Lists Home |
Date Index |
Jonathan Borden wrote:
> Yes, you mentioned linkbases last week before the onslaught and I was
> certainly meaning to respond before getting distracted (ahem)...
> Briefly, there seems to be several ways to describe a graph (of URIs):
> 1) RDF
> 2) XLink
> 3) TopicMaps (?)
> (also note that the rddl2rdf. XSLT implement's Ron's XLink -> RDF conversion
> for RDDL which uses simple XLinks).
> My thinking is that the graphs describable by RDF and XLink (including
> linkbases) are different by surface syntax but not by abstract syntax i.e.
> that both represent a collection of triples. Ron Daniel has a nice note
> describing the XLink <-> RDF conversion. There has been alot of talk about
> somehow 'merging' topic maps and RDF as well, though I think that TM may
> have some relevent distinctions between concepts like "subject" and "topic"
> that are not well captured by RDFs all encompasing "resource". In any case
> the question is then:
> 1) does any or any combination of the three meet current needs?
> 2) if not what is missing?
> In summary, I suppose it all depends on the use case for linkbases. Tell us
I haven't given much thought to the applicability of RDF or TopicMaps,
here. I like the use of XLink by RDDL and am not at all bothered by the
use of role and arcrole as some are. I've come to view the use of simple
links, though, as a bit of a straightjacket. Stepping up to extended
links seems to open up more possibilities.
The extended link approach, for example, seems to play better with the
notion of distributable libraries or XML Archives (XARs). It allows
people other than the authors of a namespace to make available packages
or collections of resources for namespaces. If someone out there
unaffiliated with the W3C comes up with some useful stylesheets or other
resources for use with XML Schema, why not have a mechanism by which
they can publish their resources and associate them with the XML Schema
namespace? XML applications could leverage linkbases and XARs in a
fashion similar to the way Java apps leverage JARs on the classpath,
rather than having to rely upon the more limited notion of having a
single container at the end of a namespace URI.
This approach would also allow authors to associate collections of
resources with resources other than namespaces. Public ids can be
mechanically converted to and from URN syntax, which means a linkbase
can associate resources with them just as RDDL does with namespace URIs.
And the author of the DTD can do this even if they are not the author of
any of the namespaces used in the document type. (And you can of course
do this with a system id that has been absolutized.) This approach could
also let you associate resources with notations and unparsed entities.
RDDL is currently limited to associating resources with XML namespaces.
Keeping with the same current approach, but stepping up to extended
links seems like it could open up a lot of doors. Toss in a PI that can
associate such a resource with a document, and it seems to me like the
key limitations of RDDL are overcome. (The PI would be a hint.
Applications would typically have other means of associating linkbases,
but could use the PI as a fallback. A model similar to that used by XML
Catalog makes sense, here.)
That's what I've been thinking lately, anyway.