Re: [Jack-Devel] Jack thread cancellation

Le 6 mai 2013 à 11:24, "MONTANARO Luciano (MM)" <[hidden]> a écrit :

> Da: Paul Davis [mailto:[hidden]]
> Inviato: venerdì 3 maggio 2013 18:04
> A: Stéphane Letz
> Cc: MONTANARO Luciano (MM); [hidden]
> Oggetto: Re: [Jack-Devel] Jack thread cancellation
> 
>> Quitting from inside a thread is one thing.
>> Cancelling a thread from outside using pthread mechanisms is unreliable. It may never hit a cancellation point.
> 
> Well, that is why I suppose the ASYNCHRONOUS flag was used. For the processing threads, the chance is it will never call a cancellation point.
> However, both the MessageBuffer thread and the SocketClient threads will happyly log messages, and if a cancellation happens at that point, a crash is the likely result.
> 

Have you tried to start JackMessageBuffer thread in PTHREAD_CANCEL_DEFERRED mode? Like with the following code in JackMessageBuffer.cpp:

JackMessageBuffer::JackMessageBuffer()
   :fInit(NULL),
   fInitArg(NULL),
   fThread(this, PTHREAD_CANCEL_DEFERRED),
   fGuard(),
   fInBuffer(0),
   fOutBuffer(0),
   fOverruns(0),
   fRunning(false)
{}

> When we noticed the problem in JackMessageBuffer, we tried to wrap the body of the Execute()  in pthread_setcancelstate(PTHREAD_CANCEL_DISABLE/ENABLE) (Actually, an RAII pattern), but the same problem is now happening elsewhere.

Where?

> 
>> Killing a thread is different, though somewhat platform dependent.
> 
> Well, thanks for the clarification.
> 
> Luciano
> 

I need to understand better the reason why your application see those problems in the first place. IRC Freenode #jack would be easier.

Stéphane 

1367832894.30383_0.ltw:2,a <36E2B99A-7948-413D-A954-6452A5B62AEE at grame dot fr>