python-3.3.5 test failures on i686 in core-updates

python-3.3.5 test failures on i686 in core-updates

[Mark H Weaver]

Hello Guix,

We're trying to get core-updates built, as it includes glibc-2.20 which
includes fixes for a couple of CVEs, but we've run into a serious snag.
Python-3.3.5 fails its test suite on i686.  See:

  http://hydra.gnu.org/build/90159
  http://hydra.gnu.org/build/90159/log/tail-reload

I guess it's probably related to the glibc-2.20 update.  Interestingly,
the problem doesn't occur on x86_64.

Any volunteers to look into it?  If you have an i686 box running Guix,
you can compile the latest core-updates and get substitutes for all of
python-3's build dependencies.

     Mark

Re: python-3.3.5 test failures on i686 in core-updates

[Mark H Weaver]

I wrote:
> Python-3.3.5 fails its test suite on i686.

This turns out to be a bug in glibc-2.20.
I've filed an upstream ticket here:

https://sourceware.org/bugzilla/show_bug.cgi?id=17398

    Mark

Re: python-3.3.5 test failures on i686 in core-updates

[Ludovic Courtès]

Mark H Weaver <mhw@netris.org> skribis:

> I wrote:
>> Python-3.3.5 fails its test suite on i686.
>
> This turns out to be a bug in glibc-2.20.
> I've filed an upstream ticket here:
>
> https://sourceware.org/bugzilla/show_bug.cgi?id=17398

It turned out to be a GCC 4.8 bug:

  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61801

So now we have two option:

  1. Patch GCC 4.8.3 in core-updates, and rebuild the whole thing.

  2. Switch to GCC 4.9 as the default.

Option #1 is the safest, I think.

Thoughts?

Ludo’.

Re: python-3.3.5 test failures on i686 in core-updates

[Mark H Weaver]

ludo@gnu.org (Ludovic Courtès) writes:

> Mark H Weaver <mhw@netris.org> skribis:
>
>> I wrote:
>>> Python-3.3.5 fails its test suite on i686.
>>
>> This turns out to be a bug in glibc-2.20.
>> I've filed an upstream ticket here:
>>
>> https://sourceware.org/bugzilla/show_bug.cgi?id=17398
>
> It turned out to be a GCC 4.8 bug:
>
>   https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61801
>
> So now we have two option:
>
>   1. Patch GCC 4.8.3 in core-updates, and rebuild the whole thing.
>
>   2. Switch to GCC 4.9 as the default.
>
> Option #1 is the safest, I think.
>
> Thoughts?

We've not yet built GCC 4.9.x successfully on mips64el, so that leads me
to prefer option #1.

    Thanks,
      Mark

Re: python-3.3.5 test failures on i686 in core-updates

[Ludovic Courtès]

Mark H Weaver <mhw@netris.org> skribis:

> ludo@gnu.org (Ludovic Courtès) writes:

[...]

>> It turned out to be a GCC 4.8 bug:
>>
>>   https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61801
>>
>> So now we have two option:
>>
>>   1. Patch GCC 4.8.3 in core-updates, and rebuild the whole thing.
>>
>>   2. Switch to GCC 4.9 as the default.
>>
>> Option #1 is the safest, I think.
>>
>> Thoughts?
>
> We've not yet built GCC 4.9.x successfully on mips64el, so that leads me
> to prefer option #1.

OK, let’s do option #1 so we can merge core-updates ASAP.

Mark: could you add the relevant patch in core-updates?

Thanks,
Ludo’.

Re: python-3.3.5 test failures on i686 in core-updates

[Mark H Weaver]

ludo@gnu.org (Ludovic Courtès) writes:

> Mark H Weaver <mhw@netris.org> skribis:
>
>> ludo@gnu.org (Ludovic Courtès) writes:
>
> [...]
>
>>> It turned out to be a GCC 4.8 bug:
>>>
>>>   https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61801
>>>
>>> So now we have two option:
>>>
>>>   1. Patch GCC 4.8.3 in core-updates, and rebuild the whole thing.
>>>
>>>   2. Switch to GCC 4.9 as the default.
>>>
>>> Option #1 is the safest, I think.
>>>
>>> Thoughts?
>>
>> We've not yet built GCC 4.9.x successfully on mips64el, so that leads me
>> to prefer option #1.
>
> OK, let’s do option #1 so we can merge core-updates ASAP.
>
> Mark: could you add the relevant patch in core-updates?

Okay, but I have other plans today and won't be able to do this for at
least 12 hours.  It would be great if someone else could do it sooner
and start another core-updates evaluation on hydra, but otherwise I'll
do it when I can.

     Mark

Re: python-3.3.5 test failures on i686 in core-updates

[Ludovic Courtès]

Commit 6904ecc applies the GCC patch.  I’ve started an evaluation of
core-updates on Hydra, so if everything goes well, it will start
rebuilding soon.

Ludo’.

Re: python-3.3.5 test failures on i686 in core-updates

[Eric Bavier]

Ludovic Courtès writes:

> Commit 6904ecc applies the GCC patch.  I’ve started an evaluation of
> core-updates on Hydra, so if everything goes well, it will start
> rebuilding soon.

Does the patch also need to be applied to gcc-4.9?  Or only gcc-4.8
because it is the one used to build glibc?

-- 
Eric Bavier

Please avoid sending me Word or PowerPoint attachments.
See http://www.gnu.org/philosophy/no-word-attachments.html

Re: python-3.3.5 test failures on i686 in core-updates

[Ludovic Courtès]

Eric Bavier <ericbavier@gmail.com> skribis:

> Ludovic Courtès writes:
>
>> Commit 6904ecc applies the GCC patch.  I’ve started an evaluation of
>> core-updates on Hydra, so if everything goes well, it will start
>> rebuilding soon.
>
> Does the patch also need to be applied to gcc-4.9?  Or only gcc-4.8
> because it is the one used to build glibc?

The latter.  GCC 4.9.2 doesn’t need the patch IIUC, and even if it does,
it can always be done afterwards.

Ludo’.

Re: python-3.3.5 test failures on i686 in core-updates

[Andreas Enge]

On Wed, Sep 17, 2014 at 12:10:24PM +0200, Ludovic Courtès wrote:
> So now we have two option:
>   1. Patch GCC 4.8.3 in core-updates, and rebuild the whole thing.
>   2. Switch to GCC 4.9 as the default.
> Option #1 is the safest, I think.

As we need to rebuild everything anyway, how about trying out option 2?

In any case, before doing that, we should merge core-updates back to activate
the security fix for glibc.

Andreas

Re: python-3.3.5 test failures on i686 in core-updates

[Mark H Weaver]

Andreas Enge <andreas@enge.fr> writes:

> On Wed, Sep 17, 2014 at 12:10:24PM +0200, Ludovic Courtès wrote:
>> So now we have two option:
>>   1. Patch GCC 4.8.3 in core-updates, and rebuild the whole thing.
>>   2. Switch to GCC 4.9 as the default.
>> Option #1 is the safest, I think.
>
> As we need to rebuild everything anyway, how about trying out option 2?

We'd need to fix the GCC 4.9 build on mips64el first.

> In any case, before doing that, we should merge core-updates back to activate
> the security fix for glibc.

Given how broken things currently are on i686 in core-updates (broken
setresuid/setresgid in glibc => python-3 test failures => hundreds of
dependency failures of other important packages including emacs), I
don't see this as a good option for i686 users.

At minimum, we'd need to disable python-3 tests for i686, but even so, I
wonder what else will be broken due to the faulty setresuid/setresgid in
glibc.

      Mark