Re: [MMUSIC] JSEP Issue #394: What appears in m= lines.

[Eric Rescorla <[e-mail]>]

On Thu, Dec 22, 2016 at 7:15 AM, Magnus Westerlund <
[e-mail]> wrote:

> Hi,
>
>
>
> See inline.
>
>
>
> Den 2016-12-16 kl. 18:05, skrev Eric Rescorla:
>
> > https://github.com/rtcweb-wg/jsep/issues/394
>
> >
>
> > Magnus writes:
>
> >
>
> >     >    For media m= sections, JSEP endpoints MUST support both the
>
> >     "UDP/TLS/
>
> >     >    RTP/SAVPF" and "TCP/DTLS/RTP/SAVPF" profiles and MUST indicate
>
> >     one of
>
> >     >     these two profiles for each media m= line they produce in an
>
> >     offer.
>
> >
>
> >
>
> >
>
> >     Do I understand this correct, we are requiring support for the
>
> >     "TCP/DTLS/RTP/SAVPF" proto so that in cases an endpoint supports the
>
> >     optional to support ICE TCP, they can indicate it, and any WebRTC
>
> >     endpoint will accept it, even if that is just one option? I do know
>
> >     that different profiles are a negotiation issue. But, wouldn't it be
>
> >     more reasonable in this case to use UDP/TLS/RTP/SAVPF in all cases
>
> >     where one offer any candidates that also use UDP? And only use
>
> >     TCP/DTLS/RTP/SAVPF in cases only TCP candidates are signalled, thus
>
> >     not forcing TCP/DTLS/RTP/SAVPF onto clients that doesn't support it?"
>
> >
>
> >
>
> > We could certainly do this, but it seems to perpetuate the idea that the
>
> > TCP/UDP distinction is
>
> >
>
> > meaningful here, which seems like the opposite direction from the one we
>
> > are going in (see
>
> >
>
> > the mmusic minutes around the topic Non-Supported Transport in m- line).
>
> > See also the note below in this paragraph about either profile being
>
> > consistent with either transport.
>
> >
>
> >
>
> > I think it would be fine to relax the requirement that implementations
>
> > *support* both UDP and TCP, but requiring them to conditionally use one
>
> > or the other depending on whether they have UDP candidates seems like
>
> > it's really going to impose a lot of pain on implementors for no good
>
> > reason, in part because you don't necessarily know at the time of offer
>
> > generation what you will have. Consider the case where you are only
>
> > offering relayed and srflx candidates. At the time of generation, you
>
> > don't know if you will be able to get a UDP candidate, because you might
>
> > be behind a firewall that blocks UDP and your TURN server might be down.
>
> >
>
>
>
> Okay, I agree that having any rules for what you should offer here based
> on what the actual outcome is not the best idea here. I think the rules
> should be based on capability and intent. So if you only are going offer
> TCP candidates, then you clearly should use TCP/…,
>

I'm going to push on this. If we've already agreed that mismatches are
normal, why should we do that? Wouldn't it be better to just effectively
deprecate this field?

-Ekr

If you have no implementation for TCP candidates then you clearly offers
> UDP/…. If the implementations intention is to offer both if they are
> determined, then one has a choice. In the context of WebRTC to WebRTC this
> does would not matter, as long as the counter part has the rule to accept
> either. However if you support both, if one you uses UDP, then it also
> doesn’t matter for implementations supporting only UDP, they will accept
> it. If the incoming offer is TCP/… then with the clarifications you propose
> below a non TCP candidate supporting implementations answering can’t fail
> immediately. It will have to wait to see if there if there ever shows up
> any UDP candidates.
>
>
>
> Gateways to legacy devices will independently have issues, as what they
> need to use will depend on the far sides capabilities and the actual set of
> candidates.
>
>
>
> >
>
> > If we make any change, I think it should just be to relax the
>
> > requirement to support TCP and then tell people to ignore the first
>
> > field here in favor of the candidate lines.
>
> >
>
>
>
> So, if I interpret the above, I would say: Set the PROTO to UDP/… and on
> reception ignore the TCP/UDP part of the PROTO string, only looking at
> candidates. That is very close to my amended suggestion. The only addition
> I have, is that if an implementation will not include any UDP candidates,
> only TCP one, then it shall use TCP.
>
>
>
>
>
> We should have defined ICE/DTLS/RTP/SAVPF to avoid this issue.
>
>
>
>
>
> Cheers
>
>
>
> Magnus Westerlund
>
>
>
> ----------------------------------------------------------------------
>
> Services, Media and Network features, Ericsson Research EAB/TXM
>
> ----------------------------------------------------------------------
>
> Ericsson AB                 | Phone  [phone]
> <+46%2010%20714%2082%2087>
>
> Färögatan 6                 | Mobile [phone]
> <+46%2073%20094%2090%2079>
>
> SE-164 80 Stockholm, Sweden | mailto: [e-mail]
>
> ----------------------------------------------------------------------
>