On Jan 18, 2011, at 8:22 PM, Marc Petit-Huguenin wrote:

> 
> On 01/18/2011 05:21 AM, Eric Rescorla wrote:
>> 
>> On Dec 6, 2010, at 2:12 PM, Marc Petit-Huguenin wrote:
>>> bytes, or 2032 bits.
>>> 
>>> A.14. Section 5.3.2  "uint32 length;"
>>> 
>>> I can assume that this length covers the Header, MessageContents and
>>> Signature,
>>> but I would guess that this was useful when Message was sent without
>>> FramingHeader over TCP (same than for the relo_token used to
>>> disambiguate STUN
>>> packets).  Now that this is useless, can't this length been only
>>> Header and
>>> MessageContents, so it is possible to find the boundary between
>>> MessageContents
>>> and Signature without having to parse MessageContents at all?
>>> 
>> 
>> I think it's a mistake to assume we'll never want messages to be
>> self-contained/unframed.
>> Is there a real advantage to this  breaking change?
> 
> A little easier to parse the message, but you are right that a message can be
> unframed.
> 
> [...]
> 
>>> A.16. Section 5.5.1.1. "IceCandidate candidates<0..2^16-1>;"
>>> 
>>> Because there must be at least one candidate, should this be:
>>> 
>>> IceCandidate candidates<1..2^16-1>;
>> 
>> It's actually quite a but longer than that, because IceCandidate is
>> fairly long.
>> However, I think it's harmless to leave it as-is and actually putting 
>> 12..2^16-1 or whatever is just confusing.
> 
> OK
> 
>>> A.19. Section 5.6.3.1.1, last paragraph
>>> 
>>> This paragraph described a lockstep algorithm, but if it is the case
>>> then there
>>> is no need for the received field, as we would always know that the
>>> previous
>>> messages have been received.
>>> 
>> 
>> Sorry, I'm not sure I understand the question. The received field is
>> there to 
>> permit the sender to unilaterally use more sophisticated algorithms.
>> 
> 
> Here's what the last paragraph says:
> 
> "Once an ACK has been received for a message, the next message can be
> sent, but the peer SHOULD ensure that there is at least 10 ms between
> sending any two messages.  The only time a value less than 10 ms can
> be used is when it is known that all nodes are on a network that can
> support retransmissions faster than 10 ms with no congestion issues."
> 
> What I understand from this text is that a message cannot be sent until the
> previous one has been acknowledged.  If other algorithms can be used (e.g. one
> that use the received field), then the text should say that the algorithm 
> above
> is the basic algorithm that must at least been implemented.
> 

Sounds good. Willdo.


>> 
>> Good point. OTOH, does this do any damage? If not, I'm tempted to write it
>> down as a misfeature and leave as-is.

Done.


> OK.  Perhaps a note somewhere about this would be useful for future 
> developers.
> 
>>> A.21. Section 10.1. "<kind name='sip-registration'><data-model>..."
>>> 
>>> Why do we need to repeat the data-model and access-control that were
>>> already in
>>> the IANA registry?  Or is it a way to override the IANA definitions as the
>>> example would suggest? (as the registration for SIP-REGISTRATION is using
>>> DICTIONARY and USER-NODE-MATCH).
>>> 
>> 
>> No, it's just to preserve uniform syntax for IANA-defined and
>> user-defined kinds.
> 
> OK, but what happen if the values in the XML are different from what is in the
> RFC describing the Kind (like it is currently in the example).  My suggestion
> would be to simply say that the values in the XML files are ignored for
> IANA-defined kinds.
> 

I've made this change, but I'd like to have other people weigh in if they don't 
like it.

-Ekr


_______________________________________________
P2PSIP mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/p2psip

Reply via email to