Re: Blank Nodes and SPARQL

On Mon, Jul 11, 2005 at 11:33:59AM -0500, Dan Connolly wrote:
> 
> On Sun, 2005-07-10 at 22:04 -0400, Ron Alford wrote:
> > After consulting with the lab, we have decided that the namespace
> > splitting proposal (_!:...) is the best way to deal with the use cases
> > that have been brought up.
> >
> > While we believe that the protocol prebinding is nice and generally
> > useful, we would prefer a syntax level solution to the bnode problem.
> > 
> > The function extension (ext:bnodelabel) comes in a distant third.  It
> > adds annoying overhead to templating queries.  We were also concerned
> > with how well implementations will deal with function extensions.
> 
> And what of the "Dynamically assign identifiers" option? I'm
> interested to know whether you find it acceptable.

I probably shouldn't answer for Ron, here, but I asked our OWL DL friends,
and they didn't like this at all.

Thinking about it a little more, it does seem to change the semantics of
every graph that gets rewritten in that way.

As Ron said, we strongly prefer the "split the bnode label space" solution
that Andy pointed out. Here's why:

1. It make sense (which is a good thing)

2. It reflects actual practice (people seem to be doing this on their own)

3. It's a relatively simple spec change, needing only some description of
the actual syntactic sugar (so that everyone does the same thing)... How
Andy chooses to do this -- we discussed just setting a convention or making
a new grammar production -- is something we're comfortable leaving to his
judgment.

4. It feels right -- i.e., it's sufficient to the problem, but not overly
powerful, such that other, hard-to-see issues might be raised.

Hope this helps elucidate our position a bit.

Kendall Clark

PS--I'm cc'ing the DAWG list.

Received on Monday, 11 July 2005 16:54:24 UTC