[Bigdata-developers] Subquery wildcard projections not rewritten

[Lee K. <le...@sw...>]

Hi,

We are currently evaluating using Blazegraph as our rdf database and have
run in the issue described at http://trac.bigdata.com/ticket/757.
The below query causes the AssertionError to be thrown:

SELECT (COUNT(*) as ?c) {
  SELECT ?uri ?graph where {
          {
            SELECT * WHERE {
              GRAPH ?graph {
                ?uri a <http://object> .
                ?uri <http://purl.org/dc/terms/title> ?title .
              }
              MINUS {
                ?uri a <http://other>
              }
            }
            ORDER BY ?title
          }
        }
}

Some debugging shows that the error is caused by the
ASTWildcardProjectionOptimizer failing to recurse into the subqueries to
rewrite the * projection. However this recursion is implemented in the
BOpUtility.postOrderIterator(BOp) method - this method uses the argIterator
to
find child operators and therefore only visits children for nodes with an
arity > 0.

The root query node for the above query has an empty 'args' collection and
all the associated components of the top-level query are stored in the
annotations map. It looks like the iterator should search through the
annotations rather than the args for query nodes.

As there are a lot of implementations of the BOp interface, it seems that
changing the postOrderIterator2(BOp) method is unlikely to be the correct
fix. It seems that either the AST query nodes should override the arity()
function to return the count of the annotations map, or the
ASTWildcardProjectionOptimizer should use its own iterator for the nodes of
the query. The latter option would be the least impactful change but I am
not familiar with the codebase to understand the correct fix.

Any help in resolving the issue would be appreciated.