Re: Blank Nodes and SPARQL

On Mon, 2005-07-11 at 15:06 -0400, Kendall Clark wrote:
> On Mon, Jul 11, 2005 at 01:46:46PM -0500, Dan Connolly wrote:
> 
> > > Care to say why you claim that it doesn't make sense?
> > 
> > I don't understand the proposal. I don't understand how
> > to relate it to what I know about logic and query languages
> > (nor our charter, nor web architecture).
> > These _!:foo things look like logical constants, to me;
> > i.e. like URIs. But you say they're different. I don't
> > understand how, except that the scope of _!:foo is
> > private to a conversation between a client and a server
> > (which seems to break webarch rule #1
> >  http://www.w3.org/TR/webarch/#pr-use-uris ).
> 
> There are violations of #pr-use-uris hereabouts, but I don't think this
> solution is one of them. (It may respond to them, it may reflect them, but I
> don't believe it *is* one of them...) Someone should yell at the FOAF & OWL
> people about their violations of #pr-use-uris, which happen to make SPARQL
> harder than it should be, IMO.
> 
> I'm curious why or how OWL-DL got past the relevant W3C review process when
> it seems to clearly mandate violation of #pr-use-uris.

Good point.

This issue has come up in some informal discussions;
http://esw.w3.org/topic/IdentifyEverything says
"It's (at least) an argument against using bNodes (BlankNodes, NodeIds)
in RDF.  If you do, then when you go to change your RDF data, you're
vaguely stuck."

The RDF Core WG had an in-depth discussion about whether bnodes
are really different from URIs at the Aug 2001 ftf meeting.
  http://www.w3.org/2001/sw/RDFCore/20010801-f2f/
(esp items "Presentation of a Model Theory for RDF by Pat Hayes"
and "Anonymous Nodes" which, alas, don't seem to have #fragids)
I suppose the relevant issue is
"Issue rdfms-identity-anon-resources: What URI if any, identifies an
anonymous resource?"
http://www.w3.org/2000/03/rdf-tracking/#rdfms-identity-anon-resources

The results of that discussion fed into the RDF model theory
and test cases (and the primer, I suppose) and became a Recommendation
in Feb 2004.

I suppose the TAG neglected to discuss how it relates
to the "Use URIs" rule. I know that TimBL spent a lot
of energy convincing me (among others) that bnodes
are really different from URIs before that Aug 2001 meeting.

Perhaps I should ask the SemWeb Best Practices WG to write
something about bnodes and URIs, and get it reviewed by the
TAG? It's a fairly fundamental architectural thing.


>  (Well, according to
> one reading of webarch.) And given that that is the state of the world and
> the specs as we find them, would we be overly restricting ourselves by not
> recognizing this systematic violation?

To me, introducing _!:foo looks as fundamental as the RDF Core WG's
decision about bnodes vs URIs.

bnodes, URIs, and literals were standardized in RDF.
Introduction of ?Variables is justified by
our #1 requirement (http://www.w3.org/TR/rdf-dawg-uc/#r3.1
3.1 RDF Graph Pattern Matching—Conjunction)
and our charter.

Introducing a new sort of term feels like something that
goes back that far.

I stipulate that bnodes are a systematic violation; but they
don't seem to meet the user requirements. _!:foo seems to
be genuinely novel w.r.t. the existing sorts of terms
(bnodes, uris, literals, universal variables).


> It's a bit of a crap situation we've inherited, and that stinks. Again, IMO.

It certainly brings to mind Henry Thompson saying
"hard cases make bad law".

> Thanks for putting yr objections on record. I think they express a
> defensible position, though I think there are other ones as well.
> 
> Kendall Clark
-- 
Dan Connolly, W3C http://www.w3.org/People/Connolly/
D3C2 887B 0F92 6005 C541  0875 0F91 96DE 6E52 C29E

Received on Monday, 11 July 2005 19:41:48 UTC