Re: [rtcweb] Proposal to break the ICE impasse

[Eric Rescorla <[e-mail]>]

On Mon, Jan 28, 2019 at 5:10 PM Ted Hardie <[e-mail]> wrote:

> In-line.
>
> On Tue, Jan 29, 2019 at 9:56 AM Eric Rescorla <[e-mail]> wrote:
>
>>
>>
>> On Mon, Jan 28, 2019 at 4:47 PM Ted Hardie <[e-mail]> wrote:
>>
>>> On Tue, Jan 29, 2019 at 9:36 AM Eric Rescorla <[e-mail]> wrote:
>>>
>>>> If we assume that this is a real problem as opposed to a specification
>>>> problem, then I agree this is reasonable. However, so far nobody has shown
>>>> me that this is a real problem, and until that this happens, I'm not in
>>>> favor.
>>>>
>>>>
>>> As a chair, I wish to confirm my understanding:  you have no technical
>>> objection to this solution, but you do not wish to include this because you
>>> have seen no evidence of the problem.  Is that correct?
>>>
>>
>> No. I object to it because it's unnecessary complexity and moves against
>> the consensus direction that we had that this field was meaningless.
>>
>> That consensus was limited to JSEP;
>

Yes, and we're talking about the JSEP spec here.


other related uses of SIP/SDP do not treat it as meaningless.
>

That seems to be assuming the answer to my question.


The additional complexity added to a system is, of course, always a
> concern.  But  I interpret your statement "If it's necessary, then I'm
> willing to live with it." as agreeing that this solution will work if the
> additional domain of interoperability which results is not empty.
>

No, that's not what I said. What I said was "A system with substantial
deployment that works properly with WebRTC if the default candidates are
UDP but not when the default candidates are TCP and the proto is UDP/foo"

-Ekr


> Ted
>
>
>
>
>>
>>
>>> On Mon, Jan 28, 2019 at 2:51 PM Justin Uberti <[e-mail]>
>>> wrote:
>>>
>>>> Posted https://github.com/rtcweb-wg/jsep/pull/863 as a stab at a
>>>> consensus. This basically says that the offerer fills in either UDP/TLS/foo
>>>> or TCP/DTLS/foo based on the current default or selected candidate, in
>>>> accordance with sip-sdp. As Adam mentioned before, this shouldn't have any
>>>> impact on JSEP functionality.
>>>>
>>>> If this looks good, I'll polish it up.
>>>>
>>>> On Sun, Jan 27, 2019 at 3:22 AM Christer Holmberg <
>>>> [e-mail]> wrote:
>>>>
>>>>> Hi,
>>>>>
>>>>>
>>>>> > I'm not yet persuaded this is needed. The alleged need here is that
>>>>> there are some ICE-implementing endpoints which will choke if
>>>>> > they see a profile that doesn't match any actual candidate. I
>>>>> recognize that this is required by 5245, but that doesn't mean anyone
>>>>> > ever did it. Can you please point me to a client which would
>>>>> interoperate with a WebRTC endpoint with this change that would not
>>>>> > if you just always sent the same profile as JSEP currently requires.
>>>>>
>>>>> I don't think it is ok to *specify* that discarding a MUST is ok as
>>>>> long as nobody can show an implementation that would break by doing so.
>>>>>
>>>>> Having said that, in order to prevent an RTCWEB shutdown I am
>>>>> generally ok with Adam's approach. I would like my pull request comments to
>>>>> be addressed, though, that is related to separation between the JSEP API
>>>>> and an application using it: an application should be allowed to act
>>>>> according to 5245/draft-ice-sdp and update the c/m line if it wishes to,
>>>>> but due to the way the JSEP API works such applications might sometimes
>>>>> always include the same value in the c/m- line.
>>>>>
>>>>> I also think it shall be explicitly written that JSEP does not update
>>>>> 5245/draft-ice-sdp, but rather deviates when it comes to the c/m- line.
>>>>>
>>>>> Regards,
>>>>>
>>>>> Christer
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> On Thu, Jan 24, 2019 at 11:12 AM Adam Roach <[e-mail]> wrote:
>>>>>
>>>>> Based on conversations in MMUSIC, as well as several offline
>>>>> conversations with interested parties, I've put together a proposed
>>>>> change to JSEP that, if accepted, will allow publication of the
>>>>> Cluster
>>>>> 238 documents to move forward.
>>>>>
>>>>> Note that this new text has no impact on existing implementations (at
>>>>> least, as far as I am able to discern), which do not currently have
>>>>> the
>>>>> capability of producing media sections consisting of exclusively TCP
>>>>> candidates. From that perspective, the change makes existing
>>>>> implementations no less compliant with JSEP than they were before.
>>>>>
>>>>> What this change does provide is both on-paper and in-the-future
>>>>> compatibility between WebRTC implementations once they finalize TCP
>>>>> candidate handling (and candidate handling in general for mid-session
>>>>> offers).
>>>>>
>>>>>    https://github.com/rtcweb-wg/jsep/pull/862/files
>>>>>
>>>>> The key insight here is that JSEP's use of ICE completely discards any
>>>>> meaning associated with the transport parameter, while SIP's use of
>>>>> ICE
>>>>> does not. The trivial change that I propose, which bears only on
>>>>> future
>>>>> WebRTC implementations -- that is, which has no as-built specification
>>>>> to point to -- allows JSEP to continue to ignore the value of the
>>>>> transport parameter, while specifying that it says the right thing for
>>>>> SIP implementations to function properly.
>>>>>
>>>>> /a
>>>>>
>>>>> _______________________________________________
>>>>> rtcweb mailing list
>>>>> [e-mail]
>>>>> https://www.ietf.org/mailman/listinfo/rtcweb
>>>>>
>>>>> _______________________________________________
>>>>> rtcweb mailing list
>>>>> [e-mail]
>>>>> https://www.ietf.org/mailman/listinfo/rtcweb
>>>>>
>>>> _______________________________________________
>>> rtcweb mailing list
>>> [e-mail]
>>> https://www.ietf.org/mailman/listinfo/rtcweb
>>>
>>