[Debian-med-packaging] Bug#951222: scikit-learn breaks python-skbio autopkgtest: skbio.stats.distance._permdisp.permdisp

[Christian Kastner]

On 2020-02-13 09:31:54 +0100 Andreas Tille <tille at debian.org> wrote:
> Control: tags -1 upstream
> Control: forwarded -1 https://github.com/biocore/scikit-bio/issues/1693
> 
> > Due to the nature of this issue, I filed this bug report
> > against both packages. Can you please investigate the situation and
> > reassign the bug to the right package?
> 
> I admit I'm not sure what package needs to be fixed but in any case I've
> forwarded the issue to python-skbio upstream to make them aware of this.
> My observation is that responses to bug reports of scikit-bio github are
> not to be expected in a short time frame.  I'm also not sure whether the
> upstream tag / forward for python-skbio might hide the issue from
> scikit-learn maintainers view - I have no idea where the issue really
> needs to be fixed.

I've been working on fixing the open issues of src:scikit-learn, and
looked into this bug.

This is odd. When executing the failing test code manually, the output
is just as expected (p = 0.35). The output is also just as expected when
the test suite is run at build time [1], as all tests pass.

It's only when autopkgtest invokes the tests that this test fails.

At build time, tests are simply run (by pybuild) with,
     python3.x -m pytest
whereas the autopkgtest calls
     python3 -m skbio.test
so I assume it has something to do with the latter.

I did not investigate this further, I just reported this back upstream
to the referenced GitHub issue.

[1]
https://buildd.debian.org/status/fetch.php?pkg=python-skbio&arch=amd64&ver=0.5.5-7&stamp=1582910349&raw=0