Benny Prijono wrote: > On 2/14/08, Michael Broughton <Michael_Broughton at advanis.ca> wrote: > >> Yes I am trying very hard to minimize my local changes. >> >> The idea is to have pjmedia_stream_destroy flips a flag somewhere to >> indicate that it is attempting to destroy a stream. In my local patch, >> the flag is a member of the stream structure. Then the DTMF callback can >> check this flag while it is trying to acquire the call lock. >> >> The benefit of this is that the DTMF callback can fail gracefully >> without waiting for a soft deadlock. The flag essentially gives it an >> early warning that a soft deadlock is going to occur. >> >> > > Wouldn't this still be prone to some race condition? The on_dtmf() > callback might get the variable before the destroy thread sets it. > > Well, it is looping while trying to acquire a lock... so if it has a reference to the flag, it can just check it on each iteration before the timeout is reached. >> Do you think it would be possible to do something like this without >> touching pjmedia? Could this be done in PJSUA? I have a feeling that >> this would be the only clean way to implement this. >> >> > > The only clean solution that I could think of is to post the destroy > request to main thread. But you've said that you didn't like this one, > so... > Destroy requests only happen from my main thread and possibly a SIP worker thread. Sorry if I was unclear about this. The problem is that in rare situations I receive DTMF events and try to drop a call at the same time. -- Michael Broughton, Advanis Unintended Recipient & Unauthorized Use of E-Mail: This message and attachments may contain confidential or privileged information that is intended only for the named recipient of this e-mail. Any unauthorized use or distribution is not permitted. If you have received this e-mail in error, deleting the e-mail and notifying the sender would be appreciated. Thank you.
