[Debian-med-packaging] Bug#820060: Debian Bug#820060: python-biom-format: broken on big-endian architectures

[Daniel McDonald]

Thanks, Greg.

Andreas, my read of the bug report suggests that the issue across
architectures is h5py or hdf5. The access being made is to an attribute of
an h5py object which happens within python and is a valid API operation;
testing locally I'd expect a different KeyError to be thrown under normal
operation.

Looking closer at one of the failing tests (offending call linked below),
you can see that this is on read of a hdf5 table that ships with the
biom-format project.

https://github.com/biocore/biom-format/blob/master/tests/test_table.py#L527-528

The implication is that this is happening when reading a table that ships
as part of the test code, suggesting that h5py or hdf5 are unable to
interact with the file as expected. My take is that the failure we're
observing is a manifestation of a bug at a lower level.

The failure observed on mipsel is a stochastic one.

Best,
Daniel

On Tue, Apr 5, 2016 at 4:40 PM, Greg Caporaso <gregcaporaso at gmail.com>
wrote:

> Thanks for sharing this.
>
> Daniel, Yoshiki - any ideas about this?
>
> Greg
> ---------- Forwarded message ----------
> From: Andreas Tille <andreas at an3as.eu>
> Date: Mon, Apr 4, 2016 at 11:18 PM
> Subject: Debian Bug#820060: python-biom-format: broken on big-endian
> architectures
> To: Greg Caporaso <gregcaporaso at gmail.com>, 820060 at bugs.debian.org
>
>
> Hi Greg,
>
> I'm hereby forwarding a bug report against the Debian packaged
> biom-format.  It would be great if you can either fix the issue (which
> would be the prefered solution) or explain that for some reason
> big-endian is not supported and we need to exclude these architectures
> from the build.
>
> Kind regards
>
>        Andreas.
>
> On Mon, Apr 04, 2016 at 10:44:51PM -0700, Steve Langasek wrote:
> > Package: python-biom-format
> > Version: 2.1.4+dfsg-1
> > Severity: serious
> >
> > As seen at http://buildd.debian.org/python-biom-format, this package
> fails
> > to build on all big-endian architectures in the archive (mips, powerpc,
> > s390x).  While this build failure has only appeared with the upload of
> > 2.1.5+dfsg-1, this is not a regression in that version of the upstream
> > source; instead, the package was already broken, but what has changed
> now is
> > that a new version of dh-python correctly finds the tests and runs them
> at
> > build time where before dh_auto_test was silently passing.
> >
> > I'm not sure if this points to breakage in python-biom-format, or in
> > something lower down the stack that it depends on.  I've confirmed that
> > python-numpy autopkgtests run ok on powerpc (except for openblas - and
> the
> > tests can't be run on s390x because openblas is not built for that
> > architecture), and h5py autopkgtests run ok on s390x.  python-scipy's
> > autopkgtests fail with SIGABRT on both powerpc and s390x, maybe related,
> > maybe not (test_random_complex_exact (test_basic.TestLstsq) ... ***
> Error in
> > `python3.5': free(): invalid pointer: 0x1172c2a8 ***).
> >
> > And hdf5 itself doesn't have any autopkgtests, and any build-time tests
> > pass; and h5py in any case should catch any bugs with its own testsuite
> > before python-biom-format.
> >
> > So it seems most likely that this is a bug in python-biom-format itself
> > rather than any of its dependencies, thus assigning here.
> >
> > --
> > Steve Langasek                   Give me a lever long enough and a Free
> OS
> > Debian Developer                   to set it on, and I can move the
> world.
> > Ubuntu Developer
> http://www.debian.org/
> > slangasek at ubuntu.com
> vorlon at debian.org
>
>
>
> > _______________________________________________
> > Debian-med-packaging mailing list
> > Debian-med-packaging at lists.alioth.debian.org
> >
> http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/debian-med-packaging
>
>
> --
> http://fam-tille.de
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/debian-med-packaging/attachments/20160405/0248c979/attachment-0001.html>