On May 15, 2007, at 12:22 PM, ned+ietf-smtp(_at_)mrochek(_dot_)com wrote:
>
> I went so far to recommend more or less this approach for the ABNF
> specification and even went so far as to generate a list of
> references and uses
> of each of ABNF's "features". This was in effect rejected by the
> IESG, however,
> in favor of documenting the existance of at least two ABNF validators.
That wasn't so much rejected as it just happened that I felt like
fooling around with ABNF validators. I learned a lot while doing it
and once done, writing the interop report went quickly. I admit this
was probably not my highest priority thing to do at the time but it
was fun.
>
> I thought this was a poor decision at the time and still do. IMO
> what matters
> about ABNF - and by extension any other document that provides
> definitional
> tools for other specifications - is whether or not it is actually
> used by
> mulitplle specifications to achieve bettter specification clarity
> and therefore
> increase the chance of creating interoperable implementations.
> Whether or not
> there happens to be a way to backstop the use of such a document
> with a
> validation tool strikes me as largely irrelevant.
Not really sure why you think it was a poor decision. I also told
other ADs my strong opinion that the real reason to move ABNF to
Internet Standard was that we use it all the time and it works for us
so "duh".
I completely agree that this really is a big "duh" in the ABNF case. It is
obvious that ABNF "works" pretty darned well and should be at Standard. My
problem with the process that was used is more general in nature: The existance
or nonexistance of validation tools really tells us nothing about the
underlying utility or lack thereof of the specification format. What needs to
be assessed is the actual use of the format in other documents and whether or
not it accomplished what it was supposed to.
Another case you might consider when thinking about this is XML Schema versus
Relax NG. It isn't an exact parallel since both XML Schema and Relax are used
operationally in addition to being specification aids. (ABNF is not used
operationally AFAIK.) But excluding the operational considerations and ignoring
personal preferences, it would not make sense to argue that XML Schema is more
suitable for a given task simply because there's more support out there for XML
Schema validation. What's supported and where is more an accident of timing and
labelling than anything else.
Ned