1.8 status and 64-bit data corruption

1.8 status and 64-bit data corruption

[Neil Jerram]

I believe the only important 1.8 bug still outstanding is the one
which causes Scheme data corruption on 64-bit platforms - and which is
usually enough to cause the build to fail in
snarf-check-and-output-texi.

Is there anything else of equal impact that I've forgotten?

I'm not sure how to proceed with debugging this (or rather with
advising someone else how to debug it, since I don't have a 64-bit box
myself), so does anyone else have ideas?  I've been assuming it's a GC
problem, i.e. data which should be marked but isn't for some reason,
but I guess it need not be; it could just be something modifying part
of the Scheme heap by mistake.

Here are references to some of the relevant threads.

http://lists.gnu.org/archive/html/guile-devel/2006-02/msg00044.html
http://lists.gnu.org/archive/html/bug-guile/2006-02/msg00009.html
http://lists.gnu.org/archive/html/bug-guile/2006-02/msg00016.html
http://lists.gnu.org/archive/html/guile-user/2006-03/msg00008.html

Regards,
        Neil



_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: 1.8 status and 64-bit data corruption

[Neil Jerram]

Neil Jerram <neil@ossau.uklinux.net> writes:

> I believe the only important 1.8 bug still outstanding is the one
> which causes Scheme data corruption on 64-bit platforms - and which is
> usually enough to cause the build to fail in
> snarf-check-and-output-texi.

Someone has been kind enough to loan me an account on a 64-bit
machine:

neil@we:~/guile-core-1.8-20060328$ uname -a
Linux we 2.6.8-11-amd64-k8-smp #1 SMP Wed Jun 1 00:01:27 CEST 2005 x86_64 GNU/Linux

But using the latest 1.8 snapshot, guile-core-1.8-20060328.tar.gz, I
can't reproduce the problem, either with straight `./configure; make'
or with `env CC="gcc -m64" ./configure; make'.

Is this because something got fixed in 1.8 CVS, or because I'm missing
some important ingredient of the builds which were reported as failed?

The gcc on this machine is:

neil@we:~/guile-core-1.8-20060328$ gcc --version
gcc (GCC) 3.3.5 (Debian 1:3.3.5-13)
Copyright (C) 2003 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

Any input much appreciated.

Regards,
    Neil



_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: 1.8 status and 64-bit data corruption

[Andy Wingo]

Hi Neil,

On Tue, 2006-03-28 at 21:56 +0100, Neil Jerram wrote:
> The gcc on this machine is:
> 
> neil@we:~/guile-core-1.8-20060328$ gcc --version
> gcc (GCC) 3.3.5 (Debian 1:3.3.5-13)

I got different results when using gcc 3.3. Can't remember exactly how
different tho :-/

The gcc I normally use is 4.0.3. I just reproduced the bug with that gcc
and latest guile, running with the default cflags (-g -O2). The error is
highly dependent on cflags tho.

Regards,
-- 
Andy Wingo
http://wingolog.org/



_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: 1.8 status and 64-bit data corruption

[Bill Schottstaedt]

I just checked Fedora Core 5 with gcc 4.1, and it's broken there in the
same way as in FC4/gcc 4.0.  

(On the Mac socklen_t bug, I can pass along the changes I made, if you
want them -- as I said before, they're not "the right thing").



_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: 1.8 status and 64-bit data corruption

[Jay Cotton]

Hi Bill,

I'd like to see the changes you made (keeping in mind your disclaimer  
about "rightness").

Thanks,
Jay

On Mar 29, 2006, at 10:12 AM, Bill Schottstaedt wrote:

> I just checked Fedora Core 5 with gcc 4.1, and it's broken there in  
> the
> same way as in FC4/gcc 4.0.
>
> (On the Mac socklen_t bug, I can pass along the changes I made, if you
> want them -- as I said before, they're not "the right thing").
>
>
>
> _______________________________________________
> Guile-devel mailing list
> Guile-devel@gnu.org
> http://lists.gnu.org/mailman/listinfo/guile-devel



_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: 1.8 status and 64-bit data corruption

[Bill Schottstaedt]

> But I don't quite follow why a remember would be wanted in
> SCM_VALIDATE_CELL.  I'd have thought it was in fact a good thing if
> the "cell" value went dead if not being checked.

Hunh!?  This whole thread is about building guile in gcc 4.n on a
64-bit machine.  It doesn't build,  or run unless you make the change
I suggested. 



_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel