Detaching from a thread about moving a particular specification from
Experimental to PS - this means I'm (quite deliberately) taking Barry and
Cyrus Daboo's quotes out of their proper context here, please bear this in
mind.
But I happened to be talking about this recently, and this has reminded me.
On 6 March 2014 16:51, Barry Leiba <barryleiba(_at_)computer(_dot_)org> wrote:
My point here is that until we have interoperable implementations on both
client and server and have caught and fixed any issues in the spec, we
shouldn't advance the status.
Why not? Interoperable implementations aren't required for Proposed
Standard. Internet Standard, sure, but not PS. The requirement for
PS is that we think we got it right.
So this made sense when Proposed Standard actually meant Proposed, rather
than Standard.
But now that Proposed Standard actually means Draft, because (to paraphrase
Scott Bradner), we've left-shifted our standards process, why aren't we
requiring implementation experience for PS?
That's one, IMO extreme, way to look at it. The other extreme is that we've
noted that the difference between draft and full was essentially meaningless
and gotten rid of it, leaving PS completely unchanged.
Both extreme views fail to capture the nuances of the shift. Indeed, since this
is an ongoing process and we have relatively little experience with our current
two level approach, any current view is going to be based on scant data.
But I think it's safe to say that PS now means more, but neverthess falls short
of what draft used to mean.
We've tightened up review a far bit, and raised the bar for spec quality in
general, but since PS is now DS by any other name, it's had the effect of
side-stepping the "two independent interoperable implementation"
requirement.
Sorry, I reject the implication this question is based on. PS is not the same
as draft used to be. Which leaves open the question of how much more
should now be required for PS.
I am comfortable with making implementation experience a SHOULD for PS. But
nothing more than that. In particular, interoperability testing of multiple
implementations is far too high a barrier for PS.
And there are going to be exception, where for various reasons implementation
prior to approval is impractical.
I would like to see more implementation experience -- it's be great to
see. That's why I asked whether there is any. But to me, the gating
factor is more about whether there's *interest* in implementing it.
Wouldn't it be great to require? Running code and all that, after all...
First, running code is not the same as multiple interoperable implementations,
which is what draft used to mean.
Second, as many others have noted, the existence of running code doesn't mean
as much as it used to. Our collective ability to make pigs fly should never be
underestimated. In fact I'll channel my inner Pete and say that simple
assertions that "I implemented this" are of no real use to me, nor, I suspect,
to anyone else.
What I really want to is is statements of the form, "I implemented foo, here's
what was eesy, here's what was hard, here's what made no sense." That
sort of thing helps me immensely when deciding whether or not something
is worth doing.
Ned