Re: Proposal to drop disjunction requirement

On Fri, Oct 01, 2004 at 01:01:44 -0400, Kendall Clark wrote:
> 
> On Thu, Sep 30, 2004 at 03:45:20PM +0100, Steve Harris wrote:
> 
> > Simple algorithm for conveting a disjuntive expression to one expressed on
> > optionals: take any common part of each branck of the disjunction, express
> > that as the must-bind part, place each of the remaining parts of the
> > disunction as an optional subgraph. You may then need to apply some value
> > constraints to reduce the set to the answer your interested in. eg:
> 
> Okay, but surely the, uh, "smart" thing to do here is for you to treat
> syntactic disjunction in this way, i.e., rewrite the query according
> to this algorithm before executing it, instead of requiring the Poor
> Dumb User (i.e., *me*) to do it?

That's a good point, but algorithm was too strong a term, the easily
implementable part only produces a superset of the results, they then need
to be winnowed, which I dont have a mechanistic transform for. There is a
(fairly) simple transform from disjuntive expressions to a set of
conjunctive ones, but it explodes the number of queries needed
dramatically.

In any case, I'm not convinced that users generally think in terms of
graph level disjunction, as I said I've not had requests for it. SQL has
no equivalent syntaxic form to graph disjunction, and I havent seen any
calls for it. It doesn have left joins though (optional), and union
queries (multiple SELECTs per query).
 
> We have 2.13 and 3.13 in UC&R that imply disjunction in SPARQL. I
> haven't heard enough to disregard that.

2.13 can be answered just as well with []'s, and 2.13 is the justification
for 3.13.

- Steve

Received on Friday, 1 October 2004 17:17:18 UTC