Sqlg query with union and tree combination returns wrong vertices

[dkarthikeyan88]

Sqlg query with union and tree combination returns incorrect vertices after upgrading to 2.0.2-SNAPSHOT. I've created the vertices for the below graph and associated the edges for all the vertices appropriately.

										 Test Cluster	
											|
											|
										 Test Service	
											|
											|  
										 Test DB			
									         /         \
									 /                      \
				        			 /                                  \
							/                                                  \
						 /	    				                        \
				   Test Schema1                    		            Test Schema2
				       /       \                                                            /       \
				     /           \                                                       /           \
				   /               \                		                           /               \ 
			  Table1             Table2                                     Table3               Table4
			  /    \                /    \                                             /    \                  /    \
			 /      \              /      \                                          /       \               /        \
			/        \            /        \                                      /           \           /            \
		Col1         Col2    Col3   Col4                                Col5   Col6      Col7        Col8

When I run the below query, I get incorrect vertices(some vertices are missing) with the latest master code.

g.traversal().V().has("public.Cluster","name","Test Cluster")
.out("has_Service").has("name","Test Service")
.out("has_Database").has("name","Test DB")
.union(
out("has_Schema").has("name",P.eq("Test Schema1")).out("has_Table").has("name",P.without("Table2")),
out("has_Schema").has("name",P.eq("Test Schema1")).out("has_Table").has("name",P.within("Table1")),
out("has_Schema").has("name",P.eq("Test Schema2")).out("has_Table").has("name",P.neq("Table4")))
.out("has_Column")
.range(0, 100).tree();

Expected result: "{Test Cluster={Test Service={Test DB={Test Schema1={Table1={Column1={}, Column2={}}}, Test Schema2={Table3={Column5={}, Column6={}}}}}}}";

Actual result: "{Test Cluster={Test Service={Test DB={Table3={Column5={}, Column6={}}, Table1={Column1={}, Column2={}}}}}}"

This works properly with 2.0.1 version though. I've added a simple test to validate the sqlg query at https://github.com/dkarthikeyan88/sqlg-version-testing/ for versions 2.0.2-SNAPSHOT and 2.0.1.

Please check if this is issue really an issue and let me know if any other information is required. Thank you!!

[JPMoresmau]

A simpler test

@Test
    public void testUnionHasPath() {
    	Vertex a1 = this.sqlgGraph.addVertex(T.label, "A","name","A1");
        Vertex b1 = this.sqlgGraph.addVertex(T.label, "B","name","B1");
        Vertex b2 = this.sqlgGraph.addVertex(T.label, "B","name","B2");
        Vertex b3 = this.sqlgGraph.addVertex(T.label, "B","name","B3");
        Vertex c1 = this.sqlgGraph.addVertex(T.label, "C","name","C1");
        Vertex c2 = this.sqlgGraph.addVertex(T.label, "C","name","C2");
        Vertex c3 = this.sqlgGraph.addVertex(T.label, "C","name","C3");
        a1.addEdge("toB", b1);
        a1.addEdge("toB", b2);
        a1.addEdge("toB", b3);
        b1.addEdge("toC", c1);
        b2.addEdge("toC", c2);
        b3.addEdge("toC", c3);
        
        GraphTraversal<Vertex, Path> traversal = this.sqlgGraph.traversal().V().has("A","name","A1")
				   .union(
				        out("toB").has("name",P.eq("B1")).out("toC"),
				        out("toB").has("name",P.eq("B2")).out("toC"))
				   .path();
        
        Set<Object> objs=new HashSet<>();
        while (traversal.hasNext()) {
        	Path p=traversal.next();
        	assertEquals(3, p.size());
        	Object root0=p.get(0);
        	assertEquals(a1,root0);
        	Object child0=p.get(1);
        	assertEquals("B",((Vertex)child0).label());
        	Object child1=p.get(2);
        	assertEquals("C",((Vertex)child1).label());
        	objs.add(child0);
        	objs.add(child1);
        	
        	
        }
        assertEquals(4,objs.size());
        assertTrue(objs.contains(b1));
        assertTrue(objs.contains(b2));
        assertTrue(objs.contains(c1));
        assertTrue(objs.contains(c2));
        
    }

Looks like the VertexStep on B don't get labels, so we don't generate output for them...

[pietermartin]

Thanks, master branch is finally passing all TinkerPop tests again, so I'll get time to look at this one soon.

[pietermartin]

Ok started on this one. Sqlg does not currently optimize the UnionStep. When the UnionStep's traversals are optimized they do not realize that they are in fact part of a path query so no labels are added to them.
I am looking at optimizing the UnionStep now seeing as I am in the vicinity.

[pietermartin]

@dkarthikeyan88 I added your test in TestUnion. Your test still actually fails but just because the order of the map is slightly different.
@JPMoresmau your test is also added.
Thanks