Re: [MMUSIC] IANA registration of SDP attributes

[Paul Kyzivat <[e-mail]>]

On 3/19/16 10:27 PM, Christian Groves wrote:
> With the current registry don't only documents that introduce NEW
> attributes get included in the registry?
>
> dcsa (MSRP) and dcsa (BFCP) don't define new attributes they use setup.
> This is similar to the fact that multiple protocols at the media level
> use a=setup but we don't add references to them in the registry.
>
> So do we now say that if a draft/RFC uses an existing media level
> attribute in a DCSA that must be added to the registry with a dcsa
> indication?

ISTM that a document that broadens the applicability of an attribute 
ought to be recorded in the registry.

I expect that this is a bit fuzzy. For instance, the use of setup with 
TCP was defined. If a new proto of 'TCP/FOO' is defined that runs over 
TCP, and simply uses setup for establishing the TCP part, then maybe it 
doesn't need to be recorded in the registry.

But if setup is used for something other than TCP, or also used for some 
semantic over and above its use for TCP, then it surely ought to be 
recorded. (For instance, when it is used to control initialization of 
some other protocol over TCP.)

I expect that this is less than clear, and may be controversial. Seems 
to need more discussion.

	Thanks,
	Paul

> Regards, Christian
>
> On 12/03/2016 3:46 AM, Paul Kyzivat wrote:
>> On 3/11/16 11:15 AM, Juergen Stoetzer-Bradler wrote:
>>> Paul,
>>>
>>> The last alternative would have the advantage that different subprotocol
>>> documents could be referenced for the same attribute. Like e.g. for the
>>> setup attribute (if there were BFCP over data channel transport specific
>>> aspects):
>>>
>>> *SDP Name*     *Level(s)*     *Reference(s)*
>>> accept-types
>>>     media,
>>> dcsa(MSRP)
>>>     [RFC4975]
>>> [draft-ietf-mmusic-msrp-usage-data-channel]
>>> cat
>>>     session
>>>     [RFC4566]
>>> fmtp
>>>     media,source     [RFC4566][RFC5576]
>>> mediaclk
>>>     session,media,source
>>>     [RFC7273]
>>> ptime
>>>     media
>>>     [RFC4566]
>>> recvonly
>>>     session,media,
>>> dcsa(MSRP)
>>>     [RFC4566][RFC4975]
>>> [draft-ietf-mmusic-msrp-usage-data-channel]
>>> setup
>>>     session, media
>>> dcsa(MSRP)
>>> dcsa(BFCP)
>>>     [RFC4145]
>>> [draft-ietf-mmusic-msrp-usage-data-channel]
>>> [draft-schwarz-mmusic-bfcp-usage-data-channel]
>>>
>>>
>>> Therefore I'd be in favor of your last alternative.
>>
>> Let's see what other comments we get, especially from Flemming.
>>
>> Then, if this is preferred direction we can work on refining it.
>>
>>     Thanks,
>>     Paul
>>
>>> Thanks,
>>> Juergen
>>>
>>>
>>> On 10.03.2016 19:48, EXT Paul Kyzivat wrote:
>>>> [splitting off from the thread on data-channel-sdpneg]
>>>>
>>>> Currently IANA has five(!) separate registries for sdp attributes:
>>>>
>>>> att-field (session level)
>>>> att-field (both session and media level)
>>>> att-field (media level only)
>>>> att-field (source level)
>>>> att-field (unknown level)
>>>>
>>>> They all have the same format:
>>>>
>>>> *Type**
>>>> *     *SDP Name**
>>>> *     *Reference**
>>>> *
>>>> att-field (session level)     cat     [RFC4566]
>>>> att-field (both session and media level)     recvonly
>>>>     [RFC4566]
>>>> att-field (both session and media level)     mediaclk
>>>>     [RFC7273]
>>>> att-field (media level only)     accept-types
>>>>     [RFC4975]
>>>> att-field (media level only)     fmtp
>>>>     [RFC4566]
>>>> att-field (source level)     fmtp
>>>>     [RFC5576]
>>>>
>>>>
>>>> This format is a pain, because it is hard to look an attribute up if
>>>> you don't know at what level(s) it is valid. It also has the potential
>>>> to allow an attribute name to be registered for unrelated purposes if
>>>> the type is different. (IMO that would be bad.)
>>>>
>>>> A long time ago (several years now), as part of the 4566bis work, I
>>>> proposed that these tables be merged into one. It was my impression
>>>> that this was agreed and would be done. But I don't recall any
>>>> agreement on the logistics of doing so.
>>>>
>>>> My thought was that the combined table would look like:
>>>>
>>>> *SDP Name*     *Level(s)*     *Reference(s)*
>>>> accept-types
>>>>     media
>>>>     [RFC4975]
>>>> cat
>>>>     session
>>>>     [RFC4566]
>>>> fmtp
>>>>     media,source     [RFC4566][RFC5576]
>>>> mediaclk
>>>>     session,media,source
>>>>     [RFC7273]
>>>> ptime
>>>>     media
>>>>     [RFC4566]
>>>> recvonly
>>>>     session,media
>>>>     [RFC4566]
>>>>
>>>>
>>>>
>>>> Then we get to data channel attributes. My thought is to incorporate
>>>> them into this table structure, as yet another "level". E.g.,
>>>>
>>>> *SDP Name*     *Level(s)*     *Reference(s)*
>>>> accept-types
>>>>     media,dcsa
>>>>     [RFC4975][draft-ietf-mmusic-msrp-usage-data-channel]
>>>> cat
>>>>     session
>>>>     [RFC4566]
>>>> fmtp
>>>>     media,source     [RFC4566][RFC5576]
>>>> mediaclk
>>>>     session,media,source
>>>>     [RFC7273]
>>>> ptime
>>>>     media
>>>>     [RFC4566]
>>>> recvonly
>>>>     session,media,dcsa
>>>>     [RFC4566][RFC4975][draft-ietf-mmusic-msrp-usage-data-channel]
>>>>
>>>>
>>>> (And this could also be extended for websockets if somebody proposes a
>>>> way to negotiate attributes for data channels too.)
>>>>
>>>> Using this format, if you want to know more than the name and the
>>>> level(s) at which it can be used you need to consult the references.
>>>> And when there are multiple references you don't know which one(s) you
>>>> need to consult. This can be "fixed" by including more information
>>>> from the reference into the registry. Conversely, we could strip it
>>>> down further and remove the levels from the registry - so you need to
>>>> consult the references for that too.
>>>>
>>>> For instance, if we wanted to simplify finding the right reference for
>>>> the level you are interested in, we could do:
>>>>
>>>> *SDP Name*     *Level(s)*     *Reference(s)*
>>>> accept-types
>>>>     media,
>>>> dcsa
>>>>     [RFC4975]
>>>> [draft-ietf-mmusic-msrp-usage-data-channel]
>>>> cat
>>>>     session
>>>>     [RFC4566]
>>>> fmtp
>>>>     media,source     [RFC4566][RFC5576]
>>>> mediaclk
>>>>     session,media,source
>>>>     [RFC7273]
>>>> ptime
>>>>     media
>>>>     [RFC4566]
>>>> recvonly
>>>>     session,media,
>>>> dcsa
>>>>     [RFC4566][RFC4975]
>>>> [draft-ietf-mmusic-msrp-usage-data-channel]
>>>>
>>>>
>>>> Or we could go further, and break the dcsa level down by subprotocol:
>>>>
>>>> *SDP Name*     *Level(s)*     *Reference(s)*
>>>> accept-types
>>>>     media,
>>>> dcsa(MSRP)
>>>>     [RFC4975]
>>>> [draft-ietf-mmusic-msrp-usage-data-channel]
>>>> cat
>>>>     session
>>>>     [RFC4566]
>>>> fmtp
>>>>     media,source     [RFC4566][RFC5576]
>>>> mediaclk
>>>>     session,media,source
>>>>     [RFC7273]
>>>> ptime
>>>>     media
>>>>     [RFC4566]
>>>> recvonly
>>>>     session,media,
>>>> dcsa(MSRP)
>>>>     [RFC4566][RFC4975]
>>>> [draft-ietf-mmusic-msrp-usage-data-channel]
>>>>
>>>>
>>>> Thoughts?
>>>>
>>>>     Thanks,
>>>>     Paul
>>>>
>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> mmusic mailing list
>>>> [e-mail]
>>>> https://www.ietf.org/mailman/listinfo/mmusic
>>>
>>>
>>>
>>> _______________________________________________
>>> mmusic mailing list
>>> [e-mail]
>>> https://www.ietf.org/mailman/listinfo/mmusic
>>>
>>
>> _______________________________________________
>> mmusic mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/mmusic
>>
>
> _______________________________________________
> mmusic mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/mmusic
>