Xunit & KnownFailure

[takluyver]

For the buildbots, Xunit currently counts KnownFailures as failures. This monkeypatches it to count them as skips, so we get a clean test run, like this:

https://jenkins.shiningpanda.com/ipython/job/ipython-xunit-test/16/

[minrk]

Looks fine to me. Are there any versioning issues with this? i.e. nose.plugins.xunit not being present?

If that's guaranteed to exist, then go ahead and merge.

[takluyver]

It looks like it was added in nose 0.11, which was out by 2010 or earlier.
Are we happy to require that?

[minrk]

0.11 is in Ubuntu 10.04 LTS, so that seems plenty conservative to me. Package folks don't break up the plugins into subpackages, do they?

[takluyver]

Debian/Ubuntu packagers don't. It doesn't look like Fedora do, though I'm
less familiar with their system:
http://koji.fedoraproject.org/koji/buildinfo?buildID=293192

[minrk]

Okay, then go ahead and merge. I think we've done due diligence here.

[minrk]

@takluyver - this seems to have broken oneiric auto-builds, if you know why that would be. I can't seem to reproduce locally on oneiric with apt-installed nose 1.0.

Output:

https://launchpadlibrarian.net/92736954/buildlog_ubuntu-oneiric-i386.ipython_0.12%2B2321-0~7~oneiric1_FAILEDTOBUILD.txt.gz

[takluyver]

I'm mostly baffled, but I think these lines in nose trunk (a few lines
above the failure) might be a clue:

        if issubclass(ec, SkipTest):
            if not hasattr(self.plugin, 'addSkip'):
                return

https://github.com/nose-devs/nose/blob/master/nose/plugins/manager.py#L324

Not sure how that could make it error, though. I didn't even think we
had any known failures in IPython.parallel. ShiningPanda (Debian 6)
doesn't reproduce the error either.

[takluyver]

(Of course ShiningPanda wouldn't replicate it - that's using nose installed by pip in a virtualenv).

[minrk]

There's exactly one KnownFailure in IPython.parallel, and it's only raised on Python 3. But I don't think that really as anything to do with this. The exception is raised as soon as the first IPython.parallel test group is finished, and nothing is skipped or fails there. What if we made a small tweak here, and only applied the monkeypatch when xunit is requested?

[takluyver]

Good idea. Check over PR #1405 and merge it if you're happy.