Re: Reducing normative dependencies

Le samedi 06 décembre 2014 à 17:21 +0100, Harald Alvestrand a écrit :
> > I don't find this use case particularly compelling.  An informative
> > reference would be enough to address the discovery issue.
>
> I agree for the discovery case.
> 
> From a W3C-process standpoint, an informative reference from PC to Stats
> is the easiest to manage.
> 
> For the case of other specs (inside or outside the W3C) choosing what
> they have normative references to when (if) they say "as part of
> conforming to this spec, you have to implement WebRTC", I'm unsure.
> 
> I'd like to hear what others think.

If we know other specs that need to reference "WebRTC" conformance, and
if we know that this conformance to be meaningful need to include stats,
then I suggest we should have a different spec that normatively
references all the pieces that would constitute such a conformance
profile.

But given that browsers don't usually implement stuff based on what is
mandated in an abstract spec but based on their own analysis of their
market, and given that our specs are targeted at browsers, I'm
personally doubtful this is a useful exercice — most previous examples
in this space have failed to achieve this.

Given that from a pure implementability perspective (the one that I
think matters when it comes to normative references), one can obviously
implement the stuff in webrtc-pc without implementing the stuff in
webrtc-stats, I concur that an informative reference from the former to
the latter is all we need, and that we should do so as we move the
getStats() partial interface to webrtc-stats.

Dom

Received on Tuesday, 9 December 2014 12:18:24 UTC