Problems building pl (swi-prolog) on ppc64

[Gérard Milmeister]

I have difficulties building pl on ppc64. The previous build (a few
releases back) has been fine, and i386, x86_64, ppc are ok, too. The
koji build page is:
http://koji.fedoraproject.org/koji/taskinfo?taskID=278425
The 64-bit ppc arch seems to cause a lot of trouble compared to the
32-bit arch and x86_64. Is there any specific reason for this? In any
case I hope someone can help fixing the issue.

BTW. I would welcome a co-maintainer for the pl package, preferably
someone with experience on archs other that i386.
-- 
Gérard Milmeister
Langackerstrasse 49
CH-8057 Zürich

[Kevin Kofler]

Gérard Milmeister <gemi <at> bluewin.ch> writes:
>  I have difficulties building pl on ppc64. The previous build (a few
>  releases back) has been fine, and i386, x86_64, ppc are ok, too. The
>  koji build page is:
>  http://koji.fedoraproject.org/koji/taskinfo?taskID=278425 
Judging by the warnings about aliasing violations, the first thing I'd try is 
adding -fno-strict-aliasing to the CFLAGS.

        Kevin Kofler

[Gérard Milmeister]

On Thu, 2007-12-06 at 11:57 +0000, Kevin Kofler wrote:
>  Judging by the warnings about aliasing violations, the first thing
> I'd
>  try is 
>  adding -fno-strict-aliasing to the CFLAGS. 
These warnings appear on all platforms,
but only ppc64 segfaults.

-- 
Gérard Milmeister
Langackerstrasse 49
CH-8057 Zürich

[Kevin Kofler]

Gérard Milmeister <gemi <at> bluewin.ch> writes:
>  These warnings appear on all platforms, but only ppc64 segfaults.
This doesn't exclude an aliasing problem, aliasing bugs often only lead to 
actually invalid code on some architectures, it all depends on how the GCC 
optimizers process the code.

        Kevin Kofler

[Andrew Haley]

Gérard Milmeister writes:
>  On Thu, 2007-12-06 at 11:57 +0000, Kevin Kofler wrote:
>  > Judging by the warnings about aliasing violations, the first thing I'd
>  > try is 
>  > adding -fno-strict-aliasing to the CFLAGS.
>  These warnings appear on all platforms, but only ppc64 segfaults. 
If you get any such warnings, your program is almost certainly broken.
-fno-strict-aliasing is a way to get your program working until it's
fixed.

Andrew.

-- 
Red Hat UK Ltd, Amberley Place, 107-111 Peascod Street, Windsor, Berkshire, SL4 1TE, UK
Registered in England and Wales No. 3798903

[Gérard Milmeister]

On Thu, 2007-12-06 at 13:28 +0000, Andrew Haley wrote:
>  If you get any such warnings, your program is almost certainly
> broken.
>  -fno-strict-aliasing is a way to get your program working until it's
>  fixed. 
I just tried to build with -fno-strict-aliasing, there are no warnings
now, but the problem on ppc64 remains.
-- 
Gérard Milmeister
Langackerstrasse 49
CH-8057 Zürich

[Mary Ellen Foster]

Gérard Milmeister <gemi <at> bluewin.ch> writes:

>  
>  On Thu, 2007-12-06 at 13:28 +0000, Andrew Haley wrote:
>  > If you get any such warnings, your program is almost certainly broken.
>  > -fno-strict-aliasing is a way to get your program working until it's
>  > fixed.
>  I just tried to build with -fno-strict-aliasing, there are no warnings
>  now, but the problem on ppc64 remains. 
Just resurrecting this thread: the problem with building swi-prolog on ppc64
remains, and neither Gérard Milmeister (the maintainer) nor I (new
co-maintainer) have any idea what's going on. I've tried a variety of CFLAGS
including "-O0" and the build always dies with a segfault when running the
newly-built pl interpreter as part of the build process. Here's a recent failed
build from Koji:

    http://koji.fedoraproject.org/koji/taskinfo?taskID=472959

Anyone with any idea what's going on?

Thanks,

MEF