If I run something like:
  ./python -m test.regrtest -R3:2: test_threading_local

Python sometimes hangs and I have to "kill -9" it.

Running through gdb shows it gets stuck at the following point:

/home/antoine/cpython/debug/Lib/threading.py (239): wait
/home/antoine/cpython/debug/Lib/threading.py (638): join
/home/antoine/cpython/debug/Lib/test/test_threading_local.py (68):
test_derived
[snip]

That is, it hangs waiting for a thread to join().

It turns out that the __del__ method in _threading_local.py tries to
call threading.enumerate() which itself takes the _active_limbo_lock.
The problem is that __del__ can be called at any point in time (because
of the GC), including at a point where the same thread has already taken
the lock. The obvious fix is to bypass enumerate().

(an alternate fix would be to use an RLock for _active_limbo_lock, but
it could have unforeseen consequences, such as performance ones)

How about defining this in threading.py:

def _enumerate():
    """Internal use only: enumerate() without the lock."""
    return _active.values() + _limbo.values()

And calling it from _threading_local instead of accessing _active and
_limbo directly.

Good point. Here is a new patch.

Looks good to me.

typed poorly on a tiny Android phone.

On Nov 4, 2009 3:24 PM, "Antoine Pitrou" <report@bugs.python.org> wrote:

Antoine Pitrou <pitrou@free.fr> added the comment:

Good point. Here is a new patch.

----------
Added file: http://bugs.python.org/file15264/threading_local2.patch

_______________________________________ Python tracker <
report@bugs.python.org> <http://bugs.python...

Committed, thanks.