I can't seem to run the Twisted tests without hitting an intermittent failure on my Windows machine

[twisted-trac]

	@glyph reported
Trac ID	trac#2981
Type	defect
Created	2008-01-07 11:37:08Z
Branch	https://github.com/twisted/twisted/tree/intermittent-2981

Although our Windows buildslave reports test successes most of the time these days, I have found that the tests will fairly reliably hit at least one intermittent failure on my desktop installation of Windows.

I don't particularly like the scope of ticket, but one of my new years' resolutions is to avoid letting process anxieties get in the way of useful work. This ticket may just be closed as "invalid" eventually. For now it should serve as an anchor point to discuss the issues that I've found and link them to the appropriate, hopefully pre-existing tickets, and to give me a ticket number so I can make a branch to put the fixes I've made.

Searchable metadata

``` trac-id__2981 2981 type__defect defect reporter__glyph glyph priority__high high milestone__ branch__branches_intermittent_2981 branches/intermittent-2981 branch_author__ status__closed closed resolution__fixed fixed component__core core keywords__discussion discussion time__1199705828000000 1199705828000000 changetime__1262318997000000 1262318997000000 version__None None owner__ cc__therve cc__Trent.Nelson cc__exarkun ```

[twisted-trac]

	Automation removed owner

[twisted-trac]

	@exarkun set status to closed

All those other tickets have been resolved. There still may be something in test_tcp that hasn't been addressed. Please open a new ticket detailing that issue.

[twisted-trac]

	@glyph commented

(In [22288]) Branching to 'intermittent-2981'

[twisted-trac]

	@therve commented

There is at least #2876 which is up to review. Also, you should run trial with the --unclean-warnings option to avoid those errors.

[twisted-trac]

	Trent.Nelson commented

What sort of failures? I've been working on all kinds of stuff trying to address these sorts of things on the Python Windows buildbots. Biggest issue: file permissions. Second biggest issue: not using SO_EXCLUSIVEADDRUSE on every TCP/IP socket. Third biggest issue: using SO_REUSEADDR on Windows for anything but true multicast UDP sockets. Having tests hard code ports was pretty common in Python's test suite, that'll cause all sorts of problems obviously when tests are executing simultaneously -- I'm told you guys use ephemeral ports for client/server testing though so it's less of a problem.

[twisted-trac]

	@exarkun commented

The branch seems to contain fixes for these issues:

• #3424
• #2878
• #2869
• #2876
• #2374
• Something in test_tcp I can't immediately associate with a ticket.

Please verify the above and then close this ticket and open a new one with details about the failures related to the test_tcp changes.

[twisted-trac]

	davidsarah commented

Replying to Trent.Nelson:

Second biggest issue: not using SO_EXCLUSIVEADDRUSE on every TCP/IP socket.

I completely agree; this is a serious security issue (now #4195).