Re: [MMUSIC] [Technical Errata Reported] RFC8864 (7805)

[Christer Holmberg <[e-mail]>]

Hi,

>>>> I am not sure what you mean be "pre-agreed data channels", but I was
>>>> referring to applications that use both DCEP and SDP.
>>>
>>> I was specifically referring to datachannels created with the "negotiated"
>>> flag set to "true".
>>
>> Gotcha.
>>
>> So, you are saying there ARE applications that use both DCEP and SDP
>> ("negotiated" flag set to "true") to negotiate data channels?
>
> Yes. We've seen it used.

Ok, that is useful input.

Do you know WHY those applications use both mechanisms?

>>>> WebRTC, as currently specified, will ignore all attempts to negotiate
>>>> datachannels using SDP, so those channels can only be created by the
>>>> application parsing the SDP, extracting the information, and calling
>>>> "createDataChannel" with "negotiated" set to true.
>>>
>>> Correct.
>>>
>>>> Also, your claim that the DCEP odd/even restriction would not apply
>>>> for certain cases is (AFAIK) not defined anywhere. Also, I don't
>>>> understand why the odd/even restriction would not apply in the case
>>>> where you have pre- agreed data channels.
>>>
>>> The WebRTC spec, at <removed link> does not
>>> mention the odd/even rule. For good reason; both ends have to call
>>> createDataChannel with the same value for "id", so it *has* to "violate"
>>> the odd/even rule.
>>
>> Both endpoints obviously have to use the same "id", but one endpoint
>> has to decide which "id" to use, and when using SDP the idea was to
>> re-use the DCEP odd/even rule.
>
> And that re-use is exactly the thing that we can't do.

I agree.

>> I DO agree that the odd/even rule does not work in SDP actpass cases
>> (where the id has to be set before the DTLS roles have been
>> determined), and should be corrected, but I am not sure whether the
>> correction suggested in the errata is the right way.
>
> I'm happy to see alternate text suggested. My suggestion was only in order 
> to
> have something specific to suggest in order to get rid of the inappropriate
> MUST.

The problem with simply removing the MUST is that it will introduce risk for 
id collision - especially if there are applications that use both DCEP and 
SDP.

We can either try to solve that some other way, or simply document the 
collision risk, but in either case we should document it.

Regards,

Christer