From:	jtgalkowski AT alum DOT mit DOT edu
Newsgroups:	comp.os.msdos.djgpp
Subject:	Re: Recently installed version of DJGPP's gcc hangs
Date:	Mon, 28 Sep 1998 02:08:57 GMT
Organization:	Deja News - The Leader in Internet Discussion
Lines:	69
Message-ID:	<6umr47$t1$1@nnrp1.dejanews.com>
NNTP-Posting-Host:	12.68.173.172
To:	djgpp AT delorie DOT com
DJ-Gateway:	from newsgroup comp.os.msdos.djgpp

This reports the failure of using DJGPP's gcc compiler
to compile a particular set of C functions, a file which
is a component of a much bigger system.

The file has been diagnosed by Mr Eli Zaretskii as "a huge
'switch' statement with a lot of cases."  Moreover, he
goes on to say "Such sources are known to require
preposterous amounts of memory while GCC compiles them,
and also a lot of stack space."

I was originally compiling the file of functions on a
25 MHz 8 MB RAM 486 system with 32 MB of disk space
available.  Juggling values for both the CWSDPMI internal
heap via CWSPARAM and the stack space using STUBEDIT
of "cc1.exe" failed in every case.  Most failed issuing
a Page Fault in RMDB.  Others apparently executed into
boot memory on the 486 and caused the system to restart.
All cases were run in the pure DOS shell without Windows
running.  I initially ran the system with a large disk
cache (Symantec's NCACHE2), but later dispensed with it
when it became clear memory was tight.

The go32-v2 interrogator on the 486 netted a space of
some 7+ MB available for DPMI memory.

When memory seemed to be a problem, I downloaded
and installed DJGPP on a Pentium 133 MHz system with
80 MB of RAM and 230 MB of disk.  I used the DPMI
provided with DJGPP and ran it under pure Windows.

The go32-v2 resulted in 65799 Kb of BPMI memory and
130389 Kb of DPMI swap.  The largest "minstack" setting
I tried was 32768 Kb for cc1.exe and 32767 for the
CWSDPMI internal heap setting.  I couldn't increase
beyond 32767 because the complement nature of that
parameter makes 32768 look like -32768.

The compilation of this program failed as well on
the Pentium, using differing sizes for these two
variables.

What's disturbing here is that there should be one
particular kind of program that brings GCC to its
knees in any situation.  While I'm now shopping around
for some other DOS-based C compiler, I wonder if
this shouldn't be put on some "things to be improved"
list someplace.

Anyone who wants a copy of the offending C source
is welcome to write me at jtgalkowski AT alum DOT mit DOT edu.
Of course, I welcome additional suggestions.  The C
source actually is in the public domain, and is
available from the Web, but I see no reason to mention
that source here, since there is nothing wrong with
that C source, it being successfully compiled by
the Cygnus GCC for Windows 95 and for several other
larger systems, including Alpha.

------------------------------------------------------
  Jan Theodore Galkowski,
   jtgalkowski AT alum DOT mit DOT edu
------------------------------------------------------
  http://www.hotwired.com/members/profile/algebraist/
------------------------------------------------------
  http://www.cwi.nl/projects/alp/
------------------------------------------------------

-----== Posted via Deja News, The Leader in Internet Discussion ==-----
http://www.dejanews.com/rg_mkgrp.xp   Create Your Own Free Member Forum

- Raw text -