Re: [Boost-users] [thread] Mysterious deadlock

From: David A. Greene (greened_at_[hidden])
Date: 2006-02-09 20:23:47

Delfin Rojas wrote:

> I don't understand very well why you have 2 mutexes inside Consumer. Anyway,

"mutex" is for access to the queue data structure.  "full_mutex" is for
signaling that the consumer is waiting on data.  Without "full_mutex"
the producer might send a notify that gets missed.  Or am I thinking
about this incorrectly?

> I left only one mutex inside Consumer which I called m_mutex and I renamed
> the condition from full_condition to m_condition and changed 
> The Consumer::push method to this:
> 
> void push(int val) {
> 	    LOG("push(): about to lock");
> 	    lock_type lock(m_mutex);
> 	    LOG("push(): locked");
> 	    queue.push(val);
> 	    m_condition.notify_all();
>       };
> 
> And all seems to work. I ran it for a little while and producer was
> generating data (ran it up to 33) and consumer was consuming data (about 17
> when producer reached 33).

Ok, I think you're right.  This should work.  As I noted in another
message, mocing the notify to the same scope as the lock protecting
the condition was the key.

> The boost::thread library includes several examples of this
> producer/consumer pattern. Look at monitor.cpp and starvephil.cpp in
> libs/thread/example

Boost.Thread has monitors?  Why isn't this documented?

                                    -Dave