Re: LC comment for SpecGL : 'what does "MUST define scope" mean?'

On Wed, 2003-04-23 at 18:37, Alex Rousskov wrote:
> On Thu, 24 Apr 2003, Dan Connolly wrote:
> 
> > "Conformance requirements: the specification MUST define the subject
> > matter of the specification"
> >
> > how can I tell whether my spec has defined the subject matter of the
> > specification or not?
> 
> This requirement is not "testable" in general, of course. In other
> words, it is impractical to give you an algorithm that will find scope
> definition given an arbitrary spec.

Yes, and since it's not testable, it seems counterproductive to
phrase it using rfc2119:MUST.

> In most cases however, it is possible to search for the word "scope"
> in the spec text, read abstract/introduction sections, or use other
> approaches to find spec's scope. Spec authors should be able to find
> the scope definition and should make it easy for others to do the
> same.

Nicely put. Please let the spec say that.

> > SpecGL uses MUST in the sense of RFC2119, but RFC2119 says, of
> > MUST/MAY/SHOULD keywords...
> >
> >    In particular, they MUST only be used where it is actually
> >    required for interoperation or to limit behavior which has
> >    potential for causing harm
> >
> > What interoperability failure results from the
> > failure of a spec to define conformance?
> 
> We are talking about scope definition, not conformance definition here
> (though a lack of any conformance definition is also bad).
> 
> The MUST in question limits behavior which has potential for causing
> harm.

If you really insist on using rfc2119:MUST, you'll have to
give me a testable criterion to use it with.

>  Potentially harmful behavior in this case is omitting scope
> definition. Absence of a [well-defined] scope is harmful both for spec
> authors (they tend to document things they should not care about,
> increasing the amount of their work and the complexity of the spec)
> and for spec users that will have to spend extra time figuring out
> whether the spec applies to their case, often making mistakes.
> 
> > Don't use MUST to constrain specs; specs aren't software agents.
> 
> RFC2119 scope is not limited to specs about software agents.

OK, so never mind software; yes, RFC2119 constraints can be applied
to any sort of agent: software, a person, a group, or some combination.
But a spec is none of these.

> Alex.

In case it's not clear, I'm not satisfied by this response
to my last call comment.

-- 
Dan Connolly, W3C http://www.w3.org/People/Connolly/

Received on Thursday, 24 April 2003 10:06:05 UTC