homepage

This issue is partly in service to issue 15376, which is to refactor test_runpy's walk_package support code into a common location.
Currently, the temp_cwd() context manager in test.support does two things: it creates a temp directory, and it changes the current working directory.
It would be useful and clearer if test.support exposed these two things as separate context managers. This will both let us simplify the implementation of temp_cwd() and give us an opportunity to increase code reuse between test.support and test.script_helper. (Currently, test.script_helper has its own version of temp_dir(), whose functionality overlaps with temp_cwd().)
I will upload a patch shortly with tests.

Attaching a patch with tests.

Fixing patch so it will apply cleanly.

Trying again to have the patch apply cleanly (for the review link).

cleaning up patch so that it will apply cleanly after applying patch from issue15494 

New changeset 4f0034477ba9 by Nick Coghlan in branch '3.3':
Close #15415: Factor out temp dir helpers to test.support
http://hg.python.org/cpython/rev/4f0034477ba9
New changeset b20a10c97e08 by Nick Coghlan in branch 'default':
Merge #15415 from 3.3
http://hg.python.org/cpython/rev/b20a10c97e08 

-def temp_cwd(name='tempcwd', quiet=False, path=None):
+def temp_cwd(name='tempcwd', quiet=False):
Since this parameter removal was pushed, the Windows buildbots consistently have this new error in this Windows-only test.
ERROR: test_empty (test.test_startfile.TestCase)
----------------------------------------------------------------------
Traceback (most recent call last):
 File "C:\buildbot.python.org3円.3.kloth-win64\build\lib\test\test_startfile.py", line 26, in test_empty
 with support.temp_cwd(path=path.dirname(sys.executable)):
 File "C:\buildbot.python.org3円.3.kloth-win64\build\lib\contextlib.py", line 115, in helper
 return _GeneratorContextManager(func, *args, **kwds)
 File "C:\buildbot.python.org3円.3.kloth-win64\build\lib\contextlib.py", line 37, in __init__
 self.gen = func(*args, **kwds)
TypeError: temp_cwd() got an unexpected keyword argument 'path'

New changeset a76693942de3 by Nick Coghlan in branch '3.3':
Attempt to fix #15415 on Windows
http://hg.python.org/cpython/rev/a76693942de3
New changeset b14308524cff by Nick Coghlan in branch 'default':
Merge #15415 Windows buildbot change from 3.3
http://hg.python.org/cpython/rev/b14308524cff 

I did see the Windows buildbot failure, but stopped scanning the log once I saw the intermittent test_signal failure :P
With the more granular API, test_startfile is a better fit for support.change_cwd, so we'll see how that goes on the Windows buildbots.

No dice: http://buildbot.python.org/all/builders/AMD64%20Windows7%20SP1%203.x/builds/2399/steps/test/logs/stdio
Could someone with a Windows build set up try this out?
I'd like to rule out a conflict with the -j setting, since being dependent on the current working directory pretty much rules out being run in parallel with other tests, but I didn't see anything in regrtest that let me say "run this one before fanning out into parallel execution".

New changeset 403b038ff075 by Nick Coghlan in branch '3.3':
Third attempt to fix #15415 on Windows
http://hg.python.org/cpython/rev/403b038ff075
New changeset fa13a159f063 by Nick Coghlan in branch 'default':
Merge #15415 from 3.3 (again)
http://hg.python.org/cpython/rev/fa13a159f063 

As of this moment, the test has run on one of the two Win7 machines and the xp buildbot. It also runs by itself on my win7 desktop with a fresh pull.

While there are some remaining buildbot failures, they don't appear to be related to this issue any more.