From:	pavenis AT lanet DOT lv
To:	"Eli Zaretskii" <eliz AT is DOT elta DOT co DOT il>
Date:	Sat, 16 Jun 2001 16:32:58 +0300
MIME-Version:	1.0
Subject:	Re: Link errors with Debug switch on
Cc:	djgpp AT delorie DOT com
Message-ID:	<3B2B8A3A.1619.432328@localhost>
In-reply-to:	<3791-Sat16Jun2001160700+0300-eliz@is.elta.co.il>
References:	<Pine DOT A41 DOT 4 DOT 05 DOT 10106161439240 DOT 106024-100000 AT ieva06 DOT lanet DOT lv> (message from Andris Pavenis on Sat, 16 Jun 2001 14:52:01 +0300 (WET))
X-mailer:	Pegasus Mail for Win32 (v3.12c)
Reply-To:	djgpp AT delorie DOT com
Errors-To:	nobody AT delorie DOT com
X-Mailing-List:	djgpp AT delorie DOT com
X-Unsubscribes-To:	listserv AT delorie DOT com

On 16 Jun 2001, at 16:07, Eli Zaretskii wrote:

> > Date: Sat, 16 Jun 2001 14:52:01 +0300 (WET)
> > From: Andris Pavenis <pavenis AT lanet DOT lv>
> >
> > andris AT hal:~$ i586-pc-msdosdjgpp-gcc -c -gstabs impar.c
> > andris AT hal:~$ i586-pc-msdosdjgpp-nm impar.o 
> > 00000090 b .bss
> > 000006f1 ? .comment
> > 00000090 d .data
> > 000000d0 ? .stab
> > 00000310 ? .stabstr
> > 00000000 t .text
> > 00000090 b _imach.0
> > 00000000 T _ipmpar
> > 000000c0 b _ipmpar.1
> 
> So the problem is still there in GCC 3.0.
> 
> > However comparing assembler outputs with -gcoff and without it it seems
> > that a bug could be in GNU assembler but not in GCC. One can try different
> > binutils versions (beginning from 2.8.1 or 2.9.1)
> 
> Ah, sorry that I didn't test different assemblers.
> 
> I now tested with Gas from Binutils 2.10, and I still see the same
> problem.  Gas 2.10 is the version I tried (on another machine) with
> GCC 2.7.2.1, where the problem didn't exist.
> 
> So it looks like Gas is not the problem.  But I admit that I'm not
> familiar with all the details of "gcc -S" output, so perhaps I'm
> missing something.
> 

It's not a first case I met when contents of .def file in assembler source
fools GAS to do something stupid. I got something similar similar
recently with testing gcc-3.0 prerelease versions from CVS. Initially I 
workarounded problem by patching gcc/sdbout.c in gcc sources,
but currently it's no more needed (gcc generates code that no more 
breaks GAS when -gcoff is being used). GAS incorrectly assumed 
external symbol to be present from COFF debug info.

So it seems (my impression) that COFF related code in binutils have 
suffered from rather severe bitrot and have to be fixed. I haven't tested
example we're talking about with current CVS version of binutils, but
I'm almost sure nothing will change if I'll test ...

Andris

- Raw text -