X-Authentication-Warning: delorie.com: mail set sender to djgpp-workers-bounces using -f
X-Recipient: djgpp-workers AT delorie DOT com
Message-ID: <50216A91.7010005@iki.fi>
Date: Tue, 07 Aug 2012 22:20:49 +0300
From: Andris Pavenis <andris DOT pavenis AT iki DOT fi>
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:14.0) Gecko/20120717 Thunderbird/14.0
MIME-Version: 1.0
To: djgpp-workers AT delorie DOT com
Subject: Re: gcc-4.8 snapshots and DJGPP
References: <50152C66 DOT 9040202 AT iki DOT fi> <201208010039 DOT 53912 DOT juan DOT guerrero AT gmx DOT de> <501ACB2A DOT 8020203 AT iki DOT fi> <201208051423 DOT 28044 DOT juan DOT guerrero AT gmx DOT de>
In-Reply-To: <201208051423.28044.juan.guerrero@gmx.de>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Reply-To: djgpp-workers AT delorie DOT com

On 08/05/2012 03:23 PM, Juan Manuel Guerrero wrote:
>
> OFYI and to finalize this test sequence, I was able to compiled successfully
> both gcc480-20120722s.zip and gcc480-20120729s.zip and to use those produced
> compilers to compile the sources a second time from scratch.  I checked this
> on WinXP Prof SP3 (32 bit), Win2K SP5 (32 bit) and Win98SE.  No memory issues
> like the ones described for Vista raised.  Due to the time it takes to compile
> the sources and that LFN support is required to compile it, it becomes pointless
> and a waste of time to try to compile the sources on MSDOS or some clones of it.
>
> Neitherless the "uncomputable CFA for frames" issue remains making the produced
> binaries useless for debugging.  This is a really serious issue.

"uncomputable CFA for frames" problem happens with -g, -ggdb, -gdwarf-3, -gdwarf-4
but not with -gdwarf-2. I do not know currently what causes that, but perhaps we
may have 2 possible suspects (gcc snapshot itself and also gdb)

About memory corruption related problems on Vista: perhaps I should
sometimes try fmalloc.c from DJGPP libc (old allocator for v2.03) instead of nmalloc
when building GCC

Additionally my memory test program + nmalloc seems to be much slower when built with
gcc-4.8.0-20120805 (freeing 10**7 - 2*10**7 blocks takes about 3 times as long)

About building under DOS: Dosemu under Linux supports LFN, but last when I checked
(some years ago) not all DOS calls was implemented and some DJGPP libc functions
failed causing build to fail. Current libc CVS version is expected to be better,
but tools may need to be rebuild for testing

Andris