To: dj AT citron DOT com
Cc: djgpp AT sun DOT soe DOT clarkson DOT edu
Subject: Problems with rebuilding GO32
Date: Thu, 24 Mar 94 09:24:29 +0200
From: eliz AT is DOT elta DOT co DOT il

  I was trying to patch GO32 these days along the lines suggested by
terra AT diku DOT dk (set dr[7] = 0 in extdebug.c) so that data read breakpoints
work with FSDB.  The following is the list of (minor) problems encountered:

  1. In the main() of coff2exe there is a fragment:

     stub-file = -open(argv[2], O-READONLY);
     if (stub-file < 0)
     {
       perror(stub-file);
       exit(1);
     }

     But stub-file is an int, and perror() wants a const char * as its
     arg.  This probably should be perror(argv[2]), right?  My BC++ 3.1
     compiler barked at this line.  I don't know, how it evaded the TCC
     that was originally used, maybe because its diagnostics was somehow
     suppressed by command-line switches.

  2. In the makefile (in the go32/ directory), the CFLAGS variable should
     include -A-, because some source files which use non-ANSI features
     won't compile otherwise.  Whoever writes makefiles for Borland products
     should keep in mind that they have configuration files which list the
     *default* switches.  Mine includes the -A switch (I like portability and
     want all my non-portable features to be kept at minimum, and then only
     added conciously).

  3. Also in the makefile, the rule for stamp.exe includes the `tcc' command
     explicitly, whereas it should be $(CC).

  4. The link rule should include -l3 switch (which gives Turbo Link the /3
     switch); otherwise TLink won't link some executable (is it GO32 itself?
     I forgot) which uses 386-specific code.

  Btw, what about other optimization-related switches of BCC?  Did anybody try
them?  Is, e.g., -O1 safe?  What about -O2?  Can I use -3 switch (produces
386 code which uses 32-bit registers)?

		Eli Zaretskii.