- From: Gregory Williams <greg@evilfunhouse.com>
- Date: Thu, 16 Jun 2016 15:38:26 -0700
- To: "Peter F. Patel-Schneider" <pfpschneider@gmail.com>
- Cc: public-sparql-dev@w3.org
On Jun 16, 2016, at 2:36 PM, Peter F. Patel-Schneider <pfpschneider@gmail.com> wrote: > > https://www.w3.org/TR/sparql11-query/#subqueries says > > Due to the bottom-up nature of SPARQL query evaluation, the subqueries are > evaluated logically first, and the results are projected up to the outer query. > > I think that this is incorrect. For example, in > > SELECT ?x WHERE { > ?x :a :b . > FILTER EXISTS { > SELECT ?x WHERE { ?x :a :b } HAVING ( COUNT(*) = 1 ) > } > } I’m not sure this specific example makes sense. The subquery involves aggregation, but attempts to project a variable that is neither aggregated nor a part of a GROUP BY clause. The semantics of EXISTS variable substitution *might* give an intuitive answer because ?x will be replaced with a constant term during evaluation, but I believe the intention of §11.1 is for this to be a syntactic restriction. FWIW, Andy’s SPARQL validator seems happy to produce an algebra for your example query, but if you try to validate just the subquery, it’ll complain about a syntax error. > The inner select is not even known until the bindings for ?x in the outer > query have been determined because EXISTS uses substitution into the inner > query. (Whether that is a good idea or not is a separate issue.) > > I ran into this issue when reading https://scirate.com/arxiv/1606.01441 > I believe that the sentence I quote above is the one that the authors indicate > that Fuseki and Blazegraph are using to support their implementation of > subqueries inside EXISTS. > > > I suggest that there be an erratum removing this sentence. The wording here is a bit awkward, but I believe the subquery evaluation is still occurring first *within the context of the evaluating the EXISTS pattern*. For example, had your filter clause looked like: ?x :a :b . FILTER EXISTS { ?x ?y ?z SELECT ?x WHERE { ?x :a :b } HAVING ( COUNT(*) = 1 ) } then the sub-query would be evaluated before the 1-triplepattern BGP {?x ?y ?z} that it joins with, but after {?x :a :b} is evaluated and the ?x variables are substituted in the EXISTS pattern body. I think calling out the bottom-up semantics here is a good thing, but the text might have benefited from discussing how evaluating EXISTS patterns is different than evaluating other type of pattern. .greg
Received on Thursday, 16 June 2016 22:38:52 UTC