[Buildbot-devel] Some questions/problems running latest build bot source on Windows

Elliot Murphy elliot.murphy at veritas.com
Tue Nov 9 16:49:21 UTC 2004


Hi Brian,
This morning the rest of your recent patches were available through CVS, so
I updated and re-ran the unit tests (without any of my local changes, and
without your tsc patch). Now I only see 5 failures, and no errors.

All 5 failures are testShell tests. Looking in _trial_temp/trial.log, I see
5 identical stack traces:

File "C:\working\cvs.sourceforge.net\buildbot\buildbot\slave\commands.py",
line 168, in _startCommand
	    usePTY=self.usePTY)
	  File "C:\Python23\Lib\site-packages\twisted\internet\default.py",
line 179, in spawnProcess
	    raise NotImplementedError, "process only available in this " \
	exceptions.NotImplementedError: process only available in this
reactor on POSIX, use win32eventreactor on Windows

After doing some more digging, it seems that this is a twisted problem with
spawnProcess not being implemented in the default reactor on Windows, not a
buildbot problem. The good news is that it seems to have been fixed a couple
of months ago:
http://www.twistedmatrix.com/users/roundup.twistd/twisted/issue591

I don't know enough about twisted to know whether reactor.spawnProcess() is
something that is trivial to work around in buildbot, or whether I just need
to wait for the next twisted release. I'll post a message over to
twisted-python to ask about the best way to get the fix for Issue591. It
seems like running commands in subprocesses might be required in order to
get even a slave to work properly. Any thoughts?

regards,
-elliot

|-----Original Message-----
|From: buildbot-devel-admin at lists.sourceforge.net [mailto:buildbot-devel-
|admin at lists.sourceforge.net] On Behalf Of Elliot Murphy
|Sent: Monday, November 08, 2004 5:03 PM
|To: Brian Warner
|Cc: buildbot-devel at lists.sourceforge.net
|Subject: RE: [Buildbot-devel] Some questions/problems running latest build
|bot source on Windows
|
|Thanks, the bb patch fixed a few of the failing tests.
|
|The tsc.diff patch did not seem to make any difference. The error message
|in
|test.log about spawnProcess seems to be asking for
|win32eventreactor.Win32Reactor to be imported instead of plain reactor
|buildbot/slavecommands.py (only when running on win32, of course). I
|temporarily tried changing the import, and it seemed to fix a few more
|tests. Dunno if it is relevant, but I'm running with Twisted 1.30.
|
|Assuming the Win32Reactor thing is the right fix, I only see with 2
|failures
|and 9 errors. I assume some of the problem could be that changes have not
|rippled through CVS yet, so I can try re-running the test suite tomorrow.
|Just in case, I've copied the fail/error messages below and am attaching
|the
|trial/test logs.
|
|testPrefix (buildbot.test.test_changes.TestChangePerspective) ... [FAIL]
|testMsgS1 (buildbot.test.test_mailparse.Test3) ... [FAIL]
|
|testMsg1p (buildbot.test.test_mailparse.Test2) ... [ERROR]
|testMsg2p (buildbot.test.test_mailparse.Test2) ... [ERROR]
|testMsg3p (buildbot.test.test_mailparse.Test2) ... [ERROR]
|testMsg4p (buildbot.test.test_mailparse.Test2) ... [ERROR]
|testMsg5p (buildbot.test.test_mailparse.Test2) ... [ERROR]
|testMsg6p (buildbot.test.test_mailparse.Test2) ... [ERROR]
|testMsg7p (buildbot.test.test_mailparse.Test2) ... [ERROR]
|testMsgS4 (buildbot.test.test_mailparse.Test3) ... [ERROR]
|testSlave (buildbot.test.test_run.Status) ... [ERROR]
|
|regards,
|-elliot
|
||-----Original Message-----
||From: Brian Warner [mailto:warner-buildbot at lothar.com]
||Sent: Monday, November 08, 2004 4:23 PM
||To: Elliot Murphy
||Cc: buildbot-devel at lists.sourceforge.net
||Subject: Re: [Buildbot-devel] Some questions/problems running latest build
||bot source on Windows
||
||> However, here is something that will work on NT-based Windows systems
||> (Win2K, XP, etc.), but not on Win95:
||> win32api.MoveFileEx(src, dst, win32con.MOVEFILE_REPLACE_EXISTING)
||
||Eyah. Ok, I think I'll go with the os.unlink approach, since the most
||likely
||failure case is when the filesystem is full, in which case the pickle will
||fail first. I've committed the change, it may be a few hours before it
||shows
||up on sf.net's anonymous-cvs, so I've attached the patch.
||
|| -Brian
|





More information about the devel mailing list