Re: debian-hppa port support

[Luk Claes]

Carlos O'Donell wrote:

> If you have any problems building a debian package for hppa, please
> contact me on debian-hppa.
>
> As a porter I've setup a debian buildd on my local hppa machine for
> reproducing build problems.
>
> Please don't hesitate to email debian-hppa.
>
> Known issues:
> * The ruby1.9 build problem still exists on hppa until we switch to
> nptl (to support the signal semantics expected by ruby) or disable the
> particular test to allow the build to complete.
> * glibc does not yet support nptl. glibc patches to enable nptl for
> hppa are complete, but usability testing is still in progress to iron
> out the remaining bugs.

Do you have an idea why petsc fails on hppa? It's one of the blockers of
the petsc/openmpi/suitesparse transition.

Cheers

Luk


--
To UNSUBSCRIBE, email to debian-hppa-REQUEST@...
with a subject of "unsubscribe". Trouble? Contact listmaster@...

[Carlos O'Donell-2]

On Sat, Jul 11, 2009 at 11:01 AM, Luk Claes<luk@...> wrote:
> Do you have an idea why petsc fails on hppa? It's one of the blockers of
> the petsc/openmpi/suitesparse transition.

I'm sbuild-ing petsc, I'll tell you what I find.

Cheers,
Carlos.


--
To UNSUBSCRIBE, email to debian-hppa-REQUEST@...
with a subject of "unsubscribe". Trouble? Contact listmaster@...

[Carlos O'Donell-2]

On Sat, Jul 11, 2009 at 11:41 AM, Carlos
O'Donell<carlos@...> wrote:
> On Sat, Jul 11, 2009 at 11:01 AM, Luk Claes<luk@...> wrote:
>> Do you have an idea why petsc fails on hppa? It's one of the blockers of
>> the petsc/openmpi/suitesparse transition.
>
> I'm sbuild-ing petsc, I'll tell you what I find.

I can reproduce this. I'm seeing the same failure as the autobuilder,
and I've seen at least one python sigsegv (note that there is a
discussion on linux-parisc about a bug in the kernel related to
corrupting memory protection bits). I've enabled core dumps in the
buildd chroot so I can examine crashes post-mortem.

https://buildd.debian.org/fetch.cgi?&pkg=petsc&ver=3.0.0.dfsg-5.1&arch=hppa&stamp=1246867119&file=log
~~~
make[1]: Entering directory `/build/buildd/petsc-3.0.0.dfsg'
/build/buildd/petsc-3.0.0.dfsg/conf/variables:118:
/build/buildd/petsc-3.0.0.dfsg/linux-gnu-c-debug/conf/petscvariables:
No such file or directory
/build/buildd/petsc-3.0.0.dfsg/conf/rules:982:
/build/buildd/petsc-3.0.0.dfsg/linux-gnu-c-debug/conf/petscrules: No
such file or directory
make[1]: *** No rule to make target
`/build/buildd/petsc-3.0.0.dfsg/linux-gnu-c-debug/conf/petscrules'.
Stop.
make[1]: Leaving directory `/build/buildd/petsc-3.0.0.dfsg'
make: *** [build-arch] Error 2
dpkg-buildpackage: error: debian/rules build gave error exit status 2
~~~

I'll build this by hand to inspect why it fails.

Cheers,
Carlos.


--
To UNSUBSCRIBE, email to debian-hppa-REQUEST@...
with a subject of "unsubscribe". Trouble? Contact listmaster@...

[Carlos O'Donell-2]

On Sat, Jul 11, 2009 at 12:22 PM, Carlos
O'Donell<carlos@...> wrote:
> I can reproduce this. I'm seeing the same failure as the autobuilder,
> and I've seen at least one python sigsegv (note that there is a
> discussion on linux-parisc about a bug in the kernel related to
> corrupting memory protection bits). I've enabled core dumps in the
> buildd chroot so I can examine crashes post-mortem.

On another topic, how do I stop sbuild from erasing the build
directory, technically it should contain the python core file, but it
looks like it's been wiped clean.

Cheers,
Carlos.


--
To UNSUBSCRIBE, email to debian-hppa-REQUEST@...
with a subject of "unsubscribe". Trouble? Contact listmaster@...

[Luk Claes]

Carlos O'Donell wrote:

> On Sat, Jul 11, 2009 at 12:22 PM, Carlos
> O'Donell<carlos@...> wrote:
>> I can reproduce this. I'm seeing the same failure as the autobuilder,
>> and I've seen at least one python sigsegv (note that there is a
>> discussion on linux-parisc about a bug in the kernel related to
>> corrupting memory protection bits). I've enabled core dumps in the
>> buildd chroot so I can examine crashes post-mortem.
>
> On another topic, how do I stop sbuild from erasing the build
> directory, technically it should contain the python core file, but it
> looks like it's been wiped clean.

In the .sbuildrc file (or /etc/sbuild.conf) you can put the following:

# When to purge the build directory afterwards; possible values are "never",
# "successful", and "always"
$purge_build_directory="always";

So you could put that on successful or never if you always want to keep
the build directory.

Cheers

Luk


--
To UNSUBSCRIBE, email to debian-hppa-REQUEST@...
with a subject of "unsubscribe". Trouble? Contact listmaster@...

[Luk Claes]

Carlos O'Donell wrote:

> On Sat, Jul 11, 2009 at 11:41 AM, Carlos
> O'Donell<carlos@...> wrote:
>> On Sat, Jul 11, 2009 at 11:01 AM, Luk Claes<luk@...> wrote:
>>> Do you have an idea why petsc fails on hppa? It's one of the blockers of
>>> the petsc/openmpi/suitesparse transition.
>> I'm sbuild-ing petsc, I'll tell you what I find.
>
> I can reproduce this. I'm seeing the same failure as the autobuilder,
> and I've seen at least one python sigsegv (note that there is a
> discussion on linux-parisc about a bug in the kernel related to
> corrupting memory protection bits). I've enabled core dumps in the
> buildd chroot so I can examine crashes post-mortem.

Ah good to know that the kernel issues are being worked on.

> I'll build this by hand to inspect why it fails.

Great, thanks already!

Cheers

Luk


--
To UNSUBSCRIBE, email to debian-hppa-REQUEST@...
with a subject of "unsubscribe". Trouble? Contact listmaster@...

[Carlos O'Donell-2]

On Sat, Jul 11, 2009 at 12:50 PM, Luk Claes<luk@...> wrote:
>> I'll build this by hand to inspect why it fails.
>
> Great, thanks already!

I get consistent lockups in python when trying to build petsc, the
lockup happens during configure and while python is running the
equivalent of pthread_cond_wait. It looks, but I haven't verified,
like a similar situation to ruby, where the semantics of Linuxthreads
vs. NPTL are different enough that the newer python2.5 code is
unreliable under Linuxthreads. I don't know how m68k keeps up without
NPTL.

I'm going to stop debugging petsc, ruby, and python for now, and I'm
going to instead focus on producing a libc6 deb (with NPTL) which I
can put into my sbuild chroot.

After I have a libc6 deb with nptl, then I'll bubble back up to see if
I can build ruby and python and look at their testsuite failures.

After that I'll bubble back up to petsc.

If you have any better suggestions, please share them.

Cheers,
Carlos.


--
To UNSUBSCRIBE, email to debian-hppa-REQUEST@...
with a subject of "unsubscribe". Trouble? Contact listmaster@...

[Carlos O'Donell-2]

On Sat, Jul 11, 2009 at 4:54 PM, Carlos O'Donell<carlos@...> wrote:
> After I have a libc6 deb with nptl, then I'll bubble back up to see if
> I can build ruby and python and look at their testsuite failures.
>
> After that I'll bubble back up to petsc.
>
> If you have any better suggestions, please share them.

I'm building glibc 2.1 experimental, still in progress.

Cheers,
Carlos.


--
To UNSUBSCRIBE, email to debian-hppa-REQUEST@...
with a subject of "unsubscribe". Trouble? Contact listmaster@...

[Carlos O'Donell-2]

On Sat, Jul 11, 2009 at 11:01 AM, Luk Claes<luk@...> wrote:
> Do you have an idea why petsc fails on hppa? It's one of the blockers of
> the petsc/openmpi/suitesparse transition.

Luk, this is a reminder that the petsc build failure is still on my plate.

The petsc build failure looks like python problems with the glibc
threading library, therefore I'm focusing on the hppa transition to
the new glibc threading library (NPTL).

Fixing the petsc build failure is next on the queue. Thanks.

Cheers,
Carlos.


--
To UNSUBSCRIBE, email to debian-hppa-REQUEST@...
with a subject of "unsubscribe". Trouble? Contact listmaster@...