Intermittent failure of twisted.web.test.test_webclient.WebClientSSLTestCase.test_timeoutTriggering

[exarkun]

On OS X, I saw this fail (while working on #3782):

[ERROR]: twisted.web.test.test_webclient.WebClientSSLTestCase.test_timeoutTriggering
Traceback (most recent call last):
Failure: twisted.trial.util.DirtyReactorAggregateError: Reactor was unclean.
Selectables:
<<class 'twisted.internet.tcp.TLSConnection'> to ('127.0.0.1', 59660) at 1345af0>

[exarkun]

(In [32326]) Branching to 'oldwebclient-cleanup-3796'

[exarkun]

I changed getPage and downloadPage to only fire their result Deferred after the underlying connection gives the HTTPClient its connectionLost notification.

I tested this change in the #5118 branch (with the icky tearDown added in that branch deleted) on my Ubuntu Karmic box to verify it actually solves the problems encountered there, and it does. As is the branch fails 5 webclient ssl tests there, with this patch it passes them all reliably.

 Build results

[glyph]

Aah, what a nice, self-contained change. Looks great. Should make naive third-party tests using getPage a bit more reliable, too. (And thanks for the pyflakes fixes.)

No further comments; please land.

[exarkun]

(In [32329]) Merge oldwebclient-cleanup-3796

Author: exarkun Reviewer: glyph Fixes: #3796

Change twisted.web.client.getPage and twisted.web.client.downloadPage to not fire their returned Deferred until the connection used for the request each issues is closed. This avoids problems in unit tests using these APIs where trial reports an unclean reactor due to these connections still being up when the test thinks it is finished.