RE: Comments on archdoc section 1.1

Larry,

I think your point of view shows a frequent interpretration of RFC2396, which IMHO is wrong.

In particular:

> I think the binary partition between 'absolute' and 'URI reference'
> isn't right. The key to RFC 2396 is to look at the BNF terms
> actually defined. There are absolute URIs (absoluteURI)
> and relative URIs (relativeURI) and URI references
> (URI-reference).

IMHO, that's wrong.

It's true that RFC2396 defines the term "relativeURI", but the first time it appears is in chapter 4, URI references. Thus, URIs are defined by the production for "absoluteURI" (see chapter 3) only, and there's no other kind of URI. 

> Actually, no. In practice, people use URIs to (uniformly)
> identify resources, and URI references with fragment identifiers
> to identify fragments of resources. The usage is pretty
> consistent. 'http://www.w3.org/2001/tag/2002/0813-archdoc'
> identifies the resource of the architectural document, and
> http://www.w3.org/2001/tag/2002/0813-archdoc#identifiers
> identifies Chapter 1 of it. To confuse resource fragments
> with resources would be a serious mistake.

Actually, if I have an absolute URI reference for something, I have a name. There's a straighforward way to define a new URI (not ref) for this "something", so there's little point in claiming that it isn't a resource...

> The term "URI" is used within RFC 2396
> informally (there is no BNF terminal labeled 'URI')

...but the chapter (3) that defines URI syntactic components defines the term "absoluteURI"...

> to refer to any of the different kinds of terms defined:
> URI-reference, absoluteURI, relativeURI. So it might
> be that the TAG is requesting that there be a more
> formal definition, and that the term 'URI' be assigned
> to 'URI-reference based on an absoluteURI'. Perhaps
> the idea is to add a BNF item:
> 
>    URI    = absoluteURI [ "#" fragment ]
> 
> but I think this term could be covered more accurately
> by "absolute URI reference".

If think the problem with RFC2396 is that it uses the name "URI reference" for two very different concepts (relative URIs and fragment identifiers). It would be good to reduce the confusion, but I'm not convinced that including "absoluteURI + fragment" into the set of URIs is the right way... Changing the meaning of "URI" requires that readers of specifications using URIs always need to lookup which kind of URI definition is in use -- that was bad enough with the changes between RFC2068 and RFC2396 -- please don't repeat this...
 

Julian

Received on Wednesday, 14 August 2002 11:16:31 UTC