I think that's right. However, what may well be missing in the mix
is input from people who actually deploy and operate our stuff, and
live with its limitations and quirks every day. We need to understand
the indirect consequences of our choices: not "can it be coded and will
it interoperate?" but "will it drive service providers and users crazy?"
I think there have been a lot of red herrings in this discussion, but this I
agree with completely.
I'm sure there are places in the IETF where implementation work prior to
standardization is wholly insufficient if not actually nonexistant. But this
doesn't generalize: There are also places where implementation work is the
norm. Consider the SIEVE working group as yet another example: This group has
10 active drafts, and I believe there are already multiple implementations of
all of them. (I have implemented 9 of them myself, as a matter of fact.)
While early implementation can help, it is no guarantee that the specification
is actually any good, especially if the people doing the implementing are also
the authors of the specification. In fact I would go so far as to say that
implementations by specification authors only protect against the most
egregious errors, errors that should be caught by any reasonable review. After
all, it is axiomatic that the authors know what they intended to say, which may
or may not be the same as what was actually said.
Moreover, as the world we develop our specifications for gets increasingly
complicated, our ability to predict how things will actually work out, which
was never all that great to begin with, is being eroded. This means we need to
look more and more at actual operational experience, not simple
implementability.
In short, I think bemoaning the lack of early implementations of our
specifications, whether true or not, misses the point almost completely. What
we need to be looking for is operational experience and better ways to
incorporate operational experience into our process.
And since most people are reluctant to try things operationally until there's
at least a published specification in place, one thing we need to be doing is
placing less emphasis on getting everything absolutely perfect for initial
publication and more emphasis on getting operational feedback and revising
things after that. And this, I regret to say, is pretty much the opposite of
where we seem to be heading.
Ned
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/ietf