[h323plus] PTimedMutex::Wait patch

[Robert Jongbloed]

I have applied this patch to the PTLib trunk. Thank you very much!


Robert Jongbloed
OPAL/OpenH323/PTLib Architect and Co-founder.


> -----Original Message-----
> From: h323plus-bounces at lists.packetizer.com [mailto:h323plus-
> bounces at lists.packetizer.com] On Behalf Of Nir Soffer
> Sent: Sunday, 21 September 2008 5:26 AM
> To: h323plus
> Subject: [h323plus] PTimedMutex::Wait patch
> 
> In PTimedMutex::Wait(const PTimeInterval & waitTime), when
> pthread_mutex_timedlock is not available or broken (it is broken on
> our arm platform), the code always sleeps before checking the finish
> time, which is incorrect.
> 
> For example, if you call Wait(0), and fail to acquire the lock, you
> will sleep for 10 milliseconds before you get back the retrun value.
> 
> This is a bigger issue if you use
> H323EndPoint::FindConnectionWithLock:
> 
>    PWaitAndSignal mutex(connectionsMutex);
> 
>    H323Connection * connection;
>    while ((connection = FindConnectionWithoutLocks(token)) != NULL) {
> 
>      // XXX this will block for 10 milliseconds if the lock can not
> be acquired!
>      // TryLock calls Wait(0) which sleeps for 10ms on failure
>      switch (connection->TryLock()) {
> 
>        case 0 :
>          return NULL;
>        case 1 :
>          return connection;
>      }
>      // Could not get connection lock, unlock the endpoint lists so a
> thread
>      // that has the connection lock gets a chance at the endpoint
> lists.
>      connectionsMutex.Signal();
>      PThread::Sleep(20);
>      connectionsMutex.Wait();
>    }
> 
>    return NULL;
> 
> The code try sleep for 20 milliseconds while the connectionsMutex is
> unlocked, letting the other thread that hold the connection a chancee
> to use the endpoint lists, but for each 30 milliseconds cycle, it
> keeps the endpoint lists locked for 10 milliseconds.
> 
> Attached a trivial fix.