I'm running 2.6.6rc1 64 bit on Windows 7

I get the following error when I try to instantiate a Pool:

>>> import multiprocessing
>>> p = Pool(processes = 2)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "c:\Python26\lib\multiprocessing\__init__.py", line 227, in Pool
    return Pool(processes, initializer, initargs)
  File "c:\Python26\lib\multiprocessing\pool.py", line 84, in __init__
    self._setup_queues()
  File "c:\Python26\lib\multiprocessing\pool.py", line 130, in _setup_queues
    from .queues import SimpleQueue
  File "c:\Python26\lib\multiprocessing\queues.py", line 22, in <module>
    from multiprocessing.synchronize import Lock, BoundedSemaphore, Semaphore, Condition
  File "c:\Python26\lib\multiprocessing\synchronize.py", line 22, in <module>
    from multiprocessing.forking import assert_spawning, Popen
  File "c:\Python26\lib\multiprocessing\forking.py", line 153, in <module>
    from ._multiprocessing import win32, Connection, PipeConnection
ImportError: No module named _multiprocessing
>>>

prior to this I already imported Pool (from multiprocessing import Pool)

I also tried the 64 bit 2.6.5 and got the same error.

I just fixed a similar problem for bpo-9513 but the problem was only ever seen when test_multiprocessing was run through regrtest on 3.x on Windows. Removing the relative import dot on _multiprocessing imports let them succeed.

I've never actually seen this cause a problem on an installed version, though. I'll check out 2.6.6 rc1 tomorrow when I get back to a Windows box to see if I can reproduce this.

Reclassifying from crash to behavior - it's just an ImportError, not actually a crash

I should correct myself. As seen on the 2.6 Win7 buildbot[0], test_multiprocessing is skipped there likely due to the "from ._multiprocessing..." ImportError. I kind of remember this now but didn't want to backport the fix to release26-maint since it was frozen, and this didn't appear to be an actual bug in the installed version. As said before, I'll investigate tomorrow.

[0] http://www.python.org/dev/buildbot/builders/x86%20Windows7%202.6/builds/547/steps/test/logs/stdio

Ok, it turns out this is in fact a regression from 2.6.5. My prior investigation for that 3.x issue must not have been on the 2.6 version I thought it was.

Barry: the fix from bpo-9513 (e.g., r83722) will correct this.

Please upload a patch specifically for 2.6.6 finally so that I can review it (and test it locally). It *seems* safe enough so if you and I both have a high confidence it won't regress anything, and it can go in before Monday, then I might approve it.

I still can't build on Windows 7 w/VS2010 though unfortunately, but I will locally test Ubuntu (Lucid, Maverick), Debian, and OS X 10.6.

Here's the patch. All four changes are one char each and live within win32-specific blocks.

Looks okay to me on *nix. You can check this in as long as test -u all passes on Windows. Please include a NEWS file entry and close this issue when it's committed. Thanks!

Committed in r84031.

Thanks for reporting this, cipater.

Notice that this is also a duplicate of bpo-9144, which was fixed for 2.7, but nobody bothered to move the fix to the other branches. This goes all back to bpo-7902, which has now introduced an incompatible change into 2.6 (otherwise, changing multiprocessing shouldn't have been necessary, despite that having been wrong all the time). I'll reopen that as release-critical for review.