Re: "a=extmap" with BUNDLE

2017-03-28 0:24 GMT+02:00 Taylor Brandstetter <[email protected]>:
>> It's just about getting the appropriate RtpReceiver
>> associated to the current RTP packet (this can be done by MID, RID,
>> SSRC or even PT, as the ORTC and JSEP "RTP Matching Rules" define).
>
>
> Right. So, if you can't get the appropriate RtpReceiver with RID, SSRC or
> PT, then you must do it with MID. And if the MID extension shares an ID with
> another extension, it becomes impossible to know the MID. For example:
>
> m=audio ...
> a=extmap:1 urn:ietf:params:rtp-hdrext:sdes:mid
> a=extmap:2 urn:foo
> m=video ...
> a=extmap:1 urn:bar
> a=extmap:2 urn:ietf:params:rtp-hdrext:sdes:mid
>
> I know it would be crazy to generate SDP like that, but there's nothing I've
> found that disallows it.

And you are 100% right here. MID ID must be unique across all the m=
lines, otherwise it is useless.

To make things even uglier, the RFC 5285 states:

   The mapping may be provided per media stream (in the media-level
   section(s) of SDP, i.e., after an "m=" line) or globally for all
   streams (i.e., before the first "m=" line, at session level).  The
   definitions MUST be either all session level or all media level; it
   is not permitted to mix the two styles.  In addition, as noted above,
   the IDs used MUST be unique for each stream type for a given media,
   or for the session for session-level declarations.

So a solution may be using just global a=extmap lines at SDP session
level. That would work, but it's a hack, because not all the IDs are
to be used within all the m= lines, not even across all the m= lines
of the same kind (audio, video).

Anyhow, the key of your argument is the "just MID for m=line
identification". It clearly requires the MID ID to be unique across
all the m= lines. One "solution" would be for the BUNDLE spec [*] to
mandate having a single MID ID, but it does not. I still prefer to
mandate that the same ID always references the same extension URI, and
that the same extension URI is just referenced by a single and unique
ID in the entire SDP, but that "breaks" RFC 5285...

So IMHO, the immediate issue is in BUNDLE spec [*] which defines the
MID stuff but does not mandate the MID ID to be unique in the whole
SDP.


[*] https://tools.ietf.org/html/draft-ietf-mmusic-sdp-bundle-negotiation-36

-- 
Iñaki Baz Castillo
<[email protected]>

_______________________________________________
mmusic mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/mmusic