[JENA-1275] TransformScopeRename does the wrong thing with FILTER NOT EXISTS

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Jena 3.1.1
• Fix Version/s: Jena 3.2.0
• Component/s: ARQ
• Assignee: Andy Seaborne
• Reporter: Rob Vesse
• Created: 11/Jan/17 12:50
• Updated: 09/Feb/17 13:09
• Resolved: 22/Jan/17 21:18

Description

I have produced the following minimal query from an originally much larger query. When the optimise is applied to this it incorrectly double renames the variables inside of the FILTER NOT EXISTS clause leading to incorrect algebra.

Query:

SELECT ?triangles ?openTriplets
  {
    {
      #subQ2: calculate #open-triplets
      SELECT (COUNT(?x) as ?openTriplets)
      WHERE {
        ?x ?a ?y .
        ?y ?b ?z .
        FILTER NOT EXISTS {?z ?c ?x}
      }
    }
  }

Output:

(project (?triangles ?openTriplets)
  (project (?openTriplets)
    (extend ((?openTriplets ?/.0))
      (group () ((?/.0 (count ?/x)))
        (filter (notexists
                   (quadpattern (quad <urn:x-arq:DefaultGraphNode> ?//z ?//c ?//x)))
          (quadpattern
            (quad <urn:x-arq:DefaultGraphNode> ?/x ?/a ?/y)
            (quad <urn:x-arq:DefaultGraphNode> ?/y ?/b ?/z)
          ))))))

Note that we apply the quad transformation prior to applying the optimiser.

Strangely enough I cannot reproduce the problem using pure Jena command line tools i.e. qparse although I note from the code that it applies quad transformation after applying optimisation. This suggests that it is a bug in how TransformScopeRename applies to quad form algebra.

I can reproduce it with a unit test like so:

    @Test
    public void filter_not_exists_scoping_03() {
        //@formatter:off
        Op orig = SSE.parseOp(StrUtils.strjoinNL("(project (?triangles ?openTriplets)",
                                       "  (project (?openTriplets)",
                                       "    (extend ((?openTriplets ?.0))",
                                       "      (group () ((?.0 (count ?x)))",
                                       "        (filter (notexists",
                                       "                   (quadpattern (quad <urn:x-arq:DefaultGraphNode> ?z ?c ?x)))",
                                       "          (quadpattern",
                                       "            (quad <urn:x-arq:DefaultGraphNode> ?x ?a ?y)",
                                       "            (quad <urn:x-arq:DefaultGraphNode> ?y ?b ?z)",
                                       "          ))))))"));
        Op expected = SSE.parseOp(StrUtils.strjoinNL("(project (?triangles ?openTriplets)",
                "  (project (?openTriplets)",
                "    (extend ((?openTriplets ?/.0))",
                "      (group () ((?/.0 (count ?/x)))",
                "        (filter (notexists",
                "                   (quadpattern (quad <urn:x-arq:DefaultGraphNode> ?/z ?/c ?/x)))",
                "          (quadpattern",
                "            (quad <urn:x-arq:DefaultGraphNode> ?/x ?/a ?/y)",
                "            (quad <urn:x-arq:DefaultGraphNode> ?/y ?/b ?/z)",
                "          ))))))"));
        //@formatter:on
        
        Op transformed = TransformScopeRename.transform(orig);
        
        Assert.assertEquals(transformed, expected);
    }
    
    @Test
    public void filter_not_exists_scoping_04() {
        //@formatter:off
        Op orig = SSE.parseOp(StrUtils.strjoinNL(
                                       "  (project (?openTriplets)",
                                       "    (extend ((?openTriplets ?.0))",
                                       "      (group () ((?.0 (count ?x)))",
                                       "        (filter (notexists",
                                       "                   (quadpattern (quad <urn:x-arq:DefaultGraphNode> ?z ?c ?x)))",
                                       "          (quadpattern",
                                       "            (quad <urn:x-arq:DefaultGraphNode> ?x ?a ?y)",
                                       "            (quad <urn:x-arq:DefaultGraphNode> ?y ?b ?z)",
                                       "          )))))"));
        Op expected = SSE.parseOp(StrUtils.strjoinNL(
                "  (project (?openTriplets)",
                "    (extend ((?openTriplets ?.0))",
                "      (group () ((?.0 (count ?x)))",
                "        (filter (notexists",
                "                   (quadpattern (quad <urn:x-arq:DefaultGraphNode> ?z ?c ?x)))",
                "          (quadpattern",
                "            (quad <urn:x-arq:DefaultGraphNode> ?x ?a ?y)",
                "            (quad <urn:x-arq:DefaultGraphNode> ?y ?b ?z)",
                "          )))))"));
        //@formatter:on
        
        Op transformed = TransformScopeRename.transform(orig);
        
        Assert.assertEquals(transformed, expected);
    }

The first test fails while the second passes. This implies that you need 2 levels of projection to hit the bug.

Issue Links

• is related to: JENA-1167 Variables used in a filter and then defined are not handled properly by the join strategy.
• links to: GitHub Pull Request #206