Message-ID:	<373967C1.7024A77F@tls.msk.ru>
Date:	Wed, 12 May 1999 15:36:33 +0400
From:	"Michael Ju. Tokarev" <mjt AT tls DOT msk DOT ru>
Organization:	Telecom Service, JSC
X-Mailer:	Mozilla 4.5 [en] (Win95; I)
X-Accept-Language:	ru,en
MIME-Version:	1.0
To:	pgcc AT delorie DOT com
CC:	Martin Kahlert <martin DOT kahlert AT mchp DOT siemens DOT de>
Subject:	Re: Strange preprocessor bug?
Reply-To:	pgcc AT delorie DOT com

Martin Kahlert wrote:

> Hi,
> a very strange bug from the snapshot (10.May):
> 
> ./xgcc -B./ -B/sw/pgcc/i686-pc-linux-gnulibc1/bin/ -DIN_GCC -DUSE_GNULIBC_1 
>     -g -W -Wall -I./include -I. -I../../pgcc/gcc -I../../pgcc/gcc/config 
>     -I../../pgcc/gcc/../include \
>   -aux-info SYSCALLS.c.X -E -S -o tmp-SYSCALLS.s SYSCALLS.c 
> In file included from SYSCALLS.c:86:
> include/stddef.h:28: machine/ansi.h: No such file or directory
> make: *** [SYSCALLS.c.X] Error 1
> 
> O.k. i thought, such things can happen.
> But then i changed the lines around 28 from gcc/include/stddef.h from
> #if defined (__BSD_NET2__) || defined (____386BSD____) || \
>     defined (__FreeBSD__) || defined (__NetBSD__)
> #include <machine/ansi.h>
> #endif
> 
> into 
> #if 0
> #include <machine/ansi.h>
> #endif
> 
> And the same problem again! This seems to indicate a cpp 
> miscompilation, since the first stage went o.k.
> 
> What is my problem here?
> 
> Bye,
> Martin.
> 
> PS: 
> My build procedure:
> ../pgcc/configure --prefix=/sw/pgcc --enable-shared --disable-debug
> gmake LANGUAGES='c c++ f77 proto' bootstrap-lean
> 
> System is a Linux-PPro/2.0.36 kernel.
> 

I found almost the same thing: "#if defined(A) || !defined(B) && 0", for example,
evaluates to 0 (false) with 'gcc -OX' (with any X from 0 to 9),
'gcc -OX -m[arch=]pentium', BUT evaluates to _1_ (that's wrong!) for
'gcc -OX -march=pentiumpro' (but it is Ok for 'gcc -O0 -march=pentiumpro').
Note also that different machines cause different results (I'm now at office
and experiment on Solaris 2.6 x86 on Intel PentiumII/322 machine; at home
I have IntelMMX/166 with Linux, and got different results (but with pentium
option ofcause, not pentiumpro!) -- sorry, I do not remember this).

Another interesting thing -- I was unable to compile pgcc at Solaris/x86 at all
starting from november, 1998 (what version it was?) -- stage2 always failed with
different results (different from version to version) if I include any optimization
for pentium/ppro; but it was compiled and worked well on Linux (but on another
hardware -- at my home). pgcc-2.90.29 980515 (egcs-1.0.3 release) is worked well
on both machines (and this is what I'm using now...).

Regards,
  Michael Ju. Tokarev,
  System Administrator,
  JCS "Telecom-Service", Moscow, Russia.

- Raw text -