- From: Harald Alvestrand <harald@alvestrand.no>
- Date: Tue, 05 Jun 2012 08:59:31 +0200
- To: public-webrtc@w3.org
- Message-ID: <4FCDAE53.7090005@alvestrand.no>
On 06/04/2012 11:23 PM, Justin Uberti wrote: > Cullen and I discussed this and agreed the onaddstream handling > mentioned here (and I presume, the contents of > PeerConnection.remoteStreams) is the right solution. > > However, this does bring up a couple other questions: > - If the caller advertises multiple media streams/tracks in its offer, > but the callee only accepts some of them (accepting audio but not > video by setting the m= port to zero), does a callback at the callee > side get fired (after setLocalDescription(answer)) indicating those > remoteStreams have now gone away? > - Same thing, but from the caller side - how does the caller find out > that some of its streams/tracks were rejected (we have no callback for > this, currently), and do those localStreams go away? My opinion: At the API level, we should keep the concept that the sender dictates the content of LocalStreams, and RemoteStreams is strictly a mirror of what's in LocalStreams. Similarly for Tracks. Thus, a track that's rejected at the SDP level should not disappear from the API; it should be present, but marked as muted at the sender's side. If SDP later renegotiates the session to allow the relevant media type, it can be unmuted. > > On Mon, Jun 4, 2012 at 10:16 AM, Harald Alvestrand > <harald@alvestrand.no <mailto:harald@alvestrand.no>> wrote: > > On 06/01/2012 09:48 PM, Eric Rescorla wrote: > > On Fri, Jun 1, 2012 at 11:53 AM, Justin > Uberti<juberti@google.com <mailto:juberti@google.com>> wrote: > > I don't know if this is even needed for this case. I would > think we would > handle this by calling setRemoteDescription("offer", sdp), > and then look at > the MediaStream (either from remoteStreams on onaddstream > to see if it has > audio or video tracks. Bob hasn't signaled anything back > to Alice at this > point, so this should be fine. > > This is what I had originally assumed one would do, and then > when Cullen > and I tried to do independent sequence diagrams, it turned out > that he > didn't agree, so I asked him if the editors could try to > propose something :) > > Cullen can provide his own arguments for why you don't want to use > onaddstream, but IIRC part of it was that the streams aren't > actually > readable at this point. > > My understanding of the movement of onaddstream from "media > received" to "media configured" is that it is intended to allow > you to ask exactly this question. > > In fact, Stefan posted a CONCLUSION message on November 30 saying: > > The conclusion of this discussion seems to be: > > 1. MediaStream objects should be created more or less > immediately as result of "getUserMedia" or "addStream" > operations (in the later case a MediaStream object on the > remote side should be created as a result of the signaling). > > 2. There is a need to, on a MediaStreamTrack level, notify the > application when live data is available (which can be a bit > later than the creation of the MediaStream object). > > The task is now on the Editors to change the drafts along > these lines. > > > It would seem natural that one can inspect the resulting > MediaStream to see which kinds of tracks it contains. > > (This is the last thread on the list archive that has > "onaddstream" in the subject.) > > Harald > > > > -Ekr > > > > >
Received on Tuesday, 5 June 2012 07:02:28 UTC