sdp-utils: add AV1 fmtp parsing

[tmatth]

This matches what is already done for VP9 and H264.

[lminiero]

Are browsers negotiating profiles for AV1 too, now? What are the options and what impact does it have?

[lminiero]

Found the new parameters documented here: https://aomediacodec.github.io/av1-rtp-spec/#72-sdp-parameters Is keeping track of the profile enough? In H.264 the profile contains info on more things, and in VP9 it's just a different flavour of the codec, while in AV1 apparently the H.264 profile equivalent is split across multiple properties (e.g., profile + level-idx).

[tmatth]

Is keeping track of the profile enough? In H.264 the profile contains info on more things, and in VP9 it's just a different flavour of the codec, while in AV1 apparently the H.264 profile equivalent is split across multiple properties (e.g., profile + level-idx).

Yeah right now I'm seeing level-idx, profile and tier, both e.g.,

a=fmtp:96 profile=0;level-idx=8;tier=0 a=fmtp:96 level-idx=8; profile=0; tier=0

I can add parsing for those as well.

[tmatth]

Also the code is simpler than I expected since the spec mentions that if the fields aren't present they must default to profile=0, level_idx=5 and tier=0 respectively.

[lminiero]

That helps with generating a valid fmtp, but I suspect would not be enough, especially since when generating answers, providing a profile (and maybe more) acts as a way to filter/find the specific payload type we want to access. We currently use the VP9/H.264 profile as a parameter to add to this request, for instance

int pt = janus_sdp_get_codec_pt_full(offer, codec, video_profile);

where we look for the payload type associated with that codec and that specific profile. Your patch adds AV1 as a potential profile for that filtering, but since the profile on its own may not be enough, this could result in the wrong payload type being selected if there's multiple AV1 fmtp lines with the same profile but different tiers/layers.

Not sure how this should be addressed: we could extend the signature of janus_sdp_get_codec_pt_full to have other optional properties that only make sense for AV1, but 1. that would be a breaking change (sdp-utiils are available to plugins too), 2. would cause a precedent (what if a new codec comes up with properties of its own? do we extend this forever?), and 3. would probably overcomplicate the filtering process (especially where some of the fmtp properties may be omitted in the offer).

I think that for now we can live with just using the profile for filtering. If it turns out it causes the problems that I mentioned (I haven't checked how many options browsers give for AV1) we can revisit this. I guess we'll also need to ensure this doesn't break AV1 for browsers that don't use profiles: e.g., what if Chrome adds one, but Firefox doesn't?

[fippo]

beware, Chromium is not behaving correctly either: https://issues.webrtc.org/issues/396434695

[JanFellner]

@fippo still beeing impress that you are able to follow stuff everywhere 🤣