=== frankban|afk is now known as frankban
[10:22]  * cjwatson crosses fingers and sends virtualenv-pip to PQM
[10:23] <cjwatson> Also have now diverged lp-source-dependencies git, so make sure to have switched to that
[10:34] <cjwatson> first buildbot run with pip predictably taking forever ...
[10:34] <cjwatson> (it's populating its wheel cache)
[10:39] <cjwatson> ... well, it built, at least ...
[10:40]  * wgrant finds the fireworks
[10:40] <wgrant> oh no fireworks yet, sad
[10:44] <cjwatson> I don't think I've seen that lp.codehosting.codeimport.tests.test_worker.TestBzrSvnImport.test_stacked error before
[10:44] <wgrant> Indeed
[10:44] <wgrant> It's a bit odd.
[10:44] <cjwatson> not obviously anything to do with pip, but ...
[10:44] <wgrant> But I built the tree in a couple of containers and ran the full test suite a few times locally
[10:45] <wgrant> And I assume you did too
[10:45] <cjwatson> Yeah
[10:47] <cjwatson> Could be something wrong with subvertpy, I suppose, but not sure why it'd only show on buildbot
[10:53] <cjwatson> hmm, what, this test fails in isolation locally too
[10:53] <cjwatson> maybe it has an implicit dependency on something else having run first?
[10:54] <cjwatson> yeah, if I run all of TestBzrSvnImport, it passes; if I run TestBzrSvnImport.test_stacked alone, it fails
[10:54] <wgrant> Aha
[10:54] <cjwatson> at least that's the less hideously painful way round for test isolation bugs to manifest
[10:57] <cjwatson> reminds me slightly of the bzrlib.plugins.git.cache thing
[11:52] <cjwatson> wgrant: This is actually much more interesting than expected
[11:53] <wgrant> cjwatson: oh ho?
[11:53] <cjwatson> Tracing it through, lp.codehosting is being imported from sitecustomize, which is apparently early enough that sys.getfilesystemencoding returns None, which causes bzrlib.i18n.load_plugin_translations to fail
[11:54] <cjwatson> So the svn plugin doesn't necessarily get correctly installed
[11:54] <wgrant> And the test requires it be in Swahili?
[11:54] <wgrant> Or does the load_plugin_translations failure actually break more interesting things silently?
[11:54] <cjwatson> It causes bzrplugins/svn/__init__.py to silently fail part-way through
[11:55] <wgrant> Ah, helpful.
[11:55] <cjwatson> Specifically, before it gets round to installing its prober
[13:15] <cjwatson> wgrant: https://code.launchpad.net/~cjwatson/launchpad/avoid-importing-bzr-plugins-from-site/+merge/335379 ?
[13:24] <cjwatson> passes that individual test, that test suite, and all of lp.code.model.tests
[13:24] <cjwatson> I haven't done a full run but it should be OK
[13:34] <cjwatson> Hmm, idle.  I think I'll self-review and go ahead with that; the test suite should catch anything overly interesting here
[13:38] <cjwatson> subsequent pip runs on buildbot indeed quite a bit faster
[14:09] <cjwatson> Still not quite right - need to adjust how bzr subprocesses are run, I think.
[15:10] <cjwatson> OK, I think it just needed a subvertpy wheel built against precise (ugh, but the proper fix isn't in a released version of subvertpy yet)
[16:55] <cjwatson> yay, it passed
[17:26] <cjwatson> ... and qastaging has updated successfully
=== frankban is now known as frankban|afk
[23:20] <wgrant> Nice