Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cs.utexas.edu!tut.cis.ohio-state.edu!ucbvax!DMZRZU71.BITNET!Ritzert
From: Ritzert@DMZRZU71.BITNET
Newsgroups: comp.sys.atari.st
Subject: (none)
Message-ID: <900130001909.312838@DMZRZU71-UNI-MAINZ--GERMANY>
Date: 30 Jan 90 00:19:00 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 50

Subject:  gcc problems

I just installed the gcc1.36 which I got from dsrgsun and tried to build
ctex (initex).  It soon turned out that gcc1.36 is not able to compile
this stuff on the ST.  Even at the first module, align.c, the error
message "virtual memory exhausted" appeared. Switching off any
optimizations didn't really help: the same error message appeared in the
module "cmds.c". Playing around with the stacksize wasn't successful,
too. I used an old gulam, no memory resident programs or accessories at
all. All auto folder programs have been disabled, including tos14fix, to
ensure it isn't a conflict with anything of my standard configuration.

Gcc1.34 is able to compile this routine but only when the optimisation
is switched off.  I guess this is a bug in the library.  I had similar
problems when I recompiled gcc1.34 with itself. It was possible but the
newbuilt compiler could only compile small programs (same error as
above). John Dunning told me that alloca would fail due to a bug in
qmult which I thought has been fixed in the meantime.

Simple lack of system memory can't be the reason --- replacing my ST4 by
a ST2 reproduced the error at the same point.

Interestingly, the Unix based cross-compiler seems to work correctly...

Has anyone fixed this in the meantime? I have lots of trouble with
ftp (via bitftp to a vm host, uploading across a strange serial
conncection with an ibmclone since neither uniterm nor ckermit work with
this connection, then transferring everything to the ST is an annoying
procedure, and it takes weeks to fetch the stuff). So is someone willing
to mail me the fixes, please? (needed: fixed gcc_cc1, fixed lib and
diffs for the lib sources.)

A few other bugs: contrary to the readme.st, the stacksize of gcc_cc1
was 128K instead of 1M. 1.34 usually runs with 0.5M.
I don't trust the utilities to manipulate the stacksize since they are
buggy.

fixstk 524288 gcc_cc1 resulted in a reported stack size of 51K !!!!!!

I had to specify 10 times the amount of memory I wanted to have in order
to get a correct report. So either the input or the output conversion
routine of the library is broken.

Again, please help me since I don't have the time to dig into the
library stuff (and lack documentation) but need a compiler I can trust.

Thanks for any help,

Michael Ritzert
mjr@dmzrzu71.bitnet