Change onssrcconflict event in RTCRtpSender by exception

[murillo128]

Currently we have an onssrcconflict in the RTCRtpSender, but I fail to see which situations would cause it to be triggered.

AFAIK the only way of having an ssrc conflict is by assigning the same ssrc to two different RTPSender on same transport, or having two different encodings with same ssrc. Or when changing a sender from one transport to another which already had another RTPSender using that same ssrc.

This can only be triggered by an API method call (either RTCRtpSender.send()or RTCRtpSender.setTransport()) so IMHO it is much better to throw an exception in this case (an InvalidParameter exception for send() and maybe a new SSRConflict exception for setTranspor()), that triggering an event.

[murillo128]

This event must be fired if an SSRC conflict is detected. On an SSRC conflict, the RTCRtpSender automatically sends an RTCP BYE on the conflicted SSRC.

More over, if we move to exceptions, this would cause the method call to fail, so we can rollback changes and leave the RTPSender on same state as before the method call, avoiding to call BYE and terminate the stream.

[robin-raymond]

I don't know any implementation (yet) that detects and throws the exception. We do need to be clear when this can happen. I don't want to have specialized exceptions needlessly or different exceptions for the same event needlessly.

This is a very problematic thing for the sender because of multiple reasons:

1. At time of send(..) it may be possible to detect a conflict of ssrc exists with other senders if both senders are attached to the same transport (and thus fire an invalid params BUT)
2. Firing an invalid exception on send based on ssrc conflict might require asynchronous processing to determine since knowing this must reach deep into sender transport objects
3. If a programmer changes the sender's transport to attach to a transport that already has the ssrc then the conflict occurs at that time of setTransport(...) and not at time of calling send(...)

So... we do have some clarifying to do on how / when this is supposed to occur...

[ibc]

If we are in the sender, SSRC conflicts can just happen due to user actions. If they cannot be detected synchronously then having send() return a Promise would be a much better approach than firing a context-less event.

Anyhow:

[robin-raymond]

LOL

then having send() return a Promise would be a much better approach than firing a context-less event.

Except a Promise is not sufficient... This can also happen later after send(...) is called by attaching a sender to a transport after send is called that already has another sender with the same ssrc.

[ibc]

This can also happen later after send(...) is called by attaching a sender to a transport after send is called that already has another sender with the same ssrc.

rtpSender.setTransport() may also return a Promise.

[robin-raymond]

@ibc I would prefer to just event ssrc conflicts; it's such a rare use case and I seriously doubt many are going to implement or concern themselves with it.

[ibc]

If somebody gets a sending SSRC conflict he deserves ugly things to happen.
Yep, there are is more important stuff to improve.