Raw

Something along these general lines came up fairly recently in IRC or somewhere but I forgot to take a reference. I'm wondering if the following is sane -

Say http://example.org/my_graph names the graph:

<http://example.com> dc:title "Something" .

(forgetting the @prefixes here) 

To tie in nicely with HTTP, doing a GET on http://example.org/my_graph will return a representation which regular RDF tools can interpret as that graph. But to avoid having to do a GET every time the graph is needed, I want a locally cached copy. A named graph/contexts/quads-capable store would be one general approach, but why not just have the following in a local model:

<http://example.org/my_graph> x:hasRepresentation

[

  x:mediaType "application/rdf+n3" ;

  rdf:value "<http://example.com> dc:title \"Something\" ."

] . 

Ok, it needs a bit of extra magic to run queries/inference on the quoted graph, but just as a substitute for getting the original, does this make sense? (The media type might make a custom datatype, but I suspect that'd complicate things).

~ 

Background. This has come up in my agent play, the ultimate aim of which is having a little framework which I can use to develop semweb applications very quickly. I'm allowing a year for this, and recently I've been making slow but steady progress. It's a major yak shaving project. Looking ahead a little I'm going to need agent (service) descriptions. They'll be the quoted graphs above if that's sane. It's no coincidence that the closest spec to what I want is the RDF Process Profile thing I roughed out ages ago, in effect I'm trying to implement what I had in mind back then with the benefit of what's come along since. RPP can have code quoted inside graphs, so I appear to be heading towards some very weird looking RDF...

@en