Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!husc6!cmcl2!rutgers!clyde!cbosgd!gwspc!cbcsta!n8emr!oink!jep
From: jep@oink.UUCP (James E. Prior)
Newsgroups: comp.sys.ibm.pc,comp.lang.c
Subject: Re: TURBO C Linking Problem
Message-ID: <190@oink.UUCP>
Date: Thu, 12-Nov-87 22:52:19 EST
Article-I.D.: oink.190
Posted: Thu Nov 12 22:52:19 1987
Date-Received: Sun, 15-Nov-87 17:08:34 EST
References: <2293@watcgl.waterloo.edu>
Reply-To: jep@oink.UUCP (James E. Prior)
Distribution: na
Organization: Random Prime Research Institute  Columbus, Ohio
Lines: 66
Keywords: TURBO C Linking
Xref: mnetor comp.sys.ibm.pc:10159 comp.lang.c:5414

I have a solution to the TURBO C Linking Problem.

In article <2293@watcgl.waterloo.edu> smvorkoetter@watmum.waterloo.edu (Stefan M. Vorkoetter) writes:
>
>I am having problems linking a program with TURBO C.  The program consists
>of about 15 source files.  Each and every one compiles, but when I link,
>I get an error when CH.LIB is being linked in that there is too much global
>data defined in a file.  None of my files defines more than 64K of global
>data.  As a matter of fact, the entire program uses less than 64K of global
>data.  Unfortunately, it won't even tell me which file is the supposed
>culprit.  I have double checked that the model is selected properly, and that
>everything else is set right.  I am running under the integrated environment.
>
>By the way, the program I am compiling is a C compiler that I am developing.
>Perhaps this is Borland's way of preventing competition :-)
>
>Any help on this matter would be appreciated.  The program compiles fine
>under 4.3 BSD.  I am not doing anything the least bit esoteric anywhere.
>
>Thanks in advance,
>Stefan Vorkoetter
>Symbolic Computation Group
>University of Waterloo
>(smvorkoetter@watmum.waterloo.edu)

I had a similar problem with Turbo C, and have a way of coping with it.

I was working on a program with much data.  I was already using the
compact memory model (64K code, 1M data).  I added more and more
arrays of initialized structures.  Eventually, the linker complained
that I had too much global data.  I split the arrays into many modules,
so that every module was well under 64K of initialized data.  It still
croaked.  I changed from Turbo's TC environment to conventional command
line invokations of tcc and tlink.  Tlink would always croak.  I switched
to the huge model.  It is supposed to handle more than 64K of 
initialized data.  It still croaked, both in the tc environment and by
tcc and tlink.

I finally called Borland's technical support number, and they knew both
the problem and the solution.  The problem is a bug in their tlink
program.  The solution is to use the ordinary link program.  Their
support staff didn't waste anyone's time.  My entire call was under five
minutes.  They were smart.  They asked the right questions.

Following are excerpts from my makefile, that *do* work for me.

# huge memory model: 1M code, 1M data
MODEL=h
CFLAGS=-Ic:\tc -m$(MODEL) -c

jem2set.exe:			\
.
. dependencies omitted
.
	rem tlink /c /m /s ..\c0$(MODEL) @obj.lst, jem2set.exe,, ..\cusrc\a$(MODEL)curses ..\emu ..\math$(MODEL) ..\C$(MODEL)
	\msc4\link /NOI /STACK:4096 /SEGMENT:256 ..\c0$(MODEL) @obj.lst, $*,$*, ..\cusrc\a$(MODEL)curses ..\emu ..\math$(MODEL) ..\C$(MODEL)

BTW, I used Polymake since I have used it for years.  I didn't feel
like messing with Turbo make, so I can't say anything good or bad about
it.

Warning: There are an amazing number of link.exe's by Microsoft.  I used
the one that came with Microsoft C 4.0 .  I doubt that some of the older
versions will work, especially the ones supplied with DOS.
-- 
James E. Prior  {ihnp4|cbosgd}!n8emr!oink!jep