Discrepancies in SPARQL implementations - Aggregate expression in SELECT and its alias in GROUP BY

[kouralex]

I have a SPARQL query that had been working a while in Fuseki but recently got a report that in case of GraphDB/RDF4J implementation where it seemingly got rejected: NatLibFi/Skosmos#1857 / NatLibFi/Skosmos#1860. From that rather convoluted case I extracted this minimal example:

PREFIX ex: <http://example.org/>
SELECT ?author (COUNT(?book) AS ?bookCount)
WHERE {
  ?book a ex:Book ;
        ex:writtenBy ?author .
}
GROUP BY ?author ?bookCount

As it is, works in Fuseki. Should it work? SPARQL 1.1 specification does not clarify this use case. A colleague of mine thinks it should not work, however, I see a point here. In any case, this would be a great example to the SPARQL test cases to bring conforming implementations in line.

[osma]

I think this query should give an error, because ?bookCount is an aggregate alias and should not be used inside GROUP BY. Conceptually, grouping happens before projection and aliasing, so ?bookCount doesn't exist yet at that stage.

[afs]

SPARQL 1.1 specification does not clarify this use case.

It may be counter-intuitive but isn't an error per se.

It's an outcome of the scoping rules: https://www.w3.org/TR/sparql11-query/#variableScope .
Scopes are usually more obvious in the query e.g. nested SELECTs.

A variable is in-scope if is potentially defined. If it is not mentioned, it isn't potentially defined, so it is not "in-scope".

The WHERE clause does not mention ?bookCount so ?bookCount is not in-scope coming out of the BGP.
If it had a OPTIONAL { ?a ?b ?bookCount . FILTER(false)} it would be potentially defined / in-scope.

The GROUP BY group keys are expressions, they don't assign and don't contribute to variables being potentially defined.
You can use AS in grouping, and that is an assignment. GROUP BY (expr AS ?var) would add ?varto the scope.

The output of the group stage can not contain a setting for ?bookCount because it is never used in a place where it can be set.
That makes it legal in AS ?bookCount.

    (project (?author ?bookCount)
      (extend ((?bookCount ?.0))
        (group (?author ?bookCount) ((?.0 (count ?book)))
          (bgp
            (triple ?book <http://www.w3.org/1999/02/22-rdf-syntax-ns#type> ex:Book)
            (triple ?book ex:writtenBy ?author)
          ))))

If you want the intuition to happen, the variable scope table would need a new row saying any variable mentioned in GROUP BY expr is in-scope. It would be the only place where expressions contribute to scope.

There is a case to made that it should be like SELECT ?v, although the ?v can not be a general expression.

[osma]

Thanks Andy. Can we at least agree that including ?bookCount in GROUP BY doesn't change the query result?

[afs]

Agreed. It should not affect the results.