Re: [Fwd: xmlns, uri+name pairs or just uris..? Clarification n eeded.]

Pierre-Antoine CHAMPIN <champin@bat710.univ-lyon1.fr> wrote:
[...]
> Let's be clearer (excuse me if I was confusing...)
> 
> As I understood XML Namespaces (XML-NS) the last time I read it,
> a namespace is identified by a URI ;

pretty much. There are some details in dispute.
In particular, the spec says "URI reference",
and the URI RFC specifies what URI any URI reference
denotes, in the context of a base URI, but the namespaces
spec doesn't seem to follow the conventional usage
of a URI reference to denote a URI. But all that
is orthogonal to the claims you make...

> the URI does not have to be related to the namespace definition.
> (anyway, no language is given in XML-NS for namespaces definitions. Note that DTDs can not do, as namespaces can define global attributes, which are out of DTDs'scope)
> 
> RDF M&S uses the namespace mechanism in the syntax it recomends, with 2 specificities:
> 
>  (a) qname are expanded to URIs by concatenating the namespace URI with the element name. This is not compliant with XML-NS, since the reverse transformation is not unique.

Huh? Which text from XML-NS says anything about how the namespace
name is used, let alone that it can't be combined with
some other piece of information via a function that's not
invertable?

>  (b) namespaces of RDF elements *must* be the URI of the schema defining those elements. This is a restriction of XML-NS.

Again, which text in the RDF spec says that?

You should be able to get some RDF statements about any
RDF property by dereferencing its URI (i.e. the concatenation
of its namespace name and its local name); this makes
the whole system self-describing. But this is a *should*,
not a *must*. There's no way for a spec to mandate the
behaviour of the network.


> I guess that (a) was motivated by the fact that URIs are more general than XML-NS pairs (namespace;tagname); the problem of the reverse transformation was implicitely solved by (reasonable) asumption that any vocabulary item has the form <schema>#<name> or <schema>/<name>, hence (b).
> 
> What I wanted to say is:
> although reasonable, that asumption is too strong. URIs (identifers) differ from URLs (locators) in that: they do not *always* allow to retrieve the correpsonding resource.

Huh? The following are all URLs *and* URIs:

	http://www.w3.org/2000/01/rdf-schema#
	mid:23o49u23oi@example.org
	uuid:23o4u2o3i4ju2o3i4ju
	file:/etc/hosts

None of them is guaranteed to "*always* allow to retrieve
the corresponding resource".

There just isn't any property of URLs that doesn't also
apply to URIs in general, nor vice versa.


> If they don't, the application should work anyway. If they do, very well, it will work better !

Right.

> By restricting the property URIs to contain the URI of their defining schemas, RDF restricts itself to "friendly" URIs, which is IMHO a loss of generality.

There is no such restriction.
	uuid:32lj4k2l3k4j23lkj23l4j
is a perfectly good RDF property identifier.

It's not a very convenient one, since there isn't a lot
of infrastructure deployed for dereferencing it. But
it conforms to the spec in every way.

> Hope this is clearer.

Not to me.



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

Received on Thursday, 10 August 2000 14:48:43 UTC