SPARQL query with COUNT and ORDER returns odd result

Question

The following query counts all triples in a store

SELECT count(*) where { ?s ?p <http://dbpedia.org/resource/Cat> }

And returns the expected results

http://dbpedia.org/sparql?default-graph-uri=http://dbpedia.org&query=select+count(*)+{+%3Fs+%3Fp+%3Chttp://dbpedia.org/resource/Cat%3E+}+&debug=on&timeout=&format=text/html&save=display&fname=

However, when I first tried it I accidentally left in an ORDER BY statement, e.g.,

select count(*) { ?s ?p <http://dbpedia.org/resource/Cat> } order by ?s

Then I get a very long list of results

http://dbpedia.org/sparql?default-graph-uri=http://dbpedia.org&query=select+count(*)+{+%3Fs+%3Fp+%3Chttp://dbpedia.org/resource/Cat%3E+}+order+by+%3Fs&debug=on&timeout=&format=text/html&save=display&fname=

Can anyone explain why this result happens and what it means? Is it maybe a bug with the Virtuoso SPARQL implementation?

Answer 1

It does look like a bug, if you run the same type of queries on a different store, i.e on http://api.talis.com/stores/bbc-backstage/services/sparql (which doesn't run virtuoso)

This first query works ...

SELECT (count(?s) as ?c)
WHERE {
?s ?p <http://purl.org/ontology/po/Version> .
}

and the second ...

SELECT (count(?s) as ?c)
WHERE {
?s ?p <http://purl.org/ontology/po/Version> .
} order by ?s

... gives the same result.

Actually counting + ordering doesn't make much sense here because ?s is not selected to be retrieved. But as you said, you tried it accidentally and ... it does look like a bug.

My recommendation is to send an email to the virtuoso-user mailing list to notify about this issue.

Answer 2

We (= OpenLink) are in trouble here. This ORDER BY ?s is formally a bug in the query: an aggregate without grouping means "aggregate on everything", there should be no variables outside aggregates at the output end of the query. However this error is not reported now: violations of this rule are so numerous that SQL compiler makes an auto-grouping and our SPARQL-to-SQL preprocessor also ignores this error if possible.

We will probably keep the current behaviour as is. If a "strict" compilation mode is added it will trigger the error reporting in cases like this.

Answer 3

This may be a bug with Virtuoso, it seems to treat queries with aggregates and an ORDER BY clause as having an implicit GROUP BY clause. I've noticed this on other Virtuoso endpoints besides the DBPedia one.

IMO this is a bug and you should report it to the Virutoso users mailing list as msalvadores suggests