
              ------------------------------------------------------------
                                        Cost Per    Number of
       ORDERING:  BULLET 2.5            License     Licenses    Extended
     -----------------------------------            --------    --------

     Bullet 2 (includes all platforms)    $89.00  x          =
     -----------------------------------           --------     ---------

                                                             =
     Bullet 2 printed manual (OPTIONAL)   $20.00  x -------     ---------

                                                 SUB-TOTAL  = $
                                                                ---------

                 North Carolina Residents add 6% Sales Tax  =  +
                                                                ---------

                                           BULLET 2  TOTAL  = $
                                                                =========
     ---------------------------------------------------------------------
     |                                                                   |
     | For credit card payment only                                      |
     |                                                                   |
     | Circle one: VISA / Master / Discover / American Express / Diners  |
     |                                                                   |
     | Credit card number:                                               |
     |                     --------------------------------------------- |
     | Expiration date:                                                  |
     |                  ------------------------------------------------ |
     |                                                                   |
     | Authorization signature:                                          |
     |                          ---------------------------------------- |
     ---------------------------------------------------------------------
     -------------------------------cut here-----------------------------


                   ORDERING FROM INSIDE GERMANY ONLY
                   =================================

Persons in Germany wishing to order shareware may also transfer funds
into our account with Deutsche Bank.   Once the money is deposited you
may either fax a confirmation to us with proof of deposit or wait until
Deutsche Bank notifies us of the transaction (usually 10-18 business days).
Account information is as follows:

Deutsche Bank / Frankfurt Branch
EmpfДnger:  Thomas Bradford / BMT Micro
Konto-Nummer: 0860221
Bankleitzahl: 500-700-10

When you make the transfer, be sure to put your name and the program you
are registering on the transfer.

Current exchange rates can be obtained by sending an email to
[e-mail].   An automated reply will return todays exchange
rates.

It is very important that you send us a completed order form by
either email or fax if you deposit money into this account for a
registration.  Fill the order form out as usual except in the credit
card number field put "DEUTSCHE BANK".   We will file the order and
use it to match against the deposit information we receive from the
bank.

                               IMPORTANT!
                               ----------
When you email us your order form, we will reply with an
acknowledgement.   If you do not get an acknowledgement within 24 hours
please send your order again in case it was lost.  This extra bit of
caution can save a lot of confusion.

If you are concerned that your order is taking too long to process, feel
free to check with us about the status of your order.   It's important
to all of us that you feel safe doing business with our company and
please feel free to suggest ways we can improve our service to you.


ΓòÉΓòÉΓòÉ 6. Tutorial ΓòÉΓòÉΓòÉ

This tutorial describes how to set up a basic database.  It describes file
layout and how to use Bullet to enter into and read from the database.  It does
not describe aspects that do not relate to Bullet, such as how to load list
boxes, create dialogs, etc.  This tutorial creates a music CD collection
database.

The tutorial follows.


ΓòÉΓòÉΓòÉ 6.1. Tutorial: Output Requirements ΓòÉΓòÉΓòÉ

The goal of this tutorial is to create a database for a music CD collection.
The method used is a suggested method; it is by no means the only way to
develop a database, and only touches the surface of database programming.  That
said...

The most important thing to do is to know what output you want.  From that, you
know what input is needed.  For the 'CD database', I've drawn up a list of
things that I want:

Title, artist, and year.  Must have those.  Track name (the song name), track
number, and play time would be essential, too.  How about who was playing on
the track and instruments they used?  Put that on your list -- it's not on this
one.

All these could be entered into a single file as a single database record, but
that would be a pretty poor design.  A single data file may seem to be the
obvious way, but in the long run, it's not very good at all.  This database
uses two data files.  This concept -- multiple data files -- is especially
useful when the database is more complex.  For example, an accounting system
would employ many physical data files.  It could be done in one, sure, but the
maintenance would be very difficult.

Two tables are used:  The Title-Artist-Year table (Title/CTAY), and the
TrackNumber-TrackName-Time table (Tracks/CTTT).  A table is a file composed of
rows and columns.  A row is a record; a column is a field.  No two rows may be
exactly alike (what would be the point?).  In addition to the fields listed
(the Title, Artist, and Year, for instance), each table includes a Code field.
TrackNumber is included as a field value in the second table since the physical
ordering of rows cannot be used for this purpose (it could, technically, but it
is unwise to depend on row order being preserved since there is nothing
enforcing the physical ordering of records).

The Code field for the Title table (CTAY) is generated by the program and is
composed of the first 4 characters of the Title, plus the first 4 characters of
the Artist, plus the year (all four characters of it).  This code is also used
in the Tracks table (CTTT).  This serves two purposes:  First, it ensures that
each Track record is unique (with pretty good certainty; without it, it is
possible that TrackNumber-TrackName-Time alone may occur more than once), and
second, it can be used as the primary key when being accessed through the
foreign key in the Title table (see Record Layout Diagrams).  By searching CTAY
(the Title Table) for a desired title, or artist, or even year, and using the
Code field from that row, you can match it to the Code in CTTT (Track Table) to
find the tracks that belong to that title row (or artist, or year -- whatever
the search was based on).  To speed up searching, indexes are maintained and
used for lookups.  For a very large CD collection (say, several thousand CD
titles, and tens of thousands of tracks), the database can still be very
quickly queried for whatever information you need (and have programmed).


ΓòÉΓòÉΓòÉ 6.2. Tutorial: Record Layout ΓòÉΓòÉΓòÉ

The Code field is generated by your program.  Other fields are input data.  The
following are the data record layouts of the DBF files (tag fields not shown).
The time field may be created as a text field with the form "mm:ss", or could
even be made into a 'seconds' field.  This database uses the more direct
"mm:ss" since that's how it's read off the CD. Note that this information is
already stored digitally on the CD itself (running time), but that would
require additional coding and is not relevant to this tutorial (but oh, how
sweet it could be).

 CTAY TABLE
ΓöîΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö¼ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö¼ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö¼ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÉ
Γöé   CODE     Γöé      TITLE      Γöé     ARTIST      ΓöéYEAR Γöé
Γöé            Γöé                 Γöé                 Γöé     Γöé
ΓööΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö┤ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö┤ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö┤ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÿ
 MexiConc1993  Mexican Moon      Concrete Blonde  1993
 No NCran1994  No Need to Argue  Cranberries, The 1994
 MachDeep1972  Machine Head      Deep Purple      1972


              CTTT TABLE
             ΓöîΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö¼ΓöÇΓöÇΓöÇΓö¼ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö¼ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÉ
             Γöé    CODE    ΓöéTRKΓöé     TRACKNAME      ΓöéTIME Γöé
             Γöé            Γöé   Γöé                    Γöé     Γöé
             ΓööΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö┤ΓöÇΓöÇΓöÇΓö┤ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö┤ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÿ
              MexiConc1993  01 Jenny I Read         05:18
              MexiConc1993  02 Mexican Moon         05:03
              MexiConc1993  03 Heal It Up           04:21
              MexiConc1993  04 Jonestown            06:06
              MexiConc1993  05 Rain                 03:28
              MexiConc1993  06 I Call It Love       05:15
              MexiConc1993  07 Jesus Forgive Me     05:17
              MexiConc1993  08 When You Smile       04:18
              MexiConc1993  09 Close To Home        03:31
              MexiConc1993  10 One Of My Kind       03:55
              MexiConc1993  11 End Of The Line      04:39
              MexiConc1993  12 Blind Ambition       06:10
              MexiConc1993  13 Bajo La Lune         05:07
              No NCran1994  01 Ode To My Family     04:30
              No NCran1994  02 I Can't Be With You  03:07
              No NCran1994  03 Twenty One           03:08
              No NCran1994  04 Zombie               05:06
              No NCran1994  05 Empty                03:26
              No NCran1994  06 Everything I Said    03:53
              No NCran1994  07 The Icicle Melts     02:54
              No NCran1994  08 Disappointment       04:14
              No NCran1994  09 Ridiculous Thoughts  04:31
              No NCran1994  10 Dreaming My Dreams   03:37
              No NCran1994  11 Yeat's Grave         02:59
              No NCran1994  12 Daffodil Lament      06:09
              No NCran1994  13 No Need To Argue     02:56
              MachDeep1972  01 Highway Star         06:05
              MachDeep1972  02 Maybe I'm A Leo      04:51
              MachDeep1972  03 Pictures Of Home     05:03
              MachDeep1972  04 Never Before         03:56
              MachDeep1972  05 Smoke On The Water   05:40
              MachDeep1972  06 Lazy                 07:19
              MachDeep1972  07 Space Truckin'       04:31


ΓòÉΓòÉΓòÉ 6.3. Tutorial: Index Files Used ΓòÉΓòÉΓòÉ

The index files created for this database are, for CTAY: Title+Artist and
Artist+Title (the + indicates a compound key made up of more than one field);
for CTTT: Code+TrackName and TrackName+Code.  With these indexes, searching can
be made on CD title, by Artist, or by TrackName.  If searching by year is
required, an additional index could be used, such as Year+Artist+Title (Title
being used as part of the key in case an Artist releases more than one title in
a year).

To search by CD title, a title is used as the key.  Even a partial title can be
used during the actual lookup.  For example, 'Mexican'.  From this, a search is
made in the Title+Artist index.  The first key starting with 'Mexican' is
returned.  The index can then be traversed to find any other keys also starting
with this (by getting the next key).  Since Bullet has high-level access
routines (GET_FIRST_XB, for example), where a key search returns the data
record, once a key is found its data is already in memory (the record is
typically read directly into a structure variable you have set up).  This lets
you move through the CTAY file getting each record in key order, one at a time
(one key and record per call to Bullet).

With the data record in memory, the Code field in the record (CTAY table) is
used as a foreign key into CTTT.  A new search needs to be done, this time on
the Code+TrackName index of the CTTT table.  With the search key set to
'MexiConc1993' (the foreign key field in CTAY of the record that first matched
'Mexican'), the Code+TrackName index for CTTT is searched.  The first key
starting with 'MexiConc1993' is returned, and its data record.  In that record
is the track name, track number, and track time.  When done with that record,
the index can be tranversed in-order (forward, or even backward), and, so long
as the first 12 characters of the Code+TrackName key match 'MexiConc1993', that
key's record is used.  When the match is no longer true in those first 12
characters, which in the example data occurs after 13 tracks have been read, it
means that all matching keys in that index file have been found, and there are
no others matching the criterion.

The TrackName part of Code+TrackName (the part of the key after the first 12
characters) is not needed as a match criterion in this case; TrackName is being
used solely for the purpose of making the key unique to that particular CD
Title (all Code fields are the same for each trackname on the CD).

A possible additional index file would be a key based on Code, for CTAY (there
is such an index already for CTTT).  This would allow indexed access to the
CTAY table when looking for the Title of a track name.  (In other words, you
have a list of track names of all CDs in the database, in track name order, and
want to know the CD Title for each track you see.)  Even without an index this
can be done by using the Code value in CTTT as a sequential search lookup value
in CTAY (searching for the matching Code value in CTAY).  This additional index
is not used here.


ΓòÉΓòÉΓòÉ 6.4. Tutorial: Creating the Data Files ΓòÉΓòÉΓòÉ

Now that the output requirements are known, and the layout of the data records
of the two data files constructed, and the index access methods designed, the
database can be created.  Bullet provides routines to create the files of the
database according to specifications that you supply.

The CTAY (Code-Title-Artist-Year) and CTTT (Code-Track-TrackName-Time) tables
are to be in this format:

  CTAY TABLE
 ΓöîΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö¼ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö¼ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö¼ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÉ
 Γöé   CODE     Γöé      TITLE      Γöé     ARTIST      ΓöéYEAR Γöé
 ΓööΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö┤ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö┤ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö┤ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÿ
  MexiConc1993  Mexican Moon      Concrete Blonde  1993   (sample record)
     :

 CODE      character data, 12 bytes
 TITLE     character data, 32 bytes
 ARTIST    character data, 35 bytes
 YEAR      character data, 4 bytes


  CTTT TABLE
 ΓöîΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö¼ΓöÇΓöÇΓöÇΓö¼ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö¼ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÉ
 Γöé    CODE    ΓöéTRKΓöé     TRACKNAME      ΓöéTIME Γöé
 ΓööΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö┤ΓöÇΓöÇΓöÇΓö┤ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓöÇΓö┤ΓöÇΓöÇΓöÇΓöÇΓöÇΓöÿ
  MexiConc1993  01 Jenny I Read         05:18   (sample record)
      :

 CODE      character data, 12 bytes
 TRK       character data, 2 bytes
 TRACKNAME character data, 40 bytes
 TIME      character data, 5 bytes

The fields are sized primarily on expected need.  Since each DBF record has a
1-byte tag field, the sum of the above record layouts, including the tag byte,
is aligned to an even 4-byte size.  An admirable goal, but not a primary design
consideration.  CTAY is 84 bytes per record; CTTT is 60 bytes per record.

Data is not entered during the create process.  That comes later.  The create
is only to make the physical files.  The following is the Bullet code required
to build the two data files.  The index files are covered later.

typedef struct _CTAY {
 CHAR tag;
 CHAR code[12];
 CHAR title[32];
 CHAR artist[35];
 CHAR year[4];
} CTAY;         // (total CTAY record length is 84 bytes)
CTAY ctayRec;

CHAR ctayName[] = "CTAY.DBF";
ULONG ctayID=0;                             // handle of CTAY file
FIELDDESCTYPE ctayFieldList[4];             // 4 fields used by the record
memset(ctayFieldList,0,sizeof(ctayFieldList));  // init unused bytes to 0

typedef struct _CTTT {
 CHAR tag;
 CHAR code[12];
 CHAR trk[2];
 CHAR trackName[40];
 CHAR time[5];
} CTTT;         // (total CTTT record length is 60 bytes)
CTTT ctttRec;

CHAR ctttName[] = "CTTT.DBF";
ULONG ctttID=0;
FIELDDESCTYPE ctttFieldList[4];
memset(ctttFieldList,0,sizeof(ctttFieldList));

// field descriptor info for CTAY

strcpy(ctayFieldList[0].fieldName, "CODE");
ctayFieldList[0].fieldType = 'C';
ctayFieldList[0].fieldLen = 12;
ctayFieldList[0].fieldDC = 0;

strcpy(ctayFieldList[1].fieldName, "TITLE");
ctayFieldList[1].fieldType = 'C';
ctayFieldList[1].fieldLen = 32;
ctayFieldList[1].fieldDC = 0;

strcpy(ctayFieldList[2].fieldName, "ARTIST");
ctayFieldList[2].fieldType = 'C';
ctayFieldList[2].fieldLen = 35;
ctayFieldList[2].fieldDC = 0;

strcpy(ctayFieldList[3].fieldName, "YEAR");
ctayFieldList[3].fieldType = 'C';
ctayFieldList[3].fieldLen = 4;
ctayFieldList[3].fieldDC = 0;

// field descriptor info for CTTT

strcpy(ctttFieldList[0].fieldName, "CODE");
ctttFieldList[0].fieldType = 'C';
ctttFieldList[0].fieldLen = 12;
ctttFieldList[0].fieldDC = 0;

strcpy(ctttFieldList[1].fieldName, "TRACK");
ctttFieldList[1].fieldType = 'C';
ctttFieldList[1].fieldLen = 2;
ctttFieldList[1].fieldDC = 0;

strcpy(ctttFieldList[2].fieldName, "TRACKNAME");
ctttFieldList[2].fieldType = 'C';
ctttFieldList[2].fieldLen = 40;
ctttFieldList[2].fieldDC = 0;

strcpy(ctttFieldList[3].fieldName, "TIME");
ctttFieldList[3].fieldType = 'C';
ctttFieldList[3].fieldLen = 5;
ctttFieldList[3].fieldDC = 0;

// create the CTAY file

CDP.func = CREATE_DATA_XB;
CDP.filenamePtr = ctayName;
CDP.noFields = 4;
CDP.fieldListPtr = ctayFieldList;
CDP.fileID = 3;
rez = BULLET(&CDP);
if (rez) {
   printf("Failed CTAY create.  Err: %d\n",rez);
   return(rez);
}

// create the CTTT file

CDP.func = CREATE_DATA_XB;
CDP.filenamePtr = ctttName;
CDP.noFields = 4;
CDP.fieldListPtr = ctttFieldList;
CDP.fileID = 3;
rez = BULLET(&CDP);
if (rez) {
   printf("Failed CTTT create.  Err: %d\n",rez);
   return(rez);
}


ΓòÉΓòÉΓòÉ 6.5. Tutorial: Opening the Data Files ΓòÉΓòÉΓòÉ

Before using a data file it must be opened.  Any DBF file can be opened,
whether created by Bullet or another program.  Before creating indexes for the
data files created here, the data files must be open.  The handles of the open
data files are used in the index file create process.


// open CTAY and store handle to ctayID

OP.func = OPEN_DATA_XB;
OP.filenamePtr = ctayData;
OP.asMode = READWRITE | DENYNONE;
rez = BULLET(&OP);
if (rez) {
   printf("Failed CTAY file open.  Err: %d\n",rez);
   return(rez);
}
ctayID = OP.handle;

// open CTTT and store handle to ctttID

OP.func = OPEN_DATA_XB;
OP.filenamePtr = ctttData;
OP.asMode = READWRITE | DENYNONE;
rez = BULLET(&OP);
if (rez) {
   printf("Failed CTTT file open.  Err: %d\n",rez);
   return(rez);
}
ctttID = OP.handle;


ΓòÉΓòÉΓòÉ 6.6. Tutorial: Creating the Index Files ΓòÉΓòÉΓòÉ

Two pairs of index files are used.  For CTAY, Title+Artist and Artist+Title;
for CTTT, Code+TrackName and TrackName+Code.  These are expected to be unique,
and so the index files are specified to not accept duplicate keys.  Also,
rather than using the full field sizes for key expressions, substrings are
specified.  Long keys affect performance, both in index file size and general
access speed (smaller is usually better), and using the entire field data does
little to differentiate the key (for the uniqueness requirement).  If
DUPS_ALLOWED were specified, even shorter keys could be used, though duplicates
generally are not desirable.  If there comes a record whose generated key is
not unique (i.e., it already belongs to another already-added record), either
modify the data entered so that its key is different, if possible, or redo your
index files to allow duplicates.  You could also play if 'safe' and permit
duplicates keys to occur from the outset, knowing that not many duplicates will
occur, where if they do, you can continue and let Bullet managed the duplicate
key with an enumerator.  If many duplicates of keys are possible, re-evaluate
your key expression to make it more unique.  This is done by using a compound
key with more field components.  Still, the fewer components in a key, and the
shorter, the better the key is.

Index file creation:

CHAR xTitleName[] = "TITLE.IX3";
CHAR xTitleExp[] = "SUBSTR(TITLE,1,8)+SUBSTR(ARTIST,1,8)";
CHAT xTitleBuffer[16];          // key buffer for later use
                                // note that if DUPS_ALLOWED, the key buffer must
                                // provide room for the enumerator (2 bytes more)

CHAR xArtistName[] = "ARTIST.IX3";
CHAR xArtistExp[] = "SUBSTR(ARTIST,1,8)+SUBSTR(TITLE,1,8)";
CHAT xArtistBuffer[16];

CHAR xCodeName[] = "CODETRK.IX3";
CHAR xCodeExp[] = "CODE+SUBSTR(TRACKNAME,1,12)";
CHAT xCodeBuffer[24];

CHAR xTrackName[] = "TRKNAME.IX3";
CHAR xTrackExp[] = "SUBSTR(TRACKNAME,1,12)+CODE";
CHAT xTrackBuffer[24];

// create two index files for CTAY

CIP.func = CREATE_INDEX_XB;
CIP.filenamePtr = xTitleName;
CIP.keyExpPtr = xTitleExp;
CIP.xbLink = ctayID;            // the handle of the open CTAY file
CIP.sortFunction = NLS_SORT;    // sort key by NLS for proper mixed-case order
CIP.codePage = 0;               // use OS-default code page
CIP.countryCode = 0;            // use OS-default country code
CIP.collatePtr = NULL;          // no need for a special collate table
CIP.nodeSize = 512;             // 512-byte node size (or 1024, 2048 bytes)
rez = BULLET(&CIP);
if (rez) {
   printf("Failed Title index create.  Err: %d\n",rez);
   return(rez);
}

CIP.filenamePtr = xArtistName;
CIP.keyExpPtr = xArtistExp;     // other values still valid from above
rez = BULLET(&CIP);
if (rez) {
   printf("Failed Artist index create.  Err: %d\n",rez);
   return(rez);
}

// create two index files for CTTT

CIP.func = CREATE_INDEX_XB;
CIP.filenamePtr = xCodeName;
CIP.keyExpPtr = xCodeExp;
CIP.xbLink = ctttID;            // as above...
CIP.sortFunction = NLS_SORT;
CIP.codePage = 0;
CIP.countryCode = 0;
CIP.collatePtr = NULL;
CIP.nodeSize = 512;
rez = BULLET(&CIP);
if (rez) {
   printf("Failed Code index create.  Err: %d\n",rez);
   return(rez);
}

CIP.filenamePtr = xTrackName;
CIP.keyExpPtr = xTrackExp;      // other values still valid from above
rez = BULLET(&CIP);
if (rez) {
   printf("Failed TrackName index create.  Err: %d\n",rez);
   return(rez);
}


ΓòÉΓòÉΓòÉ 6.7. Tutorial: Opening the Index Files ΓòÉΓòÉΓòÉ

Before using an index file it must be opened.  Only Bullet index files can be
opened.

// open Title index and store handle to xTitleID

OP.func = OPEN_INDEX_XB;
OP.filenamePtr = xTitleName;
OP.asMode = READWRITE | DENYNONE;
OP.xbLink = ctayID;                     // note xbLink
rez = BULLET(&OP);
if (rez) {
   printf("Failed Title index open.  Err: %d\n",rez);
   return(rez);
}
xTitleID = OP.handle;

// open Artist index and store handle to artistID

OP.func = OPEN_INDEX_XB;
OP.filenamePtr = xArtistName;
OP.asMode = READWRITE | DENYNONE;
OP.xbLink = ctayID;
rez = BULLET(&OP);
if (rez) {
   printf("Failed Artist index open.  Err: %d\n",rez);
   return(rez);
}
xArtistID = OP.handle;

// open Code index and store handle to xCodeID

OP.func = OPEN_INDEX_XB;
OP.filenamePtr = xCodeName;
OP.asMode = READWRITE | DENYNONE;
OP.xbLink = ctttID;
rez = BULLET(&OP);
if (rez) {
   printf("Failed Code index open.  Err: %d\n",rez);
   return(rez);
}
xCodeID = OP.handle;

// open TrackName index and store handle to xTrackNameID

OP.func = OPEN_INDEX_XB;
OP.filenamePtr = xTrackName;
OP.asMode = READWRITE | DENYNONE;
OP.xbLink = ctttID;
rez = BULLET(&OP);
if (rez) {
   printf("Failed TrackName index open.  Err: %d\n",rez);
   return(rez);
}
xTrackNameID = OP.handle;


ΓòÉΓòÉΓòÉ 6.8. Tutorial: Inserting CD Title Data ΓòÉΓòÉΓòÉ

The files have been created, and are open.  Data may now inserted into the
database.  The term 'inserted' is used in contrast to 'added' since Bullet
'adds' data only, while it 'inserts' data and key information together (and
especially, the key is inserted, in order).  The CD data entry would typically
be done via the keyboard.  This tutorial assumes that this has already been
done, and that the data items are in the following structure variables:
ctayRec for CD title data; ctttRec for track data.

First, the CD title record is added to CTAY with a key inserted into the Title
index and another key into Artist index.  Data entered by the user consists of
CD title, artist, and year.  The Code field is generated by the program code,
shown below.

// ctayRec has CD title, artist, and year -- generate Code field data

strncpy(ctayRec.code  ,ctayRec.title,4);   // first 4 of title
strncpy(ctayRec.code+4,ctayRec.artist,4);  // first 4 of artist
strncpy(ctayRec.code+8,ctayRec.year,4);    // and 4 of year make the Code field

// insert CD title data (one data record added, two index files inserted into)

AP[0].func = INSERT_XB;
AP[0].handle = xTitleID;        // handle of Title index
AP[0].recNo = 0;                // required
AP[0].recPtr = &ctayRec;        // the CD title record
AP[0].keyPtr = xTitleBuffer;    // key buffer
AP[0].nextPtr = &AP[1];         // since two index files, point to next
AP[1].handle = xArtistID:       // handle of Artist index
AP[1].recNo = 0x80000000;       // required
AP[1].recPtr = &ctayRec;        // the CD title record
AP[1].keyPtr = xArtistBuffer;   // key buffer
AP[1].nextPtr = NULL;

rez = BULLET(&AP[0]);
if (rez) {                      // rez is not error, but pack index of error
   rc = AP[abs(rez)-1].stat;
   printf("Insert failed, pack: %d, err: %d\n",rez,rc);
   return(rc);
}

A CD title has been entered.  That's the first part of the data entry.  Next,
each track on that CD needs to be entered.  It needs the Code field data
generated above, along with the general track data (track name, etc.).


ΓòÉΓòÉΓòÉ 6.9. Tutorial: Inserting CD Track Data ΓòÉΓòÉΓòÉ

During the CD title insert, a Code field value was generated.  That field value
is used for each of the track entries inserted into database for this CD.  The
actual track data has already been placed in the ctttRec structure variable (by
the main program code, not shown here), and only the Code field of ctttRec
needs to be set.  Once done, the track record is inserted, this time into the
CTTT data file and the Code and Track index files.

//
// this code section is repeated for each track on the CD
//

// copy Code field data from CTAY

strncpy(ctttRec.code,ctayRec.code,12); // use Title's Code for each Track entry

// insert track data (one data record added, two index files inserted into)

AP[0].func = INSERT_XB;
AP[0].handle = xCodeID;         // handle of Code index
AP[0].recNo = 0;                // required
AP[0].recPtr = &ctttRec;        // the track record
AP[0].keyPtr = xCodeBuffer;     // key buffer
AP[0].nextPtr = &AP[1];         // since two index files, point to next
AP[1].handle = xTrackID:        // handle of TrackName index
AP[1].recNo = 0x80000000;       // required
AP[1].recPtr = &ctttRec;        // the track record
AP[1].keyPtr = xTrackBuffer;    // key buffer
AP[1].nextPtr = NULL;

rez = BULLET(&AP[0]);
if (rez) {                      // rez is not error, but pack index of error
   rc = AP[abs(rez)-1].stat;
   printf("Insert failed, pack: %d, err: %d\n",rez,rc);
   return(rc);
}

// and repeat for each track on the CD

The database now contains a complete entry for the one CD.  This process is
repeated for each CD that is to entered into the database.  This is the data
entry process.  Data retrieval is covered next.


ΓòÉΓòÉΓòÉ 6.10. Tutorial: Retrieving Data ΓòÉΓòÉΓòÉ

The database can be accessed in several different ways.  For indexed access,
any index file can be used.  Alternatively, direct access without an index
could also be used.  Generally, since there's no guarantee that the physical
order of rows in the database is in any order at all, index access is usually
desired.  Possible retrievals are by:

1. CD title, in alphabetical order, and, from information from the CD title,
all tracks on the CD.  The tracks are read into an array and sorted by track
number.  Alternatively, a separate index could be used to order track entries
by TrackNo+TrackName(+etc.), but since there are so few tracks per CD
(generally less than 20), it's much more efficient to simply get all tracks for
a CD title into an array, and then sort that array for presentation.
Otherwise, an index file would need to be maintained on a permanent basis, and
for something that can easily be done at run-time (and then discarded).

2. Artist, in alpha order.  Access is similar to CD title, above, since it uses
the same data file (CTAY).  As above, tracks for each artist's title can then
be retrieved.  The only difference between this and the CD title access is that
this lists CDs by artist, rather than by title.

3. Track name, in alphabetical order.  This lists all tracks by track name, in
order, with different tracks intermixed with CD titles.  In other words, if
three CDs had the track called "In the Summer Time", then each of those tracks
would appear together, one after the other.  There is no index available in
this database to access the CD on which the track name appears, so you cannot
go from trackname alone to CD title (by index).  However, since the Code field
is available, and is common to each data file, the Code field value of a track
record can be used to sequentially search the Title records for a match.  This
is a non-indexed search.  Access is slower than if an index were available, but
unless you created indexes for all possible search methods, some sequential
(read: slow) searching may be required.  For smaller files (less than 10,000
records), or where only a few such searches occur, this may be perfectly
acceptable.

An index on Code exists (for CTTT, but not CTAY), and you may find a use for
this other than for use as a foreign key lookup.  Also, you may create ad hoc
index files whenever you need to, and these can be deleted when no longer
required.  Since Bullet can create an index for even a large file in just a few
seconds, this is a viable option; the few seconds needed to create an index may
be many times repaid in the time saved in sequential searches on a non-indexed
access.

The first retrieval mentioned is shown next.

// for each CD title in the database, display all its tracks

                        // ctayRec already defined
CTTT cttyRecs[59];      // store each Track record for later sorting by track#

// This would be repeated for each CD title (only the first is shown).
// Note: For the next CD title, GET_NEXT_XB would be used.  If necessary,
// the key just accessed could be stored (preserved), then laster reaccessed
// using GET_EQUAL_XB again, and then immediately followed with a GET_NEXT_XB.
// This way, you can stop processing completely, and restart up where you left
// off.

// files are locked as required (e.g., full-lock, shared)

AP.func = GET_FIRST_XB;
AP.handle = xTitleID;
AP.recPtr = &ctayRec;
AP.keyPtr = xTitleBuffer;
rez = BULLET(&AP);
while (rez==0) {
   printf("Title: %s  Artist: %s  Year: %s\n",
           ctayRec.title,
           ctayRec.artist,
           ctayRec.year);

   // get each track belonging to this CD title (same Code values)

   int trk=0;                            // counter
   memset(xCodeBuffer,0,sizeof(xCodeBuffer)); // clear it to 0
   strncpy(xCodeBuffer,ctayRec.code,12); // copy Code to search-for buffer

   AP.func = GET_EQUAL_XB;
   AP.handle = xCodeID;
   AP.recPtr = &ctttRecs[trk];
   AP.keyPtr = xCodeBuffer;             // find Code
   rez = BULLET(&AP);

   // rather than print each when gotten, collect them, sort, then display
   // if rez==0 here then Code matched exactly, no strncmp needed, otherwise,
   // there's no match on the first 12 characters of Code

   AP.func = GET_NEXT_XB;               // continue getting while same Code
   while (rez==0) {
      trk++;                            // limit to array size...
      AP.recPtr = &ctttRecs[trk];       // read into next array record
      rez = BULLET(&AP);
      if (rez==0) {

         // if code field in CTTT no longer the same as CTAY, then have them all

         if (strncmp(ctttRecs[trk].code,ctayRec.code,12)!=0) {
            rez = 1;
            break;                      // different Code fields
         }
      }
   }

   // if rez==1 or at end of file then no error, sort and display --
   // trk at this point is number of tracks that matched Code (1-based)
   // DoSortOf... routine sorts the ctttRecs array with trk count of elements

   if ((rez==1) | (rez==EXB_END_OF_FILE)) {
      DoSortOfTheTrackRecordsByTrackNumber(&ctttRecs[0],trk);

      // print each track record, now sorted by track number

      for (i=0;i < trk;i++) {
         printf("Trk: %s  Name: %s   Time: %s\n",
                 ctttRecs[i].trk,
                 ctttRecs[i].trackName,
                 ctttRecs[i].time);
      }
   else {
       printf("failed, rez: %d\n",rez);
       return(rez);
   }
}


ΓòÉΓòÉΓòÉ 6.11. Tutorial: Preparing to End Your Program ΓòÉΓòÉΓòÉ

It's proper programming practice to perform an orderly shutdown of Bullet when
ending your program.  This also extends to files, too, where even if you are
not ending your program, but are finished using a file, close that file so that
you release the resources allocated for it.

In the event that an abnormal termination occurs, and you are unable to release
outstanding locks and close all files, you may choose to call EXIT_XB and allow
that to close files (the OS releases locks when the file is closed).  If you
are unable to call even EXIT_XB, there is still another method that gets
Bullet's attention.  During Bullet initialization, the EXIT_XB routine is
registered with the OS so that whenever your program terminates (or is
terminated), EXIT_XB is called automatically.  This ensures that headers are
properly written in all but the most severe of abnormal terminations (abend),
such as a lockup.

// all outstanding locks should have been released immediately after use

// close each file that was opened

HP.func = CLOSE_DATA_XB
HP.handle = xTitleID;
if (HP.handle) rez = BULLET(&HP);
if (rez) printf("close failed, err: %d\n",rez);

HP.handle = xArtistID;
if (HP.handle) rez = BULLET(&HP);
if (rez) printf("close failed, err: %d\n",rez);

HP.handle = xCodeID;
if (HP.handle) rez = BULLET(&HP);
if (rez) printf("close failed, err: %d\n",rez);

HP.handle = xTrackID;
if (HP.handle) rez = BULLET(&HP);
if (rez) printf("close failed, err: %d\n",rez);

EP.func = EXIT_XB;
rez = BULLET(&EP);
if (rez) printf("exit failed, err: %d\n",rez);

// Bullet is now deinitialized, and must be INIT_XB'ed before further use


ΓòÉΓòÉΓòÉ 7. History of Changes ΓòÉΓòÉΓòÉ

Changes Made

2.50 25-Jun-97:

   1. Modifed Win32 DLL to work in NT4.
   2. Changed the way a Next/Prev key access works after reaching EOF/TOF.
      E.g., now when EOF is reached a Prev key access will get the second to
      last key, in order, rather than re-getting the last key.
   3. Fixed when using ATOMIC_MODE=1 in SET_SYSVARS_XB, and you have
      DUPS_ALLOWED allowed, when using INSERT_XB, UPDATE_XB, or LAST_KEY_XB,
      where error EXB_KEY_NOT_FOUND (8501) was returned when the key newly
      inserted/updated already existed.
   4. Fixed PREV_KEY/GET_PREV_XB where only the first encounter of
      EXB_TOP_OF_FILE was returned, after which out-of-order key was returned.

 2.14 29-May-97:

   1. Fix for key order bug dropping to ASCII sort even when ASCII sort was not
      picked.

 2.13 23-Nov-96:

   1. Added EQUAL_OR_GREATER_KEY/LESSER_KEY for fuzzy key access (atomic).
   2. Added GET_EQUAL_OR_GREATER/LESSER for fuzzy key+record access (atomic).
   3. Increased maximum fields from 255 to 1024 for dBASE 5 DBFs.
   4. Update of data record would not set DBF dirty flag, though effect was
      minimal (last update timestamp in DBF header not set).
   5. Fixed Bullet/2 REXX interface DLL dispatcher.

 2.12 6-Nov-96:

   1. New, lower prices!  Order direct-from-author or from BMT Micro.
   2. New source example, compact.c, to compact DBFs with attached DBT memo
      file.
   3. Release 2.12.  Code version is 2.11.

 2.11 16-Oct-96:

   1. New order option, option E, includes DOSX32, OS/2 and Win32 versions of
      Bullet all in one box.

 2.105 11-Oct-96:

   1. Added OS function override revectoring for all OS calls made, and added
      supporting query/set routines, QUERY/SET_VECTORS_XB.
   2. Added USE_ANSI_SET flag for index file collate sequence table (esp. for
      Windows), in addition to standard OEM character set, for use at
      CREATE_INDEX_XB.
   3. Added shared/exclusive region locking mode for NT.
   4. UPDATE_XB efficiency improved, esp. with large number of index files.
   5. Internal sort efficiency improved (even faster).
   6. Improved demo program samples; also added compile option for generating
      console or windowed app.
   7. Consolidated Bullet header files into one, for all platforms, with
      minimal conditionals; the file is named bullet_2.h.

 2.103 4-Oct-1996:

   1. Win32 version would fail under NT (and Win32s) because of un-saved
      register into consecutive kernel calls.
   2. Win32s now supported.
   3. Win32 version now in a single DLL (supports Watcom (old and new) and VC4,
      both, and both Win32 and Win32s).
   4. CREATE_INDEX_XB was always using OS-supplied codepage/country code.
   5. Added callback support in reindex and pack records, esp. useful for
      DOSX32 and Win32s.
   6. UPPER() in key expression would fail.

 2.101 28-Aug-1996:

   1. REINDEX_XB would fail on certain key sizes, causing an access
      violation/GPF.

 2.100 3-Aug-1996:

   1. REINDEX_XB and GET_LAST_XB could fail if DUPS_ALLOWED.
   2. Sort functions 1, 2, and 3 were operating as expected but 4-6 were using
      3.
   3. Node scanning was using the wrong sort-compare function (1 was using 3, 2
      was using 1, and 3 was using 2).
   4. UPDATE_XB was returning failed pack's AP[].stat=1 instead of true error
      code if the update failed when inserting a new key.
   5. BUILD_KEY_XB was not issuing skip tag warning, and was not passing
      .recPtr to external build key function.
   6. PACK_RECORDS_XB and DEBUMP_RECORD_XB values in bullet2.h were
      interchanged: Pack is 47; Debump is 46.
   7. DEBUMP_RECORD_XB requires FLUSH_DATA_HEADER_XB call prior to use.
   8. Documentation errata: GET_DESCRIPTOR_XB uses DP.handle for IN; includes
      DP.FD.altFieldLength for OUT (DP.fieldOffset noted previously)
   9. GET_DESCRIPTOR_XB was not returning info in DP.FD.* when accessed by
      fieldName.
  10. SDP.lastUpdate year was returned with year-256 (1996 was 1740).
  11. GET_ERROR_CLASS_XB was not returning useful info.
  12. ZAP_INDEX_HEADER_XB had been making index data start at 16KB size.
  13. DosExitList() handler is now dropped after calling EXIT_XB.

 2.051 24-Apr-96:

   1. Fixes to transaction routines InsertXB and UpdateXB, including Insert
      rollback fix with multiple packs.
   2. Update pack number return value fix.
   3. Update w/ multiple packs fix, and an Update rollback fix.
   4. Fix of GetKeyForRecordXB when DUPS_ALLOWED.

 2.050 26-Feb-96:

   1. Shareware refresh released.

 2.044 31-Jan-96:

   1. Fixed DUPS_ALLOWED bug at INSERT_XB where error EXB_TOP_OF_FILE or error
      EXB_TOO_MANY_DUPLICATES would be returned.

 2.043  3-Jan-96:

   1. Added atomic key access for NEXT_KEY, PREV_KEY, GET_NEXT, and GET_PREV
      for simpler use in multi-threaded programs.
   2. Changed EXB_SHARED_LOCK_ON in BULLET2.H (was ERR_ rather than EXB_).

 2.042 20-Dec-95:

   1. MAKE_DIR_DOS was expecting directory name in DFP.bufferPtr.  Changed to
      expect path in DFP.filenamePtr.

 2.040 28-Oct-95:

   1. BACKUP_FILE_XB extended to back up related memo file on DBF backup.
   2. [2.033] Fixed memo file problems.
   3. [2.032] Win95 version available.
   4. [2.031] DOSX32 version available.

 2.031 11-Sep-95:

   1. Added DP.fieldOffset to documentation for GET_DESCRIPTOR_XB.
   2. Instance tracker purified of DosEnterCritSec need.

 2.030 9-Sep-95:

   1. Removed erroneous '...header reload...' in online docs of RELOCK_DATA_XB.
   2. Default maximum filesizes set to 2047 MB from 2048 MB (absolute max is
      4095 MB).
   3. Fixed REINDEX_XB so that it actually uses re-evaluted key expression.
   4. ATEXIT_XB is obsolete.
   5. BREAK_XB is obsolete.

 2.020 31-Aug-95:

   1. Locality and other cache-related options can be set at file open.
   2. IP.versionBullet changed to *1000 from *100 (2020 is version 2.020).
   3. Instance tracker corrected.

 2.01  22-Aug-95:

   1. Changed index header to 1024 bytes from 768 for sector-alignment.
   2. Changed header ID to '31ch' from '30ch'.

 2.00  20-Aug-95:

 Preliminary release.


ΓòÉΓòÉΓòÉ 8. Bullet Include File ΓòÉΓòÉΓòÉ

The Bullet/2 C/C++ header file, BULLET_2.H:


/* BULLET_2.H    23-Nov-96-chh
 *
 *  Bullet header for 32-bit C/C++ (DOSX32, OS/2, and Win32s/Win32)
 *  Bullet call numbers, parameter packs, and error number equates
 *
 *  Requires PLATFORM defined and set to ON_DOSX32 (3), ON_OS2 (4),
 *  or ON_WIN32 (5) before getting here.  For example:
 *    #define PLATFORM ON_DOSX32  (ON_DOSX32 defined as 3)
 *
 */

#ifndef __BULLET_H
#define __BULLET_H

/*
 * The #pragma pack(1)/#pragma pack() is no longer required since all
 * structure members in this header will align properly -- all are
 * 32-bit size except for the structure "FieldDescType", but fieldDA
 * member (a LONG) is at a 32-bit alignment already (at byte offset +12).
 * The altFieldLength member, same structure, is also already at
 * proper alignment for a 16-bit value (at byte offset +18).  If, for
 * some reason, your compiler aligns the members differently, then you
 * must use the appropriate compiler pragma to prevent this -- the
 * FieldDescType size is 32 bytes exactly.  It is not likely that any
 * conforming compiler will alter this structure, but, now you know what
 * to do if it does.
 *
 * #pragma pack(1)
 *
 * NOTE: In your program source code, when you layout your record buffer
 * structure, you must use the #pragma pack(1)/#pragma pack() directives
 * around it since it will be, most likely, modified.  The reason is that
 * this structure MUST start with the implicit TAG field (a BYTE), and so,
 * unless you use only BYTE/CHAR members in your structure (Bullet can use
 * binary field values), or take special care to align the record layout
 * so no padding is performed by the compiler, then you will need to use
 * the pack(1) pragma.
 *
 * #pragma pack()
 */

/* Re: Bullet/X for DOSX32:
 * (refer to ccdosfn.c for more)
 * Bullet ccdosfn.c provides no OS call support to determine the
 * system country code and code page ID.  This can be coded in
 * ccdosfn.c, but it is quite compiler- and extender-dependent.
 * The current state is to supply a collate sequence table for
 * country code=1 and code page=437 (the table is statically coded).
 * For other sort tables, modify as required.  If support for this
 * at the system level is made available at run-time, you may
 * want to change the following to 0, for both CTRYCODE and CODEPAGE.
 * Until this is so coded, you cannot use 0 here (as with the Win32
 * and OS/2 versions), else error EXB_216501 (8251) is the result.
 * This table is added to each index file (if NLS or a user sort).
 */

#ifndef PLATFORM
 #error No PLATFORM specified
 #error ---------------------
#endif

#ifndef ON_DOSX32
 #define ON_DOSX32 3
 #define ON_OS2    4
 #define ON_WIN32  5
#endif

#if PLATFORM == ON_DOSX32
 #define CTRYCODE 1     /* 0 signifies default country code (at index create) */
 #define CODEPAGE 437   /* 0 signifies default code page (at index create) */
                        /* but DOS extender may not support OS call to get info */
                        /* see ccdosfn.c for making changes for DOSX32 platform */
 #define RELOCK_AVAIL 0 /* relock not supported */

 #define VOID void      /* these are already defined if Win32 or OS/2, but not DOS */
 #define SHORT short
 #define LONG long
 #define CHAR char

 typedef unsigned char BYTE;
 typedef unsigned short USHORT;
 typedef unsigned long ULONG;
 typedef unsigned char *PSZ;
 typedef VOID *PVOID;

 #define APIENTRY __cdecl

#elif PLATFORM == ON_OS2
 #define CTRYCODE 0
 #define CODEPAGE 0
 #define RELOCK_AVAIL 1    /* relock is supported */

 /* above types are assumed defined in os2def.h */

#elif PLATFORM == ON_WIN32
 #define CTRYCODE 0
 #define CODEPAGE 0     /* may be ANSI or OEM code page value, depending on USE*CHARSET flag */
 #define RELOCK_AVAIL 0 /* relock not supported */

 /* above types are assume defined in windef.h and winnt.h */

#else
 #error No PLATFORM specified
 #error ---------------------

#endif

#ifndef __BLT_DYNA    // define this if using run-time loading of BULLET*.DLL
                      // via LoadLibrary(Win32) or DosLoadModule(OS/2)
 #ifdef __cplusplus
  extern "C" LONG APIENTRY BULLET(PVOID datapack);
 #else
  extern LONG APIENTRY BULLET(PVOID datapack);
 #endif

#else

  LONG (* APIENTRY BULLET)(PVOID datapack);

#endif



/* The following on mutex-semaphore protection does not apply to Bullet/X */
/* unless the semaphore routines (in ccdosfn.c) are coded to do something */

/* All Bullet routines are mutex-semaphore protected except the following:
 *
 * MEMORY_XB            STAT_HANDLE_XB          GET_ERROR_CLASS_XB
 * QUERY_SYSVARS_XB     QUERY_VECTORS_XB        CHECK_REMOTE_XB
 * STAT_DATA_XB         STAT_INDEX_XB
 *
 * This means that any thread can call the above routines at any time.  All
 * other calls in the current process block until the previous thread exits
 * BULLET.  The default mutex wait is 0 milliseconds, and can be set via
 * SET_SYSVARS_XB using the MUTEX_SEM_TIMEOUT index.  In the case of
 * STAT_DATA_XB and STAT_INDEX_XB, these should be used only when there
 * is no chance that another thread may close that file handle while the
 * routine is working.
 *
 */



/* ************************************************************************
 *
 * xxx.func call numbers
 *
 * ************************************************************************/

#define GEN_ERR_XB              0
#define INIT_XB                 1  /* system */
#define EXIT_XB                 2
#define MEMORY_XB               4
#define BACKUP_FILE_XB          6
#define STAT_HANDLE_XB          7
#define GET_ERROR_CLASS_XB      8

#define QUERY_SYSVARS_XB        10 /* advanced system */
#define SET_SYSVARS_XB          11
#define SET_DVMON_XB            12 /* reserved */
#define QUERY_VECTORS_XB        13
#define SET_VECTORS_XB          14

#define CREATE_DATA_XB          20 /* data control mid-level */
#define OPEN_DATA_XB            21
#define CLOSE_DATA_XB           22
#define STAT_DATA_XB            23
#define READ_DATA_HEADER_XB     24
#define FLUSH_DATA_HEADER_XB    25
#define COPY_DATA_HEADER_XB     26
#define ZAP_DATA_HEADER_XB      27

#define CREATE_INDEX_XB         30 /* key control mid-level */
#define OPEN_INDEX_XB           31
#define CLOSE_INDEX_XB          32
#define STAT_INDEX_XB           33
#define READ_INDEX_HEADER_XB    34
#define FLUSH_INDEX_HEADER_XB   35
#define COPY_INDEX_HEADER_XB    36
#define ZAP_INDEX_HEADER_XB     37

#define GET_DESCRIPTOR_XB       40 /* data access mid-level */
#define GET_RECORD_XB           41
#define ADD_RECORD_XB           42
#define UPDATE_RECORD_XB        43
#define DELETE_RECORD_XB        44
#define UNDELETE_RECORD_XB      45
#define DEBUMP_RECORD_XB        46
#define PACK_RECORDS_XB         47

#define GET_MEMO_SIZE_XB        50 /* memo access mid-level */
#define GET_MEMO_XB             51
#define ADD_MEMO_XB             52
#define UPDATE_MEMO_XB          53
#define DELETE_MEMO_XB          54
#define MEMO_BYPASS_XB          59 /* see below for bypass ordinals */

#define BYPASS_CREATE_MEMO       1 /* The bypass routines are automatically */
#define BYPASS_OPEN_MEMO         2 /* performed by BULLET but can be done */
#define BYPASS_CLOSE_MEMO        3 /* manually, if needed - these numbers are */
#define BYPASS_READ_MEMO_HEADER  4 /* put in MDP.memoBypass, with MDP.func */
#define BYPASS_FLUSH_MEMO_HEADER 5 /* set to MEMO_BYPASS_XB */



#define FIRST_KEY_XB            60 /* key access mid-level */
#define EQUAL_KEY_XB            61
#define EQUAL_OR_GREATER_KEY_XB 110
#define EQUAL_OR_LESSER_KEY_XB  111
#define NEXT_KEY_XB             62
#define PREV_KEY_XB             63
#define LAST_KEY_XB             64
#define STORE_KEY_XB            65
#define DELETE_KEY_XB           66
#define BUILD_KEY_XB            67
#define GET_CURRENT_KEY_XB      68
#define GET_KEY_FOR_RECORD_XB   69

#define GET_FIRST_XB            70 /* key and data access high-level */
#define GET_EQUAL_XB            71
#define GET_EQUAL_OR_GREATER_XB 112
#define GET_EQUAL_OR_LESSER_XB  113
#define GET_NEXT_XB             72
#define GET_PREV_XB             73
#define GET_LAST_XB             74
#define INSERT_XB               75
#define UPDATE_XB               76
#define REINDEX_XB              77

#define LOCK_XB                 80 /* network control */
#define UNLOCK_XB               81
#define LOCK_INDEX_XB           82
#define UNLOCK_INDEX_XB         83
#define LOCK_DATA_XB            84
#define UNLOCK_DATA_XB          85
#define CHECK_REMOTE_XB         86
#define RELOCK_XB               87
#define RELOCK_INDEX_XB         88
#define RELOCK_DATA_XB          89

#define DELETE_FILE_DOS         90 /* DOS file I/O low-level */
#define RENAME_FILE_DOS         91
#define CREATE_FILE_DOS         92
#define OPEN_FILE_DOS           93
#define SEEK_FILE_DOS           94
#define READ_FILE_DOS           95
#define WRITE_FILE_DOS          96
#define CLOSE_FILE_DOS          97
#define ACCESS_FILE_DOS         98
#define EXPAND_FILE_DOS         99
#define MAKE_DIR_DOS            100
#define COMMIT_FILE_DOS         101

/* ************************************************************************
 *
 * operating system file I/O equates
 *
 * ************************************************************************/

#define READONLY        0x00000000 /* std file access mode */
#define WRITEONLY       0x00000001 /* no underscore used for std equates */
#define READWRITE       0x00000002

#define DENYREADWRITE   0x00000010 /* std file share mode, cannot be 0 */
#define DENYWRITE       0x00000020
#define DENYREAD        0x00000030
#define DENYNONE        0x00000040
#define NOINHERIT       0x00000080

#define NO_LOCALITY     0x00000000 /* optional cache modes */
#define SEQ_LOCALITY    0x00010000
#define RND_LOCALITY    0x00020000
#define MIX_LOCALITY    0x00030000
#define SKIP_CACHE      0x00100000 /* not inherited by child process */
#define WRITE_THROUGH   0x00400000 /* not inherited by child process */


#define LOCK_SHARED      1         /* a read-only lock, OS/2 and NT only */
#define LOCK_EXCLUSIVE   0         /* default */

/* ************************************************************************
 *
 * .sortFunction IDs, Query/SetSysVars|Vectors item IDs
 *
 * ************************************************************************/

// SORT_SET flag:  USE_*_CHARSET defaults to OEM character set, and is used
// (OEM or ANSI) if the .sortFunction is NLS or a custom sort-compare.

#define USE_OEM_CHARSET  (0 << 17) /* for DOSX32, OS/2, and Windows */
#define USE_ANSI_CHARSET (1 << 17) /* for Windows (.sortFunction flag) */

#define DUPS_ALLOWED (1 << 16) /* allow duplicate keys (.sortFunction flag) */

/* All Bullet system vars set to default values at INIT_XB */
/* Sorts 1-19 also used as CIP.sortFunction (can be OR'ed with DUPS_ALLOWED) */
/* Intrinsic sorts (1-6) are read-only (R-O) */

#define ASCII_SORT 1    /* sort by: ASCII value (R-O) */
#define NLS_SORT   2    /* NLS (R-O) */
#define S16_SORT   3    /* 16-bit signed integer (R-O) */
#define U16_SORT   4    /* 16-bit unsigned integer (R-O) */
#define S32_SORT   5    /* 32-bit signed integer (R-O) */
#define U32_SORT   6    /* 32-bit unsigned integer (R-O) */

/* sorts 7 to 9 are reserved */
/* Custom sort-compare functions are from 10 to 19 */

#define BUILD_KEY_FUNC  20      /* key build function ptr */
#define PARSER_FUNC     21      /* key expression parser function ptr */

#define MUTEX_SEM_HANDLE     29 /* handle of Bullet's mutex semaphore (R-O) */
#define LOCK_TIMEOUT         30 /* lock-wait timeout (default=0, no wait)*/
#define MUTEX_SEM_TIMEOUT    31 /* mutex semaphore-wait timeout (def=0,none) */
#define PACK_BUFFER_SIZE     32 /* pack buffer size (def=0, min autosize) */
#define REINDEX_BUFFER_SIZE  33 /* reindex buffer size (def=0, min autosize) */
#define REINDEX_PACK_PCT     34 /* reindex node pack % (default=100, max) */
#define TMP_PATH_PTR         35 /* temporary file path ptr (default=NULL) */
#define REINDEX_SKIP_TAG     36 /* index skip tag select (default=0, none) */
#define COMMIT_AT_EACH       37 /* commit each insert/update in pack (def=0) */
#define MEMO_BLOCKSIZE       38 /* memo block size (default=512 bytes) */
#define MEMO_EXTENSION       39 /* memo filename extension (default='DBT\0') */
#define MAX_DATAFILE_SIZE    40 /* max data size (default=0x7FEFFFFF=2095MB) */
#define MAX_INDEXFILE_SIZE   41 /* max index size (default=0x7FEFFFFF=2095MB)*/
#define ATOMIC_MODE          42 /* bit0=1 atomic next/prev key access (def=0)*/
#define CALLBACK_PTR         43 /* callback at reindex/pack (def=0, none) */

/* ************************************************************************
 *
 * Query/SetVectors vector IDs
 *
 * ************************************************************************/

#define VECTOR_CLOSE_FILE           2
#define VECTOR_CREATE_DIR           3
#define VECTOR_CREATE_FILE          4
#define VECTOR_CREATE_UNIQUE_FILE   5
#define VECTOR_DELETE_FILE          6
#define VECTOR_LENGTH_FILE          7
#define VECTOR_MOVE_FILE            8
#define VECTOR_OPEN_FILE            9
#define VECTOR_READ_FILE           10
#define VECTOR_SEEK_FILE           11
#define VECTOR_UPDATE_DIR_ENTRY    12
#define VECTOR_WRITE_FILE          13
#define VECTOR_LOCK_FILE           14
#define VECTOR_IS_DRIVE_REMOTE     15
#define VECTOR_IS_FILE_REMOTE      16
#define VECTOR_EXITLIST            17
#define VECTOR_REMOVE_EXITLIST     18
#define VECTOR_FREE                19
#define VECTOR_GET_SORT_TABLE      20
#define VECTOR_GET_COUNTRY_INFO    21
#define VECTOR_GET_ERROR_CLASS     22
#define VECTOR_GET_MEMORY          23
#define VECTOR_GET_TMP_DIR         24
#define VECTOR_GET_VERSION         25
#define VECTOR_MALLOC              26
#define VECTOR_SET_HANDLE_COUNT    27
#define VECTOR_GET_TIME_INFO       28
#define VECTOR_UPPERCASE           29
#define VECTOR_CLOSE_MUTEX_SEM     30
#define VECTOR_CREATE_MUTEX_SEM    31
#define VECTOR_RELEASE_MUTEX_SEM   32
#define VECTOR_REQUEST_MUTEX_SEM   33


/* ************************************************************************
 *
 * Parameter pack structures, typedefs
 *
 * ************************************************************************/

/* AP, CP, CDP, etc., are suggested variable names */

typedef struct _ACCESSPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle of Bullet file to access */
LONG  recNo;          /* IO, record number */
PVOID recPtr;         /* I, programmer's record buffer */
PVOID keyPtr;         /* I, programmer's key buffer */
PVOID nextPtr;        /* I, NULL if not xaction, else next AP in list */
} ACCESSPACK; /* AP */
typedef ACCESSPACK *PACCESSPACK;

/* CBP is the structure received by the callback procedure */
/* structure members are filled in by Bullet */

typedef struct _CALLBACKPACK {
ULONG sizeIs;         /* structure size (current 16 bytes) */
ULONG callMode;       /* 0=from reindex; 1=from DBF pack */
ULONG handle;         /* file handle */
ULONG data1;          /* for callMode=0/1: progress percent (1-99,0) */
} CALLBACKPACK; /* CBP */
typedef CALLBACKPACK *PCALLBACKPACK;

typedef struct _COPYPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle of Bullet file to copy */
PSZ   filenamePtr;    /* I, filename to use (drv+path must exist if used) */
} COPYPACK; /* CP */
typedef COPYPACK *PCOPYPACK;

typedef struct _CREATEDATAPACK {
ULONG func;
ULONG stat;
PSZ   filenamePtr;    /* I, filename to use */
ULONG noFields;       /* I, 1 to 1024 */
PVOID fieldListPtr;   /* I, descriptor list, 1 per field */
ULONG fileID;         /* I, 0x03 for standard DBF, 0x8B if memo file also */
} CREATEDATAPACK; /* CDP */
typedef CREATEDATAPACK *PCREATEDATAPACK;

typedef struct _CREATEINDEXPACK {
ULONG func;
ULONG stat;
PSZ   filenamePtr;    /* I, filename to use */
PSZ   keyExpPtr;      /* I, e.g., "SUBSTR(LNAME,1,4)+SSN" */
LONG  xbLink;         /* I, opened data file handle this indexes */
ULONG sortFunction;   /* I, 1-9 system, 10-19 custom */
ULONG codePage;       /* I, 0=use process default */
ULONG countryCode;    /* I, 0=use process default */
PVOID collatePtr;     /* I, NULL=use cc/cp else use passed table for sort */
ULONG nodeSize;       /* I, 512, 1024, or 2048 */
} CREATEINDEXPACK; /* CIP */
typedef CREATEINDEXPACK *PCREATEINDEXPACK;

typedef struct _FIELDDESCTYPE {
BYTE  fieldName[11];  /* IO, upper A-Z and _; 1-10 chars, 0-filled, 0-term */
BYTE  fieldType;      /* IO, C,D,L,N, or M */
LONG  fieldDA;        /* x, offset within record (run-time storage option) */
BYTE  fieldLen;       /* IO, C=1-255,D=8,L=1,N=1-19,M=10 */
BYTE  fieldDC;        /* IO, fieldType=N then 0-15 else 0 */
USHORT altFieldLength;/* IO, 0 */
BYTE  filler[12];     /* I, 0 */
} FIELDDESCTYPE; /* nested in _DESCRIPTORPACK */
typedef FIELDDESCTYPE *PFIELDDESCTYPE;

typedef struct _DESCRIPTORPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle of DBF file */
ULONG fieldNumber;    /* IO, first field is 1 */
ULONG fieldOffset;    /* O, offset of field within record (tag=offset 0) */
FIELDDESCTYPE FD;     /* IO FD.fieldName only, O for rest of FD */
} DESCRIPTORPACK; /* DP */
typedef DESCRIPTORPACK *PDESCRIPTORPACK;

typedef struct _DOSFILEPACK {
ULONG func;
ULONG stat;
PSZ   filenamePtr;    /* I, filename to use */
ULONG handle;         /* IO, handle of open file */
ULONG asMode;         /* I, access-sharing mode */
ULONG bytes;          /* IO, bytes to read, write, length of */
LONG  seekTo;         /* IO, seek to offset, current offset */
ULONG method;         /* I, seek method (0=start of file, 1=current, 2=end) */
PVOID bufferPtr;      /* I, buffer to read into or write from */
ULONG attr;           /* I, attribute to create file with */
PSZ   newNamePtr;     /* I, name to use on rename */
} DOSFILEPACK; /* DFP */
typedef DOSFILEPACK *PDOSFILEPACK;

typedef struct _EXITPACK {
ULONG func;
ULONG stat;
} EXITPACK; /* EP */
typedef EXITPACK *PEXITPACK;

typedef struct _HANDLEPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle of Bullet file */
} HANDLEPACK; /* HP */
typedef HANDLEPACK *PHANDLEPACK;

typedef struct _INITPACK {
ULONG func;
ULONG stat;
ULONG JFTsize;        /* I, max opened files (20-1024+) */
ULONG versionDOS;     /* O, e.g., 230 for 2.30 */
ULONG versionBullet;  /* O, e.g., 2019 for 2.019 */
ULONG versionOS;      /* O, e.g., 4=OS/2 32-bit */
PVOID exitPtr;        /* O, function pointer to EXIT_XB routine */
} INITPACK; /* IP */
typedef INITPACK *PINITPACK;

typedef struct _LOCKPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle of Bullet file to lock */
ULONG xlMode;         /* I, index lock mode (0=exclusive, 1=shared) */
ULONG dlMode;         /* I, data lock mode (0=exclusive, 1=shared) */
LONG  recStart;       /* I, if data, first record # to lock, or 0 for all */
ULONG recCount;       /* I, if data and recStart!=0, # records to lock */
PVOID nextPtr;        /* I, NULL if not xaction, else next LP in list */
} LOCKPACK; /* LP */
typedef LOCKPACK *PLOCKPACK;

typedef struct _MEMODATAPACK {
ULONG func;
ULONG stat;
ULONG dbfHandle;      /* I, handle of DBF file to which this memo file belongs */
ULONG memoBypass;     /* I, memo bypass function to do, if any */
PVOID memoPtr;        /* I, ptr to memo record buffer */
ULONG memoNo;         /* IO, memo record number (aka block number) */
ULONG memoOffset;     /* I, position within record to start read/update */
ULONG memoBytes;      /* IO, number of bytes to read/update */
} MEMODATAPACK; /* MDP */
typedef MEMODATAPACK *PMEMODATAPACK;

typedef struct _MEMORYPACK {
ULONG func;
ULONG stat;
ULONG memory;         /* O, not used in OS/2 */
} MEMORYPACK; /* MP */
typedef MEMORYPACK *PMEMORYPACK;

typedef struct _OPENPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* O, handle of file opened */
PSZ   filenamePtr;    /* I, Bullet file to open */
ULONG asMode;         /* I, access-sharing-cache mode */
LONG  xbLink;         /* I, if index open, xbLink=handle of its opened DBF */
} OPENPACK; /* OP */
typedef OPENPACK *POPENPACK;

typedef struct _QUERYSETPACK {
ULONG func;
ULONG stat;
ULONG item;           /* I, Bullet sysvar item to get/set */
ULONG itemValue;      /* IO, current/new value */
} QUERYSETPACK; /* QSP */
typedef QUERYSETPACK *PQUERYSETPACK;

typedef struct _REMOTEPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle of file, or if 0, use RP.drive */
ULONG drive;          /* I, drive (1=A,2=B,3=C,...0=current) to check */
ULONG isRemote;       /* O, =1 of handle/drive is remote, =0 if local */
ULONG flags;          /* O, 0 */
ULONG isShare;        /* O, 1 */
} REMOTEPACK; /* RP */
typedef REMOTEPACK *PREMOTEPACK;

typedef struct _STATDATAPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle to check */
ULONG fileType;       /* O, bit0=1 data file */
ULONG flags;          /* O, bit0=1 dirty, bit1=1 full-lock, bit2=1 shared */
ULONG progress;       /* O, 0,1-99% pack progress */
PVOID morePtr;        /* O, 0 */
ULONG fields;         /* O, fields per record */
ULONG asMode;         /* O, access-sharing-cache mode */
PSZ   filenamePtr;    /* O, filename used in open */
ULONG fileID;         /* O, first byte of DBF file */
ULONG lastUpdate;     /* O, high word=year,low byte=day, high byte=month */
ULONG records;        /* O, data records (including "deleted") */
ULONG recordLength;   /* O, record length */
ULONG xactionFlag;    /* O, 0 */
ULONG encryptFlag;    /* O, 0 */
PVOID herePtr;        /* O, this file's control address */
ULONG memoHandle;     /* O, handle of open memo file (0 if none) */
ULONG memoBlockSize;  /* O, memo file block size */
ULONG memoFlags;      /* O, bit0=1 dirty */
ULONG memoLastRecord; /* O, last accessed memo record (0 if none) */
ULONG memoLastSize;   /* O, size of last accessed memo record (in bytes, +8) */
ULONG lockCount;      /* O, number of full-locks in force */
} STATDATAPACK; /* SDP */
typedef STATDATAPACK *PSTATDATAPACK;

typedef struct _STATHANDLEPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle to check */
LONG  ID;             /* O, bit0=1 data file, bit0=1 index file */
} STATHANDLEPACK; /* SHP */
typedef STATHANDLEPACK *PSTATHANDLEPACK;

typedef struct _STATINDEXPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle to check */
ULONG fileType;       /* O, bit0=0 index file */
ULONG flags;          /* O, bit0=1 dirty, bit1=1 full-lock, bit2=1 shared */
ULONG progress;       /* O, 0,1-99% reindex progress */
PVOID morePtr;        /* O, 0 */
ULONG xbLink;         /* O, XB file link handle */
ULONG asMode;         /* O, access-sharing-cache mode */
PSZ   filenamePtr;    /* O, pointer to filename used in open */
ULONG fileID;         /* O, "31ch" */
PSZ   keyExpPtr;      /* O, pointer to key expression */
ULONG keys;           /* O, keys in file */
ULONG keyLength;      /* O, key length */
ULONG keyRecNo;       /* O, record number of current key */
PVOID keyPtr;         /* O, ptr to current key value (valid to keyLength) */
PVOID herePtr;        /* O, this file's control address */
ULONG codePage;       /* O, code page at create time */
ULONG countryCode;    /* O, country code at create time */
PVOID CTptr;          /* O, collate table ptr, NULL=no collate table present */
ULONG nodeSize;       /* O, node size */
ULONG sortFunction;   /* O, sort function ID */
ULONG lockCount;      /* O, number of full-locks in force */
} STATINDEXPACK; /* SIP */
typedef STATINDEXPACK *PSTATINDEXPACK;

typedef struct _XERRORPACK {
ULONG func;
ULONG stat;           /* I, error to check */
ULONG errClass;       /* O, class of error */
ULONG action;         /* O, action recommended for error */
ULONG location;       /* O, location of error */
} XERRORPACK; /* XEP */
typedef XERRORPACK *PXERRORPACK;


/* ************************************************************************
 *
 * Error codes
 *
 * ************************************************************************/

#define EXB_NOT_ENOUGH_MEMORY   8  /* cannot get memory requested */
#define EXB_INVALID_DRIVE       15 /* not a valid drive letter */
#define EXB_UNEXPECTED_EOF      38 /* unexpect EOF (bytes read != bytes asked) */
#define EXB_DISK_FULL           39 /* disk full on WriteFile */
#define EXB_FILE_EXISTS         80 /* cannot create file since it already exists */
#define EXB_SEM_OWNER_DIED      105 /* in place of Win32 error 80h (mutex) */
#define EXB_TIMEOUT             640 /* in place of Win32 error 102h (mutex) */

/* Other operating system errors are as returned by OS itself */

/* System/general error codes */

#define EXB_OR_WITH_FAULTS      8192 /* 8192+1 to +4, close-type errors */

                                     /* ERR_216501/6 are for Bullet/x only */
#define EXB_216501              8251 /* INT21/6501h not supported by DOS extender */
                                     /* (do not use default cc/cp) */
#define EXB_216506              8256 /* INT21/6506h not supported by DOS extender */
                                     /* (provide a custom collate table) */

#define EXB_ILLEGAL_CMD         8300 /* function not allowed */
#define EXB_OLD_DOS             8301 /* OS version < MIN_DOS_NEEDED */
#define EXB_NOT_INITIALIZED     8302 /* init not active, must do INIT_XB */
#define EXB_ALREADY_INITIALIZED 8303 /* init already active, must do EXIT_XB */
#define EXB_TOO_MANY_HANDLES    8304 /* more than 1024 opens requested */
#define EXB_SYSTEM_HANDLE       8305 /* Bullet won't use or close handles 0-2 */
#define EXB_FILE_NOT_OPEN       8306 /* file not open (not Bullet handle, including xbLink) */
#define EXB_FILE_IS_DIRTY       8307 /* tried to reload header but current still dirty */
#define EXB_BAD_FILETYPE        8308 /* tried key op on non-key file, data op on non... */
#define EXB_TOO_MANY_PACKS      8309 /* too many INSERT,UPDATE,REINDEX,LOCK_XB packs */
#define EXB_NULL_RECPTR         8310 /* null record pointer passed to Bullet */
#define EXB_NULL_KEYPTR         8311 /* null key pointer passed to Bullet */
#define EXB_NULL_MEMOPTR        8312 /* null memo pointer passed to Bullet */
#define EXB_EXPIRED             8313 /* evaluation time period has expired */
#define EXB_BAD_INDEX           8314 /* Query/SetSysVars index beyond last one */
#define EXB_RO_INDEX            8315 /* SetSysVar index item is read-only */
#define EXB_FILE_BOUNDS         8316 /* file size > 4GB, or > system var sets */
#define EXB_FORCE_ROLLBACK      8397 /* rollback test completed (internal use) */
#define EXB_INVALID_DLL         8398 /* DLL seems to be invalid, 8399 same */

/* Multi-access error codes */

#define EXB_BAD_LOCK_MODE       8401 /* lock mode (LP) not valid */
#define EXB_NOTHING_TO_RELOCK   8402 /* cannot relock without existing full-lock */
#define EXB_SHARED_LOCK_ON      8403 /* write access needed but lock is shared (flush on backup) */

/* Index error codes */

#define EXB_KEY_NOT_FOUND       8501 /* exact match of key not found */
#define EXB_KEY_EXISTS          8502 /* key exists already and dups not allowed */
#define EXB_END_OF_FILE         8503 /* already at last index order */
#define EXB_TOP_OF_FILE         8504 /* already at first index order */
#define EXB_EMPTY_FILE          8505 /* nothing to do since no keys */
#define EXB_CANNOT_GET_LAST     8506 /* cannot locate last key */
#define EXB_BAD_INDEX_STACK     8507 /* index file is corrupt */
#define EXB_BAD_INDEX_READ0     8508 /* index file is corrupt */
#define EXB_BAD_INDEX_WRITE0    8509 /* index file is corrupt */

#define EXB_OLD_INDEX           8521 /* old index, run through ReindexOld to update */
#define EXB_UNKNOWN_INDEX       8522 /* not a Bullet index file */
#define EXB_KEY_TOO_LONG        8523 /* keylength > 62 (or 64 if unique), or is 0 */

#define EXB_PARSER_NULL         8531 /* parser function pointer is NULL */
#define EXB_BUILDER_NULL        8532 /* build key function pointer is NULL */
#define EXB_BAD_SORT_FUNC       8533 /* CIP.sortFunction not valid */
#define EXB_BAD_NODE_SIZE       8534 /* CIP.nodeSize is not 512, 1024, or 2048 */
#define EXB_FILENAME_TOO_LONG   8535 /* CIP.filenamePtr->pathname > max path length */

#define EXB_KEYX_NULL           8541 /* expression is effectively NULL */
#define EXB_KEYX_TOO_LONG       8542 /* CIP.keyExpPtr->expression > 159 */
#define EXB_KEYX_SYM_TOO_LONG   8543 /* fieldname/funcname in expression > 10 chars */
#define EXB_KEYX_SYM_UNKNOWN    8544 /* fieldname/funcname in expression unknown */
#define EXB_KEYX_TOO_MANY_SYMS  8545 /* too many symbols/fields used in expression */
#define EXB_KEYX_BAD_SUBSTR     8546 /* invalid SUBSTR() operand in expression */
#define EXB_KEYX_BAD_SUBSTR_SZ  8547 /* SUBSTR() exceeds field's size */
#define EXB_KEYX_BAD_FORM       8548 /* didn't match expected symbol in expression */

#define EXB_NO_READS_FOR_RUN    8551 /* unlikely, use different reindex buffer size */
#define EXB_TOO_MANY_RUNS       8552 /* unlikely, too many runs (64K or more runs) */
#define EXB_TOO_MANY_RUNS_FOR_BUFFER 8553 /* unlikely, too many runs for run buffer */
#define EXB_TOO_MANY_DUPLICATES 8554 /* more than 64K "identical" keys */

#define EXB_INSERT_RECNO_BAD    8561 /* AP.recNo cannot be > 0 if inserting */
#define EXB_PREV_APPEND_EMPTY   8562 /* no prev append for insert yet AP.recNo==80000000h */
#define EXB_PREV_APPEND_MISMATCH 8563 /* prev append's xbLink does not match this */
#define EXB_INSERT_KBO_FAILED   8564 /* could not back out key at INSERT_XB */
#define EXB_INSERT_DBO_FAILED   8565 /* could not back out data records at INSERT_XB */

#define WRN_NOTHING_TO_UPDATE   8571 /* all AP.recNo=0 at UPDATE_XB */
#define EXB_INTERNAL_UPDATE     8572 /* internal error UPDATE_XB, not in hdl/rec# list */

#define EXB_FAILED_DATA_RESTORE 8573 /* could not restore original data record (*) */
#define EXB_FAILED_KEY_DELETE   8574 /* could not remove new key (*) */
#define EXB_FAILED_KEY_RESTORE  8575 /* could not restore original key(*) */
/* *original error, which forced a back-out, has been replaced by this error */
/* this error is always returned in the first AP.stat (-1 on data, 1 on index) */

/* Data error codes */

#define EXB_EXT_XBLINK          8601 /* xbLink handle is not internal DBF (is -1) */
#define EXB_FIELDNAME_TOO_LONG  8602 /* fieldname is > 10 characters */
#define EXB_RECORD_TOO_LONG     8603 /* record length is > 64K */
#define EXB_FIELD_NOT_FOUND     8604 /* fieldname not found in descriptor info */
#define EXB_BAD_FIELD_COUNT     8605 /* fields <= 0 or >= MAX_FIELDS (Init,Open) */
                                     /* and also GetDescriptor by field number */
#define EXB_BAD_HEADER          8606 /* bad header (reclen=0, etc., from LocateTo, Flush) */
#define EXB_BUFFER_TOO_SMALL    8607 /* buffer too small (pack buffer < reclen in pack) */
#define EXB_INTERNAL_PACK       8608 /* internal error in PackRecords */
#define EXB_BAD_RECNO           8609 /* record number=0 or > records in data file hdr */
                                     /* or Pack on empty data file */
#define WRN_RECORD_TAGGED       8610 /* record's tag field matches skip tag */

/* Memo error codes */

#define WRN_CANNOT_OPEN_MEMO    8701 /* DBF says memo file but memo open fails */
#define EXB_MEMO_NOT_OPEN       8702 /* no open memo file for operation */
#define EXB_BAD_BLOCKSIZE       8703 /* memo blocksize must be at least 24 bytes */
#define EXB_MEMO_DELETED        8704 /* memo is deleted */
#define EXB_MEMO_PAST_END       8705 /* memo data requested is past end of record */
#define EXB_BAD_MEMONO          8706 /* memo number is not valid */
#define EXB_MEMO_IN_USE         8707 /* memo add encountered likely corrupt memo file */
#define EXB_BAD_AVAIL_LINK      8708 /* memo avail link cannot be valid (is 0) */
#define EXB_MEMO_ZERO_SIZE      8709 /* memo data has no size */
#define EXB_MEMO_IS_SMALLER     8710 /* memo attempt to shrink but already <= size */

#endif /* ifndef __BULLET_H */


ΓòÉΓòÉΓòÉ 9. Bullet Functions ΓòÉΓòÉΓòÉ

System level                    Advanced system-level

 INIT_XB                        QUERY_SYSVARS_XB
 EXIT_XB                        SET_SYSVARS_XB
 MEMORY_XB                      SET_DVMON_XB
 BACKUP_FILE_XB                 QUERY_VECTORS_XB
 STAT_HANDLE_XB                 SET_VECTORS_XB
 GET_ERROR_CLASS_XB

 Data low-level                 Data mid-level                     Memo
                                                                   mid-level

 CREATE_DATA_XB                 GET_DESCRIPTOR_XB
                                                                   GET_MEMO_SIZE_XB
 OPEN_DATA_XB                   GET_RECORD_XB                      GET_MEMO_XB
 CLOSE_DATA_XB                  ADD_RECORD_XB                      ADD_MEMO_XB
 STAT_DATA_XB                   UPDATE_RECORD_XB
                                                                   UPDATE_MEMO_XB
 READ_DATA_HEADER_XB            DELETE_RECORD_XB
                                                                   DELETE_MEMO_XB
 FLUSH_DATA_HEADER_XB           UNDELETE_RECORD_XB
                                                                   MEMO_BYPASS_XB
 COPY_DATA_HEADER_XB            DEBUMP_RECORD_XB
 ZAP_DATA_HEADER_XB             PACK_RECORDS_XB

 Index low-level                Index mid-level                    Index
                                                                   high-level

 CREATE_INDEX_XB                FIRST_KEY_XB                       GET_FIRST_XB
 OPEN_INDEX_XB                  EQUAL_KEY_XB                       GET_EQUAL_XB
 CLOSE_INDEX                    EQUAL_OR_GREATER_KEY_XB
                                                                   GET_EQUAL_OR_GREATER_XB
 STAT_INDEX_XB                  EQUAL_OR_LESSER_KEY_XB
                                                                   GET_EQUAL_OR_LESSER_XB
 READ_INDEX_HEADER_XB           NEXT_KEY_XB                        GET_NEXT_XB
 FLUSH_INDEX_HEADER_XB          PREV_KEY_XB                        GET_PREV_XB
 COPY_INDEX_HEADER_XB           LAST_KEY_XB                        GET_LAST_XB
 ZAP_INDEX_HEADER_XB            STORE_KEY_XB                       INSERT_XB
                                DELETE_KEY_XB                      UPDATE_XB
                                BUILD_KEY_XB                       REINDEX_XB
                                GET_CURRENT_KEY_XB
                                GET_KEY_FOR_RECORD_XB

 Network level                  CP level

 LOCK_XB                        DELETE_FILE_DOS
 UNLOCK_XB                      RENAME_FILE_DOS
 LOCK_INDEX_XB                  CREATE_FILE_DOS
 UNLOCK_INDEX_XB                OPEN_FILE_DOS
 LOCK_DATA_XB                   SEEK_FILE_DOS
 UNLOCK_DATA_XB                 READ_FILE_DOS
 CHECK_REMOTE_XB                WRITE_FILE_DOS
 RELOCK_XB                      CLOSE_FILE_DOS
 RELOCK_INDEX_XB                ACCESS_FILE_DOS
 RELOCK_DATA_XB                 EXPAND_FILE_DOS
                                MAKE_DIR_DOS
                                COMMIT_FILE_DOS


ΓòÉΓòÉΓòÉ 9.1. INIT_XB ΓòÉΓòÉΓòÉ

Pack: INITPACK                 Source Example

     IN               OUT
   IP.func          IP.stat
   IP.JFTsize       IP.versionDOS
                    IP.versionBullet
                    IP.versionOS
                    IP.exitPtr

 This must be the first routine called (except for SET_VECTORS_XB).  If it has
 already been called the system variables are restored to their defaults, and
 an error is returned.  Otherwise, the entire Bullet system is initialized, and
 EXIT_XB is registered with the OS ExitList handler (DosExitList).

 For more than the default open files (generally 20), set IP.JFTsize to the
 total number of concurrently open files you need.  Depending on your version,
 Bullet manages up to 1024 Bullet files per process (total data and index; memo
 files are not counted against this total).  Setting this less than 20 does
 nothing.  This number is for Bullet files, your files, pipes -- anything using
 a handle.  If you need to account for handles that you are managing, you
 should add those to IP.JFTsize.  For example, if you need 10 data files, each
 with a memo file, and 2 index files per data file, that is 40 total Bullet
 files.  If you need to use 15 other handles, for whatever use, add that number
 to the 40 Bullet files, for a total setting of 55.  The OS also uses 3 handles
 for itself, so, for all these, IP.JFTsize=58 would be the minimum.  You can
 set it higher, but unused handles are wasted handles. In addition, if the
 current process has fewer total handles available than the number you
 specified in IP.JFTsize, Bullet sets the total available handles to IP.JFTsize
 (as the absolute number of handles required).  If the current process already
 has more total handles than IP.JFTsize, no action is taken.

 On return (where no error occurred), the operating system version is in
 IP.versionDOS (*100) and the Bullet version (*1000) in IP.versionBullet.
 IP.versionOS return is based on the following table:

      Bullet Platform   IP.versionOS
         MS-DOS 16-bit      0
          Win3x 16-bit      1
         DOSX32 32-bit      3
           OS/2 32-bit      4
          Win32 32-bit      5

 IP.exitPtr returns with the function pointer to the EXIT_XB routine. This
 function pointer is redundant unless specifically mentioned as being required
 for your platform.  It is not needed in OS/2, but is useful for other
 platforms, for example, for use with atexit(), part of C's standard library.

 Win32s, as run in Windows 3.1, returns IP.versionDOS equal to its level.  For
 example, if running Win31, using win32s version 1.25, IP.versionDOS returns as
 125, not 310 as might be expected.

 Note:  References under OUT using *AP.keyPtr or similar (note then *) are used
 throughout this manual and indicate that Bullet updates the contents at
 AP.keyPtr with data (i.e., Bullet filled the buffer).


ΓòÉΓòÉΓòÉ 9.2. EXIT_XB ΓòÉΓòÉΓòÉ

Pack: EXITPACK                 Source Example

     IN               OUT
   EP.func          EP.stat

 Call EXIT_XB before ending your program to release any remaining resources
 back to the OS.  Open files should be closed by using CLOSE_DATA_XB and
 CLOSE_INDEX_XB. EXIT_XB closes any Bullet files that are still open.

 This routine is registered with the operating system and so is called
 automatically when your program terminates in OS/2, or with the C startup code
 if atexit() is used.


ΓòÉΓòÉΓòÉ 9.3. ATEXIT_XB ΓòÉΓòÉΓòÉ

This routine is obsolete.  In OS/2, the EXIT_XB shutdown procedure is
registered with the operating system.  For those systems that do not offer this
feature in the OS, the compiler run-time routine atexit() is used immediately
after calling INIT_XB, using IP.exitPtr as the function pointer for atexit().


ΓòÉΓòÉΓòÉ 9.4. MEMORY_XB ΓòÉΓòÉΓòÉ

Pack: MEMORYPACK               Source Example

     IN               OUT
   MP.func          MP.stat
                    MP.memory

 MP.memory returns with the number of bytes in the private memory arena
 allocatable by the process according to DosQuerySysInfo for QSV_MAXPRMEM, or
 as provided by the OS.  It is for informational use only.

 Note:  This routine is not mutex-protected.


ΓòÉΓòÉΓòÉ 9.5. BACKUP_FILE_XB ΓòÉΓòÉΓòÉ

Pack: COPYPACK                 Source Example

     IN               OUT
   CP.func          CP.stat
   CP.handle
   CP.filenamePtr

 Copy an open BULLET index file or data/memo files.  BULLET repacks and
 reindexes files in place, requiring less disk space to perform the function.
 This routine allows a file to be safely copied for a possible later restore.

 This function is recommended prior to packing a data file with
 PACK_RECORDS_XB.  For index files, COPY_INDEX_HEADER_XB is sufficient since
 index files are easily recreated so long as you have the data file along with
 the index file header.

 A full-lock should be in force before copying.  A shared lock may be used.

 If CP.handle belongs to a DBF data file, and if a memo file is attached, the
 memo file backup name is as CP.filenamePtr, the backup DBF pathname, but the
 extension is always set to "._BT".  For example, if CP.handle is for a DBF
 that has a DBT memo file attached, then the current state of the DBF file is
 copied to CP.filenamePtr, say, "\CURRBACK\ACCT.DBF", and, in this case, the
 DBT memo file is copied to "\CURRBACK\ACCT._BT".  The name of the original
 DBF/DBT does not matter.  If MEMO_EXTENSION of SET_SYSVARS_XB has changed the
 default, then that extension is used on the memo copy, with '_' replacing its
 first character.

 To prevent the backing up of a DBT memo file when backing up a DBF data file,
 set CP.handle = -CP.handle (i.e., negative CP.handle).  This way, the DBT memo
 file is not copied.  To backup only a DBT file, close the DBF and copy the DBT
 by some other means.


ΓòÉΓòÉΓòÉ 9.6. STAT_HANDLE_XB ΓòÉΓòÉΓòÉ

Pack: STATHANDLEPACK           Source Example

     IN               OUT
   SHP.func         SHP.stat
   SHP.handle       SHP.ID

 Get information on a file handle number to determine if it is a BULLET file,
 and if so, its type:  index or data.

      SHP.ID   File type
         0     index, IX3  use STAT_INDEX_XB for file stats
         1     data, DBF   use STAT_DATA_XB for file stats
        -1     unknown

 Only bit0 of SHP.ID is significant if not -1.  So, if bit0=0 then the handle
 belongs to an index file.  If bit0=1 then it's a data file.

 Memo file handles return as unknown.  A DBF file's memo file handle is stored
 in the DBF file's data area, and is returned by STAT_DATA_XB in
 SDP.memoHandle.

 Note:  This routine is not mutex-protected.


ΓòÉΓòÉΓòÉ 9.7. GET_ERROR_CLASS_XB ΓòÉΓòÉΓòÉ

Pack: XERRORPACK               Source Example

     IN               OUT
   XEP.func         XEP.errClass
   XEP.stat         XEP.action
                    XEP.location

 Get the extended error information for the code passed in XEP.stat. This
 information includes the error classification, recommended action, and origin
 of the error.

 Any system error code can be specified, not necessarily the one that last
 occurred.  If a return code is not a BULLET code, then it is a system error
 code (from the CP, DosXXX routines).

 The ERRCLASS, ERRACT, and ERRLOC items below are OS/2 values, names and
 descriptions for DosErrClass().

  Error Classification

    Value   Name                Description
      1    ERRCLASS_OUTRES     Out of resources
      2    ERRCLASS_TEMPSIT    Temporary situation
      3    ERRCLASS_AUTH       Authorization failed
      4    ERRCLASS_INTRN      Internal error
      5    ERRCLASS_HRDFAIL    Device hardware failure
      6    ERRCLASS_SYSFAIL    System failure
      7    ERRCLASS_APPEAR     Probably application error
      8    ERRCLASS_NOTFND     Item not located
      9    ERRCLASS_BADFMT     Bad format for function or data
     10    ERRCLASS_LOCKED     Resource or data locked
     11    ERRCLASS_MEDIA      Incorrect media, CRC error
     12    ERRCLASS_ALREADY    Action already taken or done, or resource already exists
     13    ERRCLASS_UNK        Unclassified
     14    ERRCLASS_CANT       Cannot perform requested action
     15    ERRCLASS_TIME       Timeout


  Recommended Action

    Value   Name                Description
      1    ERRACT_RETRY        Retry immediately
      2    ERRACT_DLYRET       Delay and retry
      3    ERRACT_USER         Bad user input - get new values
      4    ERRACT_ABORT        Terminate in an orderly manner
      5    ERRACT_PANIC        Terminate immediately
      6    ERRACT_IGNORE       Ignore error
      7    ERRACT_INTRET       Retry after user intervention


  Origin

    Value   Name                Description
      1    ERRLOC_UNK          Unknown
      2    ERRLOC_DISK         Disk
      3    ERRLOC_NET          Network
      4    ERRLOC_SERDEV       Serial device
      5    ERRLOC_MEM          Memory

 Note:  This routine is not mutex-protected.


ΓòÉΓòÉΓòÉ 9.8. QUERY_SYSVARS_XB ΓòÉΓòÉΓòÉ

Pack: QUERYSETPACK             Source Example

     IN               OUT
   QSP.func         QSP.stat
   QSP.item         QSP.itemValue

 Query a BULLET system variable.

 To get the function pointers to the sort-compare functions, use:

   QSP.item           FuncPtr To
   ASCII_SORT        ASCII sort compare
   NLS_SORT          NLS sort compare
   S16_SORT          16-bit signed integer
   U16_SORT          16-bit unsigned integer
   S32_SORT          32-bit signed integer
   U32_SORT          32-bit unsigned integer

 All intrinsic sort compares (1-6) point to the same function. They cannot be
 called except by BULLET itself.  The integer compare routines are based on
 Intel byte order.  For Motorola byte order, ASCII sort can be used for
 all-positive numbers, otherwise a custom sort-compare should be used.

    QSP.item          FuncPtr To
     10-19           Custom sort-compare functions

 Before creating or opening an index file with a custom sort-compare function
 (which is specified during CREATE_INDEX_XB), that function's address must
 first be sent to BULLET using SET_SYSVARS_XB. Thereafter, that function must
 be available whenever that index file is accessed.  See Custom Sort-Compare
 Function for creating custom sort-compare functions.

 To get the function pointers to the build key and expression parser routines,
 use:

    QSP.item          FuncPtr To
   BUILD_KEY_FUNC    Build key routine
   PARSER_FUNC       Key expression parser routine

 Before creating or opening an index file with a custom build key or expression
 parser routine (which is specified at any time, but must be used in a
 consistent manner), that routine's address must first be sent to BULLET using
 SET_SYSVARS_XB.  Thereafter, that routine should be available since it may be
 required again.  See Custom Build-Key Routine for creating a custom build-key
 routine and Custom Expression Parser Routine for creating a custom key
 expression parser.

 To get the BULLET system variables' values, use:

    QSP.item             Value To
   26 (read-only)        low-word: number of xb$Malloc(), high-word: number of xb$Free()
   27 (read-only)        Max instances (2, 32, 999)
   28 (read-only)        Max files (100, 250, 1024)
   29 (read-only)        Handle of Bullet's mutex semaphore

   LOCK_TIMEOUT          Lock file bytes timeout, in milliseconds (default=0)
   MUTEX_SEM_TIMEOUT     Mutex semaphore request timeout, in milliseconds (default=0)
   PACK_BUFFER_SIZE      Pack buffer size, in bytes (default=0: autosize)
   REINDEX_BUFFER_SIZE   Reindex buffer size, in bytes (default=0: autosize)
   REINDEX_PACK_PCT      Reindex node pack percentage, 50-100% (default=100)
   TMP_PATH_PTR          Temporary file path pointer (default=NULL, where TMP= used, then .\)
   REINDEX_SKIP_TAG      Reindex tag field character to skip (default=0, no skip)
   COMMIT_AT_EACH        Commit each individual file during INSERT/UPDATE_XB (default=0, defer until flush)
   MEMO_BLOCKSIZE        Memo file block size (default=512 bytes; minimum is 24 bytes)
   MEMO_EXTENSION        Memo file extension (default is "DBT\0")
   MAX_DATAFILE_SIZE     Max data file size-1 (default=2047MB, absolute max is 4095MB)
   MAX_INDEXFILE_SIZE    Max index file size-1 (default=2047MB, absolute max is 4095MB)
   ATOMIC_MODE           Atomic mode (bit0=1 then atomic Next and Prev access, default=0)
   CALLBACK_PTR          Callback routine at reindex/pack (def=0, none)

 The timeout values determine if the kernel should wait for a pre-determined
 time before returning an error if the resource cannot be obtained.  The lock
 timeout specifies how long to wait for a lock to be obtained in case some
 other process has a lock on the same resource.  The mutex timeout specifies
 how long to wait for access to BULLET in case some other thread in this
 process is in BULLET.  Multiple processes can access BULLET at the same time,
 but only one thread in each process can be inside BULLET at any one time.

 The buffer sizes, when 0, default to a minimum reasonable size.  Performance
 is acceptable at these sizes.  For best performance, provide as much real
 memory as possible, up to 512KB.  Larger buffers can be used.

 The reindex node pack percentage determines how many keys are packed on a
 node.  100% forces as many keys as possible, minus 1.

 If the temporary file path pointer is NULL (the default), then the TMP=
 environment variable is used to locate any temporary files created by BULLET,
 or if that is not found, then the current directory is used.  The pointer
 supplied, if any, should be to a string containing an existing path (drive
 should be included; a trailing '\' is optional, but recommended).  See
 REINDEX_XB for size requirements.

 The reindex skip tag character, if encountered in the DBF record's tag field
 (the first byte of each record), causes the reindex routine to not place that
 record's key value into the index file.  Also, BUILD_KEY_XB returns a warning
 if the record supplied has a matching tag character. To disable skip tag
 processing, set it to 0.

 Inserts and Updates, by default, do not commit each file when that pack is
 processed.  Instead, it is left to the programmer to issue a FLUSH_XB to
 commit.  To force a commit after each pack file is processed, set CommitAtEach
 to 1.  This is not one single commit, but a commit for each file in the pack,
 after that file has been processed, but before the next file in the pack is.
 This will not prevent a roll-back should it be needed.

 A memo file can have at most 589,822 blocks.  At the default 512 bytes per
 block, that equates to about 288MB.  If you need more memo space, increase the
 block size.  The memo extension default is "DBT\0".  Generally, it's a good
 idea to leave it at this.

 The maximum file sizes are enforced when adding to or reading from DBF files,
 and when inserting into or reading from index files.  The default is 2047 MB
 (0x7FEFFFFF).  If your file system permits 4GB files, set the values to 4095
 MB (0xFFEFFFFF).

 The Atomic mode flag determines how key access is handled.  When bit0=0, the
 default, the key routines, NEXT_KEY_XB, PREV_KEY_XB, and GET_NEXT_XB,
 GET_PREV_XB, use the internal position of the last gotten key as their
 starting point.  In multi-threaded code, it's possible that another thread has
 since accessed the same file handle and altered the last gotten key.  By
 setting bit0=1, key access (next or previous) can now specify a starting point
 (typically already set up in AP.keyPtr), rather than starting at the last
 accessed key for that handle (which may have been changed by another thread).

 The Callback routine receives a pointer to a CALLBACK structure on the stack.
 The callback routine may clean the stack (e.g., a _syscall routine).  It may
 also leave it for the caller to clean (e.g., a _stdcall or __cdecl routine).
 See bd_rix.c on the distribution disk for an example.  When the CALLBACK_PTR
 == NULL (default), no callback is made.  Currently, the callback is made
 during REINDEX_XB and PACK_RECORDS calls, at a rate dependent on the
 *_BUFFER_SIZE.

 Note:  This routine is not mutex-protected.


ΓòÉΓòÉΓòÉ 9.9. SET_SYSVARS_XB ΓòÉΓòÉΓòÉ

Pack: QUERYSETPACK             Source Example

     IN               OUT
   QSP.func         QSP.stat
   QSP.item         QSP.itemValue
   QSP.itemValue

 Set a BULLET system variable, returning the previous value.

 To use, set QSP.item to the item to set, and QSP.itemValue with the value to
 use (function's address, variable's timeout value, etc., whatever the case may
 be).  On return, QSP.itemValue is the previous value that QSP.item was set to.

   QSP.item           FuncPtr To
   ASCII_SORT        ASCII sort compare
   NLS_SORT          NLS sort compare
   S16_SORT          16-bit signed integer
   U16_SORT          16-bit unsigned integer
   S32_SORT          32-bit signed integer
   U32_SORT          32-bit unsigned integer

 All intrinsic sort compares (1-6) point to the same function. They cannot be
 called except by BULLET itself.  They should not be overloaded with custom
 functions.  If you have a custom sort-compare, use one of the custom slots.
 The integer compare routines are based on Intel byte order.  For Motorola byte
 order, ASCII sort can be used for all-positive numbers, otherwise a custom
 sort-compare should be used.

    QSP.item          FuncPtr To
     10-19           Custom sort-compare functions

 Before creating or opening an index file with a custom sort-compare function
 (which is specified during CREATE_INDEX_XB), that function's address must
 first be sent to BULLET using this routine. Thereafter, that function must be
 available whenever that index file is accessed.  See Custom Sort-Compare
 Function for creating custom sort-compare functions.

 To set the function pointers to the build key and expression parser routines,
 use:

    QSP.item          FuncPtr To
   BUILD_KEY_FUNC    Build key routine
   PARSER_FUNC       Key expression parser routine

 Before creating or opening an index file with a custom build key or expression
 parser routine (which is specified at any time, but must be used in a
 consistent manner), that routine's address must first be sent to BULLET using
 this routine.  Thereafter, that routine should always be ready (in a callable
 state) since it may be required again.  See Custom Build-Key Routine for
 creating a custom build-key routine and Custom Expression Parser Routine for
 creating a custom key expression parser.

 To set the BULLET system variables' values, use:

    QSP.item             Value To
   LOCK_TIMEOUT          Lock file bytes timeout, in milliseconds (default=0)
   MUTEX_SEM_TIMEOUT     Mutex semaphore request timeout, in milliseconds (default=0)
   PACK_BUFFER_SIZE      Pack buffer size, in bytes (default=0: autosize)
   REINDEX_BUFFER_SIZE   Reindex buffer size, in bytes (default=0: autosize)
   REINDEX_PACK_PCT      Reindex node pack percentage, 50-100% (default=100)
   TMP_PATH_PTR          Temporary file path pointer (default=NULL, where TMP= used, then .\)
   REINDEX_SKIP_TAG      Reindex tag field character to skip (default=0, no skip)
   COMMIT_AT_EACH        Commit each individual file during INSERT/UPDATE_XB (default=0, defer until flush)
   MEMO_BLOCKSIZE        Memo file block size (default=512 bytes; minimum is 24 bytes)
   MEMO_EXTENSION        Memo file extension (default is "DBT\0")
   MAX_DATAFILE_SIZE     Max data file size-1 (default=2047MB, absolute max is 4095MB)
   MAX_INDEXFILE_SIZE    Max index file size-1 (default=2047MB, absolute max is 4095MB)
   ATOMIC_MODE           Atomic mode (bit0=1 then atomic Next and Prev access, default=0)
   CALLBACK_PTR          Callback routine at reindex/pack (def=0, none)

 The timeout values determine if the kernel should wait for a pre-determined
 time before returning an error if the resource cannot be obtained.  The lock
 timeout specifies how long to wait for a lock to be obtained in case some
 other process has a lock on the same resource.  The mutex timeout specifies
 how long to wait for access to BULLET in case some other thread in this
 process is in BULLET.  Multiple processes can access BULLET at the same time,
 but only one thread in each process can be inside BULLET at any one time.

 The buffer sizes, when 0, default to a minimum reasonable size.  Performance
 is acceptable at these sizes.  For best performance, provide as much real
 memory as possible, up to 512KB.  Larger buffers can be used.

 The reindex node pack percentage determines how many keys are packed on a
 node.  100% forces as many keys as possible, minus 1.

 If the temporary file path pointer is NULL (the default), then the TMP=
 environment variable is used to locate any temporary files created by BULLET,
 or if that is not found, then the current directory is used.  The pointer
 supplied, if any, should be to a string containing an existing path (drive
 should be included; a trailing '\' is optional, but recommended).  See
 REINDEX_XB for size requirements.

 The reindex skip tag character, if encountered in the DBF record's tag field
 (the first byte of each record), causes the reindex routine to not place that
 record's key value into the index file.  Also, BUILD_KEY_XB returns a warning
 if the record supplied has a matching tag character. To disable skip tag
 processing, set it to 0.

 Inserts and Updates, by default, do not commit each file when that pack is
 processed.  Instead, it is left to the programmer to issue a FLUSH_XB to
 commit.  To force a commit after each pack file is processed, set CommitAtEach
 to 1.  This is not one single commit, but a commit for each file in the pack,
 after that file has been processed, but before the next file in the pack is.
 This will not prevent a roll-back should it be needed.

 A memo file can have at most 589,822 blocks.  At the default 512 bytes per
 block, that equates to about 288MB.  If you need more memo space, increase the
 block size.  The memo extension default is "DBT\0".  Generally, it's a good
 idea to leave it at this.

 The maximum file sizes are enforced when adding to or reading from DBF files,
 and when inserting into or reading from index files.  The default is 2047 MB
 (0x7FEFFFFF).  If your file system permits 4GB files, set the values to 4095
 MB (0xFFEFFFFF).

 The Atomic mode flag determines how key access is handled.  When bit0=0, the
 default, the key routines, NEXT_KEY_XB, PREV_KEY_XB, and GET_NEXT_XB,
 GET_PREV_XB, use the internal position of the last gotten key as their
 starting point.  In multi-threaded code, it's possible that another thread has
 since accessed the same file handle and altered the last gotten key.  By
 setting bit0=1, key access (next or previous) can now specify a starting point
 (typically already set up in AP.keyPtr), rather than starting at the last
 accessed key for that handle (which may have been changed by another thread).

 The Callback routine receives a pointer to a CALLBACK structure on the stack.
 The callback routine may clean the stack (e.g., a _syscall routine).  It may
 also leave it for the caller to clean (e.g., a _stdcall or __cdecl routine).
 See bd_rix.c on the distribution disk for an example.  When the CALLBACK_PTR
 == NULL (default), no callback is made.  Currently, the callback is made
 during REINDEX_XB and PACK_RECORDS calls, at a rate dependent on the
 *_BUFFER_SIZE.

 Note:  Issuing INIT_XB restores all system variables (those setable via this
 routine) and function pointers, but not vectors, to their default values.
 This is done even if INIT_XB returns an error that BULLET has already been
 initialized.


ΓòÉΓòÉΓòÉ 9.10. SET_DVMON_XB ΓòÉΓòÉΓòÉ

This routine is not currently used.


ΓòÉΓòÉΓòÉ 9.11. QUERY_VECTORS_XB ΓòÉΓòÉΓòÉ

Pack: QUERYSETPACK             Source Example

     IN               OUT
   QSP.func         QSP.stat
   QSP.item         QSP.itemValue

 Query a BULLET OS API vector.

 To get the vectors that Bullet makes to access OS API calls, set QSP.item to
 the desired VECTOR_* constant below.  On return, QSP.itemValue has the
 function pointer for that item.

       QSP.item
   VECTOR_CLOSE_FILE
   VECTOR_CREATE_DIR
   VECTOR_CREATE_FILE
   VECTOR_CREATE_UNIQUE_FILE
   VECTOR_DELETE_FILE
   VECTOR_LENGTH_FILE
   VECTOR_MOVE_FILE
   VECTOR_OPEN_FILE
   VECTOR_READ_FILE
   VECTOR_SEEK_FILE
   VECTOR_UPDATE_DIR_ENTRY
   VECTOR_WRITE_FILE
   VECTOR_LOCK_FILE
   VECTOR_IS_DRIVE_REMOTE
   VECTOR_IS_FILE_REMOTE
   VECTOR_EXITLIST
   VECTOR_REMOVE_EXITLIST
   VECTOR_FREE
   VECTOR_GET_SORT_TABLE
   VECTOR_GET_COUNTRY_INFO
   VECTOR_GET_ERROR_CLASS
   VECTOR_GET_MEMORY
   VECTOR_GET_TMP_DIR
   VECTOR_GET_VERSION
   VECTOR_MALLOC
   VECTOR_SET_HANDLE_COUNT
   VECTOR_GET_TIME_INFO
   VECTOR_UPPERCASE
   VECTOR_CLOSE_MUTEX_SEM
   VECTOR_CREATE_MUTEX_SEM
   VECTOR_RELEASE_MUTEX_SEM
   VECTOR_REQUEST_MUTEX_SEM

 If QSP.itemValue is returned NULL, the default Bullet OS API call is being
 used.

 Note:  This routine is not mutex-protected.


ΓòÉΓòÉΓòÉ 9.12. SET_VECTORS_XB ΓòÉΓòÉΓòÉ

Pack: QUERYSETPACK             Source Example

     IN               OUT
   QSP.func         QSP.stat
   QSP.item         QSP.itemValue
   QSP.itemValue

 Set a BULLET OS API vector.

 To set the vectors that Bullet makes to access OS API calls, set QSP.item to
 the desired VECTOR_* constant below, and QSP.itemValue to its replacement's
 address.  On return, QSP.itemValue has the previous function pointer for that
 item.

       QSP.item
   VECTOR_CLOSE_FILE
   VECTOR_CREATE_DIR
   VECTOR_CREATE_FILE
   VECTOR_CREATE_UNIQUE_FILE
   VECTOR_DELETE_FILE
   VECTOR_LENGTH_FILE
   VECTOR_MOVE_FILE
   VECTOR_OPEN_FILE
   VECTOR_READ_FILE
   VECTOR_SEEK_FILE
   VECTOR_UPDATE_DIR_ENTRY
   VECTOR_WRITE_FILE
   VECTOR_LOCK_FILE
   VECTOR_IS_DRIVE_REMOTE
   VECTOR_IS_FILE_REMOTE
   VECTOR_EXITLIST
   VECTOR_REMOVE_EXITLIST
   VECTOR_FREE
   VECTOR_GET_SORT_TABLE
   VECTOR_GET_COUNTRY_INFO
   VECTOR_GET_ERROR_CLASS
   VECTOR_GET_MEMORY
   VECTOR_GET_TMP_DIR
   VECTOR_GET_VERSION
   VECTOR_MALLOC
   VECTOR_SET_HANDLE_COUNT
   VECTOR_GET_TIME_INFO
   VECTOR_UPPERCASE
   VECTOR_CLOSE_MUTEX_SEM
   VECTOR_CREATE_MUTEX_SEM
   VECTOR_RELEASE_MUTEX_SEM
   VECTOR_REQUEST_MUTEX_SEM

 Example replacement routines are in ccdosfn.c.  Many of the routines in
 ccdosfn.c are standard C, but some are OS-specific and must be updated for the
 OS being used.

 If QSP.itemValue is set to NULL, the default Bullet OS API call is used.

 On return from a successful call, QSP.itemValue is the value of the previous
 function pointer for that item, which may be returned NULL, indicating that
 the default Bullet OS API call was being used.


ΓòÉΓòÉΓòÉ 9.13. CREATE_DATA_XB ΓòÉΓòÉΓòÉ

Pack: CREATEDATAPACK           Source Example

     IN               OUT
   CDP.func         CDP.stat
   CDP.filenamePtr
   CDP.noFields
   CDP.fieldListPtr
   CDP.fileID

 Create a new BULLET DBF data file with the name at CDP.filenamePtr, and an
 optional DBT memo file.

 Before using this routine, allocate an array of field descriptors of type
 FIELDDESCTYPE, one for each field in the record (number of fields as set in
 CDP.noFields). It is recommended  that this allocation be zeroed before use
 since fieldnames and reserved entries must be 0-filled:

     FIELDDESCTYPE fieldList[12];  // 12 fields used in data record
       :
     memset(fieldList,0,sizeof(fieldList)); // init unused bytes to 0 (required)

 Filename

 The drive and path must exist if used as part of the filename. Long filenames
 may be used if supported by the file system in use. As with all text strings
 used by Bullet, the filename must end in a '\0'.

 Number of Fields

 The number of descriptors in the array, described next.  Each field has a
 descriptor.  The tag field is not a formal field, and so has no descriptor,
 and is not counted in the number of fields.  The maximum number of fields is
 254 according to the DBF standard.  Bullet allows 1024, but 254 should be used
 if creating a standard DBF III Plus file.

 Field Descriptors

 For each field, a descriptor is used to identify and type it.  These
 descriptors are assigned to an array; the pointer to that array is assigned to
 CDP.fieldListPtr.  The format of the descriptor follows, with a physical
 format listed in DBF File Format.

 Γûá Fieldname

 10 characters plus null byte terminator.  Valid fieldname characters are ASCII
 A-Z (upper-case) and the underscore (ASCII 95).  All bytes after the fieldname
 must be null bytes.  E.g., if the fieldname is "LNAME", five characters, the
 following six bytes (including the 11th byte) are set to 0. The eleventh byte
 is always a null byte since 10 characters is the maximum fieldname length.
 Extended ASCII characters (above 127) should not be used.

  fieldList[0].fieldname = "ANYNAME";   // see memset() above

 Γûá Field type and size

 Standard Xbase field types are C, D, L, M, and N:

 Type   Description
   C    Character field, any code page character, 1 to 255 characters.
        Null bytes are not desirable except as a string terminator.  There is
        no requirement that field data be terminated with a '\0'.  The field
        data should be left-justified within the field, but this is not
        required (in which case use leading spaces, not 0 bytes).

                       fieldList[0].fieldType = 'C';
                       fieldList[0].fieldLen = 25;
                       fieldList[0].fieldDC = 0;

   D    Date field, valid ASCII digits for date, 8 characters.
        The physical format is YYYYMMDD, where YYYY is the year (1999), MM is
        the month (1-12), and DD the day (1-31).  The date field is always 8
        bytes long, and is in ASCII digits ('19991231').  If no date, set to
        all spaces.

                       fieldList[0].fieldType = 'D';
                       fieldList[0].fieldLen = 8;
                       fieldList[0].fieldDC = 0;

   L    Logical field, < SPACE> Y N T F y n t f, 1 character.
        A single-byte field.  When not yet initialized the value will be a
        <SPACE> (ASCII 32).  This is typically displayed as a '?' to the user,
        indicating that the field has not been initialized.  Initialized values
        are variations of yes, no, true, false ('Y', 'y', etc.).

                       fieldList[0].fieldType = 'L';
                       fieldList[0].fieldLen = 1;
                       fieldList[0].fieldDC = 0;

   M    Memo field, 10 ASCII digits, 10 characters.
        Field data is used as the block number in the corresponding DBT memo
        file.  Each block is typically 512 bytes, with the first block (block
        #0) used as the memo file header.  If no block is used in the .DBT by
        this record, the field is set to <SPACES>.  The first memo block is
        stored as "0000000001".  (This description is valid for dBASE IV and
        later memo files, as created and used by BULLET.) Some Xbase versions
        use field types B and G as variations of memo files.  They are as M,
        but contain general data (as in anything), while memo files contain
        only text. BULLET supports any type data in its memo files, and you may
        use the CDP.fieldType of 'B' or 'G'.

        More than one memo field per record is permitted.  For example, you may
        need a memo for the printable address, where the address is free-form
        rather than in separate fields (i.e., you have both forms), and another
        memo for general notes, and yet a third for problem reports, and so on.
        All these, and all memos for the rest of the DBF file, are stored in
        the same DBT memo file.

        Note:  BULLET does not use the fieldType with regard to identifying
        memo field type; it is the programmer's responsibility to check the
        fieldType and act on it accordingly, such as adding memo numbers.

                       fieldList[0].fieldType = 'M';
                       fieldList[0].fieldLen = 10;
                       fieldList[0].fieldDC = 0;

   N    Numeric field, ASCII digits, 19 digits maximum (see below).
        All standard Xbase data is stored in ASCII form (for universal
        exchange).  Numeric fields are to be right-justified, with leading
        spaces, and an aligned decimal point, if any (relative this field in
        other records).  Do not end the field with a null byte.

        The total size of the numeric field is specified in .fieldLen, which
        includes any leading sign, the decimal point, and decimal digits to the
        right of the decimal point (if any decimal point). The maximum total
        size is 19 places.  If a decimal point, then the number of digits to
        the right may be from 1 to 15 digits, but must be no more than the
        total-2.

                   FieldLen.FieldDC    Example
                           8.2          " 2345.78"
                           8.2          "12345.78"
                           8.2          "-2345.78"
                           8.1          "123456.8"
                           8.0          "12345678"
                           5.3          "2.235"
                           5.4          (not valid)

                       fieldList[0].fieldType = 'N';
                       fieldList[0].fieldLen = 8;
                       fieldList[0].fieldDC = 2;

 Although not dBASE compatible, you may use binary fields in your data records.
 The Xbase standard always has ASCII data in the data fields, even if the field
 is numeric.  For example, an 'N' type field of 8.2 (total
 length.decimal-count) is stored as an ASCII text string in the data record,
 say, a string like " 1100.55".  If you want dBASE compatibility your field
 data must also be ASCII.  However, if you can forgo this requirement, you can
 use binary values in the fields.

 To do this you must specify a field type of 'Y' (actually, anything but an
 'N') and, if it is to be used as a key field, also set the sort function to
 the appropriate type (S16_SORT, etc.).  The field length
 (fieldList[x].fieldLen) for a 'Y' field type is 2 if 16-bit, and 4 if 32-bit.
 Also possible is floating-point (with a custom sort-compare function).  A
 likely field type marker for this would be 'F'.  Note that both 'Y' and 'F'
 are completely non-standard Xbase types, and only your programs will
 understand them.

 Note:  'B' should not be used as a binary field type marker since dBASE V uses
 'B' to signify a binary-data memo file field.  Bullet makes no distinction in
 its memo file data; anything can be placed in them.  Typically, your memo
 fields are marked as 'M' in Bullet, but could also be 'B' or 'G'.

 File ID

 Conventional dBASE DBF files have a CDP.fileID=3.  To create a memo file (DBT,
 dBASE IV compatible), set CDP.fileID=x8B.  For the DBT to be created, both
 bits 3 and 7 (0x88) must be set.  The other bits may be anything, and are not
 checked.  In creating your DBF files, specify CDP.fileID=3 to ensure
 compatibility across Xbase versions, and limit record length to 4000
 characters.  If creating a non-standard DBF (e.g., non-standard field types,
 extended field lengths, etc.) it's recommended to use CDP.fileID=0 or
 CDP.fileID=1.  For a standard DBF file with a memo file (dBASE IV or later),
 use CDP.fileID=0x8B (eight-bee).

 Generally, field data is space-filled.  String terminators are allowed in
 C-haracter field types, but should not be used in other fields.

 Memo File Creation

 If bits 3 and 7 are set in CDP.fileID, a memo file is created for the DBF. The
 memo filename will be the same as the DBF name except the extension. The memo
 file is created after the DBF, with a block size of 512 bytes, and filename
 extension of ".DBT".  The default block size and extension can be overridden
 (see SET_SYSVARS_XB) prior to calling this routine.

 Note:  A simple way to check that your record description in the field list
 matches your source code structure is to compare the number of bytes used by
 all fields in the field list (+1 for the delete tag byte) with the size of
 your program's structure.  They must equal.  Also verify each field's size to
 make sure they match.  By doing this you prevent the problem where data on
 disk does not match the data in your record structure.


ΓòÉΓòÉΓòÉ 9.14. OPEN_DATA_XB ΓòÉΓòÉΓòÉ

Pack: OPENPACK                 Source Example

     IN               OUT
   OP.func          OP.stat
   OP.filenamePtr   OP.handle
   OP.asMode

 Open an existing DBF data file for use.  For DBF opens, two parameters are
 specified:  the filename and the access-sharing mode.  The OP.xbLink parameter
 is used only for index opens, and so is not used here.

 The OP.asMode has optional cache mode settings.  The caching modes cover
 locality, write-through, and skip cache.  Locality is typically mostly random
 (RND_LOCALITY), but may be mostly sequential if the data file has been sorted
 and the index file recently reindexed and processing is mostly in-order (first
 to last, rather than random).  Locality is used to tune the cache.  Also,
 normally, data is written to the cache with control returning immediately to
 the program before the disk is written (an asynchronous write).  To force the
 write to take place before control is returned (a synchronous write), use the
 WRITE_THROUGH mode.  To skip the cache completely, use the SKIP_CACHE mode.
 This, as all OP.asMode settings, affects this file handle only.

 On a successful open, the file handle is returned.  Use this handle for all
 further access to this file.  If the DBF was created with a compatible memo
 file, it is also opened.  The handle of the memo file is available via
 STAT_DATA_XB, but all access to the memo file is made with the handle of the
 memo file's master DBF (the handle returned by this routine in OP.handle).
 The memo file is opened using the same OP.asMode.

 Note:  FoxPro DBF files with Fox memo files (FPT) use an ID of 0xFF.  Bullet
 does not support FoxPro memo files, and so opening a FoxPro DBF with a Fox
 memo file returns the warning message, WRN_CANNOT_OPEN_MEMO.  The DBF file is
 opened, and the warning can be ignored.

 Once open, you can get information on the data file by using STAT_DATA_XB.

 Each DBF data file opened allocates and commits at least 4K bytes for internal
 use:

   Number of Fields     Memory
        1 to 121            4KB
      122 to 249            8KB
      each 128 fields       4KB more

 This memory is released when you close the file with CLOSE_DATA_XB. or issue
 EXIT_XB.

 Note:  You must open the data file before you can open or create any of its
 index files.

 When BULLET creates a DBF, it forces all fieldnames to upper-case (it's a DBF
 requirement) and 0-fills them as well.  On data file opens (OPEN_DATA_XB), it
 also does this, and so any header copy (COPY_DATA_HEADER_XB) will have
 upper-cased fieldnames (the original file is not changed).  To prevent BULLET
 from mapping the fieldnames to upper-case (NLS mapping, though fieldnames
 should be standard ASCII characters only), set bit31 of OP.asMode to 1
 (0x80000042, for example).  This skips the case mapping.  Zero-filling always
 takes place, and starts after the first '\0' byte in the fieldname.


ΓòÉΓòÉΓòÉ 9.15. CLOSE_DATA_XB ΓòÉΓòÉΓòÉ

Pack: HANDLEPACK               Source Example

     IN               OUT
   HP.func          HP.stat
   HP.handle

 Close an existing data file.

 Closing the file updates the file header and releases the memory used by the
 file.  Any associated memo file is closed, too.  Any outstanding locks should
 be unlocked before calling this routine.

 Note:  Remaining locks belonging to this handle are released by the OS upon
 the successful close.


ΓòÉΓòÉΓòÉ 9.16. STAT_DATA_XB ΓòÉΓòÉΓòÉ

Pack: STATDATAPACK             Source Example

     IN               OUT
   SDP.func         SDP.stat         SDP.recordLength
   SDP.handle       SDP.fileType     SDP.xactionFlag
                    SDP.flags        SDP.encryptFlag
                    SDP.progress     SDP.herePtr
                    SDP.morePtr      SDP.memoHandle
                    SDP.fields       SDP.memoBlockSize
                    SDP.asMode       SDP.memoFlags
                    SDP.filenamePtr  SDP.memoLastRecord
                    SDP.fileID       SDP.memoLastSize
                    SDP.lastUpdate   SDP.lockCount
                    SDP.records

 Return information BULLET has on the DBF data file specified by SDP.handle.

   Item             Description
 stat               Return code of operation
 fileType           1 for DBF
 flags              Bit0=1 if file has changed since last flush (dirty)
                    Bit1=1 if the file has its entire region locked (full lock)
                    Bit2=1 if the file has a shared lock in use (cannot write
                    to it if so)
 progress           Percentage of pack operation completed, 1-99, or 0 if done
 morePtr            Always 0
 fields             Number of fields per record (does not included implicit tag
                    field)
 asMode             Access-sharing-cache mode as specified at open (excludes
                    NoCaseMap bit31)
 filenamePtr        Pointer to the filename as used in OPEN_DATA_XB
 fileID             ID byte used when the DBF was created (the first byte of
                    the file)
 lastUpdate         Date of last change (binary: high word=year (1999), low
                    byte=day, high byte=month)
 records            Number of records in the DBF (includes any delete-tagged
                    records)
 recordLength       Total length of a data record, including tag field
 xactionFlag        Not currently used
 encryptFlag        Not currently used
 herePtr            Pointer to the internal data control area for this file
                    handle
 memoHandle         Handle of open memo file (0 if none)
 memoBlockSize      Memo file block size (512 is typical, 24 is minimum)
 memoFlags          Bit0=1 dirty
 memoLastRecord     Last accessed memo record (0 if none; same as 'block
                    number')
 memoLastSize       Size of last accessed memo record (in bytes, including 8
                    bytes overhead)
 lockCount          Number of full-locks in force (locked on first, unlocked on
                    last)

 Typically, your program tracks whether a particular handle belongs to an index
 file or data file.  In cases where this is not possible, call the
 STAT_HANDLE_XB. routine to determine what file type a handle is.

 Note:  In network environments, you should have an exclusive lock on the data
 file (and implicitly, therefore, the memo file, if any) before using this
 routine to ensure that the information is current.  This also applies to
 multi-process environments on a single machine.  This routine is not
 mutex-protected.  During the call, the file handle must not be closed by
 another thread.


ΓòÉΓòÉΓòÉ 9.17. READ_DATA_HEADER_XB ΓòÉΓòÉΓòÉ

Pack: HANDLEPACK               Source Example

     IN               OUT
   HP.func          HP.stat
   HP.handle

 Reload the disk copy of the data header for the opened DBF data file handle,
 refreshing the in-memory copy.  Any associated memo file is refreshed, too.

 Normally, this routine is not called directly but rather is done automatically
 when you full-lock the file (LOCK_XB).  This routine does not refresh the
 header if the current state is dirty (SDP.flags, bit0=1); it returns an error
 if tried. Since it is recommended that a full-lock be in force before using
 this routine (shared or exclusive), and since a full-lock always reloads the
 header anyway, calling this routine should never be required.  If ever there
 is a reason to use this routine without having a full-lock in force, then, of
 course, you may need to.  However, it is not wise to reload the header without
 a full-lock (which locks the header).  If you are using your own lock
 routines, this call will be very useful.

 In single-user, single-tasking systems this routine is not needed.  However,
 in a multi-user or multi-tasking system it's possible, and desirable, for two
 or more programs to use the same data file.  Consider this scenario: A data
 file has 100 records.  Two programs access this data file, both opening it.
 Program 1 locks the file, adds a new record, then flushes and unlocks the
 file.  Program 1 knows that there are now 101 records in the file.  However,
 Program 2 is not aware of the changes that Program 1 made--it thinks that
 there are still 100 records in the file.  This out-of-sync situation is easily
 remedied by having Program 2 reload the data header from the file on disk.

 How does Program 2 know that it needs to reload the header? It doesn't.
 Instead, BULLET uses a simple, yet effective, approach when dealing with this.
 When BULLET full-locks a file, BULLET automatically reloads the header by
 using this routine.  When removing the full-lock, BULLET automatically flushes
 the header using FLUSH_DATA_HEADER_XB (unless the current lock is a shared
 lock (SDP.flags bit2=1)).


ΓòÉΓòÉΓòÉ 9.18. FLUSH_DATA_HEADER_XB ΓòÉΓòÉΓòÉ

Pack: HANDLEPACK               Source Example

     IN               OUT
   HP.func          HP.stat
   HP.handle

 Write the in-memory copy of the data header for the opened DBF data file
 handle to disk.  The actual write occurs only if the header has been changed
 (the dirty bit is set).  Any associated memo file is flushed, too.  This
 routine ensures that the data header on disk matches exactly the data header
 that is being maintained by BULLET.

 Normally, this routine is not called directly but rather is done automatically
 when you unlock the file (UNLOCK_XB).  This routine does not write out the
 header if the current lock state is shared (SDP.flags, bit2=1); it returns an
 error if tried.  Unlocking a full-lock performs a flush automatically, and so
 you may never need to explicitly call this routine.  Also, when relocking from
 an exclusive full-lock to a shared full-lock, an automatic flush is performed.

 Assume the following: A data file with 100 records.  Your program opens the
 data file and adds 1 record.  Physically, there are 101 records on disk.
 However, the header image of the data file on disk still reads 100 records.
 This isn't a problem, BULLET uses its internal copy of the data header and the
 internal copy does read 101 records.  But, if there were a system failure now,
 the image of the header would not get updated since the disk image is written
 only on a CLOSE_ or FLUSH_DATA_XB, or on EXIT_XB (and also prior to
 PACK_RECORDS_XB).  After the system restarts, BULLET opens the file, reads the
 header and thinks that there are 100 records.  You lost a record.  Now, if
 after that record add your program issues FLUSH_DATA_HEADER_XB, the header on
 disk is refreshed with the in-memory copy, keeping the two in sync.  This
 routine also updates the directory entry for the file, keeping things neat
 there (file size).  Still, it doesn't come without cost: flushing takes
 additional time, therefore, you may elect to flush periodically, or whenever
 the system is idle.

 Note:  You should have a full-lock on the file before using this routine.


ΓòÉΓòÉΓòÉ 9.19. COPY_DATA_HEADER_XB ΓòÉΓòÉΓòÉ

Pack: COPYPACK                 Source Example

     IN               OUT
   CP.func          CP.stat
   CP.handle
   CP.filenamePtr

 Copy the DBF file structure of an open data file to a new file.

 This routine makes it easy for you to duplicate the structure of an existing
 DBF file without having to specify all the information needed by
 CREATE_DATA_XB. The resultant DBF will be exactly like the source, including
 number of fields and field descriptions, and an empty memo file, if
 applicable. It contains 0 records.  It may be opened as a regular Bullet data
 file.

 A typical use for this is to create a work file, where only a subset of
 records is required.  For example:  You want to process all records of those
 whose last name starts with A.  Copy the header to a work file, use GET_XB
 routines to get records meeting the criterion, writing those that fit the
 criterion to the work file (using either Add/Reindex, or Insert).  A new index
 can be specified, or an existing index can be copied using
 COPY_INDEX_HEADER_XB.


ΓòÉΓòÉΓòÉ 9.20. ZAP_DATA_HEADER_XB ΓòÉΓòÉΓòÉ

Pack: HANDLEPACK               Source Example

     IN               OUT
   HP.func          HP.stat
   HP.handle

 Delete all records in a DBF data file.

 This routine is similar to COPY_DATA_HEADER_XB except for one major
 difference: All data records in the source file are physically deleted.  No
 action is performed on the DBF's memo file, if any.

 If you have a DBF file with 100 records and use ZAP_DATA_HEADER_XB on it, all
 100 records will be physically deleted and the file truncated as if no records
 were ever in the file.  All data records are lost forever.


ΓòÉΓòÉΓòÉ 9.21. CREATE_INDEX_XB ΓòÉΓòÉΓòÉ

Pack: CREATEINDEXPACK          Source Example

     IN               OUT
   CIP.func         CIP.stat
   CIP.filenamePtr
   CIP.keyExpPtr
   CIP.xbLink
   CIP.sortFunction
   CIP.codePage
   CIP.countryCode
   CIP.collatePtr
   CIP.nodeSize

 Create a new BULLET index file.

 Before you can create an index file, you must first have opened (and have
 created if necessary) the BULLET DBF data file that it is to index. To open
 the data file, use OPEN_DATA_XB.  To create the index file, you need to
 provide the name to use, the key expression, the DBF file link handle
 (obtained from the OPEN_DATA_XB call), sort function/flags, and optionally,
 the code page, country code, and collate table.  There's also a node size
 parameter.  Select 512, 1024, or 2048 bytes.

 Note:  BULLET has an optional external data mode where only indexing is done
 -- no data file link is used.  In this mode, BULLET manages the index files of
 the key and key data you provide (key data is any 32-bit item, e.g., a record
 number, offset, etc.).  This would be useful for indexing non-DBF files, even
 files with variable-length records.

 Filename

 The drive and path must exist if used as part of the filename. Long filenames
 may be used if supported by the file system in use.

 Key Expression

 The key expression is an ASCIIZ string composed of the elements that are to
 make up this index file's key.  The key can be composed of any or all of the
 fields in the DBF data record, or sub-strings within any of those fields. Up
 to 16 component parts can be used in the expression.

 Two functions are supported in evaluating a key expression. These are SUBSTR()
 and UPPER():

 SUBSTR() extracts part of a field's data starting at a particular position for
 x number of characters.  The first position is 1.

 UPPER() converts all lower-case letters to their upper-case equivalent.  Since
 BULLET supports NLS, UPPER() conversion is not required for proper sorting of
 mixed-case text strings.

 Any name used in the key expression must be a valid field name in the DBF data
 file.  Below are a few sample key expressions for the given data file
 structure:

     Name  Type Len DC
    FNAME    C   25  0
    LNAME    C   25  0
    SSN      C    9  0
    DEPT     N    5  0

 A few example key expression strings for this structure:

    keyExpression[]="LNAME";
    keyExpression[]="LNAME+FNAME";
    keyExpression[]="SUBSTR(LNAME,1,4)+SUBSTR(FNAME,1,1)+SUBSTR(SSN,6,4)";
    keyExpression[]="UPPER(LNAME+FNAME)";  // for ASCII sort function only
    keyExpression[]="DEPT+SSN";

 In the last example above, even though DEPT is a numeric field type (N), it
 can still be used as a component of a multi-part character key with SSN (whose
 type is set to character).  This because numeric fields in dBASE DBF data
 files are ASCII digits, not binary values, and are sorted according to the
 ASCII value or NLS weight.

 The key expression is parsed when the index file is created (this routine) and
 also when reindexed (REINDEX_XB.). The parser() function, which parses the key
 expression, may be replaced by a programmer-supplied function if additional
 functionality is needed.  See Custom Expression Parser Routine for details.

 DBF File Link Handle (xbLink)

 Since BULLET evaluates the key expression when the file is created (this
 routine) or during reindex, it must have access to the DBF file to verify that
 the key expression is valid.  You must, therefore, supply the OS file handle
 of the opened DBF data file.  If you later change the structure of the DBF
 data file (add new fields, remove others, etc.), you must use the reindex
 routine to re-evaluate the key expression.  If the key expression is no longer
 valid after the data file changes (key field has changed names, etc.), then
 you must create a brand new index file with this routine, supplying the new
 key expression, rather than reindexing.

 Note:  Handles 0-2 are reserved handles and should never be used for any
 BULLET routine.  Also, .xbLink of -1 is reserved by BULLET to indicate an
 external data index for index create and open routines.

 Sort Function

 The sort function specifies the sort method for the index file. Essentially,
 this defines the compare function used by the access methods employed by
 BULLET when doing any type of key access (reading and writing).  There are six
 intrinsic sort compare functions available, with an additional 10 sort compare
 functions that can be specified by the programmer (see Custom Sort-Compare
 Functions).

 While not recommended, duplicate key values are supported and managed by
 BULLET.  The flag DUPS_ALLOWED is OR'ed with the sort function value to
 specify this.  Generally, it is not acceptable to allow duplicate keys for an
 index; there should be one key identifying one record without any further
 investigation needed to determine if the key is indeed for that record.  This
 is not possible, not consistently so, when duplicate keys exist.  It is much
 simpler to define your key so that duplicates are not generated, than it is to
 deal with duplicate keys once you have them.  If an attempt to insert a key
 that already exists in the index file is made, and DUPS_ALLOWED was not
 specified when the index file was created, the insert fails (either a
 STORE_KEY_XB, an INSERT_XB, or a REINDEX_XB operation), and error
 EXB_KEY_EXISTS is returned.

 For Windows, the sort function flag, USE_ANSI_CHARSET may be specified. This
 instructs Bullet to use the ANSI character set (i.e., Windows character set)
 for the current system code page.  The default is USE_OEM_CHARSET, which is
 used by MS-DOS and OS/2.

 Only data contained within a record should be used to build a key. The
 physical record number is not part of the data of a record since it can change
 at any time without you knowing about it (during a pack, for example).  Do not
 use the record number in an attempt to generate unique keys.  Only use what is
 available in the data record itself, so that the key can be built, or rebuilt,
 at any time.

 The intrinsic sort compare functions of BULLET are:

 ASCII_SORT    1 - ASCII  (up to 16 key components)
 NLS_SORT      2 - NLS   (up to 16 key components)
 S16_SORT      3 - 16-bit signed integer (single component)
 U16_SORT      4 - 16-bit unsigned integer (single component)
 S32_SORT      5 - 32-bit signed integer (single component)
 U32_SORT      6 - 32-bit unsigned integer (single component)

 To expand on the basic functionality provided by BULLET, you can supply your
 own parser, build, and sort compare routines, and have BULLET use them
 instead.  With your own routines in place, you can have BULLET do just about
 anything with regard to the index file, including evaluating the key
 expression dynamically; using more components; allowing multi-part binary
 keys; and more.

 Generally, character data (type C) is left-justified, and unused space is
 padded with the SPACE character (ASCII 32).  It is permissible to use C-type
 strings, or to 0-fill unused space.

 Numeric data (type N) is right-justified, with leading space to be padded with
 the SPACE character.  It is not permissible to use 0-fill leading bytes
 (literal '0' can used, however).  Since the field is right-justified, it is
 not generally desirable to terminate the field with a 0 byte, either.  If a
 decimal count is specified (not 0),the decimal point location is to be the
 same for all entries in this field.  The field description must match the
 actual data:  If the field length and field decimal count was specified as
 10.2 (10 total bytes, 2 digits to the right of the decimal, then the data is
 to be formatted so that '-234567.90' is the longest data that is to be entered
 in that field.  All entries in all records for this field must be of the same
 format.  For example, " 987654.21", or "   23.01", or "   -1.99" (note the
 leading spaces).  Numeric data is indexed as ASCII values (i.e., the key
 remains character digits) unless a binary sort function is specified.

 Using one of the binary integer sort compare functions requires the following:

 1.  Single component expression.
 2.  Field type must be N if the field has ASCII digits, or if the data is
     binary, then the field type must be Y (actually, anything but N).
 3.  If ASCII digits, the value must fit into the function size (-32768 to
     32767 or 0-65535 for signed/unsigned 16-bit; 2,147,483,647 to
     -2,147,483,648 or 0-4,294,967,295 for 32-bit signed/unsigned values).

 Although not dBASE compatible, you may use binary fields in your data records.
 The Xbase standard always has ASCII data in the data fields, even if the field
 is numeric.  For example, an 'N' type field of 8.2 (total
 length.decimal-count) is stored as an ASCII text string in the data record,
 say, a string like " 1100.55" (there is no \0 string terminator).  If you want
 dBASE compatibility, your field data must be ASCII.  However, if you can forgo
 this requirement, you can use binary values in the fields.

 To do this you must specify a field type of 'Y' (actually, anything but an
 'N') and, if it is to be used as a key field, also set the sort function to
 the appropriate type (S16_SORT, etc.).  The field length
 (fieldList[x].fieldLen) for a 'Y' field type is 2 if 16-bit, and 4 if 32-bit.
 For 64-bit integers, a custom sort-copmare function is required since there is
 no intrinsic 64-bit function available.

 Note:  'B' should not be used as a binary field type marker since dBASE V uses
 'B' to signify a binary-data memo file field.  Bullet makes no distinction in
 its memo file data; anything can be place in them.  Typically, your memo
 fields are marked as 'M' in Bullet, but could also be 'B' or 'G'.

 The key expression string you specify may be up to 159 characters, and
 evaluate out to 64 bytes (62 bytes if DUPS_ALLOWED is specified).  The
 expression string must be 0-terminated, as are all strings used by BULLET
 itself (filenames, etc.).

 National Language Support (NLS)

 National Language Support is available to properly sort most languages'
 alphabets.  BULLET uses NLS to build the collate sequence table (sort table)
 that it uses to ensure proper sorting of mixed-case keys as well as the
 sorting of foreign language alphabets which use extended-ASCII.  In order for
 BULLET to use the proper collate table, it must know what code page and
 country code to use.  If not supplied, Bullet gets this information directly
 from the OS, which provides the cc/cp for the current process.  If you supply
 cc/cp, the code page must be loaded or an error is returned (see OS/2's CHCP
 command).  The collate table generated is made part of the index file so that
 all subsequent access to the index file maintains the original sort order,
 even if, say, the MIS shop is moved to another location/computer system using
 another country code/code page.  These three items are discussed below.

 Code Page

 To use the default code page of the current process, specify a code page of 0.
 The OS is queried for the current code page and this code page is then used.
 Any valid and available code page can be specified. This is used only if a
 custom sort-compare or NLS sort is specified.

 Country Code

 To use the default country code of the current process, specify a country code
 of 0.  The OS is queried for the current country code and this code is then
 used.  Any valid country code can be specified.  This is used only if a custom
 sort-compare or NLS sort is specified.

 Custom Collate Table

 If a null-pointer is specified, and a custom sort-compare or NLS sort is
 specified, BULLET queries the OS for the collate sequence table to use based
 on the code page and country code specified.  Otherwise, the supplied table is
 used.  Intrinsic sorts other than NLS use no collate table, and the country
 code or code page are not used, either.

 To use a sort weight table of your own choosing, supply a non-NULL pointer to
 this parameter.  If non-NULL, the passed table is used for sort compares.  The
 table is composed of 256 weight values, one per character.  For example, table
 position 65 ('A') and table position 97 ('a') could both be weighted 65, so
 that that each are considered equal when sorted.  If a custom sort-compare
 function was specified, this sort table may, or may not, be used -- it depends
 on whether the sort compare function uses the table (it's all up to the custom
 sort-compare function's logic).

 Typically, you set both the code page and country code = 0, and the collate
 table pointer to NULL.

 Node Size

 The index file is read and written in node-size chunks.  The larger the node
 size, the more keys are read or written per chunk.  Generally, a smaller node
 size offers better random key access, while a larger node size offers better
 sequential key access.

 Typically, an average node utilizes 66% of the node space for keys (a very
 small number may contain only a few keys, while some may be filled
 completely).  In a 512-byte node file, for a key length of 8, there is room
 for (512-5)/(keylength+8) nodes, or 31 keys.  Since a typical node is filled
 to 66%, that means about 20 keys per node.  For a 2048-byte node file, same
 parameters, there is room for (2048-5)/(keylength+8), or 127 keys.  At the
 standard 66% load, there are typically 83 keys per 2K node.  That's 3 more
 keys per 2K of disk than the 512-byte node gives for 4 nodes (20 keys*4), The
 trade-off is that each node is 4 times as large, and so requires 4 times more
 searching.  Actual performance differences may be minimal, or may be great.
 Run tests on expected data to determine the best for the data and access used.


ΓòÉΓòÉΓòÉ 9.22. OPEN_INDEX_XB ΓòÉΓòÉΓòÉ

Pack: OPENPACK                 Source Example

     IN               OUT
   OP.func          OP.stat
   OP.filenamePtr   OP.handle
   OP.asMode
   OP.xbLink

 Open an existing index file for use.  For index opens, three parameters are
 specified:  the filename, the access-sharing mode, and the handle of the open
 DBF file that this file indexes.  It is required to open the data file before
 you can open its related index file.

 Note:  Handles 0-2 are reserved handles and should never be used for any
 BULLET routine.  Also, .xbLink of -1 is reserved by BULLET to indicate an
 external data index for index create and open routines.

 On a successful open, the file handle is returned.  Use this handle for all
 further access to this file.

 Once open, you can get information on the index file by using STAT_INDEX_XB.

 Each index file that you open allocates and commits 4K bytes for internal use
 (this will vary if SET_VECTORS_XB with VECTOR_MALLOC is used).  This memory is
 released when you close the file with CLOSE_INDEX_XB or issue EXIT_XB, or your
 program terminates.

 The OP.asMode has optional cache mode settings.  The caching modes cover
 locality, write-through, and skip cache.  Locality is typically mostly random
 (RND_LOCALITY), but may be mostly sequential if the data file has been sorted
 and the index file recently reindexed and processing is mostly in-order (first
 to last, rather than random).  Locality is used to tune the cache.  Also,
 normally, data is written to the cache with control returning immediately to
 the program before the disk is written (an asynchronous write).  To force the
 write to take place before control is returned (a synchronous write), use the
 WRITE_THROUGH mode.  To skip the cache completely, use the SKIP_CACHE mode.
 This, as all OP.asMode settings, affects this file handle only.

 BULLET has an optional external data mode where only indexing is done -- no
 data file link is used.  In this mode, BULLET manages the index files of the
 key and key data you provide (key data is any 32-bit item, e.g., a record
 number, offset, etc.).  This would be useful for indexing non-DBF files, even
 files with variable-length records.  Only those routines that do not access
 the data file may be used (any routine using AP.recPtr, for example INSERT_XB,
 could not be used, but NEXT_KEY_XB, STORE_KEY, etc., may).


ΓòÉΓòÉΓòÉ 9.23. CLOSE_INDEX_XB ΓòÉΓòÉΓòÉ

Pack: HANDLEPACK               Source Example

     IN               OUT
   HP.func          HP.stat
   HP.handle

 Close an open index file.

 Closing the file updates the file header and releases the memory used by the
 file.  Any outstanding locks should be unlocked before calling this routine.

 Note:  Remaining locks belonging to this handle are released by the OS upon
 the successful close.


ΓòÉΓòÉΓòÉ 9.24. STAT_INDEX_XB ΓòÉΓòÉΓòÉ

Pack: STATINDEXPACK            Source Example

     IN               OUT
   SIP.func         SIP.stat         SIP.keyLength
   SIP.handle       SIP.fileType     SIP.keyRecNo
                    SIP.flags        SIP.keyPtr
                    SIP.progress     SIP.herePtr
                    SIP.morePtr      SIP.codePage
                    SIP.xbLink       SIP.countryCode
                    SIP.asMode       SIP.CTptr
                    SIP.filenamePtr  SIP.nodeSize
                    SIP.fileID       SIP.sortFunction
                    SIP.keyExpPtr    SIP.lockCount
                    SIP.keys

 Return information BULLET has on the index file specified by SIP.handle.

   Item             Description
 stat               Return code of operation
 fileType           0 for index, IX3
 flags              Bit0=1 if file has changed since last flush (dirty)
                    Bit1=1 if the file has its entire region locked (full lock)
                    Bit2=1 if the file has a shared full-lock in use (cannot
                    write to it if so)
 progress           Percentage of reindex operation completed, 1-99, or 0 if
                    done
 morePtr            Always 0
 xbLink             handle of the open DBF file this file indexes
 asMode             Access-sharing-cache mode as specified at open
 filenamePtr        Pointer to the filename as used in OPEN_INDEX_XB
 fileID             '31ch' (the first four bytes of the file)
 keyExpPtr          Pointer to the key expression used when the index was
                    created
 keys               Number of keys in the index file
 keyLength          Length of the key, including 2-byte enumerator if
                    DUPS_ALLOWED
 keyRecNo           The DBF record number that the last accessed key indexes
 keyPtr             Pointer to the last accessed key (valid for keyLength
                    bytes)
 herePtr            Pointer to the internal data control for this file
 codePage           Code page used when the index file was created (the actual
                    code page)
 countryCode        Country code used when the index file was created (the
                    actual country code)
 CTptr              Pointer to the collate sequence table used for NLS sorting
                    (each index file has its own sequence table) or NULL if not
                    an NLS index
 nodeSize           Size of an index node, 512, 1024, or 2048 bytes, as
                    specified during create
 sortFunction       The index sort-compare method (low word) and the sort flags
                    (high word), with sort-compare values 1-9 being intrinsic,
                    and 10-19 being custom functions; the DUPS_ALLOWED flag is
                    bit0 of the high word (allowed if set)
 lockCount          Number of full-locks in force (locked on first, unlocked on
                    last)

 Typically, your program tracks whether a particular handle belongs to an index
 file or a data file.  In cases where this is not possible, call the
 STAT_HANDLE_XB routine to determine what file type a handle is.

 Note:  In network environments, you should have an exclusive lock on the index
 file before using this routine to ensure that the information is current.
 This routine is not mutex-protected.  During the call, the file handle must
 not be closed by another thread.


ΓòÉΓòÉΓòÉ 9.25. READ_INDEX_HEADER_XB ΓòÉΓòÉΓòÉ

Pack: HANDLEPACK               Source Example

     IN               OUT
   HP.func          HP.stat
   HP.handle

 Reload the disk copy of the index header for the opened index file handle,
 refreshing the in-memory copy.

 Normally, this routine is not called directly but rather is done automatically
 when you full-lock the file (LOCK_XB).  This routine does not refresh the
 header if the current state is dirty (SIP.flags, bit0=1); it returns an error
 if tried. Since it is recommended that a full-lock be in force before using
 this routine (shared or exclusive), and since a full-lock always reloads the
 header anyway, calling this routine should never be required.  If ever there
 is a reason to use this routine without having a full-lock in force, then, of
 course, you may need to.  However, it is not wise to reload the header without
 a full-lock (which locks the header).  If you are using your own lock
 routines, this call will be very useful.

 In single-user, single-tasking systems this routine is not needed.  However,
 in a multi-user or multi-tasking system it's possible, and desirable, for two
 or more programs to use the same data file.  Consider this scenario: An index
 file has 100 keys.  Two programs access this index file, both opening it.
 Program 1 locks the file, adds a new key, then flushes and unlocks the file.
 Program 1 knows that there are now 101 keys in the file.  However, Program 2
 is not aware of the changes that Program 1 made--it thinks that there are
 still 100 keys in the file.  This out-of-sync situation is easily remedied by
 having Program 2 reload the index header from the file on disk.

 How does Program 2 know that it needs to reload the header? It doesn't.
 Instead, BULLET uses a simple, yet effective, approach when dealing with this.
 When BULLET full-locks a file, it automatically reloads the header using this
 routine.  When removing the full-lock, BULLET automatically flushes the header
 using FLUSH_INDEX_HEADER_XB (unless the current lock is a shared lock
 (SIP.flags bit2=1)).


ΓòÉΓòÉΓòÉ 9.26. FLUSH_INDEX_HEADER_XB ΓòÉΓòÉΓòÉ

Pack: HANDLEPACK               Source Example

     IN               OUT
   HP.func          HP.stat
   HP.handle

 Write the in-memory copy of the index header for the opened index file handle
 to disk.  The actual write occurs only if the header has been changed. This
 ensures that the index header on disk matches exactly the index header that is
 being maintained by BULLET.

 Normally, this routine is not called directly but rather is done automatically
 when you unlock the file (UNLOCK_XB).  This routine does not write out the
 header if the current lock state is shared (SIP.flags, bit2=1); it returns an
 error if tried.  Unlocking a full-lock performs a flush automatically, and so
 you may never need to explicitly call this routine.  Also, when relocking from
 an exclusive full-lock to a shared full-lock, an automatic flush is performed.

 Assume the following: An index file with 100 keys.  Your program opens the
 index file and adds 1 key.  Physically, there are 101 keys on disk.  However,
 the header image of the index file on disk still reads 100 keys.  This isn't a
 problem; BULLET uses its in-memory copy of the index header and the in-memory
 copy does read 101 keys.  But, if there were a system failure after the key
 add, the disk image of the header would not get updated since the disk image
 is written only on a CLOSE_ or FLUSH_INDEX_XB, or on EXIT_XB (and also prior
 to REINDEX_XB).  After the system restarts, BULLET opens the file, reads the
 header and thinks that there are 100 keys.  You lost a key.  Now, if after
 that key was added, your program issues a FLUSH_INDEX_HEADER_XB, the header on
 disk is refreshed with the in-memory copy, keeping the two in sync.  The
 routine updates the directory entry, keeping things neat there as well (file
 size).  Still, it doesn't come without cost: flushing will take additional
 time, therefore, you may elect to flush periodically, or whenever the system
 is idle.

 Note:  You should have a full-lock on the file before using this routine.


ΓòÉΓòÉΓòÉ 9.27. COPY_INDEX_HEADER_XB ΓòÉΓòÉΓòÉ

Pack: COPYPACK                 Source Example

     IN               OUT
   CP.func          CP.stat
   CP.handle
   CP.filenamePtr

 Copy the index file structure of an open index file to another file.

 This routine duplicates the structure of an existing index file without having
 to re-specify the information needed by CREATE_INDEX_XB. The resultant index
 file will be exactly like the source, including sort function and key
 expression.  It contains 0 keys.


ΓòÉΓòÉΓòÉ 9.28. ZAP_INDEX_HEADER_XB ΓòÉΓòÉΓòÉ

Pack: HANDLEPACK               Source Example

     IN               OUT
   HP.func          HP.stat
   HP.handle

 Delete all keys from an index file.

 This routine is similar to COPY_INDEX_HEADER_XB except for one major
 difference: All keys in the source file are physically deleted.

 If you have an index file with 100 keys and issue ZAP_INDEX_HEADER_XB, all 100
 keys will be physically deleted and the file truncated to 0 keys. REINDEX_XB
 can be used to rebuild the index file.

 Note:  Since BULLET reindexes in place, the use of ZAP is not typically
 needed.


ΓòÉΓòÉΓòÉ 9.29. GET_DESCRIPTOR_XB ΓòÉΓòÉΓòÉ

Pack: DESCRIPTORPACK           Source Example

     IN               OUT
   DP.func          DP.stat
   DP.handle        DP.fieldNumber
   DP.fieldNumber   DP.fieldOffset
        -or-        DP.FD.fieldName
   DP.FD.fieldName  DP.FD.fieldType
                    DP.FD.fieldLen
                    DP.FD.fieldDC
                    DP.FD.altFieldLength

 Get the field descriptor information for a field by fieldname or by field
 position.

 To get descriptor info by fieldname, set DP.fieldNumber=0 and set the
 DP.FD.fieldname member to the fieldname string.  Fieldnames must be
 0-terminated and 0-filled, and must be upper-case, with A-Z and _ valid
 fieldname characters.  If the string matches a fieldname in the DBF descriptor
 area, that field's descriptor info is returned in DP.FD, (FD is
 FIELDDESCTYPE), and its position is returned in FD.fieldNumber and
 FD.fieldOffset.

 To get descriptor info by field position (i.e., field number), set
 DP.fieldNumber to the field's position.  The first is field #1. The "delete
 tag" field is not considered a field.  If the position is valid (i.e., greater
 than 0 and not beyond the last field),that field's descriptor info is returned
 in DP.FD.

 This routine lets you work with unknown DBF files -- those created by another
 program.  By reading each field descriptor, by number, from 1 to number of
 fields (SDP.noFields), you can generate a run-time layout of the DBF file.
 Alternatively, you can get input from your user for a fieldname, and locate
 the descriptor by name.

 Note:  If you need to add or delete a field, be sure to reindex all related
 index files so that their key expressions can be re-evaluated. To do this, you
 need to create a new data file and build it as you build any other new data
 file.  Then, copy record-by-record from the old DBF to the new, using the old
 record layout for reads, and the new record layout for writes.  After this,
 reindex any index file related to the DBF file.  The old DBF file can then be
 deleted.

 If non-standard fields are used (i.e., non-char structure members to match
 non-ASCII data fields in your non-standard DBF), then be aware that your
 compiler more than likely will add padding to align on member-size boundaries.
 This will result in a mis-match between your compiler structure and your DBF
 structure (as described in fieldList[]).  To prevent this, place #pragma
 pack(1) / #pragma pack() around your structures that BULLET uses.  Consult
 your particular compiler for alternate methods if it does not support #pragma
 pack.


ΓòÉΓòÉΓòÉ 9.30. GET_RECORD_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        *AP.recPtr
   AP.recNo
   AP.recPtr

 Get the physical record from the data file into a data buffer.

 The data buffer, pointed to by AP.recPtr, is typically a struct variable
 defined as the DBF record itself is defined.  For example, if the DBF record
 has 2 fields, LNAME and FNAME, each 25 characters, then then variable would be
 typed as:

  struct rectype {
   CHAR  tag;             /* The Xbase DBF delete tag (must be included) */
   CHAR  lastName[25];    /* same length as first field's descriptor fieldLen */
   CHAR  firstName[25];   /* same length as second field's descriptor fieldLen */
  }; /* 51 */
  struct rectype recbuff;

 The first record is at AP.recNo=1.  The last is SDP.records, determined by
 STAT_DATA_XB.  The buffer must be at least as large as the record length
 (SDP.recordLength).

 This method of accessing the data file does not use any indexing.  Generally,
 this access method is not used except for special purposes (sequential
 processing where order is not required).  The preferred method to access the
 data is by one of the keyed GET_XB routines.

 If non-standard fields are used (i.e., non-char structure members to match
 non-ASCII data fields in your non-standard DBF), then be aware that your
 compiler more than likely will add padding to align on member-size boundaries.
 This will result in a mis-match between your compiler structure (rectype
 above) and your DBF structure (as described in fieldList[]).  To prevent this,
 place #pragma pack(1) / #pragma pack() around your structures that BULLET
 uses.  Consult your particular compiler for alternate methods if it does not
 support #pragma pack.


ΓòÉΓòÉΓòÉ 9.31. ADD_RECORD_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.recPtr

 Append the data record in the data buffer to the end of the DBF file.

 This method of adding a record involves no indexing.  It is typically used to
 build a data file en masse.  Indexing is deferred until all records have been
 added, and then quickly indexed using REINDEX_XB.

 Since ADD_RECORD_XB is extremely fast, if you have several thousand data
 records to be added at once, appending records with this routine and
 reindexing when all have been added using REINDEX_XB is often faster than
 using INSERT_XB for each record to add.

 The record number assigned to the record appended is determined by BULLET, and
 that record number is returned in AP.recNo.

 If non-standard fields are used (i.e., non-char structure members to match
 non-ASCII data fields in your non-standard DBF), then be aware that your
 compiler more than likely will adding padding to align on member-size
 boundaries.  This will result in a mis-match between your compiler structure
 and your DBF structure (as described in fieldList[]).  To prevent this, place
 #pragma pack(1) / #pragma pack() around your structures that BULLET uses.
 Consult your particular compiler for alternate methods if it does not support
 #pragma pack.


ΓòÉΓòÉΓòÉ 9.32. UPDATE_RECORD_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle
   AP.recNo
   AP.recPtr

 Write the updated data record to the physical record number.

 This method of updating a data record must not be used if any field being used
 as a key field (i.e., part of the key expression) is changed.

 This method of updating a record is very fast if you know that that update is
 not going to alter any field used as a key in any index file that uses it. You
 must, of course, first get the data record into the record buffer.  You can
 then change it, and write the update out to disk using this routine.

 If you need to change a field that is used as a key field, or part of one
 (e.g., SUBSTR()), use the UPDATE_XB routine.

 If you plan on reindexing with REINDEX_XB immediately after using this
 routine, you may elect to update the data file using this method even if
 changing any field used as a key, rather than UPDATE_XB.  This since UPDATE_XB
 is very disk intensive.  However, if transaction support is needed (i.e.,
 updates are dependent on other updates), then UPDATE_XB should be used.


ΓòÉΓòÉΓòÉ 9.33. DELETE_RECORD_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle
   AP.recNo

 Tag the record at the physical record number as being deleted.

 This does not tag any in-memory copies of the record so be sure to mark any
 such copies as being deleted yourself.

 The first byte of every DBF record is reserved for the tag field.  This tag is
 a space (ASCII 32) if the record is normal, or a * (ASCII 42) if it's marked
 as being deleted.  This delete tag is a reserved field in the DBF record and
 as such is not defined as a formal field with a descriptor.  Make sure that
 you define your in-memory buffers to reserve the first byte for the delete
 tag.

 The Xbase DBF standard doesn't physically remove records marked as deleted
 from the data file.  It doesn't mark them as available/reusable either.  To
 physically remove records marked as deleted use PACK_RECORDS_XB

 Records can be temporarily marked as deleted during processing and then
 recalled to normal status when completed, useful for flagging a record as
 having been processed (for example, mass updating using UPDATE_XB).  The
 GET_XB routines return the record number associated with a key (in AP.recNo),
 and that record number can be used for this routine.

 While the DELETE_RECORD_XB and UNDELETE_RECORD_XB routines provided in BULLET
 use the * and SPACE characters only, you can use whatever character you want
 in the tag field when you fill your record buffer structure's data.  Normally,
 you set the tag field to SPACE (x.tag = ' ';), but, for example, if you want
 to implement your own, program-level locking you can use the tag field as a
 marker to indicate the record is locked (by using an 'L' character, or ID with
 bit7=1, or whatever you can think of) and use the very fast UPDATE_RECORD_XB
 to set it.  Another possibility is set to aside a field to be used as this,
 say, along with the user ID of the lock owner.

 The SKIP_TAG_SELECT item in SET_SYSVARS_XB can be set to have the REINDEX_XB
 routine not place a key value into the index file if a record has a matching
 tag field.  This may be useful if you want to, say, generate an ad hoc index
 for only undeleted records.


ΓòÉΓòÉΓòÉ 9.34. UNDELETE_RECORD_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle
   AP.recNo

 Tag the record at the physical record number as being normal (not deleted).

 This does not tag any in-memory copies of the record so be sure to mark any
 such copies as being normal.

 This routine removes the * character, as put there by DELETE_RECORD_XB, in the
 tag field and replaces it with a SPACE.  The tag field is always overwritten
 with a SPACE, regardless of what it was.


ΓòÉΓòÉΓòÉ 9.35. DEBUMP_RECORD_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle
   AP.recNo

 Remove the record identified by AP.recNo from the data file if and only if the
 record is the last in the file.  The file is automatically flushed before
 debumping.

 Unlike DELETE_RECORD_XB, this routine physically removes a data record from
 the DBF file, provided that the record to delete is the last. STAT_DATA_XB can
 be used to identify the last record number (SDP.records is the last).  This,
 when used after deleting any and all keys in all index files referencing this
 record (see DELETE_KEY_XB), is useful if you are managing a transaction log
 and need to back out changes made, beyond what BULLET performs.

 If the record is not the last, alternate methods must be used. The simplest,
 and often equally as good as physically deleting the record, is to just mark
 the record as deleted using DELETE_RECORD_XB and let it remain in the file
 until the next PACK_RECORDS_XB.  Another option is to overwrite the record's
 data with SPACES, or other appropriate field data (such as HIGH-VALUES, and
 use UPDATE_XB), if necessary.  This routine is the only method available to
 physically remove a record from the file, short of using PACK_RECORDS_XB.
 Removing a record with active keys referencing that record will result in an
 access error (ERR_UNEXPECTED_EOF) when accessing that key with GET_XB
 routines, or will generate stale results.  Remove any keys that reference this
 record before deleting it.


ΓòÉΓòÉΓòÉ 9.36. PACK_RECORDS_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle

 Rebuild the open DBF file by physically removing all records marked as
 deleted.

 Packing occurs in place using the existing file.  It is recommended that you
 use BACKUP_FILE_XB to copy the current DBF file before using this routine in
 case of a failure during the pack process.

 The newly packed file is truncated to reflect the current, actual size.  All
 records with the tag field set to * are removed from the file.

 If there are index files for this DBF file, they must be reindexed after the
 pack process by using REINDEX_XB.

 Memo files are not affected by this routine.  Before packing, it is
 recommended that you traverse the data file to be packed, and for records that
 are to be deleted, check to see if there is a memo record. If there is, delete
 the memo.  Do this for each such occurrence.  This way, orphaned memo records
 will not take up permanent space in the memo file.


ΓòÉΓòÉΓòÉ 9.37. GET_MEMO_SIZE_XB ΓòÉΓòÉΓòÉ

Pack: MEMODATAPACK             Source Example

     IN               OUT
   MDP.func         MDP.stat
   MDP.dbfHandle    MDP.memoBytes
   MDP.memoNo

 Get the number of bytes used by the memo at MDP.memoNo.

 Memo file allocation is made in blocks, typically of 512 bytes each.
 Therefore, a memo of 10 bytes uses 1 allocation block, as would a 500-byte
 memo.  This size is stored with each memo record, and can be retrieved.
 Before accessing a memo record, it's a good idea to retrieve the current size
 of the memo so you know how large a buffer you may need if you intend to read
 it all in, at one time, or even to just know how much to read, in total,
 reading parts of it at a time.

 The first memo is at MDP.memoNo=1.  The last memo number cannot be easily
 determined, but generally this does not need to be known.  The memo number
 identifying the memo's location is stored in the memo field area of the DBF
 record.  It is stored as a text string (e.g., "0000000001").  This is not a C
 string; there is no zero terminator so sprintf() should be used.  This number
 is the physical block number at which the memo starts.  Memos are always
 stored in consecutive blocks, if more than a single block is needed.  For
 example, a memo of 513 bytes uses two blocks, say, #1 and #2.  The next memo
 added would use memo #3 (if #3 is available), rather than #2 since #2 was used
 by the first memo.  Memo numbers may be reassigned (see UPDATE_MEMO_XB).  The
 highest possible memo number is 589,822 (0x8FFFE).  With the standard 512-byte
 block size, this allows a memo file to be up to 288MB.  If more memo data
 space is needed, use a larger block size (e.g., 2KB block size allows over 1GB
 per memo file).

 Notice For All Memo Routines

 In multitasking environments you should have a full-lock on the DBF file that
 owns this memo file, or at least a record lock on the record that owns the
 memo number.  In BULLET, locking is not performed on the memo file.  Instead,
 the lock is implied when the lock is made on the DBF file.  This because a
 memo file is for one DBF file alone, and so if you have a lock on the DBF
 before accessing the memo file (for whatever reason), then no other process
 may lock the DBF and also access the memo.

 This works only if you restrict your access to the memo file if you have a
 lock on the DBF master file (the DBF that this DBT memo file belongs to) or on
 the DBF record.  For this routine, which only requires access to this memo
 record, a record lock is sufficient since no writing is performed.  Further, a
 shared lock is all that is required.  This because all that is required to
 keep from stepping on other process's toes is that it be known that the
 current memo header info (for this memo record), as known to this process, is
 the current state of this memo.  In other words, it must be true that the memo
 file state on disk exactly matches the memo file state in memory.  With a lock
 in place, no other process may gain write access to change this memo, "out
 from under you".  A shared lock does allow the other process to read this
 memo, and that may be used if no writing is needed.

 Each memo routine following states its lock requirements (exclusive full lock,
 shared full lock, exclusive record lock, or shared record lock).


ΓòÉΓòÉΓòÉ 9.38. GET_MEMO_XB ΓòÉΓòÉΓòÉ

Pack: MEMODATAPACK             Source Example

     IN               OUT
   MDP.func         MDP.stat
   MDP.dbfHandle    *MDP.memoPtr
   MDP.memoNo       MDP.memoBytes
   MDP.memoPtr
   MDP.memoOffset
   MDP.memoBytes

 Read the specified number of bytes of the memo, starting at the offset, into
 the buffer.  The actual number of bytes read is returned.

 Use GET_MEMO_SIZE_XB to determine that total number of bytes you may need to
 read.  With that, you can allocate a buffer of that size to read the entire
 memo into.  Or, you can read chunks of the memo, a chunk at a time, up to the
 number of bytes in the memo.

 The number of bytes actually read (and stored starting at MDP.memoPtr) is
 returned in MDP.memoBytes (overwriting the value you placed there).  If the
 number of bytes requested is not the same as the number of bytes returned, you
 attempted to read beyond the end of the memo.  BULLET does not return an error
 if you try this, which is SOP for file reads, so check the two if you need to
 verify this.  An error is returned, however, if you attempt to read at a
 starting offset beyond the end of the actual memo data (i.e., MDP.memoOffset >
 memo's data size).  The first byte of the memo data is at .memoOffset=0.

 It's recommended that a lock be in force on either the DBF (full-lock) or on
 the record that this memo belongs to.  A shared lock is okay since no writing
 is done.


ΓòÉΓòÉΓòÉ 9.39. ADD_MEMO_XB ΓòÉΓòÉΓòÉ

Pack: MEMODATAPACK             Source Example

     IN               OUT
   MDP.func         MDP.stat
   MDP.dbfHandle    MDP.memoNo
   MDP.memoPtr
   MDP.memoBytes

 Add the data from the buffer for the specified bytes to a new memo.  The memo
 number used is returned.

 Any data can be stored in a memo.  The memo number returned can be any value;
 it can even be less than the previous add's memo number.  The reason for this
 is that an avail-list is kept for the memo file, and any deleted or otherwise
 freed blocks become available for re-use.  The memo is stored in the first
 contiguous group of free blocks large enough to satisfy the request.  For
 example, if MDP.memoBytes is from 1 to (blockSize-8) bytes, the first
 available block is used.  If the size needed is greater than 1 block, then the
 avail-list is walked and the first contiguous group large enough to satisfy
 the request is used.  If none of the avail-list groups is large enough,
 ultimately, the new memo data is appended to the end of the file.  This is
 also done if there are no avail-list items at all, such as in a memo file that
 has never had deletes or updates.

 The returned memo is a binary block number (ULONG).  This value should be
 converted into an ASCII string (sprintf can be used) and stored in the DBF
 data record, in the memo field.  The string should be of the form,
 "0000000001" (for MDP.memoNo=1), with leading zeros, but no zero terminator
 (exactly 10 bytes in size).  This data record should then be written to disk
 using UPDATE_RECORD_XB. Since BULLET can be used in non-standard Xbase mode,
 where binary field values can be used, you can omit the conversion from binary
 to ASCII if a standard DBF is not required.  Likewise, when accessing a memo,
 the conversion of the memo block number from ASCII to binary would not be
 required.

 It's recommended that a lock be in force on the DBF (full-lock).  A shared
 lock may not be used since writing to the memo file, and the DBF record, is
 required.  A full lock is required since the memo file header is read and
 written.


ΓòÉΓòÉΓòÉ 9.40. UPDATE_MEMO_XB ΓòÉΓòÉΓòÉ

Pack: MEMODATAPACK             Source Example

     IN               OUT
   MDP.func         MDP.stat
   MDP.dbfHandle    MDP.memoNo
   MDP.memoNo
   MDP.memoPtr
   MDP.memoOffset
   MDP.memoBytes

 Update an existing memo.  The update can overwrite current data, append new
 data extending the current size, or it can shrink the current size.

 Appending data so that the memo is extended may result in a new memo number
 returned.  The original memo blocks are made available for reuse (deleted).
 Shrinking will not change the memo number, but unused blocks from the shrink
 are made available for reuse.

 If you want to change anything in the memo at MDP.memoNo, locate its position
 within the memo with MDP.memoOffset and set the size in MDP.memoBytes. The
 first data byte of a memo is located at MDP.offset=0.  There are 8 bytes of
 overhead per memo record (any number of blocks still has only the 8 bytes of
 overhead), but these are transparent to any memo access you do. The bytes at
 MDP.memoPtr overwrite the current memo data at the position specified.  For
 example, if you want to change the first 5 bytes of the first memo, set
 MDP.memoNo=1, MDP.memoPtr=yourNewData, MDP.memoOffset=0, and MDP.memoBytes=5.
 On return, MDP.memoNo is going to be the same as it was before the update,
 since you are not extending the memo size in this example.  Nothing further
 needs to be done; the memo is updated.

 If you want to add new memo data to an existing memo at MDP.memoNo, such as
 adding another line item, or problem report paragraph, etc., set
 MDP.memoOffset=theCurrentMemoSize (this locates to the end of the current memo
 data), MDP.memoBytes=bytesYouWantToAppend, and MDP.memoPtr=yourDataToAppend.
 If the old data size plus your newly added data still fits inside the last
 memo block previously used, MDP.memoNo is returned the same as it was on
 entry.  However, if the new data requires that more blocks be allocated, the
 entire memo is relocated to the next contiguous block group that is large
 enough to store the data.  That new block number is returned in MDP.memoNo,
 and the old block number and all its blocks are placed on the top of the
 avail-list.

 If you want to shrink the size as reported by GET_MEMO_SIZE_XB from an
 existing memo at MDP.memoNo, set MDP.memoBytes=newSizeYouWant, and
 MDP.memoPtr=NULL.  This means that you should have, before making this shrink
 call, updated the memo data that occurs within this new size to be the data
 size you want to be in the memo.  For example, if you have 10 line items, say,
 each 60 bytes long, and want to remove line item #5, you could do it by
 reading all 10 line items to memory, moving line items #6 to 10 down one (so
 they are now line items #5 to 9, effectively removing old line item #5), and
 update the memo (by using memoOffset=0 and memoBytes=9*60).  After this,
 though, you still have 10*60 bytes as the memo size (old line item #10 is now
 at #9 and still at #10).  Since you want the size to reflect the real data in
 the memo, set MDP.memoBytes=90, MDP.memoPtr=NULL, and update this memo number.
 Only the memo's size is affected by this particular update.  The size
 specified must be smaller than the original size, or an error is returned.

 It's recommended that a lock be in force on the DBF (full-lock).  A record
 lock should not be used if the update may result in blocks being moved, or the
 memo being shrunk by a full block or more.  A shared lock may not be used
 since writing to the memo file, and to the DBF record if MDP.memoNo is new, is
 required.


ΓòÉΓòÉΓòÉ 9.41. DELETE_MEMO_XB ΓòÉΓòÉΓòÉ

Pack: MEMODATAPACK             Source Example

     IN               OUT
   MDP.func         MDP.stat
   MDP.dbfHandle
   MDP.memoNo

 The memo and all its blocks are made available for reuse.

 Before using PACK_RECORDS_XB, you should run through all DBF records and check
 for those records that are deleted (record.tag='*') to be sure that any memo
 belong to those records are deleted from the memo file.  If this is not done,
 orphaned memo records -- those that do not have a DBF record memo field
 pointing to it, may be left in the memo file (forever!).

 After deleting a memo record, update the DBF record's memo field by writing
 <SPACES> (ASCII 32) to the memo field member.  Update this to disk with
 UPDATE_RECORD_XB as soon as possible (and before unlocking).  A memo field
 with no current memo record is indicated by spaces ("0000000000" should not be
 used).

 It's recommended that a lock be in force on the DBF (full-lock).  Neither a
 record lock nor a shared lock may be used since writing to the memo file
 header and the DBF record is required.


ΓòÉΓòÉΓòÉ 9.42. MEMO_BYPASS_XB ΓòÉΓòÉΓòÉ

Pack: MEMODATAPACK             Source Example

     IN               OUT
   MDP.func         MDP.stat
   MDP.dbfHandle
   MDP.memoBypass

 Memo files are created, opened, closed, and flushed/reloaded by their
 corresponding DBF data file action.  To perform these tasks asynchronously,
 this routine is used.  Bypass routines are:

     MDP.memoBypass      Value
   BYPASS_CREATE_MEMO        1
   BYPASS_OPEN_MEMO          2
   BYPASS_CLOSE_MEMO         3
   BYPASS_READ_MEMO_HEADER   4
   BYPASS_FLUSH_MEMO_HEADER  5

 All bypass routines require the handle of the DBF file that this memo is for.
 Nothing is returned here, except the result code.  The memo handle from the
 open is stored internally, but is available by using STAT_DATA_XB and checking
 SDP.memoHandle.  However, none of the BULLET memo routines use the memo handle
 directly; all access to the memo file is through the master DBF file handle.

 No data is required for input other than the DBF handle and memo bypass
 routine to perform (see table above).  All required info is obtained from the
 DBF file's information.  You may use an alternate block size, as set via
 SET_SYSVARS_XB.

 Generally, there is no need to call these routines using this bypass. However,
 if you need to create a memo file anew (say, after the initial DBF was
 created), and then open it, using these routines is the easiest way to
 proceed.

 Note:  When creating a memo via the bypass method, the file ID is altered to
 indicate that the DBF has a DBT memo file.  The file ID is the first byte of
 the DBF file.  The ID is changed by OR'ing 0x88h with the current file ID
 value.  The next flush or close updates the disk image of the DBF with the new
 file ID.  The next DBF open, then, also opens the DBT memo file created here.
 Be sure to always keep the DBT and DBF pairs in the same directory, if moved.

 Since the DBF file is already open (and must be to use any of these routines),
 you must use the open bypass routine to open the memo if you plan on using it.
 Either that, or close the DBF after you've create the memo file, and simply
 re-open the DBF, which also, now, opens the DBT memo file.

 The other available bypass routines: close, read, and flush, typically will
 not be used from this bypass routine.  These operations are done automatically
 when their corresponding DBF action is performed, and have little
 functionality used on their own.

 Before using BYPASS_READ_MEMO_HEADER or BYPASS_FLUSH_MEMO_HEADER, it's
 recommended that a lock be in force on the DBF (full-lock). A shared lock can
 be used for BYPASS_READ_MEMO_HEADER, but it must be a full lock.


ΓòÉΓòÉΓòÉ 9.43. FIRST_KEY_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.keyPtr        *AP.keyPtr

 Retrieve the first key in index order from the index file.

 This routine does not access the DBF file and so does not retrieve the data
 record.  What it does do is locate the first logical key of the index file,
 returning it, and also returning the record number within the DBF that the key
 indexes.

 To retrieve the data record you can use the GET_RECORD_XB routine.  The
 preferred method, however, is to use GET_FIRST_XB, which combines these
 operations.

 The key returned includes an enumerator if the index file allows duplicate
 keys.

 This routine is typically used to position the index file to the first key so
 as to allow forward in-order access to the keys by using NEXT_KEY_XB.

 If an external data file was specified in CREATE_INDEX_XB, the record number
 returned by this routine does not refer to a DBF record, but rather is the
 value supplied when the key was stored.  This permits index access to your
 data files (data files which are not maintained by BULLET, but by you).


ΓòÉΓòÉΓòÉ 9.44. EQUAL_KEY_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.keyPtr

 Search for the exact key in the index file.

 This routine does not access the DBF file and so does not retrieve the data
 record.  What it does do is search for the key in the index, and if found,
 returns the record number within the DBF that the key indexes.  The key must
 be an exact match, including the enumerator word if the index file is using
 non-unique keys.

 To retrieve the data record you can use the GET_RECORD_XB routine.  The
 preferred method, however, is to use GET_EQUAL_XB, which combines these
 operations.

 This routine returns no key in *keyPtr since, by definition, you already have
 the key in the key buffer if this routine succeeds.

 This routine finds only an exact match to the specified key (including the
 enumerator if applicable).


ΓòÉΓòÉΓòÉ 9.45. EQUAL_OR_GREATER_KEY_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.keyPtr        *AP.keyPtr

 Search for the exact key in the index file and, if not found, get the key that
 would have followed it.

 This routine is similar to EQUAL_KEY_XB except that this routine returns a key
 in *keyPtr (either the same as on entry, or if that is not found, then the
 next greater key).

 The main benefit of this routine is that it is an atomic operation. It differs
 from setting the atomic mode flag of SET_SYSVARS_XB in that this routine
 allows a fuzzy starting point:  the key in AP.keyPtr, on entry (IN), need not
 exist.


ΓòÉΓòÉΓòÉ 9.46. EQUAL_OR_LESSER_KEY_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.keyPtr        *AP.keyPtr

 Search for the exact key in the index file and, if not found, get the key that
 would have come before it.

 This routine is similar to EQUAL_KEY_XB except that this routine returns a key
 in *keyPtr (either the same as on entry, or if that is not found, then the
 previous, lesser key).

 The main benefit of this routine is that it is an atomic operation. It differs
 from setting the atomic mode flag of SET_SYSVARS_XB in that this routine
 allows a fuzzy starting point:  the key in AP.keyPtr, on entry (IN), need not
 exist.


ΓòÉΓòÉΓòÉ 9.47. NEXT_KEY_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.keyPtr        *AP.keyPtr

 Retrieve the next key in index order from the index file.

 This routine does not access the DBF file and so does not retrieve the data
 record.  What it does do is retrieve the next key of the index, returning it,
 and also returning the record number within the DBF that the key indexes.

 To retrieve the data record you can use the GET_RECORD_XB routine.  The
 preferred method, however, is to use GET_NEXT_XB, which combines these
 operations.

 The key returned includes an enumerator if the index file allows duplicates.

 This routine is typically called after the index file has first been
 positioned to a known key using either FIRST_KEY_XB or EQUAL_KEY_XB, or after
 a previous NEXT_KEY_XB or even PREV_KEY_XB.  What it basically does is get the
 key following the current key, and then makes that key the new current key. ________________________________________________

 If bit0 of the atomic mode flag of SET_SYSVARS_XB is set to 1, key access is
 based on a given starting point.  This simplifies index access in
 multi-threaded code, where another thread may have altered the last key
 accessed in the index file.  This mode lets you set a starting point for the
 operation by supplying in AP.keyPtr the key value to start at.

 For example, say you use GET_FIRST_XB.  On return, AP.keyPtr has the the very
 first key.  Say elsewhere in your multi-threaded program, another operation
 accesses that same index file handle, and performs some other access, where
 the last accessed key is no longer the same (i.e., not the first key).  Your
 first thread is expecting that a GET_NEXT_XB would get the second key,
 however, it very likely won't since the second thread has altered the last
 accessed key for that file handle.  By using the atomic mode for key access,
 your first thread, which has the first key value in its AP.keyPtr, can do a
 call to GET_NEXT_XB and get expected results, since the NEXT operation first
 positions to the value in AP.keyPtr and then follows up with a GET_NEXT
 operation.  This is performed within the Bullet kernel, and so won't be
 interrupted by another thread (i.e., it is an atomic operation).  For this to
 work, you must ensure that the AP.keyPtr value is set to the value of the last
 accessed key.  This will always be the case unless uninitialized, or you are
 using global variables for your threads' AP (AccessPack).  On return from the
 operation, AP.keyPtr will once again be set up for another atomic operation.

 Note:  You must supply a valid key value for this atomic access mode.
 AP.keyPtr must be at least as large as the key length in all cases, and is to
 have the starting point for the operation (i.e., the last accessed key).  You
 may, alternatively, set the first byte of the key buffer to 0 (but not
 AP.keyPtr itself to NULL).  This disables atomic mode for that access, and
 reverts to the internally-stored last key accessed as the starting point.


ΓòÉΓòÉΓòÉ 9.48. PREV_KEY_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.keyPtr        *AP.keyPtr

 Retrieve the previous key in index order from the index file.

 This routine does not access the DBF file and so does not retrieve the data
 record.  What it does do is retrieve the previous key of the index, returning
 it and also returning the record number within the DBF that the key indexes.

 To retrieve the data record you can use the GET_RECORD_XB routine.  The
 preferred method, however, is to use GET_PREV_XB, which combines these
 operations.

 The key returned includes an enumerator if the index file allows duplicates.

 This routine is typically called after the index file has first been
 positioned to a known key using either LAST_KEY_XB or EQUAL_KEY_XB, or after a
 previous PREV_KEY_XB or even NEXT_KEY_XB.  What it basically does is to get
 the key previous the current key, and then make that key the new current key. ________________________________________________

 If bit0 of the atomic mode flag of SET_SYSVARS_XB is set to 1, key access is
 based on a given starting point.  This simplifies index access in
 multi-threaded code, where another thread may have altered the last key
 accessed in the index file.  This mode lets you set a starting point for the
 operation by supplying in AP.keyPtr the key value to start at.

 For example, say you use GET_FIRST_XB.  On return, AP.keyPtr has the the very
 first key.  Say elsewhere in your multi-threaded program, another operation
 accesses that same index file handle, and performs some other access, where
 the last accessed key is no longer the same (i.e., not the first key).  Your
 first thread is expecting that a GET_NEXT_XB would get the second key,
 however, it very likely won't since the second thread has altered the last
 accessed key for that file handle.  By using the atomic mode for key access,
 your first thread, which has the first key value in its AP.keyPtr, can do a
 call to GET_NEXT_XB and get expected results, since the NEXT operation first
 positions to the value in AP.keyPtr and then follows up with a GET_NEXT
 operation.  This is performed within the Bullet kernel, and so won't be
 interrupted by another thread (i.e., it is an atomic operation).  For this to
 work, you must ensure that the AP.keyPtr value is set to the value of the last
 accessed key.  This will always be the case unless uninitialized, or you are
 using global variables for your threads' AP (AccessPack).  On return from the
 operation, AP.keyPtr will once again be set up for another atomic operation.

 Note:  You must supply a valid key value for this atomic access mode.
 AP.keyPtr must be at least as large as the key length in all cases, and is to
 have the starting point for the operation (i.e., the last accessed key).  You
 may, alternatively, set the first byte of the key buffer to 0 (but not
 AP.keyPtr itself to NULL).  This disables atomic mode for that access, and
 reverts to the internally-stored last key accessed as the starting point.


ΓòÉΓòÉΓòÉ 9.49. LAST_KEY_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.keyPtr        *AP.keyPtr

 Retrieve the last key in index order from the index file.

 This routine does not access the DBF file and so does not retrieve the data
 record.  What it does do is locate the last key of the index, returning it,
 and also returning the record number within the DBF that the key indexes.

 To retrieve the data record you can use the GET_RECORD_XB routine.  The
 preferred method, however, is to use GET_LAST_XB, which combines these
 operations.

 The key returned includes an enumerator if the index file allows duplicates.

 This routine is typically used to position the index file to the last key so
 as to allow reverse in-order access to the keys by using PREV_KEY_XB.


ΓòÉΓòÉΓòÉ 9.50. STORE_KEY_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle
   AP.recNo
   AP.keyPtr

 Insert the key into the index file in proper key order.

 This routine does not add the data record to the DBF file.  It only inserts
 the key and record number into the index file.  Use INSERT_XB instead.

 To do a complete data record and key insert, use ADD_RECORD_XB to add the data
 record to the DBF, BUILD_KEY_XB to construct the key, then STORE_KEY_XB to
 insert the key and record number information into the index file.  If that key
 already exists and the file allows duplicate keys, attach the proper
 enumerator word and retry STORE_KEY_XB.  INSERT_XB does this automatically.


ΓòÉΓòÉΓòÉ 9.51. DELETE_KEY_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle
   AP.keyPtr

 Physically remove the specified key from the index file.

 This routine requires an exact key match for all bytes of the key, including
 the enumerator word if duplicate keys are allowed.

 Typically, this routine would seldom be used since deleted DBF data records
 are only physically deleted during a PACK_RECORDS_XB operation, after which a
 REINDEX_XB is done. It is useful if you are managing a transaction log and
 need to back out changes made, beyond what BULLET performs.  Also see
 DEBUMP_RECORD_XB. If you have non-unique keys (where DUPS_ALLOWED is true),
 you may have several keys that match your criterion, and only differ in their
 enumerator.  To identify which key, then, goes to a particular DBF record,
 compare that key's AP.recNo with the number of your DBF record.  If they are
 the same, then this key belongs to that record.  Use either the KEY_XB or the
 GET_XB routines, then, before using this routine.  In other words, use this
 routine only after you have identified exactly the key to delete, and for the
 exact data record.  Once you have the record number, you can locate its key by
 using GET_KEY_FOR_RECORD_XB.


ΓòÉΓòÉΓòÉ 9.52. BUILD_KEY_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        *AP.keyPtr
   AP.recPtr
   AP.keyPtr

 Build the key for the specified data record based on the key expression for
 the index file.  If the index file allows duplicate keys, a 0-value enumerator
 is attached.

 This routine, like most of the mid-level routines, typically would not be used
 since the high-level access routines take care of this detail automatically.
 If used, it normally would be used prior to STORE_KEY_XB.

 This routine can be replaced.  See Custom Build-Key Routine.

 Note:  If DUPS_ALLOWED, this routine always sets the enumerator to \0\0.
 Enumerator management, which is used to guarantee a unique key, is performed
 only when the INSERT_XB routine is used.


ΓòÉΓòÉΓòÉ 9.53. GET_CURRENT_KEY_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.keyPtr        *AP.keyPtr

 Retrieve the current key value for the specified index file handle and also
 the data record number that it indexes. The key value includes the enumerator
 if applicable.

 This routine is useful in that it retrieves, on demand, the actual key value
 of the last accessed key in the index file (and the data record number
 associated with that key). STAT_INDEX_XB returns this information, too.


ΓòÉΓòÉΓòÉ 9.54. GET_KEY_FOR_RECORD_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        *AP.keyPtr
   AP.recNo
   AP.recPtr
   AP.keyPtr

 Retrieve the key for the record/record number pair.

 This routine would typically be used prior to using DELETE_KEY_XB and
 DEBUMP_RECORD_XB.  The key returned includes the enumerator if applicable.
 This routine sifts through any duplicate keys (if DUPS_ALLOWED) for the key
 that matches the record/record number pair, and so requires both the actual
 data record along with its physical record number (even if dups are not
 allowed).

 Typically this routine is extraneous; the key is available with a GET_XB
 routine and so can be deleted from the information provided through normal
 access.

 This routine builds a key based on the supplied record at AP.recPtr and
 searches the index for that key proper.  If found, and if DUPS_ALLOWED, each
 key matching the key proper has its record number compared to the record
 number in AP.recNo.  If that matches, too, then that is the exact key being
 sought.


ΓòÉΓòÉΓòÉ 9.55. GET_FIRST_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.recPtr        *AP.recPtr
   AP.keyPtr        *AP.keyPtr

 Retrieve the first indexed key and its data record.

 The key returned includes an enumerator if the index file uses non-unique keys
 (DUPS_ALLOWED).

 This routine is typically used to process a database in index order starting
 at the first ordered key (and its data record).  After processing this first
 entry, subsequent in-order access of the database is achieved by using
 GET_NEXT_XB, until the end of the database is reached, at which point an error
 is returned.

 This routine, like all the high-level GET_XB routines, fills in the AP.recNo
 of the record accessed.  In this case, it fills AP.recNo with the record
 number pointed to by the first key. Since this is done upon each GET_XB
 access, the AP pack is primed for an UPDATE_XB


ΓòÉΓòÉΓòÉ 9.56. GET_EQUAL_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.recPtr        *AP.recPtr
   AP.keyPtr

 Search for the exact key in the index file and return its data record.

 This routine finds only an exact match to the specified key (including the
 enumerator if applicable).

 This routine, like all the high-level GET_XB routines, fills in the AP.recNo
 of the record accessed.  In this case, it fills AP.recNo with the record
 number pointed to by the matching key. Since this is done upon each GET_XB
 access, the AP pack is primed for an UPDATE_XB


ΓòÉΓòÉΓòÉ 9.57. GET_EQUAL_OR_GREATER_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.recPtr        *AP.recPtr
   AP.keyPtr        *AP.keyPtr

 Search for the exact key in the index file and return its data record, or if
 not found, get the key and record that would have followed it.

 This routine is similar to GET_EQUAL_XB except that this routine returns a key
 in *keyPtr (either the same as on entry, or if that is not found, then the
 next greater key).

 The main benefit of this routine is that it is an atomic operation. It differs
 from setting the atomic mode flag of SET_SYSVARS_XB in that this routine
 allows a fuzzy starting point:  the key in AP.keyPtr, on entry (IN), need not
 exist.


ΓòÉΓòÉΓòÉ 9.58. GET_EQUAL_OR_LESSER_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.recPtr        *AP.recPtr
   AP.keyPtr        *AP.keyPtr

 Search for the exact key in the index file and return its data record, or if
 not found, get the key and record that would have come before it.

 This routine is similar to GET_EQUAL_XB except that this routine returns a key
 in *keyPtr (either the same as on entry, or if that is not found, then the
 previous, lesser key).

 The main benefit of this routine is that it is an atomic operation. It differs
 from setting the atomic mode flag of SET_SYSVARS_XB in that this routine
 allows a fuzzy starting point:  the key in AP.keyPtr, on entry (IN), need not
 exist.


ΓòÉΓòÉΓòÉ 9.59. GET_NEXT_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.recPtr        *AP.recPtr
   AP.keyPtr        *AP.keyPtr

 Retrieve the next indexed key and its data record.

 The key returned includes an enumerator if the index file uses non-unique keys
 (DUPS_ALLOWED).

 This routine is typically called after the index file has first been
 positioned to a known key using either GET_FIRST_XB or GET_EQUAL_XB, or after
 a previous GET_NEXT_XB or even GET_PREV_XB.  What it basically does is get the
 key and data record following the current key, and then makes that key the new
 current key.

 This routine, like all the high-level GET_XB routines, fills in the AP.recNo
 of the record accessed.  In this case, it fills AP.recNo with the record
 number pointed to by the next key (now the current key). Since this is done
 upon each GET_XB access, the AP pack is primed for an UPDATE_XB. ________________________________________________

 If bit0 of the atomic mode flag of SET_SYSVARS_XB is set to 1, key access is
 based on a given starting point.  This simplifies index access in
 multi-threaded code, where another thread may have altered the last key
 accessed in the index file.  This mode lets you set a starting point for the
 operation by supplying in AP.keyPtr the key value to start at.

 For example, say you use GET_FIRST_XB.  On return, AP.keyPtr has the the very
 first key.  Say elsewhere in your multi-threaded program, another operation
 accesses that same index file handle, and performs some other access, where
 the last accessed key is no longer the same (i.e., not the first key).  Your
 first thread is expecting that a GET_NEXT_XB would get the second key,
 however, it very likely won't since the second thread has altered the last
 accessed key for that file handle.  By using the atomic mode for key access,
 your first thread, which has the first key value in its AP.keyPtr, can do a
 call to GET_NEXT_XB and get expected results, since the NEXT operation first
 positions to the value in AP.keyPtr and then follows up with a GET_NEXT
 operation.  This is performed within the Bullet kernel, and so won't be
 interrupted by another thread (i.e., it is an atomic operation).  For this to
 work, you must ensure that the AP.keyPtr value is set to the value of the last
 accessed key.  This will always be the case unless uninitialized, or you are
 using global variables for your threads' AP (AccessPack).  On return from the
 operation, AP.keyPtr will once again be set up for another atomic operation.

 Note:  You must supply a valid key value for this atomic access mode.
 AP.keyPtr must be at least as large as the key length in all cases, and is to
 have the starting point for the operation (i.e., the last accessed key).  You
 may, alternatively, set the first byte of the key buffer to 0 (but not
 AP.keyPtr itself to NULL).  This disables atomic mode for that access, and
 reverts to the internally-stored last key accessed as the starting point.


ΓòÉΓòÉΓòÉ 9.60. GET_PREV_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.recPtr        *AP.recPtr
   AP.keyPtr        *AP.keyPtr

 Retrieve the previous indexed key and its data record.

 The key returned includes an enumerator if the index file uses non-unique keys
 (DUPS_ALLOWED).

 This routine is typically called after the index file has first been
 positioned to a known key using either GET_LAST_XB or GET_EQUAL_XB, or after a
 previous GET_PREV_XB or even GET_NEXT_XB.  What it basically does is get the
 key and data record preceding the current key, and then makes that key the new
 current key.

 This routine, like all the high-level GET_XB routines, fills in the AP.recNo
 of the record accessed.  In this case, it fills AP.recNo with the record
 number pointed to by the previous key (now the current key). Since this is
 done upon each GET_XB access, the AP pack is primed for an UPDATE_XB. ________________________________________________

 If bit0 of the atomic mode flag of SET_SYSVARS_XB is set to 1, key access is
 based on a given starting point.  This simplifies index access in
 multi-threaded code, where another thread may have altered the last key
 accessed in the index file.  This mode lets you set a starting point for the
 operation by supplying in AP.keyPtr the key value to start at.

 For example, say you use GET_FIRST_XB.  On return, AP.keyPtr has the the very
 first key.  Say elsewhere in your multi-threaded program, another operation
 accesses that same index file handle, and performs some other access, where
 the last accessed key is no longer the same (i.e., not the first key).  Your
 first thread is expecting that a GET_NEXT_XB would get the second key,
 however, it very likely won't since the second thread has altered the last
 accessed key for that file handle.  By using the atomic mode for key access,
 your first thread, which has the first key value in its AP.keyPtr, can do a
 call to GET_NEXT_XB and get expected results, since the NEXT operation first
 positions to the value in AP.keyPtr and then follows up with a GET_NEXT
 operation.  This is performed within the Bullet kernel, and so won't be
 interrupted by another thread (i.e., it is an atomic operation).  For this to
 work, you must ensure that the AP.keyPtr value is set to the value of the last
 accessed key.  This will always be the case unless uninitialized, or you are
 using global variables for your threads' AP (AccessPack).  On return from the
 operation, AP.keyPtr will once again be set up for another atomic operation.

 Note:  You must supply a valid key value for this atomic access mode.
 AP.keyPtr must be at least as large as the key length in all cases, and is to
 have the starting point for the operation (i.e., the last accessed key).  You
 may, alternatively, set the first byte of the key buffer to 0 (but not
 AP.keyPtr itself to NULL).  This disables atomic mode for that access, and
 reverts to the internally-stored last key accessed as the starting point.


ΓòÉΓòÉΓòÉ 9.61. GET_LAST_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.recPtr        *AP.recPtr
   AP.keyPtr        *AP.keyPtr

 Retrieve the last indexed key and its data record.

 The key returned includes an enumerator if the index file uses non-unique keys
 (DUPS_ALLOWED).

 This routine is typically used to process a database in reverse index order
 starting at the last ordered key (and its data record).  After processing this
 last entry, subsequent reverse-order access of the database is achieved by
 using GET_PREV_XB, until the top of the database is reached, at which point an
 error is returned.

 This routine, like all the high-level GET_XB routines, fills in the AP.recNo
 of the record accessed.  In this case, it fills AP.recNo with the record
 number pointed to by the last key. Since this is done upon each GET_XB access,
 the AP pack is primed for an UPDATE_XB


ΓòÉΓòÉΓòÉ 9.62. INSERT_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.recNo         *AP.keyPtr
   AP.recPtr
   AP.keyPtr
   AP.nextPtr

 Append the data records to data files and build and insert the related keys
 into all linked index files.  (Alternate forms are possible.)

 This routine is used to add new entries into a database.  Up to 256 index
 files may be inserted into per call, with up to 256 data files being added,
 too, for a total of 512 files managed per single INSERT_XB call.

 Note:  Bullet comes in 100, 250, and 1024-file versions and so this routine is
 able to use as many files as handles are still available. If non-standard
 fields are used (i.e., non-char structure members to match non-ASCII data
 fields in your non-standard DBF), then be aware that your compiler more than
 likely will add padding to align on member-size boundaries.  This will result
 in a mis-match between your compiler structure and your DBF structure (as
 described in fieldList[]).  To prevent this, place #pragma pack(1) / #pragma
 pack() around your structures that BULLET uses.  Consult your particular
 compiler for alternate methods if it does not support #pragma pack.

 Only index handles are listed in AP.handle.  Each index file has associated
 with it a data file, known internally to BULLET (the xbLink from OPEN_XB).
 There may be more than one index file for a data file, but there is always one
 data file per index handle specified in the list.  In other words, you can
 list five index files, each indexing the same xbLink data file, and have
 BULLET perform an atomic insert of that list.  Or, another possibility is that
 you have a single index file, indexing a single data file.  Or, you can list
 256 index files, each indexing a single data file (512 total files).

 This and several other routines are transaction-list-based.  This means that
 if a failure occurs prior to the routine's completion, all changes made to the
 database by the routine will be backed-out, and the database (data and index
 files) effectively restored to its original state.

 If the routine failed to complete, the function return value is the number
 (1-based) of the pack that caused the failure.  A positive number indicates
 the failure was from an index operation; a negative number indicates the
 failure was from a data operation.  In each case, the absolute value of the
 return code is the list item that failed (the pack index).  For example, if
 five index handles are in the list (AP[0] to AP[4]), and an error occurred on
 the last pack's index file, the return code would be positive 5, indicating
 the fifth pack (AP[4]) failed.  Since it was a positive 5, the index file was
 being processed when the error occurred.  Being processed means not only
 physical access, but verification, etc.  If the return code was -5, then
 again, the error was in the fifth pack, but since it is negative, the error
 occurred while processing the data file.  In either case, upon return, the
 database is effectively restored to the way it was before the INSERT_XB call
 was made.  Remedy the error, if possible, and INSERT_XB again.

 Each pack must include a separate key buffer.  You must not share a common key
 buffer.  Doing so disables any chance of recovering the index files in case of
 error, since it is in these buffers that BULLET places the newly built keys,
 and it is from these that BULLET, upon an error condition, deletes the keys
 (required for roll-back).

 The enumerator is automatically set up by this routine, if required
 (DUPS_ALLOWED and the key already exists with enumerator 0).  It does this by
 seeking the last possible enumerator value (0xFFFF) and then backing up to the
 previous key.  That key's enumerator is evaluated and incremented, and used as
 this key's.

 Specifying Files

 As mentioned, only the index file handles are specified in AP.handle. Data
 files are implicitly specified by their links to the index files, as specified
 when the index file was opened (OP.xbLink).  INSERT_XB can process up to 256
 index files per call.  Since each index file requires a data file, this means
 that up to 256 data files can be processed per call, as well.  Also possible
 is that all 256 index handles refer to the same, single data file.  Yet
 another possibility is that there is 1 index file, and so 1 data file.  The
 possibilities can include those and anything in between.

 Example: Specifying a single index file

 The simplest form is where a single index handle is specified.  This implies a
 single data file, too.  AccessPack setup for this is:

  AP.func = INSERT_XB;
  AP.handle = indexHandle;
  AP.recNo = 0;
  AP.recPtr = &recordStruct; // contents referred to below as *recordStruct
  AP.keyPtr = keyBuffer;
  AP.nextPtr = NULL;

 A call to BULLET with the above does the following:

   1. The data in *recordStruct is used as a new record that is appended to the
      data file.  The data file was linked to this index during the index open,
      in OP.xbLink.

   2. A key is built by BULLET, based on the data in *recordStruct, and that
      key is inserted into the index file (AP.handle).  Stored with the key is
      the record number of the record added above.

 Note:  AP.recNo must be set to 0 prior to the call.  Any positive number
 results in an error (0x80000000, and negative numbers, may be used when more
 than one AP pack is used - see below).

 Upon return, if no error, the return code is 0.  AP.recNo is set to the
 physical record number in the data file that *recordStruct was placed.  The
 key that was stored, including any enumerator, is in *keyBuffer.

 Upon return, and there was an error, the return code is either -1 or 1. If -1,
 the error was caused during processing of the data file portion, and the error
 code itself is in AP.stat.  If +1, the error was caused during processing of
 the index file, and the error code itself is in AP.stat, as well.  The return
 code is, as in all BULLET transaction-list routines, an index of the AP pack
 that generated the error -- negative if a data file error, positive if an
 index file error.  Since this example has only the single pack, only a -1 or
 +1 could be returned, or 0.

 Note:  If an error occurred after any part of the database had changed (during
 this particular call), then any and all changes that were made are backed-out,
 and the files restored to the same state as before the call.

 Example: Specifying two index files for a single data file

 Two index files, related to the same data file, would set AccessPack to:

  AP[0].func = INSERT_XB;
  AP[0].handle = indexHandle_0;
  AP[0].recNo = 0;
  AP[0].recPtr = &recordStruct;
  AP[0].keyPtr = keyBuffer_0;
  AP[0].nextPtr = AP[1];

  AP[1].handle = indexHandle_1;
  AP[1].recNo = 0x80000000;
  AP[1].recPtr = &recordStruct;
  AP[1].keyPtr = keyBuffer_1;
  AP[1].nextPtr = NULL;

 A call to BULLET with the above does the following:

   1. The data in *recordStruct is used as a new record that is appended
      (added) to the data file.

   2. A key is built by BULLET, based on the data in *recordStruct, and that
      key is inserted into the index file (AP[0].handle).  Stored with the key
      is the record number of the record added above.

   3. A second key is built by BULLET, based on the data in *recordStruct, and
      that key is inserted into the second index file (AP[1].handle).  Stored
      with the key is the record number of the record added above.

 Note:  The 0x80000000 in AP[1].recNo signifies that AP[1] is using the same
 data record that was appended during processing of AP[0].  This results in
 just the one data record being added.  AP[1].recPtr must still, however, point
 to the same data as AP[0].recPtr does.

 Upon return, if no error, the return code is 0.  AP[0].recNo is set to the
 physical record number in the data file that *recordStruct was placed.  The
 key that was stored for the first index, including any enumerator, is in the
 buffer at AP[0].keyPtr.  AP[1].recNo is set to the same physical record number
 as AP[0].recNo, except that the record number is negative: For example, if
 AP[0].recNo is 22 on return, AP[1].recNo is -22 (the original 0x80000000 value
 is overwritten).  The key that was stored for the second index, including any
 enumerator, is in the buffer at AP[1].keyPtr.

 Upon return, and there was an error, the return code can be -2, -1, 1, or 2.
 If negative, the error was caused during processing of that AP pack's data
 file portion, and the error code itself is in AP[abs(rez)-1].stat (where rez
 is the return code, and -1 since C arrays start at 0).  If the return code was
 positive, the error was caused during processing of that AP pack's index file,
 and the error code itself is in AP[rez-1].stat, as well.  The return code is,
 as in all BULLET transaction-list routines, an index of the AP pack that
 generated the error -- negative if a data file error, positive if an index
 file error.

 Note:  If an error occurred after any part of the database had changed (during
 this particular call), then any and all changes that were made are backed-out,
 and the files restored to the same state as before the call.

 Example: Specifying two index files for each of two different data files

 Four total files: two index files related to one data file, and two other
 index files related to another data file, would set AccessPack to:

  AP[0].func = INSERT_XB;
  AP[0].handle = indexHandle_0;
  AP[0].recNo = 0;
  AP[0].recPtr = &recordStruct_0;
  AP[0].keyPtr = keyBuffer_0;
  AP[0].nextPtr = AP[1];

  AP[1].handle = indexHandle_1;
  AP[1].recNo = 0x80000000;
  AP[1].recPtr = &recordStruct_0;
  AP[1].keyPtr = keyBuffer_1;
  AP[1].nextPtr = AP[2];

  AP[2].handle = indexHandle_2;
  AP[2].recNo = 0;
  AP[2].recPtr = &recordStruct_1;
  AP[2].keyPtr = keyBuffer_2;
  AP[2].nextPtr = AP[3];

  AP[3].handle = indexHandle_3;
  AP[3].recNo = 0x80000000;
  AP[3].recPtr = &recordStruct_1;
  AP[3].keyPtr = keyBuffer_3;
  AP[3].nextPtr = NULL;

 A call to BULLET with the above does the following:

   1. The data in *recordStruct_0 is used as a new record that is appended to
      the data file linked to the index file in AP[0].handle.

   2. A key is built by BULLET, based on the data in *recordStruct_0, and that
      key is inserted into the index file (AP[0].handle).  Stored with the key
      is the record number of the record added above, for _0.

   3. A second key is built by BULLET, based on the data in *recordStruct_0,
      and that key is inserted into the second index file (AP[1].handle).
      Stored with the key is the record number of the record added above, using
      *recordStruct_0.

   4. The data in *recordStruct_1 is used as a new record that is appended to
      the data file linked to the index file in AP[2].handle.

   5. A third key is built by BULLET, based on the data in *recordStruct_1, and
      that key is inserted into the index file (AP[2].handle).  Stored with the
      key is the record number of the record added above, for _1.

   6. A fourth key is built by BULLET, based on the data in *recordStruct_1,
      and that key is inserted into the fourth index file (AP[3].handle).
      Stored with the key is the record number of the record added above, using
      *recordStruct_1.

 Note:  The 0x80000000 in AP[1].recNo signifies that AP[1] is using the same
 data record that was appended during processing of AP[0].  This results in
 just the one data record being added.  AP[1].recPtr must still, however, point
 to the same data as AP[0].recPtr does.  The same applies to AP[2] and AP[3]
 (though different values, of course).

 Upon return, if no error, the return code is 0.  AP[0].recNo is set to the
 physical record number in the data file that *recordStruct_0 was placed. The
 key that was stored for the first index, including any enumerator, is in the
 buffer at AP[0].keyPtr.  AP[1].recNo is set to the same physical record number
 as AP[0].recNo, except that the record number is negative: For example, if
 AP[0].recNo is 22 on return, AP[1].recNo is -22 (the original 0x80000000 value
 is overwritten).  The key that was stored for the second index, including any
 enumerator, is in the buffer at AP[1].keyPtr.  AP[2].recNo is set to the
 physical record number in the data file that *recordStruct_1 was placed.  The
 key that was stored for the third index, including any enumerator, is in the
 buffer at AP[2].keyPtr.  AP[3].recNo is set to the same physical record number
 as AP[2].recNo, except that the record number is negative: For example, if
 AP[2].recNo is 74 on return, AP[3].recNo is -74 (the original 0x80000000 value
 is overwritten).  The key that was stored for the fourth index, including any
 enumerator, is in the buffer at AP[3].keyPtr.

 Upon return, and there was an error, the return code can be -4 to -1, or 1 to
 4.  If negative, the error was caused during processing of that AP pack's data
 file portion, and the error code itself is in AP[abs(rez)-1].stat (where rez
 is the return code, and -1 since C arrays start at 0).  If the return code was
 positive, the error was caused during processing of that AP pack's index file,
 and the error code itself is in AP[rez-1].stat, as well.  The return code is,
 as in all BULLET transaction-list routines, an index of the AP pack that
 generated the error -- negative if a data file error, positive if an index
 file error.

 Note:  If an error occurred after any part of the database had changed (during
 this particular call), then any and all changes that were made are backed-out,
 and the files restored to the same state as before the call.

 Example: Specifying two index files for two records in the same data file

 Three files: two index files related to one data file, where two data records
 are to be appended, would set AccessPack to:

  AP[0].func = INSERT_XB;
  AP[0].handle = indexHandle_0;
  AP[0].recNo = 0;
  AP[0].recPtr = &recordStruct_0;
  AP[0].keyPtr = keyBuffer_0;
  AP[0].nextPtr = AP[1];

  AP[1].handle = indexHandle_1;
  AP[1].recNo = 0x80000000;
  AP[1].recPtr = &recordStruct_0;
  AP[1].keyPtr = keyBuffer_1;
  AP[1].nextPtr = AP[2];

  AP[2].handle = indexHandle_0;
  AP[2].recNo = 0;
  AP[2].recPtr = &recordStruct_1;
  AP[2].keyPtr = keyBuffer_2;
  AP[2].nextPtr = AP[3];

  AP[3].handle = indexHandle_1;
  AP[3].recNo = 0x80000000;
  AP[3].recPtr = &recordStruct_1;
  AP[3].keyPtr = keyBuffer_3;
  AP[3].nextPtr = NULL;

 A call to BULLET with the above does the following:

   1. The data in *recordStruct_0 is used as a new record that is appended to
      the data file linked to the index file in AP[0].handle.

   2. A key is built by BULLET, based on the data in *recordStruct_0, and that
      key is inserted into the index file (AP[0].handle).  Stored with the key
      is the record number of the record added above, for _0.

   3. A second key is built by BULLET, based on the data in *recordStruct_0,
      and that key is inserted into the second index file (AP[1].handle).
      Stored with the key is the record number of the record added above, using
      *recordStruct_0.

   4. The data in *recordStruct_1 is used as a new record that is appended to
      the data file linked to the index file in AP[2].handle.  Since
      AP[2].handle is the same index file as that of AP[0].handle, this means
      it's also the same data file as was just operated on above -- a second
      data record is appended to the data file.  The net effect of this
      operation is to call INSERT_XB twice, once for one insert, then again for
      the second.  The difference is that the operation is atomic -- if one
      fails, the other is not committed; it's an "all or nothing" operation.

   5. A third key is built by BULLET, based on the data in *recordStruct_1, and
      that key is inserted into the index file (AP[2].handle).  Stored with the
      key is the record number of the record added directly above, for _1. Note
      that this index file is the same as specified in AP[0].handle.

   6. A fourth key is built by BULLET, based on the data in *recordStruct_1,
      and that key is inserted into the fourth index file (AP[3].handle).
      Stored with the key is the record number of the record added above, using
      *recordStruct_1.
 The return ritual is as described above, for "Specifying two index files each
 for two different data files".

 Example: Specifying a single index file for a previously added data record

 This form lets you insert a key without adding a data record.  This would be
 required if you were, for example, creating a temporary index of select
 records in a data file (i.e., the data records already exist, you just want to
 index them).  AccessPack setup for this is:

  AP.func = INSERT_XB;
  AP.handle = indexHandle;
  AP.recNo = -recordNumberOfExistingRecord;
  AP.recPtr = &recordStruct;
  AP.keyPtr = keyBuffer;
  AP.nextPtr = NULL;

 A call to BULLET with the above does the following:

   1. A key is built by BULLET, based on the data in *recordStruct, and that
      key is inserted into the index file (AP.handle).  Stored with the key is
      the absolute value of the record number specified in AP.recNo (which is
      set to negative record number).

 Upon return, if no error, the return code is 0.  AP.recNo is changed to
 abs(AP.recNo).  The key that was stored, including any enumerator, is in
 *keyBuffer.  No data file access is made.

 Upon return, and there was an error, the return code is either -1 or 1. If -1,
 the error was caused during processing of the data file portion, and the error
 code itself is in AP.stat.  If +1, the error was caused during processing of
 the index file, and the error code itself is in AP.stat, as well.  The return
 code is, as in all BULLET transaction-list routines, an index of the AP pack
 that generated the error -- negative if a data file error, positive if an
 index file error.  Since this example has only the single pack, only a -1 or
 +1 could be returned.

 Note:  If an error occurred after any part of the database had changed (during
 this particular call), then any and all changes that were made are backed-out,
 and the files restored to the same state as before the call.

 An example use of this INSERT_XB feature is to create an ad hoc index of, say,
 records marked as deleted.  To do this, create a new index file (say, with a
 key of NAME).  Get each data record, by record number using GET_RECORD_XB (for
 records 1 to number-of-records), and check the .tag byte.  If '*', call
 INSERT_XB with the negative value of AP.recNo.  Do this for every such marked
 record.  After all records are processed, you have an index of all deleted
 records in the data file.  Delete the index when no longer needed.  That's
 just one example.


ΓòÉΓòÉΓòÉ 9.63. UPDATE_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        *AP.keyPtr
   AP.recNo
   AP.recPtr
   AP.keyPtr
   AP.nextPtr

 Update any and all files in the transaction list if necessary, including both
 index and data files.

 This routine is used to update data records while also updating the index
 files if a key field has changed due to data record updates.  Up to 256 index
 files may be updated per call, as well as 256 data files, too, for a total of
 512 files managed per single UPDATE_XB call.

 Only index handles are listed in AP.handle.  Each index file has associated
 with it a data file, known internally to BULLET (the xbLink from OPEN_XB).
 There may be more than one index file for a data file, but there is always one
 data file per index handle specified in the list.  In other words, you can
 list five index files, each indexing the same xbLink data file, and have
 BULLET perform an atomic update of that list. Or, another possibility is that
 you have a single index file, indexing a single data file.  Or, you can list
 256 index files, each indexing a single data file (512 total files).

 This and several other routines are transaction-list-based.  This means that
 if a failure occurs prior to the routine's completion, all changes made to the
 database by the routine will be backed-out, and the database (data and index
 files) effectively restored to its original state.

 If the routine failed to complete, the function return value is the number
 (1-based) of the pack that caused the failure.  A positive number indicates
 the failure was from an index operation; a negative number indicates the
 failure was from a data operation.  In each case, the absolute value of the
 return code is the list item that failed (the pack index).  For example, if
 five index handles are in the list(AP[0] to AP[4]), and an error occurred on
 the last pack's index file, the return code would be positive 5, indicating
 the fifth pack (AP[4]) failed.  Since it was a positive 5, the index file was
 being processed when the error occurred.  Being processed means not only
 physical access, but verification, etc.  If the return code was -5, then
 again, the error was in the fifth pack, but since it is negative, the error
 occurred while processing the data file.  In either case, upon return, the
 database is restored to the way it was before the UPDATE_XB call was made.
 Remedy the error, if possible, and UPDATE_XB again.

 Each pack must include a separate key buffer.  You must not share a common key
 buffer.  Doing so disables any chance of recovering the index files in case of
 error, since it is in these buffers that BULLET places any newly built keys,
 and it is from these that BULLET, upon an error condition, deletes these keys
 (required for roll-back).

 The enumerator is automatically maintained by this routine, if required
 (DUPS_ALLOWED and the key already exists with enumerator 0).  The process is
 the same as INSERT_XB's.

 How an update works

 All data records specified in the list are read from disk into memory, except
 those with AP.recNo=0.  Therefore, a memory allocation large enough to store
 all unique data records is made upon entry to this routine (and released at
 exit).  For example, if the list includes two implicit data files, and the
 record lengths of those two data files are 2048 and 4096 bytes, an allocation
 of 6K is made.  In addition, 40KB more is allocated for workspace.  So, for
 this example, 46K is allocated (rounded up to 48KB, the next 4KB page
 boundary).  Since up to 256 unique records are possible, where a unique record
 is identified by handle/record number, be aware of the memory requirements if
 you are updating very large databases (e.g., 256 unique records, each 4KB in
 length, would have UPDATE_XB allocate a bit over 1MB of memory for this call).

 After the data records have been read from disk, each list-item is processed,
 in order.  The disk record image previously read is compared with the record
 image at AP.recPtr.  If the same, that item is skipped, and the next item in
 the list is processed.  If you know beforehand that that record is the same,
 set that item's AP.recNo=0 so you can avoid having its disk image read and
 stored (or do not include it in the list at all).  If the images differ,
 BULLET creates a key for the index file being processed, for each record image
 (the original and the one in AP.recPtr).  If the keys generated are the same,
 no index file update is needed.  If different, the original key for that
 record is deleted from that index file, and the new key inserted. Finally, the
 new record replaces the old, the new directly overwriting the original.  Note
 that the actual sequence of the update event differs somewhat from this
 description in order to optimize the process.

 Specifying Files

 As mentioned, only the index file handles are specified in AP.handle. Data
 files are implicitly specified by their links to the index files, as specified
 when the index file was opened (OP.xbLink).  UPDATE_XB can process up to 256
 index files per call.  Since each index file requires a data file, this means
 that up to 256 data files can be processed per call as well.  Also possible is
 that all 256 index handles refer to the same, single data file.  Yet another
 possibility is that there is 1 index file, and so 1 data file.  The
 possibilities can include those and anything in between.

 Example: Specifying a single index file

 The simplest form is where a single index handle is specified.  This implies a
 single data file, too.  AccessPack setup for this is:

  AP.func = UPDATE_XB;
  AP.handle = indexHandle;
  AP.recNo = recordToUpdate;
  AP.recPtr = &recordStruct;
  AP.keyPtr = keyBuffer;
  AP.nextPtr = NULL;

 A call to BULLET with the above does the following:

   1. The data in *recordStruct is used as the new record that is to replace
      the data record at AP.recNo.  The data file was linked to this index file
      (AP.handle) during the index open, in OP.xbLink.

   2. If the record data in *recordStruct is the same as the original disk
      record, nothing is done.  If the data is new, the key fields are compared
      to that belonging to the original disk record, and if the same, only the
      record data is updated.  If the new record's key differs from the
      original's, the original key for this record is removed from the index,
      and the new key inserted.

 AP.recNo must be set to the record number that you are updating. Any GET_XB
 routine (GET_EQUAL_XB, etc.) may be used to identify the number of a data
 record.  Key access has the obvious advantage of knowing the record number of
 a specific key (for example, Betty Barbar's data). Any record number, from 1
 to number of records in the data file, can be used.  In addition, a negative
 record number can be used.  This is treated exactly the same as a positive
 record number (the absolute value is used). The reason this is allowed is
 because INSERT_XB replaces 0x80000000 record numbers with the negative value
 of the previous insert.

 Upon return, if no error, the return code is 0.  If the record data was new,
 the key for that data record, including any enumerator, is in *keyBuffer.
 This is so even if key fields had not changed.

 Upon return, and there was an error, the return code is either -1 or 1. If -1,
 the error was caused during processing of the data file portion, and the error
 code itself is in AP.stat.  If +1, the error was caused during processing of
 the index file, and the error code itself is in AP.stat, as well.  The return
 code is, as in all BULLET transaction-list routines, an index of the AP pack
 that generated the error -- negative if a data file error, positive if an
 index file error.  Since this example has only the single pack, only a -1 or
 +1 could be returned.

 Note:  If an error occurred after any part of the database had changed (during
 this particular call), then any and all changes that were made are backed-out,
 and the files restored to the same state as before the call.

 Example: Specifying two index files for a single data file

 Two index files, related to the same data file, would set AccessPack to:

  AP[0].func = UPDATE_XB;
  AP[0].handle = indexHandle_0;
  AP[0].recNo = recordToUpdate;
  AP[0].recPtr = &recordStruct;
  AP[0].keyPtr = keyBuffer_0;
  AP[0].nextPtr = AP[1];

  AP[1].handle = indexHandle_1;
  AP[1].recNo = recordToUpdate;
  AP[1].recPtr = &recordStruct;
  AP[1].keyPtr = keyBuffer_1;
  AP[1].nextPtr = NULL;

 A call to BULLET with the above does the following:

   1. The data in *recordStruct is used as the new record that is to replace
      the data record at AP.recNo.  The data file was linked to this index file
      (AP.handle) during the index open, in OP.xbLink.

   2. If the record data in *recordStruct is the same as the original disk
      record, nothing is done.  If the data is new, the key fields are compared
      to that belonging to the original disk record, and if the same, only the
      record data is updated.  If the new record's key differs from the
      original's, the original key for this record is removed from the index,
      and the new key inserted.

   3. The operation performed directly above is repeated, this time for the
      second index file.  The new record data, and the record number to update
      are, for this particular example, the same.

 AP.recNo must be set to the record number that you are updating. Each
 AP[].recNo must be set to a valid record number, even if the record number is
 the same as the previous AP[] pack's (the case where you have more than one
 index file for a data file).  BULLET knows if the record number duplicates a
 number in a previous AP pack, and allocates resources for only the first
 encounter of the data record.  Subsequent encounters refer to the first.

 Upon return, if no error, the return code is 0.  If the new and original data
 records differ, the key for the new data record, including any enumerator, is
 in the buffer at AP[0].keyPtr.  This even if the key fields did not change.
 The same applies to the second index, with the new data key in AP[1].keyPtr.

 Upon return, and there was an error, the return code can be -2, -1, 1, or 2.
 If negative, the error was caused during processing of that AP pack's data
 file portion, and the error code itself is in AP[abs(rez)-1].stat (where rez
 is the return code, and -1 since C arrays start at 0).  If the return code was
 positive, the error was caused during processing of that AP pack's index file,
 and the error code itself is in AP[rez-1].stat, as well.  The return code is,
 as in all BULLET transaction-list routines, an index of the AP pack that
 generated the error -- negative if a data file error, positive if an index
 file error.

 Note:  If an error occurred after any part of the database had changed (during
 this particular call), then any and all changes that were made are backed-out,
 and the files restored to the same state as before the call.

 Example: Specifying two index files for each of two different data files

 Four total files: two index files related to one data file, and two other
 index files related to another data file, would set AccessPack to:

  AP[0].func = UPDATE_XB;
  AP[0].handle = indexHandle_0;
  AP[0].recNo = recordToUpdate_0;
  AP[0].recPtr = &recordStruct_0;
  AP[0].keyPtr = keyBuffer_0;
  AP[0].nextPtr = AP[1];

  AP[1].handle = indexHandle_1;
  AP[1].recNo = recordToUpdate_0;
  AP[1].recPtr = &recordStruct_0;
  AP[1].keyPtr = keyBuffer_1;
  AP[1].nextPtr = AP[2];

  AP[2].handle = indexHandle_2;
  AP[2].recNo = recordToUpdate_1;
  AP[2].recPtr = &recordStruct_1;
  AP[2].keyPtr = keyBuffer_2;
  AP[2].nextPtr = AP[3];

  AP[3].handle = indexHandle_3;
  AP[3].recNo = recordToUpdate_1;
  AP[3].recPtr = &recordStruct_1;
  AP[3].keyPtr = keyBuffer_3;
  AP[3].nextPtr = NULL;

 A call to BULLET with the above does the following:

   1. The data in *recordStruct_0 is used as the new record that is to replace
      the data record at AP[0].recNo in the data file linked to the index file
      in AP[0].handle.

   2. If the record data in *recordStruct_0 is the same as the original disk
      record, nothing is done.  If the data is new, the key fields are compared
      to that belonging to the original disk record, and if the same, only the
      record data is updated.  If the new record's key differs from the
      original's, the original key for this record is removed from the index,
      and the new key inserted.

   3. The operation performed directly above is repeated, this time for the
      second index file.  The new record data, and the record number to update,
      are for this particular example, the same.

   4. The data in *recordStruct_1 is used as the new record that is to replace
      the data record at AP[2].recNo in the data file linked to the index file
      in AP[2].handle.

   5. If the record data in *recordStruct_1 is the same as the original disk
      record, nothing is done.  If the data is new, the key fields are compared
      to that belonging to the original disk record, and if the same, only the
      record data is updated.  If the new record's key differs from the
      original's, the original key for this record is removed from the index,
      and the new key inserted.

   6. The operation performed directly above is repeated, this time for the
      fourth index file.  The new record data, and the record number to update
      are, for this particular example, the same.

 Upon return, if no error, the return code is 0.  If the new and original data
 records differ, the keys for the new data records, including any enumerators,
 are in the buffers at AP[0].keyPtr to AP[3].keyPtr.  This even if the key
 fields did not change.  If one, or all, of the new data records matched the
 original data record, nothing is placed in *keyBuffer for that index.

 Upon return, and there was an error, the return code can be -4 to -1, or 1 to
 4.  If negative, the error was caused during processing of that AP pack's data
 file portion, and the error code itself is in AP[abs(rez)-1].stat (where rez
 is the return code, and rez-1 since C arrays start at 0).  If the return code
 was positive, the error was caused during processing of that AP pack's index
 file, and the error code itself is in AP[rez-1].stat, as well.  The return
 code is, as in all BULLET transaction-list routines, an index of the AP pack
 that generated the error -- negative if a data file error, positive if an
 index file error.

 Note:  If an error occurred after any part of the database had changed (during
 this particular call), then any and all changes that were made are backed-out,
 and the files restored to the same state as before the call.

 Example: Specifying two index files for two records in the same data file

 Three files: two index files related to one data file, where two data records
 are to be updated, would set AccessPack to:

  AP[0].func = UPDATE_XB;
  AP[0].handle = indexHandle_0;
  AP[0].recNo = recordToUpdate_0;
  AP[0].recPtr = &recordStruct_0;
  AP[0].keyPtr = keyBuffer_0;
  AP[0].nextPtr = AP[1];

  AP[1].handle = indexHandle_1;
  AP[1].recNo = recordToUpdate_0;
  AP[1].recPtr = &recordStruct_0;
  AP[1].keyPtr = keyBuffer_1;
  AP[1].nextPtr = AP[2];

  AP[2].handle = indexHandle_0;
  AP[2].recNo = recordToUpdate_1;
  AP[2].recPtr = &recordStruct_1;
  AP[2].keyPtr = keyBuffer_2;
  AP[2].nextPtr = AP[3];

  AP[3].handle = indexHandle_1;
  AP[3].recNo = recordToUpdate_1;
  AP[3].recPtr = &recordStruct_1;
  AP[3].keyPtr = keyBuffer_3;
  AP[3].nextPtr = NULL;

 A call to BULLET with the above does the following:

   1. The data in *recordStruct_0 is used as the new record that is to replace
      the data record at AP[0].recNo in the data file linked to the index file
      in AP[0].handle.

   2. If the record data in *recordStruct_0 is the same as the original disk
      record, nothing is done.  If the data is new, the key fields are compared
      to that belonging to the original disk record, and if the same, only the
      record data is updated.  If the new record's key differs from the
      original's, the original key for this record is removed from the index,
      and the new key inserted.

   3. The operation performed directly above is repeated, this time for the
      second index file.  The new record data, and the record number to update,
      are for this particular example, the same.

   4. The data in *recordStruct_1 is used as the new record that is to replace
      the data record at AP[2].recNo in the data file linked to the index file
      in AP[2].handle.  This is the same index file as the first AP pack, and
      also the same data file.  However, this is a different record number.

   5. If the record data in *recordStruct_1 is the same as the original disk
      record, nothing is done.  If the data is new, the key fields are compared
      to that belonging to the original disk record, and if the same, only the
      record data is updated.  If the new record's key differs from the
      original's, the original key for this record is removed from the index,
      and the new key inserted.

   6. The operation performed directly above is repeated, this time for the
      fourth index file. The new record data, and the record number to update
      are, for this particular example, the same.
 The return ritual is as described above, for "Specifying two index files each
 for two different data files".


ΓòÉΓòÉΓòÉ 9.64. REINDEX_XB ΓòÉΓòÉΓòÉ

Pack: ACCESSPACK               Source Example

     IN               OUT
   AP.func          AP.stat
   AP.handle        AP.recNo
   AP.keyPtr        *AP.keyPtr
   AP.nextPtr

 Reindex all files in the transaction list, re-evaluating the key expression in
 the process.

 This routine is used to reindex up to 256 index files per call. The index
 files must already exist and be open.  Any existing key values are overwritten
 by new key data.  In other words, if you have a 100MB index file, REINDEX_XB
 uses the same file space, building new keys over old.  This results in a less
 fragmented disk and also minimizes disk space needed.  You can also create a
 new, empty index file and reindex to that.  This would be useful, for
 instance, if you needed to create a temporary index file -- something that
 you'd use for a report, say, then delete after the report.  Another use for
 creating a new index file and reindexing to that is to, after creating it
 (COPY_INDEX_HEADER_XB can be used), use EXPAND_FILE_DOS and expand it to the
 expected size.  This has the benefit of ensuring that this file allocation is
 as contiguous as the file system allows (without relying on OS/API-specific
 calls).

 If the routine failed to complete, the function return value is the number
 (1-based) of the pack that caused the failure.  A positive number indicates
 the failure was from an index operation; a negative number indicates the
 failure was from a data operation (reading the data file).  In each case, the
 absolute value of the return code is the list item that failed (the pack
 index).  For example, if five index handles are in the list(AP[0] to AP[4]),
 and an error occurred on the last pack's index file, the return code would be
 positive 5, indicating the fifth pack (AP[4]) failed.  Since it was a positive
 5, the index file was being processed when the error occurred.  Being
 processed means not only physical access, but verification, etc.  If the
 return code was -5, then again, the error was in the fifth pack, but since it
 is negative, the error occurred while processing the data file.

 Unlike INSERT_XB and UPDATE_XB, each pack need not include a separate key
 buffer; you may share a common key buffer.  If duplicate keys are generated in
 the reindex process and the sort function does not flag DUPS_ALLOWED, an error
 is returned.  The duplicate key is in *AP.keyPtr and the record number it was
 generated from in AP.recNo (both these are valid only on error).  Since no
 roll-back is performed, there is only a real need for a single key buffer.
 You may use separate ones, too.

 This routine creates a temporary work file in either the current directory or,
 if the environment variable TMP is defined, in the directory pointed to by
 TMP=.  The path used for this temporary file may also be specified at run-time
 by using the TMP_PATH_PTR item for SET_SYSVARS_XB.  If TMP_PATH_PTR is NULL
 (default), then TMP= is used, or if that is not found, then the current
 directory is used.  The size of this temporary file is, in bytes,
 approximately (keylength+4) * number of records in the data file.  The
 resultant index files are, by default, optimized for minimum size and maximum
 retrieval speed.  This full-node packing leaves one empty key per node, which
 means b-tree splitting will occur almost immediately upon inserting data (with
 INSERT_XB or STORE_KEY_XB).

 This behaviour can be modified with the REINDEX_PACK_PCT item for
 SET_SYSVARS_XB so that less packing is done.  Less packing would improve
 subsequent INSERT_XB performance since all nodes are not almost full as they
 are with a full pack.  File size and retrieval times increase, though, but
 perhaps not noticeably.

 During the reindex process, each record is checked for a matching skip-tag
 value, as set in SET_SYSVARS_XB.  The skip-tag is set to 0 by default, where
 no check is done and keys for all records in the data file are inserted into
 the index file.


ΓòÉΓòÉΓòÉ 9.65. LOCK_XB ΓòÉΓòÉΓòÉ

Pack: LOCKPACK                 Source Example

     IN               OUT
   LP.func          LP.stat
   LP.handle
   LP.xlMode
   LP.dlMode
   LP.recStart=0
   LP.nextPtr

 Lock all bytes of the files in the list for exclusive use by the current
 process, and reload file headers from disk.  LP.recStart must be 0 for each
 pack.

 This routine is used to lock the database for either exclusive use by this
 process, or shared access (allowing any process to read, but not write, to the
 files).  Up to 256 index files may be locked per call, as well as 256 data
 files, too, for a total of 512 files per single LOCK_XB call.  Shared-access
 locking prevents all processes from writing to the file while a shared lock is
 in force, including this process.  To relock in exclusive lock mode, without
 unlocking first, use: RELOCK_XB.

 Only index handles are listed in LP.handle.  Each index file has associated
 with it a data file, known internally to BULLET (the xbLink from OPEN_XB).
 There may be more than one index file for a data file, but there is always one
 data file per index handle specified in the list.  For example, you can list
 five index files, each indexing the same xbLink data file, and have BULLET
 perform an atomic lock of that list.

 LP.xlMode is set to 1 to perform a shared lock on the index file.  Set to 0
 for an exclusive lock.  A shared lock allows only reading.

 LP.dlMode is set to 1 to perform a shared lock on the data file.  Set to 0 for
 an exclusive lock.  A shared lock allows only reading.

 The lock mode (shared <-> exclusive) can be changed using RELOCK_XB.

 Bullet maintains a per-handle lock count, and does a physical region lock only
 upon the first lock request (or on a relock request).  It is only on this
 first lock request that the header is reloaded.  When the lock count returns
 to 0 (from UNLOCK_XB calls), it is at that time the header is flushed, if
 required.  Only full-locks are maintained in this way.  The number of
 outstanding locks can be determined from the SIP and SDP structures, from the
 STAT_XB routines.  Note that individual LOCK_INDEX_XB and UNLOCK_INDEX_XB
 routines, as well as the data ones, also act upon this lock count.  Therefore,
 you can lock a file 100 times in a row, but only on the first lock are any
 operations actually performed, and only on the last unlock are any performed.
 Other lock/unlock calls (other than the first lock or last unlock) simply
 increment or decrement the lock count for that handle.

 This and several other routines are transaction-list-based.  This means that
 if a failure occurs prior to the routine's completion, all locks made to the
 database by this routine will be unlocked.

 If the routine failed to complete, the function return value is the number
 (1-based) of the pack that caused the failure.  A positive number indicates
 the failure was from an index operation; a negative number indicates the
 failure was from a data operation.  In each case, the absolute value of the
 return code is the list item that failed (the pack index).  For example, if
 five index handles are in the list(AP[0] to AP[4]), and an error occurred on
 the last pack's index file, the return code would be positive 5, indicating
 the fifth pack (AP[4]) failed.  Since it was a positive 5, the index file was
 being processed when the error occurred.  Being processed means not only
 physical access, but verification, etc.  If the return code was -5, then
 again, the error was in the fifth pack, but since it is negative, the error
 occurred while processing the data file.  In either case, upon return, any
 files locked during this call are unlocked before returning.

 The advantage of using region locks (LOCK_XB locks entire file regions) to
 control file access is that the file does not need to be opened/closed using
 the Deny Read/Write sharing attribute.  Opening the file for Deny None, and
 controlling subsequent access with region locks, allows for faster processing
 since files do not need to be constantly opened and closed, as they would if
 access were controlled by opening with Deny Read/Write.

 Using the operating system to control access also prevents other processes
 from accessing the files.  Other methods, such as internal locking (such as
 using 'L' in the tag field as a program-aware in-use flag), work fine so long
 as each process accessing the files knows about this internal "locking".
 However, since it's proprietary, other processes may not know about it.  Any
 locking system that is not commonly shared throughout the system is not
 effective when it comes to preventing other processes from corrupting files.

 Note:  Region locking prevents other processes from both writing and reading
 the locked file.  For operating systems that do not provide shared locks, and
 read-access is required at all times, you may specify this type access with
 the access-sharing mode when the BULLET file is opened.  Once opened for this
 (R/W, DenyWrite) then only the current process can write to the file until it
 is closed.  Other processes must open the file for Read-Only access.  For
 small networks (two or three nodes), this may be suitable.  Otherwise, region
 locking is much preferred, and very much faster, since files do not have to be
 opened and closed every time the access state needs to change.


ΓòÉΓòÉΓòÉ 9.66. UNLOCK_XB ΓòÉΓòÉΓòÉ

Pack: LOCKPACK                 Source Example

     IN               OUT
   LP.func          LP.stat
   LP.handle
   LP.recStart=0
   LP.nextPtr

 Unlock all bytes in the specified files (previously locked) and flush the
 files' headers to disk (the flush is done before the locks are released). Also
 unlock all bytes in the related data file and flush the data file header to
 disk.  LP.recStart must be 0 for each pack.

 Note:  If a shared-lock is active for this handle (as set by this process),
 the flush is not performed.  This because writing to the locked region is not
 permitted (nor is the flush required since nothing could have been changed).

 If the routine failed to complete, the function return value is the number
 (1-based) of the pack that caused the failure.  A positive number indicates
 the failure was from an index operation; a negative number indicates the
 failure was from a data operation.  In each case, the absolute value of the
 return code is the list item that failed (the pack index).  For example, if
 five index handles are in the list(AP[0] to AP[4]), and an error occurred on
 the last pack's index file, the return code would be positive 5, indicating
 the fifth pack (AP[4]) failed.  Since it was a positive 5, the index file was
 being processed when the error occurred.  Being processed means not only
 physical access, but verification, etc.  If the return code was -5, then
 again, the error was in the fifth pack, but since it is negative, the error
 occurred while processing the data file.

 This routine does not attempt to re-lock those files unlocked successfully if
 an error occurs in the transaction.  If an error does occur (unlikely), the
 remaining files must be manually unlocked with the UNLOCK_KEY_XB and
 UNLOCK_DATA_XB routines.

 Bullet maintains a per-handle lock count, and does a physical region lock only
 upon the first lock request (or on a relock request).  It is only on this
 first lock request that the header is reloaded.  When the lock count returns
 to 0 (from UNLOCK_XB calls), it is at that time the header is flushed, if
 required.  Only full-locks are maintained in this way.  The number of
 outstanding locks can be determined from the SIP and SDP structures, from the
 STAT_XB routines.  Note that individual LOCK_INDEX_XB and UNLOCK_INDEX_XB
 routines, as well as the data ones, also act upon this lock count.  Therefore,
 you can lock a file 100 times in a row, but only on the first lock are any
 operations actually performed, and only on the last unlock are any performed.
 Other lock/unlock calls (other than the first lock or last unlock) simply
 increment or decrement the lock count for that handle.


ΓòÉΓòÉΓòÉ 9.67. LOCK_INDEX_XB ΓòÉΓòÉΓòÉ

Pack: LOCKPACK                 Source Example

     IN               OUT
   LP.func          LP.stat
   LP.handle
   LP.xlMode

 Lock all bytes of the index file for exclusive use by the current process and
 reload the index file's header from disk.

 LP.xlMode is set to 1 to perform a shared lock.  Set to 0 for an exclusive
 lock.  A shared lock allows only reading.  The lock mode (shared <->
 exclusive) can be changed using RELOCK_INDEX_XB.

 Bullet maintains a per-handle lock count, and does a physical region lock only
 upon the first lock request (or on a relock request).  It is only on this
 first lock request that the header is reloaded.  When the lock count returns
 to 0 (from UNLOCK_XB calls), it is at that time the header is flushed, if
 required.  Only full-locks are maintained in this way.  The number of
 outstanding locks can be determined from the SIP and SDP structures, from the
 STAT_XB routines.  Note that individual LOCK_INDEX_XB and UNLOCK_INDEX_XB
 routines, as well as the data ones, also act upon this lock count.  Therefore,
 you can lock a file 100 times in a row, but only on the first lock are any
 operations actually performed, and only on the last unlock are any performed.
 Other lock/unlock calls (other than the first lock or last unlock) simply
 increment or decrement the lock count for that handle.


ΓòÉΓòÉΓòÉ 9.68. UNLOCK_INDEX_XB ΓòÉΓòÉΓòÉ

Pack: LOCKPACK                 Source Example

     IN               OUT
   LP.func          LP.stat
   LP.handle

 Unlock all bytes in the specified file (previously locked) and flush the
 file's header to disk (the flush is done before the locks are released).

 If the current lock state is shared, the flush is not performed.

 Bullet maintains a per-handle lock count, and does a physical region lock only
 upon the first lock request (or on a relock request).  It is only on this
 first lock request that the header is reloaded.  When the lock count returns
 to 0 (from UNLOCK_XB calls), it is at that time the header is flushed, if
 required.  Only full-locks are maintained in this way.  The number of
 outstanding locks can be determined from the SIP and SDP structures, from the
 STAT_XB routines.  Note that individual LOCK_INDEX_XB and UNLOCK_INDEX_XB
 routines, as well as the data ones, also act upon this lock count.  Therefore,
 you can lock a file 100 times in a row, but only on the first lock are any
 operations actually performed, and only on the last unlock are any performed.
 Other lock/unlock calls (other than the first lock or last unlock) simply
 increment or decrement the lock count for that handle.


ΓòÉΓòÉΓòÉ 9.69. LOCK_DATA_XB ΓòÉΓòÉΓòÉ

Pack: LOCKPACK                 Source Example

     IN               OUT
   LP.func          LP.stat
   LP.handle
   LP.dlMode
   LP.recStart
   LP.recCount

 Lock all bytes of the data file specified in LP.handle for exclusive use by
 the current process.  It also reloads the data file header from disk. You must
 set LP.recStart=0 to lock all bytes.  To lock a single record, or a set of
 contiguous records, set LP.recStart to the first record to lock and
 LP.recCount to the number of records to lock.

 Header re-loading is performed only if locking all bytes.

 If LP.recStart is not 0, be aware that the header is not locked, nor is it
 re-loaded.  Also, maintaining a lock on the single record prevents any other
 process from doing a full lock on that data file, thereby preventing write
 access to the file from any other BULLET process using LOCK_XB, but not
 necessarily preventing other applications from writing to that file.  That may
 or may not be good.  It does not prevent other BULLET processes from reading
 that file (except for that locked record).

 Multiple single records are allowed, but each must then be unlocked
 individually, in the same format (start, count) as the lock.

 LP.dlMode is set to 1 to perform a shared lock.  Set to 0 for an exclusive
 lock.  A shared lock allows only reading.  The lock mode (shared <->
 exclusive) can be changed using RELOCK_DATA_XB.

 Bullet maintains a per-handle lock count, and does a physical region lock only
 upon the first lock request (or on a relock request).  It is only on this
 first lock request that the header is reloaded.  When the lock count returns
 to 0 (from UNLOCK_XB calls), it is at that time the header is flushed, if
 required.  Only full-locks are maintained in this way.  The number of
 outstanding locks can be determined from the SIP and SDP structures, from the
 STAT_XB routines.  Note that individual LOCK_INDEX_XB and UNLOCK_INDEX_XB
 routines, as well as the data ones if a full lock, also act upon this lock
 count.  Therefore, you can lock a file 100 times in a row, but only on the
 first lock are any operations actually performed, and only on the last unlock
 are any performed.  Other lock/unlock calls (other than the first lock or last
 unlock) simply increment or decrement the lock count for that handle.


ΓòÉΓòÉΓòÉ 9.70. UNLOCK_DATA_XB ΓòÉΓòÉΓòÉ

Pack: LOCKPACK                 Source Example

     IN               OUT
   LP.func          LP.stat
   LP.handle
   LP.recStart
   LP.recCount

 Unlock all bytes in the specified file handle (previously locked) and flush
 the data file header to disk (the flush is done before the lock is released).
 You must set LP.recStart=0 to unlock all bytes.  To unlock a single record, or
 a set of contiguous records, set LP.recStart to the first record to unlock and
 LP.recCount to the number of records to unlock.

 Header flushing is performed only if unlocking a full lock.

 An unlock must exactly mimic its corresponding lock.  This means if you lock
 several records singly, you must unlock each of those records -- you cannot
 use LP.recStart=0 to unlock singly-locked records.

 Bullet maintains a per-handle lock count, and does a physical region lock only
 upon the first lock request (or on a relock request).  It is only on this
 first lock request that the header is reloaded.  When the lock count returns
 to 0 (from UNLOCK_XB calls), it is at that time the header is flushed, if
 required.  Only full-locks are maintained in this way.  The number of
 outstanding locks can be determined from the SIP and SDP structures, from the
 STAT_XB routines.  Note that individual LOCK_INDEX_XB and UNLOCK_INDEX_XB
 routines, as well as the data ones if a full lock, also act upon this lock
 count.  Therefore, you can lock a file 100 times in a row, but only on the
 first lock are any operations actually performed, and only on the last unlock
 are any performed.  Other lock/unlock calls (other than the first lock or last
 unlock) simply increment or decrement the lock count for that handle.


ΓòÉΓòÉΓòÉ 9.71. CHECK_REMOTE_XB ΓòÉΓòÉΓòÉ

Pack: REMOTEPACK               Source Example

     IN               OUT
   RP.func          RP.stat
   RP.handle        RP.isRemote
      -or-          RP.flags=0
   RP.drive         RP.isShare=1

 If RP.handle is non-zero, determine if the specified handle of a file or
 device is remote.  If the handle is local (e.g., not a network file or
 device), RP.isRemote returns 0, otherwise it is remote.  RP.flags=0 and
 RP.isShare=1 on return for either a handle or drive check under OS/2.

 If RP.handle is zero, determine if the drive specified in RP.drive is remote.
 Drive A: is 1, B: is 2, C: is 3, and so on.  To check the default drive use 0
 (the default drive is the current drive).  If the drive is local (e.g., not a
 network drive), RP.isRemote returns 0, otherwise it is remote.

 The significance of the remote-ness is less important in a multitasking
 environment since sharing of resources (files, in particular) must always be
 managed, compared to single-tasking environments where, typically, sharing
 (locking mechanisms) need only be performed when the resource is able to be
 accessed by another process (ie is a 'network' drive).  Note that the resource
 need not be located elsewhere to be classified as remote: Drives or devices or
 files on the same machine may be classified as remote if the network software
 is redirecting local access (such as on a server).

 Note:  This routine is not mutex-protected.


ΓòÉΓòÉΓòÉ 9.72. RELOCK_XB ΓòÉΓòÉΓòÉ

Pack: LOCKPACK                 Source Example

     IN               OUT
   LP.func          LP.stat
   LP.handle
   LP.xlMode
   LP.dlMode
   LP.recStart=0
   LP.nextPtr

 Relock all bytes of the index files for the mode specified in LP.xlMode (index
 files) and LP.dlMode (data files).  Also relock all bytes in the related data
 file.  LP.recStart must be 0 for each pack.

 Set LP.xlMode=1 to select a shared lock for the index file; set to 0 for an
 exclusive lock.  Set LP.dlMode=1 to select a shared lock for the data file;
 set to 0 for an exclusive lock.

 If the lock mode is from exclusive to shared, the file is flushed before the
 shared state is set.  BULLET maintains the current lock state and knows which
 direction the lock is going in.  The lock state (shared or exclusive) can be
 determined by the SIP and SDP structures from the STAT_XB routines.  This
 routine does not affect the lock count, nor are the headers reloaded (nor
 should they be).

 The lock state is on a file handle basis, not on an LP[] pack basis. This
 means the file, as identified by the handle, is in the lock state last set.

 Note:  The lock switch is made atomic:  Rather than unlocking, and then
 locking again in the new state, this performs all operations without the
 possibility that another process can grab the lock away, if supported by the
 OS.  If relock is not supported by the OS, and shared (read-only) locks are
 supported, the relock will be performed by Bullet rather than the OS.  If
 Bullet performs the relock, another, non-Bullet process may be able to steal
 away the lock since it is no longer guaranteed to be atomic.


ΓòÉΓòÉΓòÉ 9.73. RELOCK_INDEX_XB ΓòÉΓòÉΓòÉ

Pack: LOCKPACK                 Source Example

     IN               OUT
   LP.func          LP.stat
   LP.handle
   LP.xlMode

 Relock all bytes of the index file for the mode specified in LP.xlMode and
 reload the header.

 Set LP.xlMode=1 to select a shared lock; set to 0 for an exclusive lock. If
 the lock mode is from exclusive to shared, the file is flushed before the
 shared state is set.  BULLET maintains the current lock state and knows which
 direction the lock is going in.  The lock state (shared or exclusive) can be
 determined by the SIP structure from the STAT_INDEX_XB routine.  This routine
 does not affect the lock count, nor is the header reloaded (nor should it be).

 Note:  The lock switch is made atomic:  Rather than unlocking, and then
 locking again in the new state, this performs all operations without the
 possibility that another process can grab the lock away, if supported by the
 OS.  If relock is not supported by the OS, and shared (read-only) locks are
 supported, the relock will be performed by Bullet rather than the OS.  If
 Bullet performs the relock, another, non-Bullet process may be able to steal
 away the lock since it is no longer guaranteed to be atomic.


ΓòÉΓòÉΓòÉ 9.74. RELOCK_DATA_XB ΓòÉΓòÉΓòÉ

Pack: LOCKPACK                 Source Example

     IN               OUT
   LP.func          LP.stat
   LP.handle
   LP.dlMode
   LP.recStart
   LP.recCount

 Relock all bytes of the data file for the mode specified in LP.dlMode and
 reload the header, unless LP.recStart is non-zero.

 If the lock mode is from exclusive to shared, the file is flushed before the
 shared state is set.  BULLET maintains the current lock state and knows which
 direction the lock is going in.  The lock state (shared or exclusive) can be
 determined by the SDP structure from the STAT_DATA_XB routine.  This routine
 does not affect the lock count, nor is the header reloaded (nor should it be).

 You must set LP.recStart=0 to relock all bytes.  To relock a single record, or
 set of contiguous records, set LP.recStart=record# to relock and LP.recCount
 to the number of records to relock.

 Note:  The lock switch is made atomic:  Rather than unlocking, and then
 locking again in the new state, this performs all operations without the
 possibility that another process can grab the lock away, if supported by the
 OS.  If relock is not supported by the OS, and shared (read-only) locks are
 supported, the relock will be performed by Bullet rather than the OS.  If
 Bullet performs the relock, another, non-Bullet process may be able to steal
 away the lock since it is no longer guaranteed to be atomic.


ΓòÉΓòÉΓòÉ 9.75. DELETE_FILE_DOS ΓòÉΓòÉΓòÉ

Pack: DOSFILEPACK              Source Example

     IN               OUT
   DFP.func         DFP.stat
   DFP.filenamePtr

 Delete the specified file.

 Note:  OS/2 DosForceDelete is used so the file is not recoverable with the
 UNDELETE command.


ΓòÉΓòÉΓòÉ 9.76. RENAME_FILE_DOS ΓòÉΓòÉΓòÉ

Pack: DOSFILEPACK              Source Example

     IN               OUT
   DFP.func         DFP.stat
   DFP.filenamePtr
   DFP.newNamePtr

 Rename a file.  May also be used to move the file to a new directory within
 the partition.

 If the specified directory differs from the file's directory, the file's
 directory entry is moved to the new directory.

 For example, if the filenamePtr filename is /LP100/PROJ94A.INF and the
 newFilenamePtr filename is /ARCH/PROJ93A.INA, the file is essentially renamed
 and also moved to the /ARCH directory.


ΓòÉΓòÉΓòÉ 9.77. CREATE_FILE_DOS ΓòÉΓòÉΓòÉ

Pack: DOSFILEPACK              Source Example

     IN               OUT
   DFP.func         DFP.stat
   DFP.filenamePtr
   DFP.attr

 Create a new file.

 The specified filename/pathname must not already exist.

 The file created is not left open.  You must OPEN_FILE_DOS to use it.

 The attribute used during the create can be:

   Attribute  Value   Meaning
    Normal       0      normal access permitted to file
    Read-Only    1      read-only access permitted to file
    Hidden       2      file does not appear in directory listing
    System       4      file is a system file
    SubDir      10h     FILENAME is a subdirectory
    Archive     20h     file is marked for archiving

 Note:  Use MAKE_DIR_DOS to create a subdirectory.


ΓòÉΓòÉΓòÉ 9.78. OPEN_FILE_DOS ΓòÉΓòÉΓòÉ

Pack: DOSFILEPACK              Source Example

     IN               OUT
   DFP.func         DFP.stat
   DFP.filenamePtr  DFP.handle
   DFP.asMode

 Open the file with the access-sharing mode, returning the handle on success.


ΓòÉΓòÉΓòÉ 9.79. SEEK_FILE_DOS ΓòÉΓòÉΓòÉ

Pack: DOSFILEPACK              Source Example

     IN               OUT
   DFP.func         DFP.stat
   DFP.handle       DFP.seekTo
   DFP.seekTo
   DFP.method

 Position the file pointer of the file to the seekTo position based on the
 method specified.

 The position is a 32-bit value and is relative to either the start of the
 file, the current file pointer position, or the end of the file.

  Method   Meaning
    0      start move from the start of file (offset is a 32-bit unsigned value)
    1      start move at the current position (offset a signed value)
    2      start move at the end of file (offset a signed value)
 For example, to move to the last byte of a sector (512th byte, but offset
 511), set the offset value to 511 and use Method 0.  On return, the absolute
 offset value of the new position is returned.  This return value is useful
 with Method 2 since you can specify an offset of 0 and have the file length
 returned.  To move to the start of the file, use method 0, offset 0.  To move
 to the first byte of the second sector, use offset 512.

 Note:  Never position the file pointer to before the start of file.


ΓòÉΓòÉΓòÉ 9.80. READ_FILE_DOS ΓòÉΓòÉΓòÉ

Pack: DOSFILEPACK              Source Example

     IN               OUT
   DFP.func         DFP.stat
   DFP.handle       DFP.bytes
   DFP.bytes
   DFP.bufferPtr

 Read from the file or device the specified number of bytes into a buffer.

 On block devices (such as disks) input starts at the current file position and
 the file pointer is repositioned to the last byte read +1.

 It is possible to read less than the bytes specified without an error being
 generated.  Compare the bytes to read with the returned bytes read value.  If
 less then end of file was reached during the read.  If 0 then file was already
 at EOF.


ΓòÉΓòÉΓòÉ 9.81. WRITE_FILE_DOS ΓòÉΓòÉΓòÉ

Pack: DOSFILEPACK              Source Example

     IN               OUT
   DFP.func         DFP.stat
   DFP.handle       DFP.bytes
   DFP.bytes
   DFP.bufferPtr

 Write to the file or device the specified number of bytes from a buffer.

 If the number of bytes written is less than the specified bytes, this routine
 returns an error.

 On block devices (such as disk) output starts at the current file position,
 and the file pointer is repositioned to the last byte written +1.

 Note:  If the specified bytes to write is 0, the file is truncated at the
 current file pointer position.


ΓòÉΓòÉΓòÉ 9.82. CLOSE_FILE_DOS ΓòÉΓòÉΓòÉ

Pack: DOSFILEPACK              Source Example

     IN               OUT
   DFP.func         DFP.stat
   DFP.handle

 Close the file flushing any internal buffers, releasing any locked regions,
 and updating the directory entry to the correct size, date, and time.


ΓòÉΓòÉΓòÉ 9.83. ACCESS_FILE_DOS ΓòÉΓòÉΓòÉ

Pack: DOSFILEPACK              Source Example

     IN               OUT
   DFP.func         DFP.stat
   DFP.filenamePtr
   DFP.asMode

 Determine if the specified file can be accessed with the specified
 access-sharing mode.

 Basically, a Does-File-Exist routine.  It uses the specified access-sharing
 mode when trying to open the file.  For example, if you specify DFP.attr =
 0x0042 (R/W access + Deny None sharing) and issue ACCESS_FILE_DOS on a
 Read-Only file, an error is returned.  A sharing mode must be specified; it
 cannot be left 0.


ΓòÉΓòÉΓòÉ 9.84. EXPAND_FILE_DOS ΓòÉΓòÉΓòÉ

Pack: DOSFILEPACK              Source Example

     IN               OUT
   DFP.func         DFP.stat
   DFP.handle
   DFP.bytes

 Expands the file by the number of bytes beyond its current size.

 This routine is useful in pre-allocating disk space.  By reserving disk space
 in advance you can guarantee that enough disk space will be available for a
 future operation (especially if more than 1 process is running). You'll also
 be able ensure that the disk space that a file does use is as contiguous as
 possible.

 Database systems are dynamic and their files typically allocate new space on
 an as-needed basis.  This dynamic allocation can cause parts of a file to be
 located throughout the disk system, possibly affecting performance
 drastically.  By pre-allocating the disk space you can be assured of
 consistent throughput performance since the file is contiguous.

 Note:  The file space is not initialized.

 * ---------------------------------------------------------------------------


ΓòÉΓòÉΓòÉ 9.85. MAKE_DIR_DOS ΓòÉΓòÉΓòÉ

Pack: DOSFILEPACK              Source Example

     IN               OUT
   DFP.func         DFP.stat
   DFP.filenamePtr

 Create a new subdirectory.


ΓòÉΓòÉΓòÉ 9.86. COMMIT_FILE_DOS ΓòÉΓòÉΓòÉ

Pack: DOSFILEPACK              Source Example

     IN               OUT
   DFP.func         DFP.stat
   DFP.handle

 Flushes the OS system buffers for the handle, and updates the directory entry
 for size.


ΓòÉΓòÉΓòÉ 10. Bullet Source Examples ΓòÉΓòÉΓòÉ

Bullet source example excerpts for the C language follow.  Complete program
source examples are provide on the included disk.

In the examples, consider that

        CHAR keyBuffer[64];
        AP.keyPtr = keyBuffer;

provides the compiler with a pointer whenever 'keyBuffer' is referenced, hence
no need to use &keyBuffer.  Contrast this with a structure definition

        struct yourRecordLayout yourRecord;
        AP.recPtr = &yourRecord;

where &yourRecord is required.

Be aware of how you use zero-terminated strings, especially for fields that are
not to be 0T'ed (DATE and NUMERIC fields, for example).  Use sscanf() and
sprintf(), as required.


ΓòÉΓòÉΓòÉ 10.1. Bullet Initialization ΓòÉΓòÉΓòÉ

/* All example source uses minimal error checking/handling throughout */
// comment marks are used throughout

// To simplify the source examples, error handling is typically of the form:
//
//   if (rez) goto ErrorHandler;
//
// Use suitable coding as you would normally handle the error.

#include "bullet_2.h"

INITPACK IP;                    // packs used here

IP.func = INIT_XB;              // start up Bullet
IP.JFTsize = 30;                // allow at least 30 handles to be open
rez = BULLET(&IP);
if (rez) return(rez);


ΓòÉΓòÉΓòÉ 10.2. Bullet Shutdown ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

EXITPACK EP;                    // packs used here

EP.func = EXIT_XB;              // shutdown Bullet
rez = BULLET(&EP);
if (rez) return(rez);


ΓòÉΓòÉΓòÉ 10.3. Memory Available ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

MEMORYPACK EP;                  // packs used here

MP.func = MEMORY_XB;
rez = BULLET(&MP);
if (rez==0) {
   printf("Private memory arena space right now is %d bytes\n",MP.memory);
}


ΓòÉΓòÉΓòÉ 10.4. File Backup Procedure ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

// code determines file type of handle, and selects appropriate backup method
// this type of work (copy/backup) would be a prelude to reindexing an index
// file or performing a pack on a data file since both of those routines
// overwrite existing data

COPYPACK CP;
LOCKPACK LP;
STATHANDLEPACK SHP;
STATDATAPACK SDP;
STATINDEXPACK SIP;

CHAR CopyOfIndexHdrName[260];
CHAR BackupDataName[260];

// Given a handle, find out its type and, if index, generate a copy
// of the index file's header, or if data, backup the entire data file.
// Files are locked to ensure that access if permissible.

SHP.func = STAT_HANDLE_XB;
SHP.handle = passedHandle;
rez = BULLET(&SHP);
if (SHP.ID==-1)
   puts("Handle is not a Bullet data or index file.  Use DosCopy() API call.");
else {
   if (SHP.ID==0) {

      // Since the copy takes place on open files, ensure a full-lock is in place
      // A shared lock is okay since this file is only read

      // locking the entire file also RELOADS the index header

      LP.func = LOCK_INDEX_XB;
      LP.handle = passedHandle;
      LP.xlMode = LOCK_SHARED;          // in bullet_2.h
      rez = BULLET(&LP);
      if (rez) goto ErrorHandler;

      SIP.func = STAT_INDEX_XB;
      SIP.handle = passedHandle;
      rez = BULLET(&SIP);
      if (rez) goto ErrorHandler;       // must unlock file in handler

      // SIP.filenamePtr -> pathname of this handle, from which you can
      // derive a suitable name for the index header written next -- in
      // case it's not obvious, this function is not part of Bullet

      DeriveSuitableName(SIP.filenamePtr,CopyOfIndexHdrName);

      CP.func = COPY_INDEX_HEADER_XB;
      CP.handle = passedHandle;
      CP.filenamePtr = CopyOfIndexHdrName;
      rez = BULLET(&CP);
      if (rez) goto ErrorHandler;       // must unlock file in handler

      // unlocking also flushes the index header if not LOCK_SHARED (and if required)

      LP.func = UNLOCK_INDEX_XB;
      LP.handle = passedHandle;
      rez = BULLET(&LP);
      if (rez) goto ErrorHandler;
   }
   else {

      // locking the entire file also RELOADS the data header

      LP.func = LOCK_DATA_XB;
      LP.handle = passedHandle;
      LP.dlMode = LOCK_SHARED;
      LP.startRec = 0;                  // lock entire data file
      rez = BULLET(&LP);
      if (rez) goto ErrorHandler;

      SDP.func = STAT_DATA_XB;
      SDP.handle = passedHandle;
      rez = BULLET(&SDP);
      if (rez) goto ErrorHandler;       // must unlock file in handler

      DeriveSuitableName(SDP.filenamePtr,BackupDataName);

      CP.func = BACKUP_FILE_XB;
      CP.handle = passedHandle;         // set to -passedHandle to skip memo backup
      CP.filenamePtr = BackupDataName;
      rez = BULLET(&CP);
      if (rez) goto ErrorHandler;       // must unlock file in handler

      // unlocking also flushes the data header if not LOCK_SHARED (and if required)

      LP.func = UNLOCK_DATA_XB;
      LP.handle = passedHandle;
      rez = BULLET(&LP);
      if (rez) goto ErrorHandler;
   }
}


ΓòÉΓòÉΓòÉ 10.5. Get Error Class ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

XERRORPACK XEP;                 // packs used here

// Bullet errors range from 8192 to 8999, any other return code
// indicates the error number was generated by OS/2 itself

if (rez < 8192) | (rez > 8999) {
   XEP.func = GET_ERROR_CLASS_XB;
   XEP.stat = rez;
   rez = BULLET(&XEP);

   // here XEP.errClass, .action, and .location are set
   // this call is the same as OS/2 API DosErrClass()


ΓòÉΓòÉΓòÉ 10.6. Query System Variables ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

QUERYSETPACK  QSP;              // packs used here

// Query a Bullet system variable

QSP.func = QUERY_SYSVARS_XB;
QSP.item = MUTEX_SEM_HANDLE;    // in bullet_2.h
rez = BULLET(&QSP);
if (rez==0)
   printf("Bullet/2 mutex handle is %d\n",QSP.itemValue);


ΓòÉΓòÉΓòÉ 10.7. Set System Variables ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

QUERYSETPACK  QSP;              // packs used here

// Set a Bullet system variable

QSP.func = SET_SYSVARS_XB;
QSP.item = REINDEX_BUFFER_SIZE; // in bullet_2.h
QSP.itemValue = 384*1024;       // set to 384KB
rez = BULLET(&QSP);
if (rez==0) {
   printf("Reindex buffer sized changed to 384K.\n");
   printf("Previous setting was %d.\n",QSP.itemValue);

   // For REINDEX_BUFFER_SIZE, a value of 0 represents 'autosize', for which
   // Bullet selects a minimum usuable size (often 144KB).  The minimum size
   // that you can use for REINDEX_BUFFER_SIZE is 48KB.


ΓòÉΓòÉΓòÉ 10.8. Set Dual-video Monitor ΓòÉΓòÉΓòÉ


// SET_DVMON_XB is not currently used


ΓòÉΓòÉΓòÉ 10.9. Query Vectors ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

QUERYSETPACK  QSP;              // packs used here

// Query a Bullet vector

QSP.func = QUERY_VECTORS_XB;
QSP.item = VECTOR_GET_SORT_TABLE;  // in bullet_2.h
rez = BULLET(&QSP);
if (rez==0)
   printf("Bullet get-sort-table vector is %x\n",QSP.itemValue);


ΓòÉΓòÉΓòÉ 10.10. Set Vectors ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

QUERYSETPACK  QSP;              // packs used here

// Set a Bullet vector

// external routine (in ccdosfn.c, for example)
LONG __cdecl BulletMalloc(ULONG bytes, VOID **basePtr);

QSP.func = SET_VECTORS_XB;
QSP.item = VECTOR_MALLOC;
QSP.itemValue = (ULONG) &BulletMalloc;
rez = BULLET(&QSP);
if (rez==0) {
   printf("Memory allocation changed to new routine.\n");
   printf("Previous routine's address was %x.\n",QSP.itemValue);


ΓòÉΓòÉΓòÉ 10.11. Create, Open, and Close Data and Index Files ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

CREATEDATAPACK  CDP;
CREATEINDEXPACK CIP;
OPENPACK        OP;
HANDLEPACK      HP;             // packs used here

// create the data file

#pragma pack(1)         // ensure compiler does not pad Bullet-used structures
                        // (not needed here since all members are CHAR)

// generally, only left-justified strings should be 0-terminated, numbers or
// date fields should not be 0-terminated

typedef struct _RECTYPE {
 CHAR tag;              // record tag, init to SPACE, '*' means deleted
 CHAR userSSN[9];       // first field in DBF (not 0-terminated in this case)
 CHAR userScore[6];     // second field (also not 0T)
} RECTYPE;              // (total record length is 16 bytes)
RECTYPE ourRecord;

#pragma pack()

CHAR nameIX3[] = "INDEX.IX3";   // index pathname
CHAR keyExpression[] = "SSN";   // key is built from field named 'SSN'
ULONG indexID=0;                // handle of opened index file
CHAR keyBuffer[68];             // buffer to store/receive key values

CHAR nameData[] = "DATA.DBF";   // data pathname
ULONG dataID=0;                 // handle of opened data file
FIELDDESCTYPE fieldList[2];     // 2 fields used in data record (SSN and SCORE)

// the field descriptor info must have unused entries set to 0

memset(fieldList,0,sizeof(fieldList));  // init unused bytes to 0 (required)

// fill in the field descriptor info for the data file you want to create
// this descriptor must match the layout of ourRecord, above (ourRecord.tag
// is implicit and is not a formal field, and so is not in fieldList[])

strcpy(fieldList[0].fieldName, "SSN");  // field names must be upper-case
fieldList[0].fieldType = 'C';           // field types must be upper-case
fieldList[0].fieldLen = 9;              // * Note that the .fieldname here
fieldList[0].fieldDC = 0;               // * matches the keyExpression

strcpy(fieldList[1].fieldName, "SCORE");
fieldList[1].fieldType = 'C';
fieldList[1].fieldLen = 6;              // 6 is total size of field
fieldList[1].fieldDC = 0;

// Create the data file as defined in fieldList above
// To create only a DBF, set CDP.fileID=3
// To create both a DBF and a DBT memo file, set CDP.fileID=0x8B

CDP.func = CREATE_DATA_XB;
CDP.filenamePtr = nameData;
CDP.noFields = 2;
CDP.fieldListPtr = fieldList;
CDP.fileID = 3;
rez = BULLET(&CDP);
if (rez) {
   printf("Failed data file create.  Err: %d\n",rez);
   return(rez);
}

// Open the data file (required before creating an index file for it)

OP.func = OPEN_DATA_XB;
OP.filenamePtr = nameData;
OP.asMode = READWRITE | DENYNONE;
rez = BULLET(&OP);
if (rez) {
   printf("Failed data file open.  Err: %d\n",rez);
   return(rez);
}
dataID = OP.handle;

// Create an index file for the data file opened above.
// This example uses a simple primary key: the SSN field.
// Since it is assumed to be unique, DUPS_ALLOWED is not
// OR'ed with the .sortFunction.

CIP.func = CREATE_INDEX_XB;
CIP.filenamePtr = nameIX3;
CIP.keyExpPtr = keyExpression;
CIP.xbLink = dataID;            // the handle of the data file
CIP.sortFunction = ASCII_SORT;  // sort key by ASCII (fine for SSN ordering)
CIP.codePage = 0;               // use OS-default code page
CIP.countryCode = 0;            // use OS-default country code
CIP.collatePtr = NULL;          // no need for a special collate table
CIP.nodeSize = 512;             // 512-byte node size (or 1024, 2048 bytes)
rez = BULLET(&CIP);
if (rez) {
   printf("Failed index file create.  Err: %d\n",rez);
   return(rez);
}

// Open the index file (what we just created above).
// As with the index-create, the index-open requires the handle of the data
// file which this index file indexes.

OP.func = OPEN_INDEX_XB;
OP.filenamePtr = nameIX3;
OP.asMode = READWRITE | DENYNONE;
OP.xbLink = dataID;
rez = BULLET(&OP);
if (rez) {
   printf("Failed index file open.  Err: %d\n",rez);
   return(rez);
}
indexID = OP.handle;

// at this point, both the data and index files are open and accessible

// |
// | do work as required, then, when done, close them
// |

if (indexID) {
   HP.func = CLOSE_INDEX_XB;
   HP.handle = indexID;
   rez = BULLET(&HP);
   if (rez)
      printf("Failed index file close.  Err: %d\n",rez);
}

if (dataID) {
   HP.func = CLOSE_DATA_XB;
   HP.handle = dataID;
   rez = BULLET(&HP);
   if (rez)
      printf("Failed data file close.  Err: %d\n",rez);
}


ΓòÉΓòÉΓòÉ 10.12. Read Data and Index Header ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

HANDLEPACK HP;                  // packs used here

// Since it is recommended that a full-lock be in force before reloading
// data or index headers, and since performing a full lock from BULLET does
// itself reload the header, this routine would not normally be used.
// However, if you are doing your own locking, then you need to call this.

rez = YourExternalControlLockRoutine;  // so you have your own locks...

if (rez==0) {

   if (youWantDataReload) {
      HP.func = READ_DATA_HEADER_XB;
      HP.func = dataID;
   }
   else {
      HP.func = READ_INDEX_HEADER_XB;
      HP.func = indexID;
   }
   rez = BULLET(&HP);

   // since locked, release lock regardless rez value

   rez2 = YourExternalControlUnlockRoutine;
}

// Be sure to read the explanation above


ΓòÉΓòÉΓòÉ 10.13. Flush Data and Index Header ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

HANDLEPACK HP;                  // packs used here

// While BULLET automatically flushes data and index info whenever
// a BULLET file is unlocked from a full-lock (and if it
// is needed), you may to flush more frequently.

// It is recommended that you have a full-lock on the file before
// flushing.  In normal use, you would always have a full-lock when
// writing to a file (the only time you need to flush is if the file
// has been written to).  If you are, for whatever reason, keeping
// the file locked, and are updating it repeatedly, and have no
// intention of unlocking (and thereby flushing it) any time soon,
// you may do a manual flush to ensure that the disk image matches
// the memory image (in case the power goes out and you have no UPS)

if (youWantDataFlush) {
   HP.func = FLUSH_DATA_HEADER_XB;
   HP.func = dataID;
}
else {
   HP.func = FLUSH_INDEX_HEADER_XB;
   HP.func = indexID;
}
rez = BULLET(&HP);

// Only if the file has been changed does a flush actually write to disk.
// You should have an exclusive full-lock; a shared full-lock cannot be
// used since a shared lock does not allow writing to the file.


ΓòÉΓòÉΓòÉ 10.14. Copy (Add) a Subset of Records to a New File ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

ACCESSPACK APo;                 // used when accessing original
ACCESSPACK APn;                 // used when accessing new
COPYPACK   CP;
HANDLEPACK HP;
OPENPACK   OP;                  // packs used here

RECTYPE ourRecord;              // as defined in Create example

// Assume an open, locked DBF data file, handle in dataID, and that you
// want to copy those records that are marked as deleted, one at a time, to a
// new file, prior to packing the database.  An extension to this procedure is
// to access each record in key order, and to copy each record to the new
// file, thereby giving a SORTED data file (also known as a clustered file).
// The lock must be a full-lock, but may be a shared full-lock.  A ZAP is done
// on the new file if the copy did not complete as expected.

// No indexed access is used in this example.

CP.func = COPY_DATA_HEADER_XB;
CP.handle = dataID;             // dataID is the original data file handle
CP.filenamePtr = "DELRECS.DBF"; // new file for deleted records
rez = BULLET(&CP);
if (rez) {
   printf("Failed header copy.  Err: %d\n",rez);
   return(rez);
}

// DELRECS.DBF now is exactly like the original, but has 0 records.  Since
// we're building this file, open it for exclusive use (DENYREADWRTE) --
// this is different from using the LOCK_XB routines since the lock is done
// at the file open level (no other process may even open it), whereas
// the LOCK_XB routines are at the access level (other processes may
// open the file, but may or may not access it).  You could instead use
// LOCK_DATA_XB for an exclusive full-lock, but this is simpler.

OP.func = OPEN_DATA_XB;
OP.filenamePtr = "DELRECS.DBF";         // open it
OP.asMode = READWRITE | DENYREADWRITE;  // for exclusive use
rez = BULLET(&OP);
if (rez) {
   printf("Failed new file open.  Err: %d\n",rez);
   return(rez);
}

// Now have two files open:  the original, with all the records, and
// the new, with no records.  The procedure here is to get all original
// records, check each for being deleted, if so, add that record (copy it)
// to the new file.  After copying, the original file is packed so that
// all deleted records are removed.  Then an index to it is reindexed.

// most BULLET pack members will be invariant within loops... (low overhead!)

APn.func = ADD_RECORD_XB;       // AccessPack for the new file
APn.handle = OP.handle;         // the handle just opened above
APn.recPtr = &ourRecord;

APo.func = GET_RECORD_XB;       // AccessPack for the original file
APo.handle = dataID;
APo.recNo = 1;                  // start at the first record
APo.recPtr = &ourRecord;        // read data into this
rez = BULLET(&APo);             // read the first data record

while (rez==0) {

   // check if this record is deleted, if so copy it else continue

   if (ourRecord.tag = '*') {
      rez = BULLET(&APn);       // add it to the new file
      if (rez) break;
      // here APn.recNo is the record number used for the just-added record
   }

   // This while() loop could have been a for() loop if we had used
   // STAT_DATA_XB to get the number of records, but in this example,
   // the original file is read until EXB_BAD_RECNO is returned,
   // indicating that the last record has been passed.

   APo.recNo++;                 // get next original record
   rez = BULLET(&APo);          // everything else is already setup
}

// the expected rez here is EXB_BAD_RECNO, any other then quit

if (rez != EXB_BAD_RECNO) {
    printf("Failed the while() loop, ZAPing new file.  Err: %d\n",rez);

    // As an example, if the copy failed to complete as expected, the
    // new file is ZAP'ed, removing any records that may have been copied
    // up to the point that the error occurred -- investigate failure and
    // restart process (you may want to just delete, rather than ZAP/CLOSE).

    HP.func = ZAP_DATA_HEADER_XB;
    HP.handle = APn.handle;     // the new file handle
    rez2 = BULLET(&HP);         // using 'rez2' to preserve initial error code
    if (rez2)
       printf("Failed ZAP!  Err: %d\n",rez2);

    HP.func = CLOSE_DATA_XB;    // close
    rez2 = BULLET(&HP);
    if (rez2)
       printf("Failed CLOSE!  Err: %d\n",rez2);

    return(rez);                // return initial error code
}

// done with the new file
// it contains all the records in the original marked as deleted

HP.func = CLOSE_DATA_XB;        // always use the correct pack for the routine
HP.handle = APn.handle;         // -- do not try to use AP when closing a file
rez = BULLET(&HP);
if (rez) return(rez);

// Pack the original data file, now that we've "saved" the delete-marked recs.
// Note: You may want to use BACKUP_XB before using this next routine
// in case a serious error occurs.  BULLET packs in place!  Also, before
// packing, memos belonging to records about to be deleted should have
// have been deleted before packing.  Neither is shown here.

APo.func = PACK_RECORDS_XB;
rez = BULLET(&APo);
if (rez==0) {

   // After a pack, you must reindex any related index files.
   // Assume here one index file, handle in indexID

   APo.func = REINDEX_XB;
   APo.handle = indexID;
   rez = BULLET(&APo);  // rez is not errc, but is pack # than failed
   if (rez)
      printf("Failed reindex.  Err: %d\n",APo.stat);
}
else {
   printf("Failed packed!  Recommend restore from backup.  Err: %d\n",rez);
   printf("Note: Unless error is known to not be severe.\n");
}

// Here and rez==0 then everything went as planned.

return(rez);


ΓòÉΓòÉΓòÉ 10.15. Zap Index File ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

HANDLEPACK HP;                  // packs used here

// Since BULLET reindexes in place, there's typically no need to use
// ZAP to reduce disk requirements (there won't be two separate file
// spaces since the reindex copies right over the old key data).

HP.func = ZAP_INDEX_HEADER_XB;
HP.handle = indexToZap;
rez = BULLET(&HP);
if (rez) return(rez);


ΓòÉΓòÉΓòÉ 10.16. Get Descriptor, Using a DBF Not Created by Bullet ΓòÉΓòÉΓòÉ

#include <os2.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "bullet_2.h"

int main(int argc,char *argv[]) {

INITPACK IP;
EXITPACK EP;
DESCRIPTORPACK DP;
OPENPACK OP;
HANDLEPACK HP;
STATDATAPACK SDP;
ACCESSPACK AP;          // packs used here

PFIELDDESCTYPE fieldDescPtr;    // pointer to field descriptor base allocation
PFIELDDESCTYPE fdPtr;           // roving pointer to any field's descriptor

CHAR dataRec[8192];     // 'unknown' record layout since reading "any" DBF
CHAR  fmt[32];          // printf() fmt string for on-the-fly formatting
ULONG dataID;           // handle of DBF
LONG  recNo;            // loop counter
BYTE  fldNo;            // loop counter2
int   rez;              // primary op return code
int   rez2;             // secondary op return code (so to perserve primary rc)

if (argc < 2) {
   puts("Use: C>progname anyfile.dbf");
   return(1);
}

// init Bullet

IP.func = INIT_XB;
IP.JFTsize = 20;                // 20 handles is all we need here
rez = BULLET(&IP);
if (rez!=0) {
   printf("INIT_XB failed: %d\n",rez);
   return(1);
}

// open existing DBF as named in command line

OP.func = OPEN_DATA_XB;
OP.filenamePtr = argv[1];
OP.asMode = READWRITE | DENYNONE;
rez = BULLET(&OP);
if (rez==0) {

   dataID = OP.handle;

   SDP.func = STAT_DATA_XB;
   SDP.handle = dataID;
   rez = BULLET(&SDP);
   if (rez==0) {

      // allocate field descriptors needed (SDP.fields is number needed)
      // calloc() used since 0-filled storage is required

      fieldDescPtr = calloc(SDP.fields,sizeof(FIELDDESCTYPE));

      if (fieldDescPtr != NULL) {

         fdPtr = fieldDescPtr;          // fdPtr->each descriptor

         // read each field descriptor from Bullet, storing to our program
         // show each for display

         //      1234567890-123456789-123456789-12345
         printf("FLD#   FIELDNAME  T  LEN.DEC  OFFSET\n");

         DP.func = GET_DESCRIPTOR_XB;
         DP.handle = dataID;
         for (fldNo=1;fldNo <= SDP.fields;fldNo++) {

            DP.fieldNumber = fldNo;
            rez = BULLET(&DP);
            if (rez==0) {

               strcpy(fdPtr->fieldName, DP.FD.fieldName);
               fdPtr->fieldType = DP.FD.fieldType;
               fdPtr->fieldLen = DP.FD.fieldLen;
               fdPtr->fieldDC = DP.FD.fieldDC;
               fdPtr->fieldDA = DP.fieldOffset;
               printf("%3u   %-10s  %c  %3u.%1u     %4u\n",
                  fldNo,
                  fdPtr->fieldName,
                  fdPtr->fieldType,
                  (ULONG) fdPtr->fieldLen,
                  (ULONG) fdPtr->fieldDC,
                  fdPtr->fieldDA);
               fdPtr++;                 // next field descriptor
            }
            else
               break;
         }

         // An interesting item above is where fdPtr->fieldDA is set to
         // DP.fieldOffset.  fieldDA is a run-time storage area that in
         // dBASE is used to directly access the field (DA="direct access").
         // It has no meaning except for that particular run (it is a memory
         // address). In this program example I use it to store the offset
         // of the field, relative the start of the record buffer (where the
         // tag byte = offset 0).  You could just as easily use some of the
         // 12 reserved bytes left over in the descriptor, as I do for the
         // alternate field length.  But, since fieldDA is already there, and
         // not used otherwise, it makes sense to use it.

         // Now have all we need to know about the DBF fields, having just
         // read and stored the field descriptors.  For this example, we
         // grab the first nine records and spit them out, by field, in record
         // number order (no indexing used).

         if (SDP.records != 0) {

            AP.func = GET_RECORD_XB;
            AP.handle = dataID;
            AP.recPtr = &dataRec;

            for (recNo=1;recNo <= 9; recNo++) {

               printf("\nrecNo %u: ",recNo);    // show line number
               AP.recNo = recNo;                // get this record #...
               rez = BULLET(&AP);               // ...to dataRec buffer
               if (rez==0) {
                  printf("%.1s ",(CHAR *) dataRec); // show if deleted or not

                  fdPtr = fieldDescPtr; // fdPtr->first field descriptor

                  for (fldNo=1;fldNo <= SDP.fields;fldNo++) {

                     // No special formatting is done on this output for this
                     // example -- since standard DBF data is always in pure
                     // ASCII form, all is printable.

                     switch (fdPtr->fieldType) {
                     case 'C':  // text
                     case 'D':  // date, show as-is
                     case 'L':  // logical, show as-is
                     case 'M':  // memo field (block number in ASCII)
                     case 'N':  // numeric (ASCII)

                        // make fmt[] string like this: "%xx.xxs"
                        // where xx is field length for this field

                        sprintf(fmt,"%%-%i.%is ",
                                fdPtr->fieldLen,
                                fdPtr->fieldLen);

                        // fdPtr->fieldDA=offset of the field within the record
                        // so it plus dataRec (buffer base) results in the
                        // offset of the current field we are processing

                        printf(fmt,dataRec+fdPtr->fieldDA);
                        break;
                     default:
                        printf("\nUnknown field type: %c\n",fdPtr->fieldType);
                     } // switch

                     fdPtr += 1;                  // next field's descriptor

                  } // for fields
               } // if record read

               else {
                  if (rez==EXB_BAD_RECNO)       // if < for-count records in DBF
                     rez=0;                     // then would get this error
                  else
                     printf("Failed GET_RECORD_XB, err: %d\n",rez);
                  break;                        // break for any ELSE case
               }

            } // for records
            if (rez==0) printf("\nDone.\n"); // all FOR recs done
         }
         else
            printf("No records in file\n");

         free(fieldDescPtr);
      }
      else
         printf("calloc failed!\n");
   }
   else
      printf("STAT_DATA_XB failed: %d\n",rez);

   HP.func = CLOSE_DATA_XB;
   HP.handle = dataID;
   rez2 = BULLET(&HP);
}
else
   printf("OPEN_DATA_XB failed: %d\n",rez);

EP.func = EXIT_XB;
rez2=BULLET(&EP);

printf("\nPress ENTER to exit");
getchar();
if (rez==0) rez=rez2;  // rez is more important, but if 0 use rez2 result
return(rez);
}

The above is a complete program.  Running it against a sample DBF results in
the following output:

FLD#   FIELDNAME  T  LEN.DEC  OFFSET
  1   SSN         C    9.0        1
  2   LNAME       C   16.0       10
  3   FNAME       C   16.0       26
  4   HIRED       D    8.0       42
  5   DEPT_ID     C    6.0       50

recNo 1:   465309999 Que              Barbie           19900131 BOSS
recNo 2:   445038888 Stewart          Jackie           19910228 ACC
recNo 3:   760443232 Whitman          Kelly            19920414 HUM
recNo 4:   845309944 Beatty           Leslie           19940122 PRG
recNo 5:   555033388 Jasper           Amy              19930230 PRG
recNo 6:   430443222 Hauntos          Poco             19920414 PRG
recNo 7:   365502949 Hopkins          Lisa             19910121 PRG
recNo 8:   685733868 Leonard          Rosina           19850218 PRG
recNo 9:   500945242 Morton           Holly            19950406 PHY
Done.

Press ENTER to exit


ΓòÉΓòÉΓòÉ 10.17. Update a Data Record ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

ACCESSPACK AP;                  // packs used here

typedef struct _RECTYPE {
 CHAR tag;              // record tag, init to SPACE, '*' means deleted
 CHAR userSSN[9];       // first field in DBF (not 0-terminated in this case)
 CHAR userScore[6];     // second field (also not 0T)
} RECTYPE;              // (total record length is 16 bytes)
RECTYPE ourRecord;

// This excerpt demonstrates how to update (change) a record.
// The idea is to get the current contents of a record (by record
// number since no index is used for this update routine), change
// what needs changing, then write it back.  Under no circumstances
// should you change any field that is used as a key by an index, or
// as a foreign key, or in any other way removes the referential
// integrity of the database.  If you need to change a key field, then
// you must use UPDATE_XB.

AP.func = GET_RECORD_XB;
AP.handle = dataID;
AP.recNo = recordToUpdate;
AP.recPtr = &ourRecord;
rez = BULLET(&AP);
if (rez) return(rez);

// ourRecord has data stored at record number recordToUpdate -- since
// userScore is not used as a key field, this routine may be used to
// modify the contents of that field.  Since numbers are stored as ASCII
// text in compatible DBF files, must convert to binary, perform needed
// math, then convert back to text:

//t = atol(ourRecord.userScore);  // use scanf() since not 0T
sscanf(t,"%6u",&ourRecord.userScore);
t = t + ClassCurve;             // increase each score by curve value
sprintf(ourRecord.userScore,"%6.6u",t)

AP.func = UPDATE_RECORD_XB;     // other AP values already set up from GET
rez = BULLET(&AP);              // write out the record with the new score
if (rez) return(rez);


ΓòÉΓòÉΓòÉ 10.18. Delete, Undelete, 'Debump' a Data Record ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

ACCESSPACK AP;                  // packs used here

// delete or undelete or remove (debump)

switch(*requestMsg) {
case 'delete':
   AP.func = DELETE_RECORD_XB;  // places a '*' in .tag byte
   break;
case 'undelete':
   AP.func = UNDELETE_RECORD_XB; // places a SPACE in .tag byte
   break;
case 'debump':
   AP.func = DEBUMP_RECORD_XB;  // physically removes record from file, but
   break;                       // fails if AP.recNo is not last record number
}

AP.handle = dataID;             // same for all three
AP.recNo = recordNumber;
rez = BULLET(&AP);
return(rez);


ΓòÉΓòÉΓòÉ 10.19. Reading Memo Records ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

ACCESSPACK AP;
LOCKPACK LP;
MEMODATAPACK MDP;       // packs used here

typedef struct _RECTYPE {
 CHAR tag;              // record tag, init to SPACE, '*' means deleted
 CHAR userSSN[9];       // first field in DBF (not 0-terminated in this case)
 CHAR userMemo[10];     // second field (also not 0T), is memo field type
} RECTYPE;
RECTYPE someRecord;

// a minimum shared record lock is required on the DBF record that owns
// memoNumber in this example, since only reading of the single memo is done

LP.func = LOCK_DATA_XB;
LP.handle = dataID;
LP.dlMode = LOCK_SHARED;        // only reading here, shared record lock is okay
LP.recStart = recordToGet;      // lock this record
LP.recCount = 1;                // and only this record
rez = BULLET(&LP);
if (rez) return(rez);

// hereafter, must unlock before exiting

AP.func = GET_RECORD_XB;
AP.handle = dataID;
AP.recNo = recordToGet;
AP.recPtr = &someRecord;        // load someRecord with data on disk
rez = BULLET(&AP);
if (rez) return(rez);           // UNLOCK! before doing this exit

// memoNumber = atol(someRecord.userMemo); // the memo number (0 if none)
sscanf(memoNumber,"%10u",&someRecord.userMemo);

// this code reads the number of data bytes in the memo and allocates a
// run-time buffer to read that memo into

MDP.func = GET_MEMO_SIZE_XB;
MDP.dbfHandle = dataID;         // handle of the DBF this memo belongs to
MDP.memoNo = memoNumber;        // memo number to get size of (1=first)
rez = BULLET(&MDP);             // (returns an error if memoNumber is 0)
if (rez==0) {

   // BULLET does maintain 0-sized memo records, so you may want to check
   // if MDP.memoBytes from GET_MEMO_SIZE_XB is 0, and skip processing if so

   memoBytesToRead = MDP.memoBytes;         // since overwritten by next call
   memoBufferPtr = malloc(memoBytesToRead); // assuming you want it all at once

   if (memoBufferPtr) {

      MDP.func = GET_MEMO_XB;
      MDP.dbfHandle = dataID;      // same as before, as is .memoNo
      MDP.memoNo = memoNumber;
      MDP.memoPtr = memoBufferPtr; // memo disk data is loaded into this buffer
      MDP.memoOffset = 0;          // read from very first memo data byte
      MDP.memoBytes = memoBytesToRead;

      // MDP.memoBytes is already set to the total data size -- you may read
      // fewer bytes, and you may use .offset to move through the memo data
      // chunks at a time, rather than all at once -- the above simply reloads
      // it with the same count, since here memoBytesToRead==MDP.memoBytes

      rez = BULLET(&MDP);          // returns with MDP.memoBytes= bytes read
      if (rez==0) {

         // if (MDP.memoBytes != memoBytesToRead)
         //   printf("Could not read all bytes requested - probably at end of memo\n");
         // above would not happen in this case since the exact size was requested

         // process as required (here passes buffer ptr and bytes actually read)

         DoWhatYouWillWithThisMemoData(memoBufferPtr,MDP.memoBytes);
      }
      free(memoBufferPtr);
   }
   else rez=8;  // malloc failed, return 8=not enough memory
}

LP.func = UNLOCK_DATA_XB;
LP.handle = dataID;
LP.recStart = recordToGet;      // unlock this record
LP.recCount = 1;                // and only this record
rez2 = BULLET(&LP);

if (rez==0) rez=rez2;
return(rez);


ΓòÉΓòÉΓòÉ 10.20. Add, Update, Delete a Memo Record ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

ACCESSPACK AP;
LOCKPACK LP;
MEMODATAPACK MDP;       // packs used here

typedef struct _RECTYPE {
 CHAR tag;
 CHAR userSSN[9];
 CHAR userMemo[10];
} RECTYPE;
RECTYPE someRecord;

// an exclusive lock is required on the DBF file that owns this memo file
// since writing is done to the memo file (memo file header, especially)

LP.func = LOCK_DATA_XB;
LP.handle = dataID;
LP.dlMode = LOCK_EXCLUSIVE;     // writing here, exclusive full lock required
LP.recStart = 0;                // lock all
rez = BULLET(&LP);
if (rez) return(rez);

// hereafter, must unlock before exiting

AP.func = GET_RECORD_XB;
AP.handle = dataID;
AP.recNo = recordToGet;
AP.recPtr = &someRecord;        // load someRecord with data on disk
rez = BULLET(&AP);
if (rez) return(rez);           // UNLOCK! before doing this exit

// memoNumber = atol(someRecord.userMemo); // the memo number (0 if none)
sscanf(memoNumber,"%10u",&someRecord.userMemo);

// if there is no current memo, this example adds one
// if there is, this example updates it by:
//   - changing the first 16 bytes to "Kilroy was here."
//   - adding the text "Was updated." to the end of the current memo data
// the example then deletes the memo

CHAR kilroyStr[] = "Kilroy was here.";
CHAR updateStr[] = "Was updated.";
CHAR newStr[] = "New.";

if (memoNumber) {

   // modify the current memo -- first, the text "Kilroy was here." is
   // placed at the start of the memo, then the memo size is gotten (in
   // case the original memo size were less than the size of "Kilroy...").

   MDP.func = UPDATE_MEMO_XB;
   MDP.dbfHandle = dataID;      // handle of the DBF this memo belongs to
   MDP.memoNo = memoNumber;     // memo number to update (1=first)
   MDP.memoPtr = kilroyStr;     // data to write
   MDP.memoOffset = 0;          // start write at first byte of memo
   MDP.memoBytes = strlen(kilroyStr); // bytes to write
   rez = BULLET(&MDP);

      // the first 16 bytes of the memo now say kilroyStr (overwrote what was there)
      // the memo size changes only if the original memo was < 16 bytes, in
      // which case the size is now 16

   if (rez==0) {
      printf("%s overwrote first 16 bytes\n",kilroyStr);

      // must check if the update resulted in a new memo block being
      // used (if the update required more allocation blocks), and if
      // so must update the DBF field storing the memo number
      // MDP.memoNo is the memo number returned, memoNumber the original number

      // IT CAN BE ASSUMED IN THIS PARTICULAR example that this will never
      // be needed since the update modified the first 16 bytes of the
      // memo only, and so would never have required any more blocks --
      // however, unless you know before-hand that the update will not need
      // more allocation blocks (not difficult, if you know the block size,
      // overhead bytes (8), and your offset and bytes to write), it should
      // be checked -- hint: as with all Bullet routines, the idea is to
      // wrap up these separate operations into nice, neat callable routines
      // that take care of your particular need; there are no doubt 1000s of
      // variations -- pick one you can deal with.

      if (MDP.memoNo != memoNumber) {
         sprintf(someRecord.userMemo,"%10.10u",MDP.memoNo)
         memoNumber = MDP.memoNo;       // set original to new for next update

         // rather than updating here, and possibly again below, you
         // may elect to set a flag and then do the DBF update at the end, once
         // again -- this updates the _data_ record, in the DBF file:

         AP.func = UPDATE_RECORD_XB;    // this updates the data record only
         AP.handle = dataID;
         AP.recNo = recordToGet;
         AP.recPtr = &someRecord;       // write the new data
         rez = BULLET(&AP);
         if (rez) return(rez);          // UNLOCK! before doing this exit
      }

      MDP.func = GET_MEMO_SIZE_XB;  // other MDP members already set above
      rez = BULLET(&MDP);
      if (rez=0) {              // size of memo is at least 16 (from kilroyStr)
                                // but may be bigger if was bigger before
         MDP.func = UPDATE_MEMO_XB;
         MDP.memoPtr = updateStr;

         // the current memo size is used as the offset for the appending of
         // the updateStr bytes (offset is 0-based, so using MDP.memoBytes
         // results in the offset being the current size + 1)

         MDP.memoOffset = MDP.memoBytes;
         MDP.memoBytes = strlen(updateStr);
         rez = BULLET(&MDP);   // returns with MDP.memoNo
                               // bytes written==MDP.memoBytes always
         if (rez==0)

            printf("%s appended to memo\n",updateStr);
            // _minimum_ memo contents now is kilroyStr plus updateStr
            // more if original memo was > 16 bytes

            if (MDP.memoNo != memoNumber) {
               sprintf(someRecord.userMemo,"%10.10u",MDP.memoNo)
               AP.func = UPDATE_RECORD_XB;         //
               AP.handle = dataID;                 // as explained above
               AP.recNo = recordToGet;             //
               AP.recPtr = &someRecord;
               rez = BULLET(&AP);
               if (rez) return(rez);    // actually must unlock before exit!
            }
         } // memo update #2 failed
      } // memo size failed
   }
   else
      printf("update failed, err: %d\n",rez); // disk full probably
}
else {

   // no current memo, add one

   MDP.func = ADD_MEMO_XB;
   MDP.dbfHandle = dataID;      // handle of the DBF this memo belongs to
   MDP.memoPtr = newStr;        // data to write
   MDP.memoBytes = strlen(newStr); // bytes to write
   rez = BULLET(&MDP);
   if (rez==0) {
      sprintf(someRecord.userMemo,"%10.10u",MDP.memoNo)
      AP.func = UPDATE_RECORD_XB;         //
      AP.handle = dataID;                 // as explained above
      AP.recNo = recordToGet;             //
      AP.recPtr = &someRecord;
      rez = BULLET(&AP);
   }
   else
     printf("add failed, err: %d\n",rez);
}

// delete the memo just operated on

if (rez==0) {

   MDP.func = DELETE_MEMO_XB;
   // MDP.memoNo is already set
   rez = BULLET(&MDP);
   if (rez==0) {
      strcpy(someRecord.userMemo,"          ");// DBF memo field to all spaces
      AP.func = UPDATE_RECORD_XB;         //
      AP.handle = dataID;                 // as explained above
      AP.recNo = recordToGet;             //
      AP.recPtr = &someRecord;
      rez = BULLET(&AP);
   }
   else
     printf("delete failed, err: %d\n",rez);
}

LP.func = UNLOCK_DATA_XB;
LP.handle = dataID;
LP.recStart = 0;                // unlock all
rez2 = BULLET(&LP);

if (rez==0) rez=rez2;
return(rez);


ΓòÉΓòÉΓòÉ 10.21. Memo Bypass (Memo Create, Open, Close, Read/Flush Header) ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

MEMODATAPACK MDP;               // packs used here


//  These five routines are normally performed automatically, as described
//  in the main documentation, and seldom would need to be called directly.


MDP.func = MEMO_BYPASS_XB;
MDP.dbfHandle = dataID;         // handle of DBF
MDP.memoBypass = BypassRoutineToDo;

// where BypassRoutineToDo is one of the following:
//
//   BYPASS_CREATE_MEMO
//   BYPASS_OPEN_MEMO
//   BYPASS_CLOSE_MEMO
//   BYPASS_READ_MEMO_HEADER
//   BYPASS_FLUSH_MEMO_HEADER

rez = BULLET(&MDP);
if (rez) return(rez);


ΓòÉΓòÉΓòÉ 10.22. Key Access Without Data Record Read ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

CHAR keyBuffer[64];             // enough for the largest possible key

ACCESSPACK AP;                  // packs used here

// -----------------------------------------------------------
// this section starts at the first in-order key and reads all
// keys in the index file in order

AP.func = FIRST_KEY_XB;
AP.handle = indexID;
AP.keyPtr = keyBuffer;
rez = BULLET(&AP);
while (rez==0) {

   // show first 8 bytes of key, and the record number the key is for
   printf("%8.8s  %9.9lu\r", keyBuffer, AP.recNo);

   AP.func = NEXT_KEY_XB;       // and get the next key...
   rez = BULLET(&AP);           // until all keys accessed
};
if (rez!=EXB_END_OF_FILE) return(rez); // expected rez is EXB_END_OF_FILE


// ----------------------------------------------------------
// this section starts at the last in-order key and reads all
// keys in the index file in reverse order

AP.func = LAST_KEY_XB;
AP.handle = indexID;
AP.keyPtr = keyBuffer;
rez = BULLET(&AP);
while (rez==0) {

   // show first 8 bytes of key, and the record number the key is for
   printf("%8.8s  %9.9lu\r", keyBuffer, AP.recNo);

   AP.func = PREV_KEY_XB;       // and get the previous key...
   rez = BULLET(&AP);           // until all keys accessed
};
if (rez!=EXB_TOP_OF_FILE) return(rez); // expected rez is EXB_TOP_OF_FILE

// this section starts at the first key that starts with "SM", or if
// no keys do, the first key after "SM", and gets that key
// keys in the index file in reverse order

memset(keyBuffer,0,64);         // ensure remaining bytes are \0 (required)
strcpy(keyBuffer,"SM");         // get key of "SM", if present (not likely)
AP.func = EQUAL_KEY_XB;
AP.handle = indexID;
AP.keyPtr = keyBuffer;
rez = BULLET(&AP);

// since "SM" is only being used as a partial key search criterion,
// it won't be found (though, of course, it is a valid key) in this
// example -- however, by using NEXT_KEY_XB, the next key is accessed,
// say, for example, "SMITH"...  It could also be "TIMBU" if there were
// no keys with values greater than "SM" and less than "TIMBU".

// Be aware that if another thread in this process (repeat:
// this process!) is accessing this index file (it's not likely that
// you will write your could so that this would happen), then you
// can no longer rely on any multi-call key access to be an atomic
// operation.  If you need to have more than one thread access the
// same index (and you require NEXT_KEY_XB or PREV_KEY_XB), then you
// must semaphore protect your code so you don't try to access the same
// index file.

if (rez!=EXB_KEY_NOT_FOUND) {   // expected
   AP.func = NEXT_KEY_XB;       // so get the first key after "SM"
   AP.handle = indexID;
   AP.keyPtr = keyBuffer;
   rez = BULLET(&AP);
   if (rez==0)
      printf("The first key >= SM is %s\n",AP.keyBuffer);
}


ΓòÉΓòÉΓòÉ 10.23. Building and Storing Raw Key ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

ACCESSPACK AP;                  // packs used here

// this example shows a simple database insert process
// INSERT_XB should be used instead since it does all this and then some
// files should be locked (not shown)

AP.func = ADD_RECORD_XB;
AP.handle = dataID;
AP.recPtr = &yourRecord;
rez = BULLET(&AP);
if (rez) return(rez);

// AP.recNo is returned by Bullet and will be used later

AP.func = BUILD_KEY_XB;
AP.handle = indexID;
AP.recPtr = &yourRecord;
AP.keyPtr = keyBuffer;          // CHAR keyBuffer[64];
rez = BULLET(&AP);
if (rez) return(rez);

// keyBuffer filled with key to store
// a \0\0 enumerator is attached if the index file was created with DUPS_ALLOWED

AP.func = STORE_KEY_XB;
AP.handle = indexID;
// AP.recNo is already set from the ADD_RECORD_XB call
AP.keyPtr = keyBuffer;
rez = BULLET(&AP);
if (rez) return(rez);

// if no error, the key was inserted in the index file and the record
// number was associated with that key -- next time you access that
// key, the record number is returned (along with the key itself) --
// and with that record number you access the data file

// when using this routine, you must check the error for a EXB_KEY_EXISTS
// error and if DUPS_ALLOWED, you must manage your own enumerator --
// INSERT_XB is the only routine that does this automatically so unless
// you have a real desire to manage this yourself (among other things)
// use INSERT_XB instead of all this.


ΓòÉΓòÉΓòÉ 10.24. Getting Current Key, Key for Rec/RecNo Pair, Deleting Key ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

ACCESSPACK AP;
LOCKPACK LP;                    // packs used here

AP.func = GET_CURRENT_KEY_XB;
AP.handle = indexID;
AP.keyPtr = keyBuffer;          // current key placed here by Bullet
rez = BULLET(&AP);
if (rez) return(rez);
printf("The last accessed key for indexID is in keyBuffer, including any enumerator\n");

// This next example assumes that you are maintaining your own method of
// transaction rollback, and are about to delete the last item added to the
// database: in this case, the last data record is removed from the one
// DBF data file, and the key for that record is removed from the index file
// -- normally, you wouldn't do this, but if you have the need...

rez = InsertToDatabaseHoweverYouDoIt(yourPtr);

// Assume the above called succeeded, but you've decided, for whatever reason,
// that you want to backout the insert... normally, if you used INSERT_XB to
// insert a record/key into a database, you'd already have the record number
// used AND the key used for the record (for each if more than one) -- but for
// this example, assume that only the record number is known.  Steps done
// to remove the record and key for this would be:
//
//  1. Exclusive full-lock files
//  2. Get the record data at the record number (GET_RECORD_XB)
//  3. Get the key for this record/recNo pair (GET_KEY_FOR_RECORD_XB)
//  4. Delete the key (DELETE_KEY_XB) (delete the key before deleting the record)
//  5. Delete the record data (DEBUMP_RECORD_XB) (record must be last record in file)
//  6. Unlock files

// lock files being processed

LP.func = LOCK_XB;
LP.handle = indexID;            // also locks indexID's owner (its DBF)
LP.xlMode = LOCK_EXCLUSIVE;     // exclusive lock for index
LP.dlMode = LOCK_EXCLUSIVE;     // exclusive lock for data
LP.nextPtr = NULL;              // only one pack
rez = BULLET(&LP);
if (rez) return(LP.stat);       // rez for transaction-list is NOT the error

// get actual data record for record number

AP.func = GET_RECORD_XB;
AP.handle = dataID;
AP.recNo = recNoToDelete;       // get this record to...
AP.recPtr = &yourRecord;        // ...this data record buffer (a structure var)
rez = BULLET(&AP);
if (rez) goto MustAlwaysUnlock;

// now have the data record/data number pair --
// it must be the last physical record in the data file -- if this cannot
// be known, use STAT_DATA_XB to verify that recNoToDelete==SDP.records

AP.func = GET_KEY_FOR_RECORD_XB;
AP.handle = indexID;
AP.recNo = recNoToDelete;
AP.recPtr = &yourRecord;
AP.keyPtr = keyBuffer;          // CHAR keyBuffer[64]; key returned here
rez = BULLET(&AP);
if (rez) goto MustAlwaysUnlock;

// now have the key (with any attached enumerator) in keyBuffer, delete it

AP.func = DELETE_KEY_XB;
AP.handle = indexID;
AP.keyPtr = keyBuffer;          // was set with key by GET_KEY_FOR_RECORD_XB
rez = BULLET(&AP);
if (rez) goto MustAlwaysUnlock;

// and delete (physically remove) the data record

AP.func = DEBUMP_RECORD_XB;
AP.handle = dataID;
AP.recNo = recNoToDelete;
rez = BULLET(&AP);
// if (rez) goto MustAlwaysUnlock;

MustAlwaysUnlock:   // unlock ALWAYS required if lock succeeded

LP.func = UNLOCK_XB;
LP.handle = indexID;
LP.nextPtr = NULL;
rez2 = BULLET(&LP);
if (rez2) rez2=LP.stat; // rez for transaction-list is NOT the error

if (rez==0) rez=rez2;
return(rez);


ΓòÉΓòÉΓòÉ 10.25. Get Data by Key Order ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

ACCESSPACK AP;                  // packs used here

// First example starts at first in-order data and moves through to last

AP.func = GET_FIRST_XB;
AP.handle = indexID;
AP.recPtr = &yourRecord;        // struct yourStructure yourRecord;
AP.keyPtr = keyBuffer;          // CHAR keyBuffer[64];
rez = BULLET(&AP);
if (rez) return(rez);
printf("The first in-order key is in keyBuffer and its record in yourRecord\n");

AP.func = GET_NEXT_XB;          // other parm same as set above
while (rez==0) {
   rez = BULLET(&AP);
   if (rez) break;
   printf("The next in-order key is in keyBuffer and its record in yourRecord\n");
}
if (rez==EXB_END_OF_FILE) rez==0; // expected rez after end of file
if (rez) return(rez);


// Second example starts at last in-order data and moves through to first
// note: since above already is past last, the call to GET_LAST_XB in
//       this example would not be necessary -- a call to GET_PREV_XB
//       could have been made directly -- however, it doesn't matter

AP.func = GET_LAST_XB;
AP.handle = indexID;
AP.recPtr = &yourRecord;
AP.keyPtr = keyBuffer;
rez = BULLET(&AP);
if (rez) return(rez);
printf("The last in-order key is in keyBuffer and its record in yourRecord\n");

AP.func = GET_PREV_XB;          // other parms same as set above
while (rez==0) {
   rez = BULLET(&AP);
   if (rez) break;
   printf("The previous in-order key is in keyBuffer and its record in yourRecord\n");
}
if (rez==EXB_TOP_OF_FILE) rez==0; // expected rez before at beginning of file
if (rez) return(rez);


// Third example performs a GET_EQUAL_OR_GREATER operation, typically
// used to locate to a key based on a partial search criterion

AP.func = GET_EQUAL_XB;
AP.handle = indexID;
AP.recPtr = &yourRecord;    // to be filled on return, if found

memset(keyBuffer,0,sizeof(keyBuffer);   // clear it out (required)
strcpy(keyBuffer,"KING");               // find first key starting with 'KING'
AP.keyPtr = keyBuffer;

rez = BULLET(&AP);
if (rez==0)
   printf("Matched search key in keyBuffer EXACTLY -- its record is in yourRecord\n");
else if (rez==EXB_KEY_NOT_FOUND) {

   // since not found, get the following in-order one (say, 'KINGSTON')

   AP.func = GET_NEXT_XB;
   rez = BULLET(&AP);
   if (rez) return(rez);

   printf("Not exact, but next greater key is in keyBuffer and its record is in yourRecord\n");
}
else
   return(rez);


ΓòÉΓòÉΓòÉ 10.26. Insert Data Record with Key Into Database ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

ACCESSPACK AP;
LOCKPACK LP;                    // packs used here

LP.func = LOCK_XB;
LP.handle = indexID;            // also locks indexID's owner (its DBF)
LP.xlMode = LOCK_EXCLUSIVE;     // exclusive lock for index
LP.dlMode = LOCK_EXCLUSIVE;     // exclusive lock for data
LP.nextPtr = NULL;              // only one pack
rez = BULLET(&LP);
if (rez) return(LP.stat);       // rez for transaction-list is NOT the error

AP.func = INSERT_XB;
AP.handle = indexID;
AP.recNo = 0;                   // must be zero
AP.recPtr = &yourRecord;        // contains data record
AP.keyPtr = keyBuffer;          // empty, on return has key stored
AP.nextPtr = NULL;              // only the single pack
rez = BULLET(&AP);

// on return, as on all transaction-list routines, rez is not the return
// code but is the pack item that failed (neg if data, pos if index)

if (rez==0)
   printf("okay\n");
else if (rez < 0)
   printf("insert failed with data, err: %d\n",AP.stat);
else
   printf("insert failed with index, err: %d\n",AP.stat);


// if locked, MUST unlock!

LP.func = UNLOCK_XB;
LP.handle = indexID;
LP.nextPtr = NULL;
rez2 = BULLET(&LP);
if (rez2) rez2=LP.stat; // rez for transaction-list is NOT the error
if (rez==0) rez=rez2    // return the most interesting error, if any


ΓòÉΓòÉΓòÉ 10.27. Update Data Record with Key Update ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

ACCESSPACK AP;
LOCKPACK LP;                    // packs used here

LP.func = LOCK_XB;
LP.handle = indexID;            // also locks indexID's owner (its DBF)
LP.xlMode = LOCK_SHARED;        // shared lock for index
LP.dlMode = LOCK_SHARED;        // shared lock for data
LP.nextPtr = NULL;              // only one pack
rez = BULLET(&LP);
if (rez) return(LP.stat);       // rez for transaction-list is NOT the error

AP.func = GET_FIRST_XB;         // get first key's data record (and its recNo)
AP.handle = indexID;
AP.recPtr = &yourRecord;
AP.keyPtr = keyBuffer;
rez = BULLET(&AP);
if (rez==0) {

   // assume that we want to change this entry now (as if we weren't sure
   // that we wanted to initially, hence the initial shared lock) --
   // this requires write access so relock to allow write access

   LP.func = RELOCK_XB;
   LP.handle = indexID;
   LP.xlMode = LOCK_EXCLUSIVE;  // exclusive lock for index
   LP.dlMode = LOCK_EXCLUSIVE;  // exclusive lock for data
   LP.recStart = 0;
   LP.nextPtr = NULL;
   rez = BULLET(&LP);
   if (rez) rez=LP.stat         // xaction-list routine so use LP.stat
   if (rez==0) {

      strcpy(yourRecord.someField,"new field data");

      // AP.recNo has been set by Bullet GET_XB call above

      AP.func = UPDATE_XB;
      AP.nextPtr = NULL;        // all other AP members set above
      rez = BULLET(&AP);

      // on return, as on all transaction-list routines, rez is not the return
      // code but is the pack item that failed (neg if data, pos if index)

      if (rez==0)
         printf("okay\n");
      else if (rez < 0)
         printf("update failed with data, err: %d\n",AP.stat);
      else
         printf("update failed with index, err: %d\n",AP.stat);
   }
   else
      printf("relock failed, rez: %d  err: %d\n",rez,LP.stat);
}

// if locked, MUST unlock!

LP.func = UNLOCK_XB;
LP.handle = indexID;
LP.nextPtr = NULL;
rez2 = BULLET(&LP);
if (rez2) rez2=LP.stat; // rez for transaction-list is NOT the error
if (rez==0) rez=rez2    // return the most interesting error, if any


ΓòÉΓòÉΓòÉ 10.28. Remote Drive, File/Device Check ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

REMOTEPACK RP;                  // packs used here

// check if file in handle (or device handle) is on a 'network' drive

RP.func = CHECK_REMOTE_XB;
RP.handle = indexID;            // check if indexID handle is on a network
rez = BULLET(&RP);
if (rez) return(rez);
if (RP.isRemote)
   printf("handle is on a network drive\n");

// check if drive is a 'network' drive

RP.func = CHECK_REMOTE_XB;
RP.handle = 0;                  // set RP.handle to 0 to check drive
RP.drive = 0;                   // check if current drive is network drive
rez = BULLET(&RP);              // to check drive C:, set RP.drive=3
if (rez) return(rez);           // D: is RP.drive=4, and so on
if (RP.isRemote)
   printf("current drive is a network drive\n");


ΓòÉΓòÉΓòÉ 10.29. Relock Individual File ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

LOCKPACK LP;
STATHANDLEPACK SHP;
STATDATAPACK SDP;
STATINDEXPACK SIP;              // packs used here

// given a handle, determine if it's DBF or index
// check if currently locked
// if locked, determine if shared or exclusive lock
// if exclusive lock, relock to shared
// if not locked, do nothing

SHP.func = STAT_HANDLE_XB;
SHP.handle = passedHandle;
rez = BULLET(&SHP);
if (SHP.ID==-1)
   puts("Handle is not a Bullet data or index file\n");
else {
   if (SHP.ID==0) {

      // normally, you lock before calling this routine but since
      // SIP.lockCount and SIP.flags are all that is being checked,
      // and since nothing is to be done if the handle is not locked,
      // it's okay in this instance to use this routine without
      // first explicitly locking the handle

      SIP.func = STAT_INDEX_XB;
      SIP.handle = passedHandle;
      rez = BULLET(&SIP);
      if (rez) goto ErrorHandler;

      // if locked, check if the lock is exclusive

      if (SIP.lockCount) {              // count of active full locks

         if ((SIP.flags & 4)==0) {      // bit2=0 means lock is not shared

            // currently exclusive, make it shared

            LP.func = RELOCK_INDEX_XB;
            LP.handle = passedHandle;
            LP.xlMode = LOCK_SHARED;
            rez = BULLET(&LP);
            if (rez) goto ErrorHandler;
         }
      }
   }
   else {

      SDP.func = STAT_DATA_XB;
      SDP.handle = passedHandle;
      rez = BULLET(&SDP);
      if (rez) goto ErrorHandler;

      if (SDP.lockCount) {              // count of active full locks

         if ((SDP.flags & 4)==0) {      // bit2=0 means lock is not shared

            // currently exclusive, make it shared

            LP.func = RELOCK_DATA_XB;
            LP.handle = passedHandle;
            LP.dlMode = LOCK_SHARED;
            LP.startRec = 0;            // entire file
            rez = BULLET(&LP);
            if (rez) goto ErrorHandler;
         }
      }
   }
}


ΓòÉΓòÉΓòÉ 10.30. DOS Disk Routines Through Bullet ΓòÉΓòÉΓòÉ

#include "bullet_2.h"

DOSFILEPACK DFP;                // packs used here

CHAR dataDirname[] = "under";
CHAR dataFilename[]= "under\\data.dbf";
CHAR newFilename[] = "under\\newdata.dbf";

// check if pathame can be access for read/write denynone (expected not here)

DFP.func = ACCESS_FILE_DOS;
DFP.filenamePtr = dataFilename;
DFP.asMode = 0x42;
rez = BULLET(&DFP);
if (rez==0) return(0);          // file already exists
if (rez==5) return(5);          // access denied (exists but in use)
// other errors possible, too, do the Make and Create and go by those results

// make directory for file

DFP.func = MAKE_DIR_DOS;
DFP.filenamePtr = dataDirname;
rez = BULLET(&DFP);
if (rez) rez=0;                 // can't create subdirectory (already exists?)

// create file

DFP.func = CREATE_FILE_DOS;
DFP.filenamePtr = dataFilename;
DFP.attr = 0;                   // 'normal' attributes
rez = BULLET(&DFP);
if (rez) return(rez);           // can't create file

// open file

DFP.func = OPEN_FILE_DOS;
DFP.filenamePtr = dataFilename;
DFP.asMode = 0x42;
rez = BULLET(&DFP);
if (rez) return(rez);           // can't open file

// DFP.handle is set by OPEN_FILE_DOS call above
// pre-allocate file space

DFP.func = EXPAND_FILE_DOS;
// DFP.handle already set
// DFP.asMode already set
DFP.bytes = 128*1024;           // set filesize to 128KB
rez = BULLET(&DFP);
if (rez) return(rez);           // can't do it (close file before return)

CHAR writeStuff[]="Write this string to offset 888";

DFP.func = SEEK_FILE_DOS;
// DFP.handle already set
DFP.seekTo = 888;
DFP.method = 0;                 // from start of file
rez = BULLET(&DFP);
if (rez) return(rez);           // can't do it (close...)

// write the string to disk

DFP.func = WRITE_FILE_DOS;
// DFP.handle already set
DFP.bytes = strlen(writeStuff);
DFP.bufferPtr = writeStuff;
rez = BULLET(&DFP);
if (rez) return(rez);

// commit to the deep

DFP.func = COMMIT_FILE_DOS;
// DFP.handle already set
rez = BULLET(&DFP);
if (rez) return(rez);

// reposition to where write started so can read what was written

DFP.func = SEEK_FILE_DOS;
// DFP.handle already set
DFP.seekTo = 888;
DFP.method = 0;
rez = BULLET(&DFP);
if (rez) return(rez);

CHAR readBuffer[128];

DFP.func = READ_FILE_DOS;
// DFP.handle already set
DFP.bytes = strlen(writeStuff); // read it back
DFP.bufferPtr = readBuffer;
rez = BULLET(&DFP);
if (rez) return(rez);
if (DFP.bytes != strlen(writeStuff))
   printf("read came up short!\n");

DFP.func = CLOSE_FILE_DOS;
// DFP.handle already set
rez = BULLET(&DFP);
if (rez) return(rez);
DFP.handle = 0;                 // it's gone (closed, anyway)

// rename the file

DFP.func = RENAME_FILE_DOS;
DFP.filenamePtr = dataFilename;
DFP.newFilenamePtr = newFilename;
rez = BULLET(&DFP);
if (rez) return(rez);

// and get rid of it (known now as newFilename)

DFP.func = DELETE_FILE_DOS;
DFP.filenamePtr = newFilename;
rez = BULLET(&DFP);
return(rez);

// Why the DosXXX routines when they're standard API?  Not all
// language tools are able to directly call the API, but can
// call them indirectly through the Bullet DLL.


ΓòÉΓòÉΓòÉ 11. Bullet Errors ΓòÉΓòÉΓòÉ

Bullet error codes are numbered so that they do not overlap OS error codes.
The first general Bullet error number is 8193, but Bullet also returns select
system error code numbers instead of duplicating system codes (those listed
below less than 8192).  In addition, if the error is reported by the OS (for
example, attempting to access a locked file), then the OS error in returned by
Bullet.  For a list of OS errors, see OS/2 Dos API Errors.

System Error Codes

   8    EXB_NOT_ENOUGH_MEMORY
        cannot get memory requested

  15    EXB_INVALID_DRIVE
        not a valid drive letter

  38    EXB_UNEXPECTED_EOF
        unexpected end-of-file where bytes requested for read exceeded EOF

  39    EXB_DISK_FULL
        disk full on WriteFile

  80    EXB_FILE_EXISTS
        cannot create file since it already exists

 105    EXB_SEM_OWNER_DIED
        owner of mutex semaphore died  (used in place of Win32 80h)

 640    EXB_TIMEOUT
        mutex semaphore timed out (used in place of Win32 102h)


8192+   EXB_OR_WITH_FAULTS
        1=flush failed on handle close
        2=free memory failed on handle close
        4=failed memo handle close (data handle only)

        During a CLOSE_XB routine, the close process continues
        regardless of errors, and so the errors are accumulated.
        For example, 8193 means the flush failed, and 8195 means
        both the flush and the free failed (8192+1+2=8195).
        If the error occurred in the actual DosClose() API call,
        only that error is returned (it will be an OS error code).

8251    EXB_216501
        INT21/6501h not supported by DOS extender (see ccdosfn.c)

8256    EXB_216506
        INT21/6506h not supported by DOS extender (see ccdosfn.c)


8300    EXB_ILLEGAL_CMD
        function not allowed

8301    EXB_OLD_DOS
        OS version < MIN_DOS_NEEDED

8302    EXB_NOT_INITIALIZED
        init not active, must do INIT_XB before using Bullet

8303    EXB_ALREADY_INITIALIZED
        init already active, must do EXIT_XB first

8304    EXB_TOO_MANY_HANDLES
        more than 1024 opens requested,
        or more than license permits (100, 250, 1024)

8305    EXB_SYSTEM_HANDLE
        Bullet won't use or close handles 0-2

8306    EXB_FILE_NOT_OPEN
        the handle is not a Bullet handle, including the
        handle supplied in OP.xbLink

8307    EXB_FILE_IS_DIRTY
        tried to reload header but current still dirty;
        flush the file before reloading the header

8308    EXB_BAD_FILETYPE
        attempted to do a key file operation on non-key file,
        or a data operation on a non-data file

8309    EXB_TOO_MANY_PACKS
        too many INSERT, UPDATE, REINDEX, LOCK_XB packs (more than 256)

8310    EXB_NULL_RECPTR
        null record pointer passed to Bullet (.recPtr==NULL)

8311    EXB_NULL_KEYPTR
        null key pointer passed to Bullet (.keyPtr==NULL)

8312    EXB_NULL_MEMOPTR
        null memo pointer passed to Bullet (.memoPtr==NULL)

8313    EXB_EVALUATION_LIMIT
        evaluation limit of 300 records exceeded (remedy: buy a license)

8314    EXB_BAD_INDEX
        Query/SetSysVars index selection is beyond the last one

8315    EXB_RO_INDEX
        SetSysVars index item is read-only

8316    EXB_FILE_BOUNDS
        file size > 4GB, or greater than the SetSysVars value


8397    EXB_FORCE_ROLLBACK
        rollback test completed (last AP[].nextPtr=-1 for Insert/Update) (test-use only)

8398    EXB_INVALID_DLL
        DLL seems to be invalid (8399 similar)


Multi-access Error Codes

8401    EXB_BAD_LOCK_MODE
        lock mode (LP) not valid, must be 0 or 1

8402    EXB_NOTHING_TO_RELOCK
        cannot relock without existing full-lock

8403    EXB_SHARED_LOCK_ON
        unlikely error, write access needed for flush, but lock is shared


Index Error Codes

8501    EXB_KEY_NOT_FOUND
        exact match of key not found

8502    EXB_KEY_EXISTS
        key exists already and dups not allowed

8503    EXB_END_OF_FILE
        already at last index order

8504    EXB_TOP_OF_FILE
        already at first index order

8505    EXB_EMPTY_FILE
        nothing to do since no keys

8506    EXB_CANNOT_GET_LAST
        cannot locate last key

8507    EXB_BAD_INDEX_STACK
        index file is corrupt

8508    EXB_BAD_INDEX_READ0
        index file is corrupt

8509    EXB_BAD_INDEX_WRITE0
        index file is corrupt


8521    EXB_OLD_INDEX
        incompatible Bullet index, use ReindexOld subroutine, if available

8522    EXB_UNKNOWN_INDEX
        not a Bullet index file

8523    EXB_KEY_TOO_LONG
        keylength > 62 (or 64 if unique), or is 0


8531    EXB_PARSER_NULL
        parser function pointer is NULL

8532    EXB_BUILDER_NULL
        build key function pointer is NULL

8533    EXB_BAD_SORT_FUNC
        CIP.sortFunction not valid (not 1-6, or a custom sort-compare)

8534    EXB_BAD_NODE_SIZE
        CIP.nodeSize is not 512, 1024, or 2048

8535    EXB_FILENAME_TOO_LONG
        CIP.filenamePtr->pathname greater than file system allows

        This error is detected only for the file system installed,
        and does not detect using pathnames greater than 80 on a FAT
        system if HPFS is installed.  The OS returns its own error
        in this case, after the fact.


8541    EXB_KEYX_NULL
        key expression is effectively NULL

8542    EXB_KEYX_TOO_LONG
        CIP.keyExpPtr->expression is greater than 159 bytes

8543    EXB_KEYX_SYM_TOO_LONG
        fieldname/funcname in expression is longer than 10 chars

8544    EXB_KEYX_SYM_UNKNOWN
        fieldname/funcname in expression is unknown or misspelled

8545    EXB_KEYX_TOO_MANY_SYMS
        too many symbols/fields used in expression (16 max)

8546    EXB_KEYX_BAD_SUBSTR
        invalid SUBSTR() operand in expression

8547    EXB_KEYX_BAD_SUBSTR_SZ
        SUBSTR() exceeds field's size

8548    EXB_KEYX_BAD_FORM
        didn't match expected symbol in expression (missing paren, etc.)


8551    EXB_NO_READS_FOR_RUN
        unlikely error, use different reindex buffer size to fix

8552    EXB_TOO_MANY_RUNS
        unlikely error, too many runs (64K or more runs)

8553    EXB_TOO_MANY_RUNS_FOR_BUFFER
        unlikely error, too many runs for run buffer

8554    EXB_TOO_MANY_DUPLICATES
        more than 64K "identical" keys since the last enumerator used
        was 0xFFFF -- if ever you have this error, REINDEX_XB should
        be used to resequence the enumerators


8561    EXB_INSERT_RECNO_BAD
        AP.recNo cannot be > 0 if inserting with INSERT_XB

8562    EXB_PREV_APPEND_EMPTY
        no previous append for INSERT_XB yet AP.recNo==0x80000000

8563    EXB_PREV_APPEND_MISMATCH
        previous append's xbLink does not match this
        -- if this pack's AP.recNo=0x80000000 then this pack's AP.handle
        must be the same handle as that of the last pack that added a record

8564    EXB_INSERT_KBO_FAILED
        could not back out key at INSERT_XB

8565    EXB_INSERT_DBO_FAILED
        could not back out data records at INSERT_XB


8571    WRN_NOTHING_TO_UPDATE
        all AP.recNo=0 at UPDATE_XB so nothing to do

8572    EXB_INTERNAL_UPDATE
        internal error UPDATE_XB, not in handle/record# list

8573    EXB_FAILED_DATA_RESTORE
        could not restore original data record (*)

8574    EXB_FAILED_KEY_DELETE
        could not remove new key (*)

8575    EXB_FAILED_KEY_RESTORE
        could not restore original key(*)

        (*) original error, which forced a back-out, has been
        replaced by this error -- this error is always returned
        in the first AP.stat (-1 on data, 1 on index)


Data Error Codes

8601    EXB_EXT_XBLINK
        xbLink handle is not an internal DBF, as was specified during the
        index file's creation -- the Bullet routine called requires a Bullet
        DBF data file (instead use index-only access methods like NEXT_KEY_XB).

8602    EXB_FIELDNAME_TOO_LONG
        fieldname is > 10 characters

8603    EXB_RECORD_TOO_LONG
        record length is > 64K

8604    EXB_FIELD_NOT_FOUND
        fieldname not found in descriptor info

8605    EXB_BAD_FIELD_COUNT
        fields <= 0 or >= MAX_FIELDS; also use of a field number which
        is beyond the last field

8606    EXB_BAD_HEADER
        bad header (reclen=0, etc.)

8607    EXB_BUFFER_TOO_SMALL
        buffer too small (pack buffer < record length)

8608    EXB_INTERNAL_PACK
        internal error in PackRecords

8609    EXB_BAD_RECNO
        record number=0 or > records in data file header, or
        pack attempt on empty data file

8610    WRN_RECORD_TAGGED
        record's tag field matches skip tag


Memo Error Codes

8701    WRN_CANNOT_OPEN_MEMO
        the DBF header has bits 3 & 7 set, which indicates that a memo
        file is attached to this DBF, but the DBT memo file failed to open
        -- the DBF open continues, with this warning code returned

8702    EXB_MEMO_NOT_OPEN
        no open memo file for operation

8703    EXB_BAD_BLOCKSIZE
        memo blocksize must be at least 24 bytes

8704    EXB_MEMO_DELETED
        memo is deleted

8705    EXB_MEMO_PAST_END
        memo data requested is past end of record

8706    EXB_BAD_MEMONO
        memo number is not valid

8707    EXB_MEMO_IN_USE
        memo add encountered likely corrupt memo file
        -- avail list indicates this memo record is deleted, but the memoAvail
        link for the memo indicates it is use (memoAvail link==0x8FFFF)

8708    EXB_BAD_AVAIL_LINK
        memo avail link cannot be valid (e.g., memoAvail==0)

8709    EXB_MEMO_ZERO_SIZE
        memo data has no size (size is 0)

8710    EXB_MEMO_IS_SMALLER
        memo attempt to shrink but memo size is already <= size requested


ΓòÉΓòÉΓòÉ 12. OS/2 Dos API Errors ΓòÉΓòÉΓòÉ


  0   NO_ERROR
          No error occurred.

  1   ERROR_INVALID_FUNCTION
          Invalid function number.

  2   ERROR_FILE_NOT_FOUND
          File not found.

  3   ERROR_PATH_NOT_FOUND
          Path not found.

  4   ERROR_TOO_MANY_OPEN_FILES
          Too many open files (no handles left).

  5   ERROR_ACCESS_DENIED
          Access denied.

  6   ERROR_INVALID_HANDLE
          Invalid handle.

  7   ERROR_ARENA_TRASHED
          Memory control blocks destroyed.

  8   ERROR_NOT_ENOUGH_MEMORY
          Insufficient memory.

  9   ERROR_INVALID_BLOCK
          Invalid memory-block address.

  10   ERROR_BAD_ENVIRONMENT
          Invalid environment.

  11   ERROR_BAD_FORMAT
          Invalid format.

  12   ERROR_INVALID_ACCESS
          Invalid access code.

  13   ERROR_INVALID_DATA
          Invalid data.

  14   Reserved.


  15   ERROR_INVALID_DRIVE
          Invalid drive specified.

  16   ERROR_CURRENT_DIRECTORY
          Attempting to remove current directory.

  17   ERROR_NOT_SAME_DEVICE
          Not same device.

  18   ERROR_NO_MORE_FILES
          No more files.

  19   ERROR_WRITE_PROTECT
          Attempt to write on write-protected diskette.

  20   ERROR_BAD_UNIT
          Unknown unit.

  21   ERROR_NOT_READY
          Drive not ready.

  22   ERROR_BAD_COMMAND
          Unknown command.

  23   ERROR_CRC
          Data error - cyclic redundancy check.

  24   ERROR_BAD_LENGTH
          Invalid request structure length.

  25   ERROR_SEEK
          Seek error.

  26   ERROR_NOT_DOS_DISK
          Unknown media type.

  27   ERROR_SECTOR_NOT_FOUND
          Sector not found.

  28   ERROR_OUT_OF_PAPER
          Printer is out of paper.

  29   ERROR_WRITE FAULT
          Write fault.

  30   ERROR_READ_FAULT
          Read fault.

  31   ERROR_GEN_FAILURE
          General failure.

  32   ERROR_SHARING_VIOLATION
          Sharing violation.

  33   ERROR_LOCK_VIOLATION
          Lock violation.

  34   ERROR_WRONG_DISK
          Invalid disk change.

  35   ERROR_FCB_UNAVAILABLE
          FCB unavailable.

  36   ERROR_SHARING_BUFFER_EXCEEDED
          Sharing buffer overflow.

  37   ERROR_CODE_PAGE_MISMATCHED
          Code page does not match.

  38   ERROR_HANDLE_EOF
          End of file reached.

  39   ERROR_HANDLE_DISK_FULL
          Disk is full.

  40-49   Reserved.


  50   ERROR_NOT_SUPPORTED
          Network request not supported.

  51   ERROR_REM_NOT_LIST
          Remote network node is not online.

  52   ERROR_DUP_NAME
          Duplicate file name in network.

  53   ERROR_BAD_NETPATH
          Network path not found.

  54   ERROR_NETWORK_BUSY
          Network is busy.

  55   ERROR_DEV_NOT_EXIST
          Device is not installed in network.

  56   ERROR_TOO_MANY_CMDS
          Network command limit reached.

  57   ERROR_ADAP_HDW_ERR
          Network adapter hardware error.

  58   ERROR_BAD_NET_RESP
          Incorrect response in network.

  59   ERROR_UNEXP_NET_ERR
          Unexpected error in network.

  60   ERROR_BAD_REM_ADAP
          Remote network adapter error.

  61   ERROR_PRINTQ_FULL
          Network printer queue is full.

  62   ERROR_NO_SPOOL_SPACE
          No space in print spool file.

  63   ERROR_PRINT_CANCELLED
          Print spool file deleted.

  64   ERROR_NETNAME_DELETED
          Network name deleted.

  65   ERROR_NETWORK_ACCESS_DENIED
          Access to network denied.

  66   ERROR_BAD_DEV_TYPE
          Device type invalid for network.

  67   ERROR_BAD_NET_NAME
          Network name not found.

  68   ERROR_TOO_MANY_NAMES
          Network name limit exceeded.

  69   ERROR_TOO_MANY_SESS
          Network session limit exceeded.

  70   ERROR_SHARING_PAUSED
          Temporary pause in network.

  71   ERROR_REQ_NOT_ACCEP
          Network request denied.

  72   ERROR_REDIR_PAUSED
          Pause in network print disk redirection.

  73   ERROR_SBCS_ATT_WRITE_PROT
          Attempted write on protected disk.

  74   ERROR_SBCS_GENERAL_FAILURE
          General failure, single-byte character set.

  75-79   Reserved.


  80   ERROR_FILE_EXISTS
          File exists.

  81   ERROR_DUP_FCB
          Reserved.

  82   ERROR_CANNOT_MAKE
          Cannot make directory entry.

  83   ERROR_FAIL_I24
          Failure on INT 24.

  84   ERROR_OUT_OF_STRUCTURES
          Too many redirections.

  85   ERROR_ALREADY_ASSIGNED
          Duplicate redirection.

  86   ERROR_INVALID_PASSWORD
          Invalid password.

  87   ERROR_INVALID_PARAMETER
          Invalid parameter.

  88   ERROR_NET_WRITE_FAULT
          Network device fault.

  89   ERROR_NO_PROC_SLOTS
          No process slots available.

  90   ERROR_NOT_FROZEN
          System error.

  91   ERR_TSTOVFL
          Timer service table overflow.

  92   ERR_TSTDUP
          Timer service table duplicate.

  93   ERROR_NO_ITEMS
          No items to work on.

  95   ERROR_INTERRUPT
          Interrupted system call.

  99   ERROR_DEVICE_IN_USE
          Device in use.

  100   ERROR_TOO_MANY_SEMAPHORES
          User/system open semaphore limit reached.

  101   ERROR_EXCL_SEM_ALREADY_OWNED
          Exclusive semaphore already owned.

  102   ERROR_SEM_IS_SET
          DosCloseSem found semaphore set.

  103   ERROR_TOO_MANY_SEM_REQUESTS
          Too many exclusive semaphore requests.

  104   ERROR_INVALID_AT_INTERRUPT_TIME
          Operation invalid at interrupt time.

  105   ERROR_SEM_OWNER_DIED
          Previous semaphore owner terminated without freeing semaphore.

  106   ERROR_SEM_USER_LIMIT
          Semaphore limit exceeded.

  107   ERROR_DISK_CHANGE
          Insert drive B disk into drive A.

  108   ERROR_DRIVE_LOCKED
          Drive locked by another process.

  109   ERROR_BROKEN_PIPE
          Write on pipe with no reader.

  110   ERROR_OPEN_FAILED
          Open/create failed due to explicit fail command.

  111   ERROR_BUFFER_OVERFLOW
          Buffer passed to system call too small to hold return data.

  112   ERROR_DISK_FULL
          Not enough space on the disk.

  113   ERROR_NO_MORE_SEARCH_HANDLES
          Cannot allocate another search structure and handle.

  114   ERROR_INVALID_TARGET_HANDLE
          Target handle in DosDupHandle invalid.

  115   ERROR_PROTECTION_VIOLATION
          Invalid user virtual address.

  116   ERROR_VIOKBD_REQUEST
          Error on display write or keyboard read.

  117   ERROR_INVALID_CATEGORY
          Category for DevIOCtl not defined.

  118   ERROR_INVALID_VERIFY_SWITCH
          Invalid value passed for verify flag.

  119   ERROR_BAD_DRIVER_LEVEL
          Level four driver not found.

  120   ERROR_CALL_NOT_IMPLEMENTED
          Invalid function called.

  121   ERROR_SEM_TIMEOUT
          Time-out occurred from semaphore API function.

  122   ERROR_INSUFFICIENT_BUFFER
          Data buffer too small.

  123   ERROR_INVALID_NAME
          Illegal character or invalid file-system name.

  124   ERROR_INVALID_LEVEL
          Non-implemented level for information retrieval or setting.

  125   ERROR_NO_VOLUME_LABEL
          No volume label found with DosQueryFSInfo function.

  126   ERROR_MOD_NOT_FOUND
          Module handle not found with DosQueryProcAddr(),
          DosQueryModAddr().

  127   ERROR_PROC_NOT_FOUND
          Procedure address not found with DosQueryProcAddr().

  128   ERROR_WAIT_NO_CHILDREN
          DosWaitChild finds no children.

  129   ERROR_CHILD_NOT_COMPLETE
          DosWaitChild children not terminated.

  130   ERROR_DIRECT_ACCESS_HANDLE
          Handle operation invalid for direct disk-access handles.

  131   ERROR_NEGATIVE_SEEK
          Attempting seek to negative offset.

  132   ERROR_SEEK_ON_DEVICE
          Application trying to seek on device or pipe.

  133   ERROR_IS_JOIN_TARGET
          Drive has previously joined drives.

  134   ERROR_IS_JOINED
          Drive is already joined.

  135   ERROR_IS_SUBSTED
          Drive is already substituted.

  136   ERROR_NOT_JOINED
          Cannot delete drive that is not joined.

  137   ERROR_NOT_SUBSTED
          Cannot delete drive that is not substituted.

  138   ERROR_JOIN_TO_JOIN
          Cannot join to a joined drive.

  139   ERROR_SUBST_TO_SUBST
          Cannot substitute to a substituted drive.

  140   ERROR_JOIN_TO_SUBST
          Cannot join to a substituted drive.

  141   ERROR_SUBST_TO_JOIN
          Cannot substitute to a joined drive.

  142   ERROR_BUSY_DRIVE
          Specified drive is busy.

  143   ERROR_SAME_DRIVE
          Cannot join or substitute a drive to a directory on the same drive.

  144   ERROR_DIR_NOT_ROOT
          Directory must be a subdirectory of the root.

  145   ERROR_DIR_NOT_EMPTY
          Directory must be empty to use join command.

  146   ERROR_IS_SUBST_PATH
          Path specified is being used in a substitute.

  147   ERROR_IS_JOIN_PATH
          Path specified is being used in a join.

  148   ERROR_PATH_BUSY
          Path specified is being used by another process.

  149   ERROR_IS_SUBST_TARGET
          Cannot join or substitute a drive that has a directory that is the
          target of a previous substitute.

  150   ERROR_SYSTEM_TRACE
          System trace error.

  151   ERROR_INVALID_EVENT_COUNT
          DosWaitMuxWaitSem errors.

  152   ERROR_TOO_MANY_MUXWAITERS
          System limit of 100 entries reached.

  153   ERROR_INVALID_LIST_FORMAT
          Invalid list format.

  154   ERROR_LABEL_TOO_LONG
          Volume label too big.

  155   ERROR_TOO_MANY_TCBS
          Cannot create another TCB.

  156   ERROR_SIGNAL_REFUSED
          Signal refused.

  157   ERROR_DISCARDED
          Segment is discarded.

  158   ERROR_NOT_LOCKED
          Segment is not locked.

  159   ERROR_BAD_THREADID_ADDR
          Invalid thread-identity address.

  160   ERROR_BAD_ARGUMENTS
          Invalid environment pointer.

  161   ERROR_BAD_PATHNAME
          Invalid path name passed to exec.

  162   ERROR_SIGNAL_PENDING
          Signal already pending.

  163   ERROR_UNCERTAIN_MEDIA
          Error with INT 24 mapping.

  164   ERROR_MAX_THRDS_REACHED
          No more process slots.

  165   ERROR_MONITORS_NOT_SUPPORTED
          Error with INT 24 mapping.

  166   ERROR_UNC_DRIVER_NOT_INSTALLED
          Default redirection return code.

  167   ERROR_LOCK_FAILED
          Locking failed.

  168   ERROR_SWAPIO_FAILED
          Swap I/O failed.

  169   ERROR_SWAPIN_FAILED
          Swap in failed.

  170   ERROR_BUSY
          Segment is busy.

  171-172   Reserved.


  173   ERROR_CANCEL_VIOLATION
          A lock request is not outstanding for the specified file range, or the
          range length is zero.

  174   ERROR_ATOMIC_LOCK_NOT_SUPPORTED
          The file-system driver (FSD) does not support atomic lock operations.
          Versions of OS/2 prior to version 2.00 do not support atomic lock
          operations.

  175   ERROR_READ_LOCKS_NOT_SUPPORTED
          The file system driver (FSD) does not support shared read locks.

  176-179   Reserved.


  180   ERROR_INVALID_SEGMENT_NUMBER
          Invalid segment number.

  181   ERROR_INVALID_CALLGATE
          Invalid call gate.

  182   ERROR_INVALID_ORDINAL
          Invalid ordinal.

  183   ERROR_ALREADY_EXISTS
          Shared segment already exists.

  184   ERROR_NO_CHILD_PROCESS
          No child process to wait for.

  185   ERROR_CHILD_ALIVE_NOWAIT
          NoWait specified and child alive.

  186   ERROR_INVALID_FLAG_NUMBER
          Invalid flag number.

  187   ERROR_SEM_NOT_FOUND
          Semaphore does not exist.

  188   ERROR_INVALID_STARTING_CODESEG
          Invalid starting code segment, incorrect END (label) directive.

  189   ERROR_INVALID_STACKSEG
          Invalid stack segment.

  190   ERROR_INVALID_MODULETYPE
          Invalid module type - dynamic-link library file cannot be used as an
          application. Application cannot be used as a dynamic-link library.

  191   ERROR_INVALID_EXE_SIGNATURE
          Invalid EXE signature - file is a DOS mode program or an improper
          program.

  192   ERROR_EXE_MARKED_INVALID
          EXE marked invalid - link detected errors when the application was
          created.

  193   ERROR_BAD_EXE_FORMAT
          Invalid EXE format - file is a DOS mode program or an improper
          program.

  194   ERROR_ITERATED_DATA_EXCEEDS_64k
          Iterated data exceeds 64KB - there is more than 64KB of data in one
          of the segments of the file.

  195   ERROR_INVALID_MINALLOCSIZE
          Invalid minimum allocation size - the size is specified to be less than
          the size of the segment data in the file.

  196   ERROR_DYNLINK_FROM_INVALID_RING
          Dynamic link from invalid privilege level - privilege level 2 routine
          cannot link to dynamic-link libraries.

  197   ERROR_IOPL_NOT_ENABLED
          IOPL not enabled - IOPL set to NO in CONFIG.SYS.

  198   ERROR_INVALID_SEGDPL
          Invalid segment descriptor privilege level - can only have privilege
          levels of 2 and 3.

  199   ERROR_AUTODATASEG_EXCEEDS_64k
          Automatic data segment exceeds 64KB.

  200   ERROR_RING2SEG_MUST_BE_MOVABLE
          Privilege level 2 segment must be movable.

  201   ERROR_RELOC_CHAIN_XEEDS_SEGLIM
          Relocation chain exceeds segment limit.

  202   ERROR_INFLOOP_IN_RELOC_CHAIN
          Infinite loop in relocation chain segment.

  203   ERROR_ENVVAR_NOT_FOUND
          Environment variable not found.

  204   ERROR_NOT_CURRENT_CTRY
          Not current country.

  205   ERROR_NO_SIGNAL_SENT
          No signal sent - no process in the command subtree has a signal
          handler.

  206   ERROR_FILENAME_EXCED_RANGE
          File name or extension is greater than 8.3 characters.

  207   ERROR_RING2_STACK_IN_USE
          Privilege level 2 stack is in use.

  208   ERROR_META_EXPANSION_TOO_LONG
          Meta (global) expansion is too long.

  209   ERROR_INVALID_SIGNAL_NUMBER
          Invalid signal number.

  210   ERROR_THREAD_1_INACTIVE
          Inactive thread.

  211   ERROR_INFO_NOT_AVAIL
          File system information is not available for this file.

  212   ERROR_LOCKED
          Locked error.

  213   ERROR_BAD_DYNALINK
          Attempted to execute a non-family API in DOS mode.

  214   ERROR_TOO_MANY_MODULES
          Too many modules.

  215   ERROR_NESTING_NOT_ALLOWED
          Nesting is not allowed.

  217   ERROR_ZOMBIE_PROCESS
          Zombie process.

  218   ERROR_STACK_IN_HIGH_MEMORY
          Stack is in high memory.

  219   ERROR_INVALID_EXITROUTINE_RING
          Invalid exit routine ring.

  220   ERROR_GETBUF_FAILED
          Get buffer failed.

  221   ERROR_FLUSHBUF_FAILED
          Flush buffer failed.

  222   ERROR_TRANSFER_TOO_LONG
          Transfer is too long.

  224   ERROR_SMG_NO_TARGET_WINDOW
          The application window was created without the FCF_TASKLIST
          style, or the application window not yet been created or has already
          been destroyed.

  228   ERROR_NO_CHILDREN
          No child process.

  229   ERROR_INVALID_SCREEN_GROUP
          Invalid session.

  230   ERROR_BAD_PIPE
          Non-existent pipe or invalid operation.

  231   ERROR_PIPE_BUSY
          Pipe is busy.

  232   ERROR_NO_DATA
          No data available on non-blocking read.

  233   ERROR_PIPE_NOT_CONNECTED
          Pipe was disconnected by server.

  234   ERROR_MORE_DATA
          More data is available.

  240   ERROR_VC_DISCONNECTED
          Session was dropped due to errors.

  250   ERROR_CIRCULARITY_REQUESTED
          Renaming a directory that would cause a circularity problem.

  251   ERROR_DIRECTORY_IN_CDS
          Renaming a directory that is in use.

  252   ERROR_INVALID_FSD_NAME
          Trying to access nonexistent FSD.

  253   ERROR_INVALID_PATH
          Invalid pseudo device.

  254   ERROR_INVALID_EA_NAME
          Invalid character in name, or invalid cbName.

  255   ERROR_EA_LIST_INCONSISTENT
          List does not match its size, or there are invalid EAs in the list.

  256   ERROR_EA_LIST_TOO_LONG
          FEAList is longer than 64K-1 bytes.

  257   ERROR_NO_META_MATCH
          String does not match expression.

  259   ERROR_NO_MORE_ITEMS
          DosQueryFSAttach ordinal query.

  260   ERROR_SEARCH_STRUC_REUSED
          DOS mode findfirst/next search structure reused.

  261   ERROR_CHAR_NOT_FOUND
          Character not found.

  262   ERROR_TOO_MUCH_STACK
          Stack request exceeds system limit.

  263   ERROR_INVALID_ATTR
          Invalid attribute.

  264   ERROR_INVALID_STARTING_RING
          Invalid starting ring.

  265   ERROR_INVALID_DLL_INIT_RING
          Invalid DLL INIT ring.

  266   ERROR_CANNOT_COPY
          Cannot copy.

  267   ERROR_DIRECTORY
          Used by DOSCOPY in doscall1.

  268   ERROR_OPLOCKED_FILE
          Oplocked file.

  269   ERROR_OPLOCK_THREAD_EXISTS
          Oplock thread exists.

  270   ERROR_VOLUME_CHANGED
          Volume changed.

  271-273   Reserved.


  274   ERROR_ALREADY_SHUTDOWN
          System is already shut down.

  275   ERROR_EAS_DIDNT_FIT
          Buffer is not big enough to hold the EAs.

  276   ERROR_EA_FILE_CORRUPT
          EA file has been damaged.

  277   ERROR_EA_TABLE_FULL
          EA table is full.

  278   ERROR_INVALID_EA_HANDLE
          EA handle is invalid.

  279   ERROR_NO_CLUSTER
          No cluster.

  280   ERROR_CREATE_EA_FILE
          Cannot create the EA file.

  281   ERROR_CANNOT_OPEN_EA_FILE
          Cannot open the EA file.

  282   ERROR_EAS_NOT_SUPPORTED
          Destination file system does not support EAs.

  283   ERROR_NEED_EAS_FOUND
          Destination file system does not support EAs, and the source file's
          EAs contain a need EA.

  284   ERROR_DUPLICATE_HANDLE
          The handle already exists.

  285   ERROR_DUPLICATE_NAME
          The name already exists.

  286   ERROR_EMPTY_MUXWAIT
          The list of semaphores in a muxwait semaphore is empty.

  287   ERROR_MUTEX_OWNED
          The calling thread owns one or more of the mutex semaphores in the
          list.

  288   ERROR_NOT_OWNER
          Caller does not own the semaphore.

  289   ERROR_PARAM_TOO_SMALL
          Parameter is not large enough to contain all of the semaphore
          records in the muxwait semaphore.

  290   ERROR_TOO_MANY_HANDLES
          Limit reached for number of handles.

  291   ERROR_TOO_MANY_OPENS
          There are too many files or semaphores open.

  292   ERROR_WRONG_TYPE
          Attempted to create wrong type of semaphore.

  293   ERROR_UNUSED_CODE
          Code is not used.

  294   ERROR_THREAD_NOT_TERMINATED
          Thread has not terminated.

  295   ERROR_INIT_ROUTINE_FAILED
          Initialization routine failed.

  296   ERROR_MODULE_IN_USE
          Module is in use.

  297   ERROR_NOT_ENOUGH_WATCHPOINTS
          There are not enough watchpoints.

  298   ERROR_TOO_MANY_POSTS
          Post count limit was reached for an event semaphore.

  299   ERROR_ALREADY_POSTED
          Event semaphore is already posted.

  300   ERROR_ALREADY_RESET
          Event semaphore is already reset.

  301   ERROR_SEM_BUSY
          Semaphore is busy.

  302   Reserved


  303   ERROR_INVALID_PROCID
          Invalid process identity.

  304   ERROR_INVALID_PDELTA
          Invalid priority delta.

  305   ERROR_NOT_DESCENDANT
          Not descendant.

  306   ERROR_NOT_SESSION_MANAGER
          Requestor not session manager.

  307   ERROR_INVALID_PCLASS
          Invalid P class.

  308   ERROR_INVALID_SCOPE
          Invalid scope.

  309   ERROR_INVALID_THREADID
          Invalid thread identity.

  310   ERROR_DOSSUB_SHRINK
          Cannot shrink segment - DosSubSetMem.

  311   ERROR_DOSSUB_NOMEM
          No memory to satisfy request - DosSubAllocMem.

  312   ERROR_DOSSUB_OVERLAP
          Overlap of the specified block with a block of allocated memory -
          DosSubFreeMem.

  313   ERROR_DOSSUB_BADSIZE
          Invalid size parameter - DosSubAllocMem or DosSubFreeMem.

  314   ERROR_DOSSUB_BADFLAG
          Invalid flag parameter - DosSubSetMem.

  315   ERROR_DOSSUB_BADSELECTOR
          Invalid segment selector.

  316   ERROR_MR_MSG_TOO_LONG
          Message too long for buffer.

  317   ERROR_MR_MID_NOT_FOUND
          Message identity number not found.

  318   ERROR_MR_UN_ACC_MSGF
          Unable to access message file.

  319   ERROR_MR_INV_MSGF_FORMAT
          Invalid message file format.

  320   ERROR_MR_INV_IVCOUNT
          Invalid insertion variable count.

  321   ERROR_MR_UN_PERFORM
          Unable to perform function.

  322   ERROR_TS_WAKEUP
          Unable to wake up.

  323   ERROR_TS_SEMHANDLE
          Invalid system semaphore.

  324   ERROR_TS_NOTIMER
          No timers available.

  326   ERROR_TS_HANDLE
          Invalid timer handle.

  327   ERROR_TS_DATETIME
          Date or time invalid.

  328   ERROR_SYS_INTERNAL
          Internal system error.

  329   ERROR_QUE_CURRENT_NAME
          Current queue name does not exist.

  330   ERROR_QUE_PROC_NOT_OWNED
          Current process does not own queue.

  331   ERROR_QUE_PROC_OWNED
          Current process owns queue.

  332   ERROR_QUE_DUPLICATE
          Duplicate queue name.

  333   ERROR_QUE_ELEMENT_NOT_EXIST
          Queue element does not exist.

  334   ERROR_QUE_NO_MEMORY
          Inadequate queue memory.

  335   ERROR_QUE_INVALID_NAME
          Invalid queue name.

  336   ERROR_QUE_INVALID_PRIORITY
          Invalid queue priority parameter.

  337   ERROR_QUE_INVALID_HANDLE
          Invalid queue handle.

  338   ERROR_QUE_LINK_NOT_FOUND
          Queue link not found.

  339   ERROR_QUE_MEMORY_ERROR
          Queue memory error.

  340   ERROR_QUE_PREV_AT_END
          Previous queue element was at end of queue.

  341   ERROR_QUE_PROC_NO_ACCESS
          Process does not have access to queues.

  342   ERROR_QUE_EMPTY
          Queue is empty.

  343   ERROR_QUE_NAME_NOT_EXIST
          Queue name does not exist.

  344   ERROR_QUE_NOT_INITIALIZED
          Queues not initialized.

  345   ERROR_QUE_UNABLE_TO_ACCESS
          Unable to access queues.

  346   ERROR_QUE_UNABLE_TO_ADD
          Unable to add new queue.

  347   ERROR_QUE_UNABLE_TO_INIT
          Unable to initialize queues.

  349   ERROR_VIO_INVALID_MASK
          Invalid function replaced.

  350   ERROR_VIO_PTR
          Invalid pointer to parameter.

  351   ERROR_VIO_APTR
          Invalid pointer to attribute.

  352   ERROR_VIO_RPTR
          Invalid pointer to row.

  353   ERROR_VIO_CPTR
          Invalid pointer to column.

  354   ERROR_VIO_LPTR
          Invalid pointer to length.

  355   ERROR_VIO_MODE
          Unsupported screen mode.

  356   ERROR_VIO_WIDTH
          Invalid cursor width value.

  357   ERROR_VIO_ATTR
          Invalid cursor attribute value.

  358   ERROR_VIO_ROW
          Invalid row value.

  359   ERROR_VIO_COL
          Invalid column value.

  360   ERROR_VIO_TOPROW
          Invalid TopRow value.

  361   ERROR_VIO_BOTROW
          Invalid BotRow value.

  362   ERROR_VIO_RIGHTCOL
          Invalid right column value.

  363   ERROR_VIO_LEFTCOL
          Invalid left column value.

  364   ERROR_SCS_CALL
          Call issued by other than session manager.

  365   ERROR_SCS_VALUE
          Value is not for save or restore.

  366   ERROR_VIO_WAIT_FLAG
          Invalid wait flag setting.

  367   ERROR_VIO_UNLOCK
          Screen not previously locked.

  368   ERROR_SGS_NOT_SESSION_MGR
          Caller not session manager.

  369   ERROR_SMG_INVALID_SGID
          Invalid session identity.

  369   ERROR_SMG_INVALID_SESSION_ID
          Invalid session ID.

  370   ERROR_SMG_NOSG
          No sessions available.

  370   ERROR_SMG_NO_SESSIONS
          No sessions available.

  371   ERROR_SMG_GRP_NOT_FOUND
          Session not found.

  371   ERROR_SMG_SESSION_NOT_FOUND
          Session not found.

  372   ERROR_SMG_SET_TITLE
          Title sent by shell or parent cannot be changed.

  373   ERROR_KBD_PARAMETER
          Invalid parameter to keyboard.

  374   ERROR_KBD_NO_DEVICE
          No device.

  375   ERROR_KBD_INVALID_IOWAIT
          Invalid I/O wait specified.

  376   ERROR_KBD_INVALID_LENGTH
          Invalid length for keyboard.

  377   ERROR_KBD_INVALID_ECHO_MASK
          Invalid echo mode mask.

  378   ERROR_KBD_INVALID_INPUT_MASK
          Invalid input mode mask.

  379   ERROR_MON_INVALID_PARMS
          Invalid parameters to DosMon.

  380   ERROR_MON_INVALID_DEVNAME
          Invalid device name string.

  381   ERROR_MON_INVALID_HANDLE
          Invalid device handle.

  382   ERROR_MON_BUFFER_TOO_SMALL
          Buffer too small.

  383   ERROR_MON_BUFFER_EMPTY
          Buffer is empty.

  384   ERROR_MON_DATA_TOO_LARGE
          Data record is too large.

  385   ERROR_MOUSE_NO_DEVICE
          Mouse device closed (invalid device handle).

  386   ERROR_MOUSE_INV_HANDLE
          Mouse device closed (invalid device handle).

  387   ERROR_MOUSE_INV_PARMS
          Parameters invalid for display mode.

  388   ERROR_MOUSE_CANT_RESET
          Function assigned and cannot be reset.

  389   ERROR_MOUSE_DISPLAY_PARMS
          Parameters invalid for display mode.

  390   ERROR_MOUSE_INV_MODULE
          Module not valid.

  391   ERROR_MOUSE_INV_ENTRY_PT
          Entry point not valid.

  392   ERROR_MOUSE_INV_MASK
          Function mask invalid.

  393   NO_ERROR_MOUSE_NO_DATA
          No valid data.

  394   NO_ERROR_MOUSE_PTR_DRAWN
          Pointer drawn.

  395   ERROR_INVALID_FREQUENCY
          Invalid frequency for beep.

  396   ERROR_NLS_NO_COUNTRY_FILE
          Cannot find COUNTRY.SYS file.

  397   ERROR_NLS_OPEN_FAILED
          Cannot open COUNTRY.SYS file.

  398   ERROR_NLS_NO_CTRY_CODE
          Country code not found.

  398   ERROR_NO_COUNTRY_OR_CODEPAGE
          Country code not found.

  399   ERROR_NLS_TABLE_TRUNCATED
          Table returned information truncated, buffer is too small.

  400   ERROR_NLS_BAD_TYPE
          Selected type does not exist.

  401   ERROR_NLS_TYPE_NOT_FOUND
          Selected type is not in file.

  402   ERROR_VIO_SMG_ONLY
          Valid from session manager only.

  403   ERROR_VIO_INVALID_ASCIIZ
          Invalid ASCIIZ length.

  404   ERROR_VIO_DEREGISTER
          VioDeRegister not allowed.

  405   ERROR_VIO_NO_POPUP
          Pop-up window not allocated.

  406   ERROR_VIO_EXISTING_POPUP
          Pop-up window on screen (NoWait).

  407   ERROR_KBD_SMG_ONLY
          Valid from session manager only.

  408   ERROR_KBD_INVALID_ASCIIZ
          Invalid ASCIIZ length.

  409   ERROR_KBD_INVALID_MASK
          Invalid replacement mask.

  410   ERROR_KBD_REGISTER
          KbdRegister not allowed.

  411   ERROR_KBD_DEREGISTER
          KbdDeRegister not allowed.

  412   ERROR_MOUSE_SMG_ONLY
          Valid from session manager only.

  413   ERROR_MOUSE_INVALID_ASCIIZ
          Invalid ASCIIZ length.

  414   ERROR_MOUSE_INVALID_MASK
          Invalid replacement mask.

  415   ERROR_MOUSE_REGISTER
          Mouse register not allowed.

  416   ERROR_MOUSE_DEREGISTER
          Mouse deregister not allowed.

  417   ERROR_SMG_BAD_ACTION
          Invalid action specified.

  418   ERROR_SMG_INVALID_CALL
          INIT called more than once, or invalid session identity.

  419   ERROR_SCS_SG_NOTFOUND
          New session number.

  420   ERROR_SCS_NOT_SHELL
          Caller is not shell.

  421   ERROR_VIO_INVALID_PARMS
          Invalid parameters passed.

  422   ERROR_VIO_FUNCTION_OWNED
          Save/restore already owned.

  423   ERROR_VIO_RETURN
          Non-destruct return (undo).

  424   ERROR_SCS_INVALID_FUNCTION
          Caller invalid function.

  425   ERROR_SCS_NOT_SESSION_MGR
          Caller not session manager.

  426   ERROR_VIO_REGISTER
          Vio register not allowed.

  427   ERROR_VIO_NO_MODE_THREAD
          No mode restore thread in SG.

  428   ERROR_VIO_NO_SAVE_RESTORE_THD
          No save/restore thread in SG.

  429   ERROR_VIO_IN_BG
          Function invalid in background.

  430   ERROR_VIO_ILLEGAL_DURING_POPUP
          Function not allowed during pop-up window.

  431   ERROR_SMG_NOT_BASESHELL
          Caller is not the base shell.

  432   ERROR_SMG_BAD_STATUSREQ
          Invalid status requested.

  433   ERROR_QUE_INVALID_WAIT
          NoWait parameter out of bounds.

  434   ERROR_VIO_LOCK
          Error returned from Scroll Lock.

  435   ERROR_MOUSE_INVALID_IOWAIT
          Invalid parameters for IOWait.

  436   ERROR_VIO_INVALID_HANDLE
          Invalid VIO handle.

  437   ERROR_VIO_ILLEGAL_DURING_LOCK
          Function not allowed during screen lock.

  438   ERROR_VIO_INVALID_LENGTH
          Invalid VIO length.

  439   ERROR_KBD_INVALID_HANDLE
          Invalid KBD handle.

  440   ERROR_KBD_NO_MORE_HANDLE
          Ran out of handles.

  441   ERROR_KBD_CANNOT_CREATE_KCB
          Unable to create kcb.

  442   ERROR_KBD_CODEPAGE_LOAD_INCOMPL
          Unsuccessful code-page load.

  443   ERROR_KBD_INVALID_CODEPAGE_ID
          Invalid code-page identity.

  444   ERROR_KBD_NO_CODEPAGE_SUPPORT
          No code page support.

  445   ERROR_KBD_FOCUS_REQUIRED
          Keyboard focus required.

  446   ERROR_KBD_FOCUS_ALREADY_ACTIVE
          Calling thread has an outstanding focus.

  447   ERROR_KBD_KEYBOARD_BUSY
          Keyboard is busy.

  448   ERROR_KBD_INVALID_CODEPAGE
          Invalid code page.

  449   ERROR_KBD_UNABLE_TO_FOCUS
          Focus attempt failed.

  450   ERROR_SMG_SESSION_NON_SELECT
          Session is not selectable.

  451   ERROR_SMG_SESSION_NOT_FOREGRND
          Parent/child session is not foreground.

  452   ERROR_SMG_SESSION_NOT_PARENT
          Not parent of requested child.

  453   ERROR_SMG_INVALID_START_MODE
          Invalid session start mode.

  454   ERROR_SMG_INVALID_RELATED_OPT
          Invalid session start related option.

  455   ERROR_SMG_INVALID_BOND_OPTION
          Invalid session bond option.

  456   ERROR_SMG_INVALID_SELECT_OPT
          Invalid session select option.

  457   ERROR_SMG_START_IN_BACKGROUND
          Session started in background.

  458   ERROR_SMG_INVALID_STOP_OPTION
          Invalid session stop option.

  459   ERROR_SMG_BAD_RESERVE
          Reserved parameters are not zero.

  460   ERROR_SMG_PROCESS_NOT_PARENT
          Session parent process already exists.

  461   ERROR_SMG_INVALID_DATA_LENGTH
          Invalid data length.

  462   ERROR_SMG_NOT_BOUND
          Parent is not bound.

  463   ERROR_SMG_RETRY_SUB_ALLOC
          Retry request block allocation.

  464   ERROR_KBD_DETACHED
          This call is not allowed for a detached PID.

  465   ERROR_VIO_DETACHED
          This call is not allowed for a detached PID.

  466   ERROR_MOU_DETACHED
          This call is not allowed for a detached PID.

  467   ERROR_VIO_FONT
          No font is available to support the mode.

  468   ERROR_VIO_USER_FONT
          User font is active.

  469   ERROR_VIO_BAD_CP
          Invalid code page specified.

  470   ERROR_VIO_NO_CP
          System displays do not support code page.

  471   ERROR_VIO_NA_CP
          Current display does not support code page.

  472   ERROR_INVALID_CODE_PAGE
          Invalid code page.

  473   ERROR_CPLIST_TOO_SMALL
          Code page list is too small.

  474   ERROR_CP_NOT_MOVED
          Code page was not moved.

  475   ERROR_MODE_SWITCH_INIT
          Mode switch initialization error.

  476   ERROR_CODE_PAGE_NOT_FOUND
          Code page was not found.

  477   ERROR_UNEXPECTED_SLOT_RETURNED
          Internal error.

  478   ERROR_SMG_INVALID_TRACE_OPTION
          Invalid start session trace indicator.

  479   ERROR_VIO_INTERNAL_RESOURCE
          VIO internal resource error.

  480   ERROR_VIO_SHELL_INIT
          VIO shell initialization error.

  481   ERROR_SMG_NO_HARD_ERRORS
          No session manager hard errors.

  482   ERROR_CP_SWITCH_INCOMPLETE
          DosSetProcessCp is unable to set a KBD or VIO code page.

  483   ERROR_VIO_TRANSPARENT_POPUP
          Error during VIO pop-up window.

  484   ERROR_CRITSEC_OVERFLOW
          Critical section overflow.

  485   ERROR_CRITSEC_UNDERFLOW
          Critical section underflow.

  486   ERROR_VIO_BAD_RESERVE
          Reserved parameter is not zero.

  487   ERROR_INVALID_ADDRESS
          Invalid physical address.

  488   ERROR_ZERO_SELECTORS_REQUESTED
          At least one selector must be requested.

  489   ERROR_NOT_ENOUGH_SELECTORS_AVA
          Not enough GDT selectors to satisfy request.

  490   ERROR_INVALID_SELECTOR
          Not a GDT selector.

  491   ERROR_SMG_INVALID_PROGRAM_TYPE
          Invalid program type.

  492   ERROR_SMG_INVALID_PGM_CONTROL
          Invalid program control.

  493   ERROR_SMG_INVALID_INHERIT_OPT
          Invalid inherit option.

  494   ERROR_VIO_EXTENDED_SG


  495   ERROR_VIO_NOT_PRES_MGR_SG


  496   ERROR_VIO_SHIELD_OWNED


  497   ERROR_VIO_NO_MORE_HANDLES


  498   ERROR_VIO_SEE_ERROR_LOG


  499   ERROR_VIO_ASSOCIATED_DC


  500   ERROR_KBD_NO_CONSOLE


  501   ERROR_MOUSE_NO_CONSOLE


  502   ERROR_MOUSE_INVALID_HANDLE


  503   ERROR_SMG_INVALID_DEBUG_PARMS


  504   ERROR_KBD_EXTENDED_SG


  505   ERROR_MOU_EXTENDED_SG


  506   ERROR_SMG_INVALID_ICON_FILE


  507   ERROR_TRC_PID_NON_EXISTENT


  508   ERROR_TRC_COUNT_ACTIVE


  509   ERROR_TRC_SUSPENDED_BY_COUNT


  510   ERROR_TRC_COUNT_INACTIVE


  511   ERROR_TRC_COUNT_REACHED


  512   ERROR_NO_MC_TRACE


  513   ERROR_MC_TRACE


  514   ERROR_TRC_COUNT_ZERO


  515   ERROR_SMG_TOO_MANY_DDS


  516   ERROR_SMG_INVALID_NOTIFICATION


  517   ERROR_LF_INVALID_FUNCTION


  518   ERROR_LF_NOT_AVAIL


  519   ERROR_LF_SUSPENDED


  520   ERROR_LF_BUF_TOO_SMALL


  521   ERROR_LF_BUFFER_CORRUPTED


  521   ERROR_LF_BUFFER_FULL


  522   ERROR_LF_INVALID_DAEMON


  522   ERROR_LF_INVALID_RECORD


  523   ERROR_LF_INVALID_TEMPL


  523   ERROR_LF_INVALID_SERVICE


  524   ERROR_LF_GENERAL_FAILURE


  525   ERROR_LF_INVALID_ID


  526   ERROR_LF_INVALID_HANDLE


  527   ERROR_LF_NO_ID_AVAIL


  528   ERROR_LF_TEMPLATE_AREA_FULL


  529   ERROR_LF_ID_IN_USE


  530   ERROR_MOU_NOT_INITIALIZED


  531   ERROR_MOUINITREAL_DONE


  532   ERROR_DOSSUB_CORRUPTED


  533   ERROR_MOUSE_CALLER_NOT_SUBSYS


  534   ERROR_ARITHMETIC_OVERFLOW


  535   ERROR_TMR_NO_DEVICE


  536   ERROR_TMR_INVALID_TIME


  537   ERROR_PVW_INVALID_ENTITY


  538   ERROR_PVW_INVALID_ENTITY_TYPE


  539   ERROR_PVW_INVALID_SPEC


  540   ERROR_PVW_INVALID_RANGE_TYPE


  541   ERROR_PVW_INVALID_COUNTER_BLK


  542   ERROR_PVW_INVALID_TEXT_BLK


  543   ERROR_PRF_NOT_INITIALIZED


  544   ERROR_PRF_ALREADY_INITIALIZED


  545   ERROR_PRF_NOT_STARTED


  546   ERROR_PRF_ALREADY_STARTED


  547   ERROR_PRF_TIMER_OUT_OF_RANGE


  548   ERROR_PRF_TIMER_RESET


  639   ERROR_VDD_LOCK_USEAGE_DENIED


  640   ERROR_TIMEOUT


  641   ERROR_VDM_DOWN


  642   ERROR_VDM_LIMIT


  643   ERROR_VDD_NOT_FOUND


  644   ERROR_INVALID_CALLER


  645   ERROR_PID_MISMATCH


  646   ERROR_INVALID_VDD_HANDLE


  647   ERROR_VLPT_NO_SPOOLER


  648   ERROR_VCOM_DEVICE_BUSY


  649   ERROR_VLPT_DEVICE_BUSY


  650   ERROR_NESTING_TOO_DEEP


  651   ERROR_VDD_MISSING


  691   ERROR_IMP_INVALID_PARM


  692   ERROR_IMP_INVALID_LENGTH


  693   MSG_HPFS_DISK_ERROR_WARN


  730   ERROR_MON_BAD_BUFFER


  731   ERROR_MODULE_CORRUPTED


  2055   ERROR_LF_TIMEOUT


  2057   ERROR_LF_SUSPEND_SUCCESS


  2058   ERROR_LF_RESUME_SUCCESS


  2059   ERROR_LF_REDIRECT_SUCCESS


  2060   ERROR_LF_REDIRECT_FAILURE


  32768   ERROR_SWAPPER_NOT_ACTIVE


  32769   ERROR_INVALID_SWAPID


  32770   ERROR_IOERR_SWAP_FILE


  32771   ERROR_SWAP_TABLE_FULL


  32772   ERROR_SWAP_FILE_FULL


  32773   ERROR_CANT_INIT_SWAPPER


  32774   ERROR_SWAPPER_ALREADY_INIT


  32775   ERROR_PMM_INSUFFICIENT_MEMORY


  32776   ERROR_PMM_INVALID_FLAGS


  32777   ERROR_PMM_INVALID_ADDRESS


  32778   ERROR_PMM_LOCK_FAILED


  32779   ERROR_PMM_UNLOCK_FAILED


  32780   ERROR_PMM_MOVE_INCOMPLETE


  32781   ERROR_UCOM_DRIVE_RENAMED


  32782   ERROR_UCOM_FILENAME_TRUNCATED


  32783   ERROR_UCOM_BUFFER_LENGTH


  32784   ERROR_MON_CHAIN_HANDLE


  32785   ERROR_MON_NOT_REGISTERED


  32786   ERROR_SMG_ALREADY_TOP


  32787   ERROR_PMM_ARENA_MODIFIED


  32788   ERROR_SMG_PRINTER_OPEN


  32789   ERROR_PMM_SET_FLAGS_FAILED


  32790   ERROR_INVALID_DOS_DD


  32791   ERROR_BLOCKED


  32792   ERROR_NOBLOCK


  32793   ERROR_INSTANCE_SHARED


  32794   ERROR_NO_OBJECT


  32795   ERROR_PARTIAL_ATTACH


  32796   ERROR_INCACHE


  32797   ERROR_SWAP_IO_PROBLEMS


  32798   ERROR_CROSSES_OBJECT_BOUNDARY


  32799   ERROR_LONGLOCK


  32800   ERROR_SHORTLOCK


  32801   ERROR_UVIRTLOCK


  32802   ERROR_ALIASLOCK


  32803   ERROR_ALIAS


  32804   ERROR_NO_MORE_HANDLES


  32805   ERROR_SCAN_TERMINATED


  32806   ERROR_TERMINATOR_NOT_FOUND


  32807   ERROR_NOT_DIRECT_CHILD


  32808   ERROR_DELAY_FREE


  32809   ERROR_GUARDPAGE


  32900   ERROR_SWAPERROR


  32901   ERROR_LDRERROR


  32902   ERROR_NOMEMORY


  32903   ERROR_NOACCESS


  32904   ERROR_NO_DLL_TERM


  65026   ERROR_CPSIO_CODE_PAGE_INVALID


  65027   ERROR_CPSIO_NO_SPOOLER


  65028   ERROR_CPSIO_FONT_ID_INVALID


  65033   ERROR_CPSIO_INTERNAL_ERROR


  65034   ERROR_CPSIO_INVALID_PTR_NAME


  65037   ERROR_CPSIO_NOT_ACTIVE


  65039   ERROR_CPSIO_PID_FULL


  65040   ERROR_CPSIO_PID_NOT_FOUND


  65043   ERROR_CPSIO_READ_CTL_SEQ


  65045   ERROR_CPSIO_READ_FNT_DEF


  65047   ERROR_CPSIO_WRITE_ERROR


  65048   ERROR_CPSIO_WRITE_FULL_ERROR


  65049   ERROR_CPSIO_WRITE_HANDLE_BAD


  65074   ERROR_CPSIO_SWIT_LOAD


  65077   ERROR_CPSIO_INV_COMMAND


  65078   ERROR_CPSIO_NO_FONT_SWIT


  65079   ERROR_ENTRY_IS_CALLGATE


ΓòÉΓòÉΓòÉ 13. Specifications ΓòÉΓòÉΓòÉ

Specifications covered:

Γûá API Calls Made

Γûá Memory Usage

Γûá IX3 File Format

Γûá DBF File Format

Γûá DBT File Format

Γûá Custom Sort-Compare Function

Γûá Custom Build-Key

Γûá Custom Expression Parser


ΓòÉΓòÉΓòÉ 13.1. OS/2 API Calls Made ΓòÉΓòÉΓòÉ

The following are the API calls made by Bullet.

 DosAllocMem             DosClose                 DosCopy
 DosCloseMutexSem        DosCreateDir             DosCreateMutexSem
 DosDelete (*)           DosErrClass              DosExitList
 DosForceDelete          DosFreeMem               DosGetDateTime
 DosMapCase              DosOpen                  DosQueryCollate
 DosQueryCp              DosQueryCtryInfo         DosQueryCurrentDisk
 DosQueryFSAttach        DosQueryHType            DosQuerySysInfo
 DosMove                 DosRead                  DosReleaseMutexSem
 DosRequestMutexSem      DosResetBuffer           DosScanEnv
 DosSetFileLocks         DosSetFilePtr            DosSetFileSize
 DosSetMaxFH             DosSetRelMaxFH           DosWrite

 (*) DosForceDelete is used in favor of DosDelete.


ΓòÉΓòÉΓòÉ 13.2. Bullet Memory Usage ΓòÉΓòÉΓòÉ

Memory is committed when allocated, using the PAG_COMMIT and the PAG_WRITE
flags.  This is memory allocated by Bullet itself.  Additional memory needs are
made by your code, such as parameter pack data, key buffers, and data record
buffers.

Code

Bullet uses 8 pages for code, or less than 32KB.

Data

Γûá  Shared Data

A single page of shared memory is used by all Bullet processes.

Γûá  Instance Data

One page of private memory is used by each Bullet processes.

Γûá  Handle Data

One page of private memory is used by each open Bullet index file.  For open
data files, one page of private memory is used for files with 121 or fewer
fields.  Two pages are used for files with 249 or fewer fields. Thereafter, one
page for each additional 128 fields, up to 1024 max fields.

For example, if one machine is running 2 Bullet processes, each with 10 open
data files with 12 fields each, and 10 index files (one for each data file),
its total memory usage is:

     Total code is 32KB.
     Shared data is 4KB.
     Instance data is 2 processes * 4KB, or 8KB (plus INIT_XB use shown
      below).
     DBF file data is 2 * 10 files * 4KB, or 80KB.
     Index file data is 2 * 10 files * 4KB, or 80KB.

 Total memory committed by Bullet for the above is 200KB, plus code and data of
 your two applications (or your single application, if the same application is
 being run twice).  With no files open, for example when starting your program,
 only 40KB is committed.  Thereafter, 4KB per open file.  The memory is freed
 when the file is closed.

 Γûá  Temporary Data

 Additional memory is allocated on a temporary basis, where the allocation is
 requested on entry to, and is freed upon exiting from, the routine called.
 INIT_XB's allocation can be considered permanent since its allocations are not
 released until the program has ended or EXIT_XB is issued.  The following are
 the routines and the single requested amount:

     Routine          Memory Allocated, in KB

  INIT_XB             8 for 1024 MAX_FILES version; 4KB for 100 and 250 MAX_FILES versions
  BACKUP_FILE_XB      8
  CREATE_DATA_XB      4 for 1-121 fields, 8 for 122-249 fields, ...
  CREATE_INDEX_XB     4
  PACK_RECORDS_XB     adjustable, 128 default (less if file is smaller)
  REINDEX_XB          adjustable, 144 default (minimum size is 48KB)
  UPDATE_XB           varies:      40 + sum of record lengths where AP[].recNo!=0

 Γûá  Stack Data

 Stack requirements are 8KB minimum for Bullet.  No single stack allocation
 requests more than 4KB at a time (ie all changes to ESP (the CPU stack
 pointer) are less than 4KB at any one time), but some routines nest and
 require up to the minimum 8KB in total.  The minimum recommended stack size
 for your Bullet application is 16KB.  It's likely that you need to use a much
 larger size for your main program's stack use.  If you have any doubt about
 stack space, double it, twice even.


ΓòÉΓòÉΓòÉ 13.3. IX3 File Format ΓòÉΓòÉΓòÉ

The IX3 index file is composed of a header followed by node data.  The header
layout is detailed below, followed by the node format.

Index Header

// nnn, where nnn is the offset of that item relative to the start of the file

CHAR fileID[4];     // 000 file id = '31ch'
ULONG nodeSize;     // 004 size of a node, in bytes
ULONG rootNode;     // 008 root node (1-based)
ULONG noKeys;       // 012 total number of keys
ULONG availNode;    // 016 next available node (link to, 0 if none, 1-based)
ULONG freeNode;     // 020 next free node
ULONG keyLength;    // 024 length of key
ULONG maxKeys;      // 028 maximum number of keys on a node
ULONG codePage;     // 032 code page from CreateIndexFile
ULONG countryCode;  // 036 country code from CreateIndexFile
ULONG sortFunction; // 040 system (1-9) or custom (10-19)
                    //     high word has flags: bit0=1 dups allowed
                    //                          bit1-15 rez

// Translated key expression as done by Parser during CreateIndex and Reindex.
// More details on this is in the Custom Expression Parser Specifications.
// For each key part in KH.expression a 4-byte structure is used in XLATEX:

typedef struct _XLATEX {
 CHAR  ftype;   // field type (C,N,L, etc.),if bit7=1 and C then do UPPER key
 CHAR  length;  // bytes to use starting at offset (never > 64)
 SHORT offset;  // byte offset into data record that length bytes are to be used
} XLATEX;

ULONG xlateCount;         // 044 number of key fields (64/4=16 max fields)
XLATEX xlateExpression[16]; // 048 key construct info (16 dword's worth)
CHAR  miscWorkspace[236]; // 112-347 B-tree workspace
CHAR  expression[160];    // 348 key expression, user (0-Terminated)
ULONG CTsize;             // 508 size of collate table following
CHAR  collateTable[256];  // 512 collate table, fill at CreateIndexXB
CHAR  rez[256];           // 768 to 1023 reserved (header size=1024 bytes)

Node Data

Directly after the header the node data starts.  Each node is either 512, 1024,
or 2048 bytes long.  Each node contains a key count, indicating the number of
active keys on the node, followed by key data.

// nnn, where nnn is the offset of that item relative to the start of the node

CHAR  keyCount;         // 000 1 to maxKeys (in header above)
ULONG backNode;         // 001 previous node page, or 0 if this node is a leaf
XNODE node[maxKeys];    // 005...

For each key on the node:

typedef struct _XNODE {
CHAR  keyValue[keyLength]; // 005   actual key   (keyLength from header)
ULONG recordNo;            // 005+keyLength   record number for key
ULONG fwdNode;             // 005+keyLength+4   next node page, or 0 if leaf
} XNODE;

backNode and fwdNode are node numbers.  The first node is 1, and is located
directly after the header.  The last node used is at header:freeNode-1.  Each
fwdNode of a key is also the next key's backNode.  If the node has had all keys
removed, its node number is placed on the top of the header:availNode list, and
the first 4 bytes of the node are used as a link to the previous list top.


ΓòÉΓòÉΓòÉ 13.4. DBF File Format ΓòÉΓòÉΓòÉ

The DBF data file is composed of a header, field descriptors, one per field,
and the actual record data.  The header layout is detailed below, followed by
the field descriptor layout and then the description of the data record. The
standard DBF file has a record length limit of 4000 bytes.  Creating record
lengths greater than 4000 is allowed in Bullet, but these files may not be
recognized by other applications if you do so.

DBF Header

// nnn, where nnn is the offset of that item relative to the start of the file

CHAR  fileID;           // 000 file id byte
CHAR  lastUpdateYR;     // 001 binary year-1900
CHAR  lastUpdateMO;     // 002 binary month (1-12)
CHAR  lastUpdateDA;     // 003 binary day (1-31)
ULONG noRecords;        // 004 total number of records
SHORT headerLength;     // 008 length of data header
SHORT recordLength;     // 010 record length
SHORT nada;             // 012 reserved
CHAR  xactionFlag;      // 014 flag indicating incomplete dBASE transaction (n/a)
CHAR  encryptFlag;      // 015 flag indicating encrypted (n/a)
CHAR  filler[16];       // 016 fill to 32 bytes

Field Descriptors

For each field, a descriptor is stored in the DBF.  The first descriptor starts
directly after the header, at file offset 32 (the 33rd byte).  Each descriptor
is 32 bytes.  After the last descriptor, a byte with ASCII value 13 (0x0D) is
stored.  Following this byte, the record data starts.

// nnn, where nnn is offset of item relative to the start of the descriptor

CHAR  fieldName[11];    // 000 ASCII, UPPER, underscore, zero-filled, (0T)
CHAR  fieldType;        // 011 UPPER C,N,D,L,M
ULONG fieldDA;          // 012 not used
CHAR  fieldLength;      // 016 1-255 bytes, depending on fieldType
CHAR  fieldDC;          // 017 places right of decimal point
SHORT altFieldLength;   // 018 alternate field length when fieldLength==0
CHAR  filler[12];       // 020 not used

// altFieldLength is proprietary to Bullet, and can be used if Xbase
// compatibility is not required and fields need to be larger than 255 bytes.
// To use it, set fieldLength=0 and altFieldLength to > 255 bytes.

Record Data

The DBF data are free-form, fixed-length records.  Each data record starts with
a one-byte 'tag' field, which is implicitly defined for all records (hence, it
is not a formal field and has no descriptor).  Following the tag field is the
first field of the record, and following that field (whose length is described
in the field's descriptor) is the next field, and so on.  No separators are
used between fields.  After the very last data record in the file, DBF
specification dictates that an end of file marker be placed, so at the end of
the file is a byte of value ASCII 26 (0x1A).

Record layout is as you define in your application.  It must match the layout
as described in the field descriptors, byte-for-byte.

Increasing DBF Performance

Records are stored in the order they were written.  To improve performance,
especially indexed-sequential access, the data file may be sorted, or
clustered, by reading each record in primary key order, then writing that
record to a new DBF data file.  Repeat for each record.  After all records have
been written, reindex the newly created DBF data file (and all related index
files).  After this, delete the old files (data and index), and rename the new
ones to the filenames required.  This technique maximizes cache efficiency, and
can easily offer 10x performance increase in access speed.


ΓòÉΓòÉΓòÉ 13.5. DBT File Format ΓòÉΓòÉΓòÉ

The DBT memo file is composed of a header followed by memo data, stored in one
or more blocks.  The header layout is detailed below, followed by the memo
record.

DBT Header

// nnn, where nnn is the offset of that item relative to the start of the file

ULONG memoAvailBlock;   // 000 next available block (header is block 0)
ULONG memoRez;          // 004 not used
CHAR  memoFilename[8];  // 008 filename proper (first 8 of filename proper)
ULONG memoRez2;         // 016 not used (apparently)
ULONG memoBlockSize;    // 020 block size, must be at least 24

// the rest of the header block (to block size bytes) is unused

Memo Record

// nnn, where nnn is offset of item relative to the start of the memo record

ULONG memoAvail;        // 000 next available link
ULONG memoSize;         // 004 size of data (including this and memoAvail)
CHAR  memoData;         // 008 for as many bytes as memoSize, less 8

A memo may use one or more blocks (each block is a fixed size), but allocations
are always contiguous.  Unused bytes after the memo data (to the end of the
last block allocated to that memo record) are undefined.  memoAvail is 0x8FFFF
for all active memo records.  For deleted memo records, memoAvail is used as a
link in the memoAvail list.  memoSize is the total bytes used by the memo,
including the memoAvail and memoSize data, so it is the size of the real data +
8 bytes.

Removing Memo Records

A similar technique to that above could be used to remove memo records from a
DBT.  Instead of using PACK_RECORDS_XB, read each DBF record in sequence, and
if not to be deleted, write that record to a new DBF data file and its memo
records to a new DBT memo file.  If the record read is to be deleted, skip it
(and its memo record(s)) and continue with the next DBF data record.  Repeat
while more records. Delete or archive the old DBF/DBT pair, and use the new
files in their place.  Reindex.  Also see compact.c on the distribution disk
for an example technique of compacting a database.


ΓòÉΓòÉΓòÉ 13.6. Custom Sort-Compare Function ΓòÉΓòÉΓòÉ

Bullet provides 10 custom sort-compare functions, in addition to the 6
intrinsic sort-compare functions (ASCII, NLS, and the four integer compares).
The custom function you supply is not actually a sort function, as the name
implies, but a compare function.  Basically, two strings are supplied and your
function determines string1's relation to string2 (<, >, or ==).

The strings supplied (via pointers) are not C strings, and they are not
(necessarily) 0-terminated.  A count value is passed, indicating the number of
bytes to compare.  The handle of the index file for which this compare is being
done is also supplied, so that you can interrogate the index file state
(STAT_INDEX_XB) for any additional information required.

In addition to the compare function this routine performs, a special-case call
is made to this routine requesting a pointer to a string of HIGH-VALUES for
this sort compare.  The pointer must be to a static memory area that exists for
as long as the index file is open, and must be at least as long as count.  This
special-case call is indicated by both string pointers==NULL.

To use a custom sort-compare function, first use SET_SYSVARS_XB to assign the
custom sort ID (10 to 19) with the function's address pointer.  Once assigned,
an index file may be created with its CIP.sortFunction set to the sort ID
(10-19).  Also, any previously created index file with a custom sort ID may now
be opened (but only after you used SET_SYSVARS_XB to assign the sort-compare
function pointer).  During the index file create, the sort ID you specified for
the create is stored in the index file.  When that index file is later opened,
that same sort ID is used, and so requires that the custom sort-compare
function already be assigned (with SET_SYSVARS_XB) before opening the index
file.  This means that you need to be consistent in your custom sort ID
numbering, since each index created forever uses that sort ID you specified.

It's simple to create a custom sort-compare function.  The calling convention
is APIENTRY (DOSX32 = __cdecl, OS/2= _syscall, Win32=_stdcall), and the
parameters are passed to your function on the stack (by Bullet).  A sample
prototype for a custom sort-compare function follows:

LONG APIENTRY YourCustomSort10(PVOID str1,
                               PVOID str2,
                               ULONG count,
                               ULONG handle);

If the pointers are not NULL, your routine is to compare str1 to str2, for
count bytes, and is to return:

        -1 if str1 is less than str2
         0 if str1 is equal to str2
         1 if str1 is greater than str2

str is not a C string, but is of type void.  Cast as required,
depending on the data expected.

If str1 and str2 are both NULL, your routine must return a pointer to a static
object that contains HIGH-VALUES for the object type.  For example, if the
sort-compare is for IEEE floating-point, then the function is to return a
pointer to a static data area filled with the highest floating-point value.
Depending on your sort-compare routine's functionality, you may need just a
single high-value, or multiple high-values, one after the other (e.g., if you
are supporting compound key values for binary keys).  The count parameter
indicates the total bytes needed, so divide by the object size to get the
number of objects required.  Be aware that the object size (in count) is +2
bytes for the enumerator if DUPS_ALLOWED was specified when the index file was
created.  This high-values object is used in the REINDEX_XB routine, and also
the LAST_KEY_XB and GET_LAST_XB routines.


ΓòÉΓòÉΓòÉ 13.7. Custom Build-Key ΓòÉΓòÉΓòÉ

Bullet provides an internal build-key routine that constructs the key from the
data record supplied.  The internal routine can be overloaded by your custom
build-key routine if you need additional functionality.  It may be used in
conjunction with a custom sort-compare function, or an intrinsic Bullet
sort-compare.

Developing a custom build-key routine requires delving into the internal Bullet
data structures.  It is more complicated than a custom sort-compare function,
but not really any more complex.  The handle of the index file is passed, and
using this, STAT_INDEX_XB is called to get the SIP.herePtr pointer.  This is
the pointer to the internal Bullet data structure for this index file.  What
needs to be accessed in this structure is the translated key expression.  From
this, you have the starting offset in the data record, and the byte count to
use, for each key component (up to 16 components per key).  The offset value as
stored in the XLATEX structure does not include the tag field byte.  Therefore,
to locate to the correct offset, add 1 to the value in offset.  For example,
XLATEX.offset=0 means to use the first field, which is the first byte after the
tag field byte, but the physical offset, as referenced to recPtr, is not at
offset=0, but is at offset=1.

This translated key expression structure is:

// (This is an excerpt from the IX3 header format)

// Translated key expression as done by Parser during CreateIndex and Reindex.
// More details on this is in the Custom Expression Parser Specifications.
// For each key part in KH.expression a 4-byte structure is used:

typedef struct _XLATEX {
 CHAR  ftype;   // field type (C,N,L, etc.),if bit7=1 and C then do UPPER key
 CHAR  length;  // bytes to use starting at offset (never > 64)
 SHORT offset;  // byte offset into data record that length bytes are to be used
} XLATEX;       // (note that offset does not count tag field byte)

ULONG xlateCount;           // 044 number of key fields (64/4=16 max fields)
XLATEX xlateExpression[16]; // 048 key construct info (16 dword's worth)

xlateExpression is at offset +48 relative the IX3 index header.  However,
SIP.herePtr points to -384 relative the IX3 index header start.  Therefore, to
locate to xlateExpression, you must add 384 to 48.  This means that
xlateExpression[0].ftype is located at SIP.herePtr+432.  The number of valid
key components in xlateExpression is stored in xlateCount (at SIP.herePtr+428).

The calling convention for your custom build-key function is APIENTRY (or
_System, or __syscall for some compilers), and the parameters are passed to
your function on the stack (by Bullet).  A sample prototype for a build-key
function follows:

ULONG APIENTRY YourBuildKey(ULONG handle,
                            PVOID recordPtr,
                            PVOID keyPtr,
                            PULONG keyLenPtr
                            PULONG sortFuncPtr);

Using the data from xlateExpression, you are to build a key from the data
record located at the passed pointer, recordPtr, and are store the built key in
the buffer located at keyPtr.  For each key component, you copy from the data
record xlateExpression[].length bytes starting at xlateExpression[].offset+1
(given the 1-byte tag field which is not accounted for otherwise), and build
other key components after previously build parts.  If the index file allows
duplicate keys (DUPS_ALLOWED is flagged in SIP.sortFunction), then append an
enumerator to the end of the key proper.  The handle of the index file is
passed, which is used when calling STAT_INDEX_XB (to get SIP.herePtr).  The
return is 0 if successful, or an appropriate Bullet error code (EXB_) should be
used.  In addition, the key length is placed in the ULONG data pointed to by
keyLenPtr (SIP.keyLength may be used), and the sort-compare function is placed
in the ULONG data pointed to by sortFuncPtr (SIP.sortFunction may be used).

The routine is also to check if the tag field of the data record matches the
skip-tag value, as set by SET_SYSVARS_XB.  If the tag field matches,
WRN_SKIP_KEY is to be returned as the 'error' code.  The key is built
regardless of a match.


ΓòÉΓòÉΓòÉ 13.8. Custom Expression Parser ΓòÉΓòÉΓòÉ

Bullet provides an internal key expression parser routine that constructs the
translated key expression stored in the index file header.  The internal
routine can be overloaded by your custom expression parser routine if you need
additional functionality.  It may be used in conjunction with a custom
sort-compare function, with a custom build-key routine, or with an intrinsic
Bullet sort-compare.

Developing a custom expression parser routine requires delving into the
internal Bullet data structures.  It is more complicated than a custom
sort-compare function, and it is also much more complex.  Unlike the custom
sort-compare and build-key functions, no handle is passed to the parser.  This
is because, rather than using the handle to get the SIP.herePtr, this pointer
is passed directly to this routine.  This is the pointer to the internal Bullet
data structure for this index file.  What needs to be accessed in this
structure is the translated key expression location, as well as the text
version of the key expression, as supplied by the programmer/user.  To the
XLATEX data you place the starting offset in the data record, and the byte
count to use, for each key component you parse from the key expression (up to
16 components per key).  The offset value as stored in the XLATEX structure
does not include the tag field byte.  Therefore, the correct offset to store is
the physical offset within the record, minus 1.  For example, XLATEX.offset=0
should be used for the offset of the first field, which is the first byte after
the tag field byte.  For each component parsed, an XLATEX data structure is
added to the xlateExpression data area (up to 16).  Unused XLATEX components
must be set to 0.  When all components have been stored, the xlateCount value
is set to the number of key components stored.

DHDptr is the data header pointer.  It is -352 bytes relative the DBF data
header.  However, rather than using absolute addressing to locate field
descriptor data (needed for parsing), it's recommended that the DBF handle be
obtained from the KHptr structure.  Since no file handles are passed, you must
read the xbLink handle value from the index file header.  The xbLink handle is
stored at KHptr+12.  With this handle, you call the GET_DESCRIPTOR_XB routine
to obtain field descriptor info for each field.

This translated key expression structure, and text expression location, is at:

// (This is an excerpt from the IX3 header format)

// Translated key expression as done by Parser during CreateIndex and Reindex.
// For each key part in KH.expression a 4-byte structure is used:

typedef struct _XLATEX {
 CHAR  ftype;   // field type (C,N,L, etc.),if bit7=1 and C then do UPPER key
 CHAR  length;  // bytes to use starting at offset (never > 64)
 SHORT offset;  // byte offset into data record that length bytes are to be used
} XLATEX;       // (note that offset does not count tag field byte)

ULONG xlateCount;           // 044 number of key fields (64/4=16 max fields)
XLATEX xlateExpression[16]; // 048 key construct info (16 dword's worth)
      :                     // 112-347   :
CHAR  expression[160];      // 348 key expression, user (0-Terminated)

xlateExpression is at offset +48 relative the IX3 index header.  However,
KHptr, passed to this routine, points to -384 relative the IX3 index header
start.  Therefore, to locate to xlateExpression, you must add 384 to 48.  This
means that xlateExpression[0].ftype is located at KHptr+432.  The number of
valid key components in xlateExpression is stored in xlateCount (at KHptr+428).
To text key expression string, which you are to parse, is located at KHptr+732.
This is identical to the expression passed during CREATE_INDEX_XB (and it is
CREATE_INDEX_XB that calls this parser routine).

The calling convention for your custom key expression parser function is
APIENTRY (or _System, or __syscall for some compilers), and the parameters are
passed to your function on the stack (by Bullet).  A sample prototype for a
build-key function follows:

ULONG APIENTRY YourKeyExpressionParser(PVOID DHDptr,
                                       PVOID KHptr,
                                       PULONG keyLenPtr);

You are to parse the text key expression at KHptr+732 and store the key
component XLATEX structure values to the XLATEX structure, one for each key
component parsed.  In addition, the key length (the sum of the XLATEX.length
fields) is placed in the ULONG data pointed to by keyLenPtr.  The keylength may
not exceed 64 bytes.  If DUPS_ALLOWED is flagged, add two to the sum of the
XLATEX.length fields for the enumerator word.

Note:  The key expression has been mapped to upper-case and 0-filled by the
time this routine is called.

This is probably the most difficult part of customizing Bullet.  However, the
difficulty lies not with Bullet, but how you parse.  The idea is simple -- you
are to generate a xlateCount value, and for each key component (ie
non-contiguous, non-same-type run in the data record), an XLATEX variable
describing the method to build that key component out of the data record (type,
length, and starting offset) is stored.  The text key expression is available
in the index header, and the destination to write to is there, also.  You do
need to read the index header at KHptr+12 (ULONG) to obtain the DBF handle for
this index file before you can parse the expression.  This because you need to
know about the record field names, types, and lengths before you can parse the
key expression.  The matter not covered here is that of parsing the expression,
which is left to the programmer.  Any lexical parser algorithm may be used, or
you may even do no parsing at all, and simply hard-code values into the XLATEX
structures.

If you've gotten this far, you may find the following data structures useful.
The numbers at // nnn are offsets relative the SIP.herePtr and SDP.herePtr
pointers.  For example, at SIP.herePtr+352 is a ULONG of the number of key
searches requested.  These could be monitored in a separate thread.

Relative SIP.herePtr:

ULONG fType;            // 000 bit0=0 for index file, btree
ULONG flags;            // 004 bit0=1 is dirty
                        //     bit1=1 full lock (count stored in KH.lockCount)
                        //     bit2=1 shared lock (if bit1=1)
                        //     bit3-14 reserved (=0)
                        //     bit15=1 no coalesce on key delete
                        // 006 BYTE, progress of reindex (0,1-99)
                        // 007 BYTE, 0
PVOID morePtr;          // 008 ptr to additional header info, if ever needed
ULONG xbLink;           // 012 related XB data file handle
ULONG asMode;           // 016 access-sharing-cache mode of open
CHAR  filename[260];    // 020 filename at open (0T)
ULONG currKeyRecNo;     // 280 current rec number assigned to KH.currKey
CHAR  currKey[64];      // 284 current key value
ULONG rez0;             // 348 allow for 0-terminated string
ULONG searches;         // 352 keys searched for
ULONG seeks;            // 356 nodes seeked
ULONG hits;             // 360 seeks satisfied without disk access
ULONG keysDeleted;      // 364 keys deleted since last zeroed
ULONG keysStored;       // 368 keys added since last zeroed
ULONG nodesSplit;       // 372 splits needed on insert since last zeroed
ULONG nodesMadeAvail;   // 376 nodes made available from deleting keys
ULONG lockCount;        // 380 active full-lock count

// the IX3 index header follows at 384+

Relative SDP.herePtr:

ULONG fType;            // 000 bit0=1 for DBF data file, XB
ULONG flags;            // 004 bit0=1 is dirty
                        //     bit1=1 full lock
                        //     bit2=1 shared lock (if bit1=1)
                        //     bit3-15 reserved (=0)
                        // 006 BYTE, progress of pack (0,1-99)
                        // 007 BYTE, 0
PVOID morePtr;          // 008 ptr to additional header info, if ever needed
ULONG noFields;         // 012 number of fields in this data file
ULONG asMode;           // 016 access-sharing-cache mode of open
CHAR  filename[260];    // 020 filename at open (0T)
ULONG lockCount;        // 280 only when dec'ed to 0 do full unlock
ULONG memoAvailBlock;   // 284 next available block (header is block 0)
ULONG memoUnk1;         // 288 not used
CHAR  memoFilename[8];  // 292 filename proper (first 8 of filename proper)
ULONG memoUnk2;         // 300 not used (apparently)
ULONG memoBlockSize;    // 304 block size, must be at least 24 to cover header!
ULONG memoHandle;       // 308 handle of open memo file
ULONG memoFlags;        // 312 bit0=1 is dirty
ULONG memoLastNo;       // 316 last accessed memo number (if not 0)
ULONG memoLastLink;     // 320 link data for last accessed memo
ULONG memoLastSize;     // 324 size of last accessed memo (in bytes, w/OH)
ULONG align32[6];       // 328 (align to even32)

// the DBF data header follows at +352


ΓòÉΓòÉΓòÉ 14. Bullet Is... ΓòÉΓòÉΓòÉ

Bullet is a thread-safe, multi-process capable database engine toolkit for
OS/2, DOSX32, and Win32.  It provides pre-built and tested access methods to
data and index files for application programmers.  It is not an end-user
Database Management System (DBMS), but it is a tool that could be used to
develop one.  Bullet is compact, efficient, and very fast.  It can be
configured to use custom key-build, sort-compare functions, and expression
parser routines to extend the built-in functionality, and even
programmer-supplied OS API calls to replace the internal ones.

The standard data format is DBF (dBASE 3+ and later).  The supported memo
format is DBT (dBASE 4 and later).  Index-only support can be enabled and with
this any data file format may be used (the data maintained by the programmer
then).  Also, the DBF standard may be extended by using binary field values and
fields larger than 255 bytes.  Index files are NLS-compatible and use an
efficient b-tree structure.  Files may be any size supported by the OS, up to
4GB.  Up to 1024 files may be opened and in use by any one process, with any
number of processes active.

The Bullet API consists of a wide assortment of routines, from low-level OS
calls to high-level transaction-list routines that can process hundreds of
files per transaction, with roll-back on error.  Network support is included,
and makes use of operating system features such as atomic re-locking, and
shared locks that allow other processes read-access to region-locked files.

Bullet is simple to use, and may easily be modified by using function wrappers
around groups of related Bullet routines.  If you don't like working with the
parameter packs, use a wrapper and call as you like.  Bullet works the way you
are used to working.

This online manual is a complete introduction and programmer reference for
Bullet.  Sample code is included, with more still on disk.


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

A foreign key is a column value in one table used as a primary key in a second
table.  For example, if one table has two fields:  employee name (primary key)
and department code, and a second table has two fields:  department code
(primary key) and department name, then department code in the first table is
considered a foreign key, since it may be used as the primary key value when
searching the second table.  Joins and views make use of foreign keys.


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

Null pack


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _ACCESSPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle of Bullet file to access */
LONG  recNo;          /* IO, record number */
PVOID recPtr;         /* I, programmer's record buffer */
PVOID keyPtr;         /* I, programmer's key buffer */
PVOID nextPtr;        /* I, NULL if not xaction, else next AP in list */
} ACCESSPACK; /* AP */
typedef ACCESSPACK *PACCESSPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _CALLBACKPACK {
ULONG sizeIs;         /* structure size (current 16 bytes) */
ULONG callMode;       /* 0=from reindex; 1=from DBF pack */
ULONG handle;         /* file handle */
ULONG data1;          /* for callMode=0/1: progress percent (1-99,0) */
} CALLBACKPACK; /* CBP */
typedef CALLBACKPACK *PCALLBACKPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _COPYPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle of Bullet file to copy */
PSZ   filenamePtr;    /* I, filename to use (drv:path must exist if used) */
} COPYPACK; /* CP */
typedef COPYPACK *PCOPYPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _CREATEDATAPACK {
ULONG func;
ULONG stat;
PSZ   filenamePtr;    /* I, filename to use */
ULONG noFields;       /* I, 1 to 254 */
PFIELDDESCTYPE fieldListPtr;   /* I, descriptor list, 1 per field */
ULONG fileID;         /* I, 0x03 for std DBF, 0x8B for DBF+DBT */
} CREATEDATAPACK; /* CDP */
typedef CREATEDATAPACK *PCREATEDATAPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _CREATEINDEXPACK {
ULONG func;
ULONG stat;
PSZ   filenamePtr;    /* I, filename to use */
PSZ   keyExpPtr;      /* I, e.g., "SUBSTR(LNAME,1,4)+SSN" */
LONG  xbLink;         /* I, opened data file handle this indexes */
ULONG sortFunction;   /* I, 1-9 system, 10-19 custom */
ULONG codePage;       /* I, 0=use process default */
ULONG countryCode;    /* I, 0=use process default */
PVOID collatePtr;     /* I, NULL=use cc/cp else use passed table for sort */
ULONG nodeSize;       /* I, 512, 1024, or 2048 */
} CREATEINDEXPACK; /* CIP */
typedef CREATEINDEXPACK *PCREATEINDEXPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _FIELDDESCTYPE {
BYTE  fieldName[11];  /* IO, upper A-Z and _; 1-10 chars, 0-filled, 0-term */
BYTE  fieldType;      /* IO, C,D,L,N, or M */
LONG  fieldDA;        /* x, offset within record (run-time storage option) */
BYTE  fieldLen;       /* IO, C=1-255,D=8,L=1,N=1-19,M=10 */
BYTE  fieldDC;        /* IO, fieldType=N then 0-15 else 0 */
USHORT altFieldLength;/* IO, 0 */
BYTE  filler[12];     /* I, 0 */
} FIELDDESCTYPE; /* nested in DESCRIPTORPACK */
typedef FIELDDESCTYPE *PFIELDDESCTYPE;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _DESCRIPTORPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle of DBF file */
ULONG fieldNumber;    /* IO, first field is 1 */
ULONG fieldOffset;    /* O, offset of field within record (delete tag=offset 0) */
FIELDDESCTYPE FD;  /* IO FD.fieldName, O rest of FD */
} DESCRIPTORPACK; /* DP */
typedef DESCRIPTORPACK *PDESCRIPTORPACK;
};


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _DOSFILEPACK {
ULONG func;
ULONG stat;
PSZ   filenamePtr;    /* I, filename to use */
ULONG handle;         /* IO, handle of open file */
ULONG asMode;         /* I, access-sharing-cache mode */
ULONG bytes;          /* IO, bytes to read, write, length of */
LONG  seekTo;         /* IO, seek to offset, current offset */
ULONG method;         /* I, seek method (0=start of file, 1=current, 2=end) */
PVOID bufferPtr;      /* I, buffer to read into or write from */
ULONG attr;           /* I, attribute to create file with */
PSZ   newNamePtr;     /* I, name to use on rename */
} DOSFILEPACK; /* DFP */
typedef DOSFILEPACK *PDOSFILEPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _EXITPACK {
ULONG func;
ULONG stat;
} EXITPACK; /* EP */
typedef EXITPACK *PEXITPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _HANDLEPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle of Bullet file */
} HANDLEPACK; /* HP */
typedef HANDLEPACK *PHANDLEPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _INITPACK {
ULONG func;
ULONG stat;
ULONG JFTsize;        /* I, max opened files (20-1024+) */
ULONG versionDOS;     /* O, e.g., 230 for 2.30 */
ULONG versionBullet;  /* O, e.g., 2019 for 2.019 */
ULONG versionOS;      /* O, e.g., 4=OS/2 32-bit */
PVOID exitPtr;        /* O, function pointer to EXIT_XB routine */
} INITPACK; /* IP */
typedef INITPACK *PINITPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _LOCKPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle of Bullet file to lock */
ULONG xlMode;         /* I, index lock mode (0=exclusive, 1=shared) */
ULONG dlMode;         /* I, data lock mode (0=exclusive, 1=shared) */
LONG  recStart;       /* I, if data, first record # to lock, or 0 for all */
ULONG recCount;       /* I, if data and recStart!=0, # records to lock */
PVOID nextPtr;        /* I, NULL if not xaction, else next LP in list */
} LOCKPACK; /* LP */
typedef LOCKPACK *PLOCKPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _MEMODATAPACK {
ULONG func;
ULONG stat;
ULONG dbfHandle;      /* I, handle of DBF file to which this memo file belongs */
ULONG memoBypass;     /* I, memo bypass function to do, if any */
PVOID memoPtr;        /* I, ptr to memo record buffer */
ULONG memoNo;         /* IO, memo record number (aka block number) */
ULONG memoOffset;     /* I, position within record to start read/update */
ULONG memoBytes;      /* IO, number of bytes to read/update */
} MEMODATAPACK; /* MDP */
typedef MEMODATAPACK *PMEMODATAPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _MEMORYPACK {
ULONG func;
ULONG stat;
ULONG memory;         /* O, not used in OS/2 */
} MEMORYPACK; /* MP */
typedef MEMORYPACK *PMEMORYPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _OPENPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* O, handle of file opened */
PSZ   filenamePtr;    /* I, Bullet file to open */
ULONG asMode;         /* I, access-sharing-cache mode */
LONG  xbLink;         /* I, if index open, xbLink=handle of its opened DBF */
} OPENPACK; /* OP */
typedef OPENPACK *POPENPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _QUERYSETPACK {
ULONG func;
ULONG stat;
ULONG item;           /* I, Bullet sysvar item to get/set */
ULONG itemValue;      /* IO, current/new value */
} QUERYSETPACK; /* QSP */
typedef QUERYSETPACK *PQUERYSETPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _REMOTEPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle of file, or if 0, use RP.drive */
ULONG drive;          /* I, drive (1=A,2=B,3=C,...0=current) to check */
ULONG isRemote;       /* O, =1 of handle/drive is remote, =0 if local */
ULONG flags;          /* O, 0 */
ULONG isShare;        /* O, 1 */
} REMOTEPACK; /* RP */
typedef REMOTEPACK *PREMOTEPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

struct StatDataPack {
typedef struct _STATDATAPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle to check */
ULONG fileType;       /* O, bit0=1 data file */
ULONG flags;          /* O, bit0=1 dirty, bit1=1 full-lock, bit2=1 shared */
ULONG progress;       /* O, 0,1-99% pack progress */
PVOID morePtr;        /* O, 0 */
ULONG fields;         /* O, fields per record */
ULONG asMode;         /* O, access-sharing-cache mode */
PSZ   filenamePtr;    /* O, filename used in open */
ULONG fileID;         /* O, first byte of DBF file */
ULONG lastUpdate;     /* O, high word=year,low byte=day, high byte=month */
ULONG records;        /* O, data records (including "deleted") */
ULONG recordLength;   /* O, record length */
ULONG xactionFlag;    /* O, 0 */
ULONG encryptFlag;    /* O, 0 */
PVOID herePtr;        /* O, this file's control address */
ULONG memoHandle;     /* O, handle of open memo file (0 if none) */
ULONG memoBlockSize;  /* O, memo file block size */
ULONG memoFlags;      /* O, bit0=1 dirty */
ULONG memoLastRecord; /* O, last accessed memo record (0 if none) */
ULONG memoLastSize;   /* O, size of last accessed memo record (in bytes, +8) */
ULONG lockCount;      /* O, number of full-locks in force */
} STATDATAPACK; /* SDP */
typedef STATDATAPACK *PSTATDATAPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _STATHANDLEPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle to check */
LONG  ID;             /* O, bit0=1 data file, bit0=1 index file */
} STATHANDLEPACK; /* SHP */
typedef STATHANDLEPACK *PSTATHANDLEPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _STATINDEXPACK {
ULONG func;
ULONG stat;
ULONG handle;         /* I, handle to check */
ULONG fileType;       /* O, bit0=0 index file */
ULONG flags;          /* O, bit0=1 dirty, bit1=1 full-lock, bit2=1 shared */
ULONG progress;       /* O, 0,1-99% reindex progress */
PVOID morePtr;        /* O, 0 */
ULONG xbLink;         /* O, XB file link handle */
ULONG asMode;         /* O, access-sharing-cache mode */
PSZ   filenamePtr;    /* O, pointer to filename used in open */
ULONG fileID;         /* O, "31ch" */
PSZ   keyExpPtr;      /* O, pointer to key expression */
ULONG keys;           /* O, keys in file */
ULONG keyLength;      /* O, key length */
ULONG keyRecNo;       /* O, record number of current key */
PVOID keyPtr;         /* O, ptr to current key value (valid to keyLength) */
PVOID herePtr;        /* O, this file's control address */
ULONG codePage;       /* O, code page at create time */
ULONG countryCode;    /* O, country code at create time */
PVOID CTptr;          /* O, collate table ptr, NULL=no collate table present */
ULONG nodeSize;       /* O, node size */
ULONG sortFunction;   /* O, sort function ID */
ULONG lockCount;      /* O, number of full-locks in force */
} STATINDEXPACK; /* SIP */
typedef STATINDEXPACK *PSTATINDEXPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

typedef struct _XERRORPACK {
ULONG func;
ULONG stat;           /* I, error to check */
ULONG errClass;       /* O, class of error */
ULONG action;         /* O, action recommended for error */
ULONG location;       /* O, location of error */
} XERRORPACK; /* XEP */
typedef XERRORPACK *PXERRORPACK;


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

 Access-Mode (required)
  READONLY        0x00000000   read-only access open
  WRITEONLY       0x00000001   write-only access open
  READWRITE       0x00000002   read/write access open

 Share-Mode (required)
  DENYREADWRITE   0x00000010   no other process may share file
  DENYWRITE       0x00000020   no other process may share file for write
  DENYREAD        0x00000030   no other process may share file for read
  DENYNONE        0x00000040   any process may share file

 Inherit
  NOINHERIT       0x00000080   child process does not inherit file handles

 Cache
  NO_LOCALITY     0x00000000  locality is not known
  SEQ_LOCALITY    0x00010000  access will be mainly sequential
  RND_LOCALITY    0x00020000  access will be mainly random
  MIX_LOCALITY    0x00030000  access will be random with some sequential
  SKIP_CACHE      0x00100000  I/O is not to go through the cache
  WRITE_THROUGH   0x00400000  control returns only after disk is written to

Access- and Share-Mode values not explicitly listed are not valid.  The file
access mode is a combination of ACCESS + SHARE + INHERIT + CACHE.  Typical data
and index asMode is 0x00000042, though locality may be set accordingly (e.g.,
0x00020042 for mostly random access to the file).  All values are mapped to the
appropriate flags for the OS being used.

The Cache mode options are valid for OPEN_DATA_XB and OPEN_INDEX_XB only; for
OPEN_FILE_DOS, the Cache values must be right-shifted by 8.  The 'Skip Cache'
and 'Write Through' options are not inherited.


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

The enumerator is a big-endian 16-bit value that serves to differentiate up to
65536 "identical", non- unique keys.  It is attached to all keys of
DUPS_ALLOWED flagged index files (set at CREATE_INDEX_XB), and occupies the
last two bytes of the key.  The first key of the type uses \0\0, the second
uses \0\1, and so on.  This ordering of bytes is the reverse of x86 Intel
words, which uses little-endian format.


ΓòÉΓòÉΓòÉ <hidden>  ΓòÉΓòÉΓòÉ

HIGH-VALUES signifies a sort value that is the highest possible (sorts last).
HIGH-VALUES for a character key would be 0xFF for each byte, or the 256th byte
of the collate-sequence table if an NLS sort (which is 0xFF also).  For 16-bit
signed integer values, 0x7FFF is the highest.  And so on...