l...@gnu.org (Ludovic Courtès) writes:
> Mark H Weaver <m...@netris.org> skribis:
>
>> + ;; setting 'create-workers!' to a no-op is an optimization, but it is
>> + ;; still possible for '%create-workers!' to be called more than once
>> + ;; from different threads. Therefore, to avoid creating %workers more
>> + ;; than once (and thus creating too many threads), we check to make
>> + ;; sure %workers is empty within the critical section.
>
> Do you have a scenario where this happens?
I'd hoped that the comment above was clear, but let me try again.
If two threads simultaneously call 'make-future' before the worker pool
has been created, then both threads can call 'create-workers!' (before
it has been rebound), and from there both can enter '%create-workers!'.
At that point, unless one of the threads is killed, it is guaranteed
that both will enter the critical section: first one, and then the other
(after the first one has unlocked the mutex). Thus, two worker pools
will be created (though the first worker pool will become inaccessible).
Mark
