> I get the impression that there is a flaw in the shutdown sequence of
> JACK. Whenever shutting down jackd before an active client, both would
> hang, seemingly deadlocked. So I did a little debugging, and it seems
> to me the heart of the problem is that jackd writes an event to the
> client and expects a reply, _but_ the client process thread (which is
> responsible for polling event_fd am I right?) has already exited in
> response to the shutdown of jackd. So there will be noone to
> acknowledge the event, and the main jackd thread will be stuck
> listening on event_fd, while the server thread will hang waiting for
> a graph lock (in response to a request from the client). Does this
> analysis sound sensible (or is the root of my problem somewhere else)?
Did you ever get an answer to this question? It sounds like a rather
deep problem. Can you still reproduce it? How complicated is the
test sequence? Is there a problem report in Mantis for it?