Re: W3C MAWG meeting agenda, 2009-03-10 (unsigned!) - regrets

Raphaël Troncy wrote:
> Therefore, I don't see what are the remaining issues of
> creating an rdf schema of the XMP metadata model. Could you point me one?

I read this document a little while ago, but 2 things bothered in the
way XMP uses RDF:

- canonical representations are specified in terms of the XML/RDF
syntax, rather than in terms of the abstract (graph) syntax. This is
bothering because most (all?) RDF toolkits do not (and should not, in my
view) give fine-grained control on how the data is serialized. I think
that all these constraints *could* be translated in terms of the
abstract syntax, but that requires interpretation from the implementer...

- namespaces in QNames are not interpreted in XMP the way they are in
RDF; namely, XMP sticks to the standard XML interpretation (a pair
containing the namespace URI and the QName suffix), while RDF
concatenates both to produce a URI. As a consequence

  <foo xmlns="http://example.com/">the value</foo>

and

  <oo xmlns="http://example.com/f">the value</foo>

are equivalent from the point of view of RDF, but *not* from the point
of view of XMP. For this reason, the "RDF Issues" section discourages
namespaces not ending with "/" or "#" (which is considered as a bad idea
by RDF people too) but does not forbid it.

As a consequence:
- an RDF serializer may generate invalid XMP
- an RDF parser may parse XMP incorrectly


Those problems can be dealt with, provided a fair amount of caution.
But I can't help but thinking that the XMP people overstate a little
their compliance with RDF...

  pa
Raphaël Troncy a écrit :
> Dear Felix,
> 
>> I have a high preference to stick to the canonical representation of
>> XMP, since it opens or rather keeps doors to three processing
>> scenarios (XMP specific, XMP, RDF), and I hope that the door to RDF
>> processing does not rely on the non-XML serialization.
> 
> Serialization is a different issue, butI was not suggesting to use a
> different syntax than XML/RDF (I'm all for having an XML/RDF
> serialization, this is the official syntax ;-)). I like also the
> canonical representation of XMP, I didn't say we should not stick on that.
> I just say that when there are _multiple_ ways of encoding structured
> lists, we should pick one (from the canonical representation) to solve
> the ambiguity. Therefore, I don't see what are the remaining issues of
> creating an rdf schema of the XMP metadata model. Could you point me one?
> 
>   Raphaël
>