Re: SPARQL, named graphs and default graph

I've come to a few conclusions regarding this debate and my own
implementation.

First of all, even though the SPARQL spec doesn't allow it, I have
decided that support for blank nodes as graph names is
mandatory. There seems to be a large number of things that directly or
indirectly require this. For example, the TriX serialization format
uses it.

But, that is a separate thing from modelling the "unknown origin" as a
named graph.

And for that I somewhat clarified my thoughts. The triples without an
origin in my case are not so much triples that are explicitly from an
unknown origin, but triples for which the user of the API doesn't care
to specify the origin. If there is possibility for blank nodes as
graph names, the user may specify new blank node graph names as much
as he likes, explicitly. So the prime requirement for these is simply
the ease of use.

And this means means that I need to have the graph, which holds all
statements for which no origin was explicitly specified, selectable
directly as:

  SELECT ?s ?p ?o
  WHERE { GRAPH <something> { ?s ?p ?o } }

Now, according to the current SPARQL syntax, this pretty much means
that I need to have the something be an IRI.

What kind of an IRI, though, I don't know. A standard UUID URN, an OID
URN, some normal HTTP URL from a purl.org namespace - or perhaps
something entirely different, outside the scope of these.

So, any thoughts on this?

-- Naked

Received on Sunday, 17 September 2006 19:55:15 UTC