Re: [MMUSIC] Benjamin Kaduk's No Objection on draft-ietf-mmusic-data-channel-sdpneg-25: (with COMMENT)

[Benjamin Kaduk <[e-mail]>]

On Tue, Apr 23, 2019 at 12:47:10PM +0000, Roni Even (A) wrote:
> Hi,
> Thanks for the review, see inline
> Roni
> 
> -----Original Message-----
> From: Benjamin Kaduk via Datatracker [mailto:[e-mail]] 
> Sent: Thursday, April 11, 2019 5:53 AM
> To: The IESG
> Cc: [e-mail]; Bo Burman; [e-mail]; [e-mail]; [e-mail]
> Subject: Benjamin Kaduk's No Objection on draft-ietf-mmusic-data-channel-sdpneg-25: (with COMMENT)
> 
> Benjamin Kaduk has entered the following ballot position for
> draft-ietf-mmusic-data-channel-sdpneg-25: No Objection
> 
> When responding, please keep the subject line intact and reply to all email addresses included in the To and CC lines. (Feel free to cut this introductory paragraph, however.)
> 
> 
> Please refer to https://www.ietf.org/iesg/statement/discuss-criteria.html
> for more information about IESG DISCUSS and COMMENT positions.
> 
> 
> The document, along with other ballot positions, can be found here:
> https://datatracker.ietf.org/doc/draft-ietf-mmusic-data-channel-sdpneg/
> 
> 
> 
> ----------------------------------------------------------------------
> COMMENT:
> ----------------------------------------------------------------------
> 
> Section 4
> 
>    The mechanism in this document only applies to the Session
>    Description Protocol (SDP) [I-D.ietf-mmusic-rfc4566bis] when used
>    together with the SDP offer/answer mechanism [RFC3264].  Declarative
>    usage of SDP is out of scope of this document, and is thus undefined.
> 
> nit: this text doesn't actually scope itself to "this mechanism" or the RTCWeb data channel, and seems to be saying that all declarative SDP is undefined, globally.
> 
> RE: change to " Declarative  usage of SDP is out of scope for this document, and is thus undefined."
> 
> Section 5.1.7
> 
>                                 The ordered parameter is optional and
>    takes two values: "true" for ordered and "false" for unordered
>    delivery with "true" as the default value.  Any other value is
>    ignored and default "ordered=true" is assumed.  [...]
> 
> This seems to be saying that offers or answers that do not conform to the ABNF will be special-cased as being ignored (instead of rejected) when they appear where ordering-value should appear.  It's not clear to me why we need a special case here, since in Section 8 we explicitly say that peers MUST close the relevant data channel when such error cases occur.
> 
> RE: this statement only applies to the ordered parameter, I am not sure why you think it is for any parameter.

I do not think it is for any parameter.
I'm just not sure why the 'ordered' parameter is so important that we need
to specify the behavior on receipt of ordered="notaboolean", and to some
extent, why the behavior that we do specify for that case is different from
the behavior we would otherwise expect (rejecting the offer, IIUC).  In
short, the text seems superfluous given my understanding at the time  I
read id, and I wanted to understand why it was there.

> Section 5.1.8
> 
> If we chase links to draft-ietf-rtcweb-data-channel we find that larger values indicate higher priority.  Do we want to short-circuit that and say so explicitly here?
> 
> RE: I prefer to keep the link since otherwise any change in the rtcweb document will require a change here.
> 
> Section 5.2
> 
>                                                                    Each
>    of these attributes is represented by one new attribute line, and it
>    includes the contents of a media-level SDP attribute already defined
>    for use with this (sub)protocol in another IETF (Internet Engineering
>    Task Force) document.  [...]
> 
> Do we need to limit the data-channel subprotocols to those specified by the IETF?  Or is it just the media-level SDP attributes that we are trying to maintain change control over?
> 
> RE: it is just the SDP attributes, this is the dcsa section. Sub protocols are specified for dcmap

Okay, thanks for confirming.

>    The detailed offer/answer procedures for the dcsa attribute are
>    dependent on the associated sub-protocol.  If no offer/answer
>    procedures exist for the sub-protocol when used outside of the dcsa
>    attribute, no specification is needed for use with dcsa.  The IANA
>    registration procedures for the WebSocket Subprotocol Name Registry
>    (Section 9.1) do not strictly require a specification of the offer/
>    answer procedures for the sub-protocol when used with dcsa.  If the
>    sub-protocol has defined offer/answer procedures when used outside of
>    dcsa, such a specification is encouraged to ensure interoperability.
>    If the sub-protocol has defined offer/answer procedures when used
>    outside of dcsa, but no specification exists for the offer/answer
>    procedures for the sub-protocol when used with dcsa, implementations
>    SHOULD assume the use of the default values for all otherwise-
>    negotiable and applicable sub-protocol parameters.
> 
> I'm not entirely sure I understand what this paragraph is trying to say.
> The first sentence is clear, but the second sentence seems to be saying that if there are no O/A procedures for a given subprotocol outside the datachannel, it's just a free for all and I can use it in the data channel and make up my own procedures for using the dcsa attribute.
> I understand that IANA doesn't require a specification for the subprotocol when allocating the subprotocol name, but that just means that there may not be a specification available whether or not I want to use it with dcsa -- the IANA registry doesn't care directly about use with dcsa.  If a subprotocol has defined O/A procedures outside of dcsa, doesn't that basically mean there's a specification?  Why do we need to go out of our way to encourage a specification in that case (and is it a specification of the O/A semantics when the subprotocol is used outside dcsa or with dcsa)?  In the last sentence, I think we're talking about using defaults for parameters that are not explicitly negotiated using dcsa, but we don't actually say that.
> 
> RE- Since subprotocols can be registered with IANA on FCFS we cannot assume that there will be a document that will address the offer answer but we would like to encourage it.
> 
> Section 5.2.1
> 
>                                                Note however that not all
>    SDP attributes are suitable as a "a=dcsa:" parameter.  IANA SDP
>    parameters contains the lists of IANA (Internet Assigned Numbers
>    Authority) registered session and media level or media level only SDP
>    attributes.
> 
> Am I supposed to infer that these are the attributes that are suitable for a=dcsa: parameters?
> 
> RE: added the following text to this section:
> 
> " SDP attributes that are only defined for use at the
>    dcsa usage level, SHALL use the dcsa usage level when registering the
>    attribute.  If existing media attributes are used in a datachannel
>    subprotocol specific way, then a new dcsa usage level
>    MUST be defined for the existing media attribute.  Where the SDP
>   attribute is applicable to a particular subprotocol/s this SHALL also
>    be registered by indicating the applicable subprotocol identifiers
>    (see https://tools.ietf.org/html/draft-ietf-mmusic-rfc4566bis-34#section-8.5) along with the dcsa usage level.
> "
> Note that rfc4566bis changes the registration table in IANA
> 
>    As opposed to the data channel "a=dcmap:" attribute parameters, these
>    parameters are subject to offer/answer negotiation following the
>    procedures defined in the subprotocol specific documents.
> 
> I am not sure where we really say that the a=dcmap: parameters are not subject to O/A negotiation.
> 
> RE: see section 6.4 , dcmap is either accepted as is or rejected.
> 
> Section 6.1
> 
>    SCTP stream identifiers associated with data channels that have been
>    negotiated using DCEP MUST NOT be included in SDP offers and answers.
> 
> This sentence appears as a dedicated paragraph and is duplicated at the end of the previous paragraph; presumably we only need one copy thereof...
> 
> RE: yes, thanks
> 
> Section 6.2
> 
>                                          If an SDP answer contains both
>    of these parameters then the offerer MUST treat the associated SDP
>    offer/answer failed.
> 
> nit: "as failed", I think.
> 
> RE: OK
> 
> Section 6.4
> 
> We only mention that the dcmap-stream-id, max-retr, and max-time values need to match the offer, leaving the ordering/subprotocol/label/priority
> to be omitted because they are conveyed at the SCTP layer?  Perhaps we should state explicitly that they need to be absent...
> 
> RE: they can be different in each direction

Ah, thanks.

> Section 6.6
> 
> Do we want to say anything about the appropriate response to receiving changed values for the SDP attributes/attribute values is to close the corresponding data channel?
> 
> RE: Added text in the -26 version about this being application dependent
> 
> Section 6.6.1
> 
> It seems like the second bullet point only makes sense after the first has completed, as the tsvart reviewer noted?  Specifically, the "; or"
> at the end of the first bullet point doesn't seem to make much sense.
> 
> RE: after the data channel is reset you can use a new data channel with a new dcmap value 
> 
> Section 9.3
> 
>                If existing media attributes are used in a datachannel
>    subprotocol specific way (Section 5.2.1), then a new dcsa usage level
>    MUST be defined for the existing media attribute.  [...]
> 
> I'm not entirely sure that I understand what this means, though I'm not really an expert in SDP.
> 
> RE: this section was removed see my comment on section 5.2.1
> 
> Appendix A.1
> 
> I'm not sure that I know what "glare situations" are; is there a reasonable reference to include?
> 
> RE: this is an offer/answer term in RFC3264 and known to anyone who is doing offer/answer (I can add a reference to RFC3264)

(Probably no need to add the reference just for me; my apologies for not
having checked the background material.)

Thanks for all the clarifications,

Ben

>    Which set of stream identifiers is owned by which endpoint is
>    determined by convention or other means.
> 
> (specific to the application?)
> 
>       Note:For data channels negotiated via different protocol from
>       DCEP, no convention is defined by default.
> 
> That seems a bit misleading, since Section 6.1 says that the even/odd split applies to the SDP O/A negotiation case.
> 
> RE: this is true and the note says that for non DCEP you need to specify the convention which is done in section 6.1
> 
>