test.support.threading_cleanup() now logs a warning on failure

[vstinner]

The @reap_threads decorator and the threading_cleanup() function of
test.support now log a warning if they fail to clenaup threads.

The log may help to debug such other warning seen on the AMD64
FreeBSD CURRENT Non-Debug 3.x buildbot:

Warning -- threading._dangling was modified by test_logging

[mention-bot]

@Haypo, thanks for your PR! By analyzing the history of the files in this pull request, we identified @Yhg1s, @ncoghlan and @benjaminp to be potential reviewers.

[vstinner]

Oh, good job Travis! The GCC job passed, but the Clang job catched a bug in test_urllib2_localnet doesn't come from my change. It's just that my change makes the bug more obvious :-) It's now fixed as well.

[serhiy-storchaka]

I got following strange warning:

0:06:50 [178/404] test_threading passed
Warning -- threading_cleanup() failed to cleanup 0 threads after 5 sec (count: 0, dangling: 21)

Is this alright? I mean warning about 0 threads.

[vstinner]

Warning -- threading_cleanup() failed to cleanup 0 threads after 5 sec (count: 0, dangling: 21)

I am also surprised by the warning, but I also saw it. I don't understand well how the function works. It's something about "dangling threads". If I understood correctly, if the number of dangling threads is not zero, a test "leaks" threads and should be fixed to not have side effect on following tests.

[vstinner]

I added the warning because I saw that test_quick_connect() of test_asyncore took much longer than expected sometimes but only on FreeBSD. It was a bug which was quiet. The warning should help to detect bugs.

[vstinner]

I also proposed a buildbot change to catch more warnings on buildbot: python/buildmaster-config#1 (comment)