During a {Send|Recv}Transport instance destruction, its Handler also gets destroyed.
The Handler stores a PeerConnection and a map of pointers to RTP transceivers.
When the transport gets closed, the PC gets closed but the transceivers are leaked.
On the m112 branch the RTP transceiver map entries are now RTC smart pointers. This results in a fatal process hang during the destruction of the map and its entries, which occurs after the destruction of the pc instance variable.
RTPTransceiver destruction relies on waiting indefinitely for an event signaled by a thread which was owned by the now destroyed PC.
This fix simply clears the stored transceivers map before the PC is closed. This is sufficient (and necessary) for the m112 branch, it does not address the current leaky behaviour.
During a
{Send|Recv}Transportinstance destruction, itsHandleralso gets destroyed. The Handler stores aPeerConnectionand a map of pointers to RTP transceivers.When the transport gets closed, the PC gets closed but the transceivers are leaked.
On the
m112branch the RTP transceiver map entries are now RTC smart pointers. This results in a fatal process hang during the destruction of the map and its entries, which occurs after the destruction of thepcinstance variable. RTPTransceiver destruction relies on waiting indefinitely for an event signaled by a thread which was owned by the now destroyed PC.This fix simply clears the stored transceivers map before the PC is closed. This is sufficient (and necessary) for the
m112branch, it does not address the current leaky behaviour.