Message6124
Ahh, this looks very interesting in socket.py, because it's not going to be threadsafe. We need to either synchronize the ref counting, or probably better yet just use an AtomicInteger. This definitely could explain the slow leak. I would work on it now, but I'm about to go to the zoo, so posting it here for now:
def close(self):
try:
if self._sock:
self.flush()
finally:
if self._sock:
if isinstance(self._sock, _nonblocking_api_mixin):
self._sock.reference_count -= 1
if not self._sock.reference_count or self._close:
self._sock.close()
elif self._close:
self._sock.close()
self._sock = None |
|
Date |
User |
Action |
Args |
2010-10-03 16:41:28 | zyasoft | set | messageid: <1286124088.13.0.369999475321.issue1660@psf.upfronthosting.co.za> |
2010-10-03 16:41:28 | zyasoft | set | recipients:
+ zyasoft, amak, gdoutch |
2010-10-03 16:41:28 | zyasoft | link | issue1660 messages |
2010-10-03 16:41:27 | zyasoft | create | |
|