MIME-Version: 1.0
In-Reply-To: <50152C66.9040202@iki.fi>
References: <50152C66 DOT 9040202 AT iki DOT fi>
Date: Tue, 31 Jul 2012 07:54:01 -0500
Message-ID: <CAA-ihx_+Bt+JHXtp-cS9NRmuRi_a8DGAFThJK42fRN_YYWvBFw@mail.gmail.com>
Subject: Re: gcc-4.8 snapshots and DJGPP
From: Rugxulo <rugxulo AT gmail DOT com>
To: djgpp-workers AT delorie DOT com
Content-Type: text/plain; charset=ISO-8859-1
Reply-To: djgpp-workers AT delorie DOT com
Errors-To: nobody AT delorie DOT com
Precedence: bulk

Hi,

On Sun, Jul 29, 2012 at 7:28 AM, Andris Pavenis <andris DOT pavenis AT iki DOT fi> wrote:
> Tried to build the recent gcc-4.8 snapshots for DJGPP. Needed
> to update patch-set for source patching to work of course.

What changed in 4.8.x? Anything worthwhile? Their online status
doesn't list very much!

http://gcc.gnu.org/gcc-4.8/changes.html

> As earlier started with building Linux to DJGPP cross-compiler RPMs
> (with DJGPP source archive as side product).
> - all seems to be OK under Fedora 17 x86_64 (cross-compiler RPMs
>   gets built without problems)
> - tried to do the same under CentOS 5.8 i386 (really chroot under
>   the same Fedora installation). This time no success. Perhaps
>   gcc version 4.1.2 20080704 (Red Hat 4.1.2-52) present there is
>   too old. That could perhaps be fixed by taking some intermediate
>   version.
> - perhaps I should create CentOS 6.3 i386 chroot and retry

BTW, unrelated, and please don't feel like I'm actually asking anyone
to do this, but I can't for the life of me get GM2 (GCC 4.1.2) to
build. I tried a week or two ago to build a Linux-hosted
DJGPP-targeted version, but it only halfway built and didn't exactly
work (obviously).

Sometimes I think they don't (or can't) test certain combinations.
It's clear that various bugs are still there.

> After that built native compiler for DJGPP (tried only DJGPP v2.04pre).
> It built practically "out of box" except of random compiler errors
> time at time similarly as with some earlier stable version.
>
> I think that one of suspects is nmalloc used instead of DJGPP own memory
> allocation (as DJGPP free() is too slow). I had some stability problems
> earlier running my own test program (linked with nmalloc) which allocates
> large numbers of random memory blocks and after that frees them in random
> order. That memory allocation test program caused random NTVDM crashes
> on Windows Vista.

It could be nmalloc's bug, who knows, but I don't trust Vista as far
as I can throw it re: DPMI. I would not put it past them as their bug,
even with the DPMI limit override. Seriously. Try testing in pure DOS,
if possible, and see if it's still buggy.

> Not much testing done with built GCC snapshot though. Built GMP-5.0.5, but
> one of
> test-suite programs (mpz/t-io_raw.exe) failed. I did not have time
> to study why it fails yet.

Needs setmode(O_BINARY) ???