Issues when Condition.notify() is interrupted

[serhiy-storchaka]

If threading.Condition.notify() is interrupted just after it releases the waiter lock, but before remove it from the queue, the following calls of notify() will fail with RuntimeError: cannot release un-acquired lock. It can block the waiter threads if they do not use timeout.

cpython/Lib/threading.py

Lines 375 to 380 in 5bc2390

	for waiter in waiters_to_notify:
	waiter.release()
	try:
	all_waiters.remove(waiter)
	except ValueError:
	pass

The simplest solution would be to silence a RuntimeError in waiter.release(). There may be similar issues in other parts of the code.

[serhiy-storchaka]

There was a similar issue in #89437, although it is difficult to fix it correctly in that case. @vstinner

[bensimner]

In #90882 the discussion led to a solution of adding threading.Thread.acquire_and_release method which would solve #89437, but I don't think a PR for that was ever created. This solution probably doesn't help here.

Simply ignoring the RuntimeError here would mean that in those cases notify(n) would wake up fewer than n threads even if n threads were available to wake. Possibly that's not an issue since condition vars don't make guarantees about how many threads will be awoken.