Re: '#' exit document space? (was: Documents, Cars, Hills, and Valleys)

At 08:16 AM 4/11/02 -0400, Tim Berners-Lee wrote:
>The # is a huge syntactic device -- because the document identifier
>on the left is so different and the whateveritis on the right are
>so different, defined by different specs.  The former hands off
>immeditely to the protocol spec, the later to the format spec.

[grinding sound as mental gears shift to accommodate a shift of emphasis]

I'll want to ponder this, but I suppose one could view '#' in URIs a bit 
like '@' in email addresses.

> > It seems, reading your comments, that one SHOULD NOT use a bare HTTP URL
>to
> > identify, say, an XML namespace because that is not a document.  I suppose
> > it has a certain elegance, because the document:
> >
> >      http://www.w3.org/1999/02/22-rdf-syntax-ns
> >
> > can be viewed as describing the namespace:
> >
> >      http://www.w3.org/1999/02/22-rdf-syntax-ns#
> >
> > which seems to be how things work out in practice (for RDF, at least).
>
>Well, when there is a one-one correspndence between two things, the ID
>of one can be used to identify the other.  So we could just say that
>the namespace is identified by giving the URI of the namespace document.

OK, by way of exploration, let's consider a scenario:

I define a document format based on XML, such that it has both an XML 
schema and an RDF schema (there exists an example at 
<http://www.ninebynine.org/IETF/Messaging/draft-klyne-message-rfc822-xml-02.txt> 
-- it's flawed and incomplete, but I think it illustrates this is 
potentially a practical case).

Within that document, I use a namespace, let's call it 
http://id.mimesweeper.com/IANA/Namespaces/rfc822/, to avoid the urn debate 
for now.  (Yes, it does dereference.)  Or should that be 
http://id.mimesweeper.com/iana/namespaces/rfc822/#?

Now, I want to use the (bare) URI to access:
(a) a human description of what the namespace is used for.
(b) an XML schema for validating the syntax of the file format.
(c) an RDF schema for keying some RDF inferences that can be made about the 
information thus represented.

I'll assume, without discussion, that some mechanism is available to 
achieve this.

Now, with what is the one-to-one correspondence?  Are the HTML, XML schema 
and RDF schema different _representations_ of the same abstract document 
(note, they have quite different stated purposes).  Does the meaning of 
http://id.mimesweeper.com/iana/namespaces/rfc822/#from depend on which 
version of the document at 
http://id.mimesweeper.com/iana/namespaces/rfc822/ I retrieve?

These aren't trick questions, and I don't have any particular axe to 
grind.  I'm just looking for a framework that is self-consistent and also 
meets most people's expectations.

>The big hole we have is the way RDF makes URIrefs by attaching localnames,
>which is not shared.  Indeed, xml schema can't give you the URIref
>for an element type for example.

There's also a (possible) discontinuity in that RDF, according to the 
current model theory draft, does not recognize any fundamental difference 
between parts of identifiers before and after the '#' - they are all just 
identifiers that denote.  I don't think this is a serious problem, but 
we've yet to be absolutely clear about the relationship between treatment 
(of fragments) for RDF denotations and for Web references.

#g


-------------------
Graham Klyne
<GK@NineByNine.org>

Received on Thursday, 11 April 2002 12:28:15 UTC