Github user dalaro commented on the issue: https://github.com/apache/tinkerpop/pull/372 I pulled master (@ fc79de813c788ee141175bbf1ea6d0cdefa94a0f) and confirmed that the bug still existed. I didn't apply your proposed change right away. Here a `gremlin.sh` session that shows the buggy behavior. I'm also pushing a unit test that I developed first, but it's easier to see what's going wrong in this `gremlin.sh` session than in the source code for a unit test. ``` $ `find gremlin-console/target -name gremlin.sh` \,,,/ (o o) -----oOOo-(3)-oOOo----- plugin activated: tinkerpop.server plugin activated: tinkerpop.utilities plugin activated: tinkerpop.tinkergraph gremlin> sg = org.apache.tinkerpop.gremlin.structure.util.star.StarGraph.open() ==>stargraph[starOf:null] gremlin> v = sg.addVertex('name', 'foo') ==>v[0] gremlin> v.addEdge('self', v) ==>e[2][0-self->0] gremlin> sg.traversal().V().inE('self') ==>e[2][0-self->0] gremlin> sg.traversal().V().outE('self') ==>e[2][0-self->0] gremlin> sg.traversal().V().bothE('self') ==>e[2][0-self->0] ==>e[2][0-self->0] // NOTE: StarOutEdge in the inEdges datastructure gremlin> v.inEdges.values().iterator().next().get(0).getClass() ==>class org.apache.tinkerpop.gremlin.structure.util.star.StarGraph$StarOutEdge gremlin> filter = new org.apache.tinkerpop.gremlin.process.computer.GraphFilter() ==>graphfilter[none] gremlin> filter.setEdgeFilter(__.bothE("self")) ==>null gremlin> sg.applyGraphFilter(filter) ==>Optional[stargraph[starOf:v[0]]] // NOTE: no inE results, double outE results after GraphFilter is applied gremlin> sg.traversal().V().inE('self') gremlin> sg.traversal().V().outE('self') ==>e[2][0-self->0] ==>e[2][0-self->0] gremlin> sg.traversal().V().bothE('self') ==>e[2][0-self->0] ==>e[2][0-self->0] gremlin> ``` There are a couple of things worth noting above: * the `inEdges` datastructure contains a `StarOutEdge` when we construct the graph this way * the final post-filtering `inE` and `outE` are incorrect (an indirect consequence of the first point) These steps produce different behavior depending on whether I invoke `StarGraph.open()` or call `StarGraph.of(<some self-looped tinkervertex>)`. If I do the latter, I get a `StarInEdge` in the `inEdges` map and a `StarOutEdge` in the `outEdges` map, and in turn avoid this particular filtering bug. **Good news**: your change makes sense and fixes the bug I initially reported (both-filtering). **Bad news**: I think there are more graph filtering bugs on StarGraph that can create half-edges When I filter a self-loop star graph for either in or out, I get stuff like this (symmetrical for `.inE`): ``` gremlin> filter = new org.apache.tinkerpop.gremlin.process.computer.GraphFilter() ==>graphfilter[none] gremlin> filter.setEdgeFilter(__.outE("self")) ==>null gremlin> sg.applyGraphFilter(filter) ==>Optional[stargraph[starOf:v[0]]] gremlin> sg.traversal().V() ==>v[0] gremlin> sg.traversal().V().inE('self') gremlin> sg.traversal().V().outE('self') ==>e[2][0-self->0] gremlin> sg.traversal().V().bothE('self') ==>e[2][0-self->0] gremlin> ``` @okram: This doesn't look correct, but then again, I don't really know what it should do. Should it completely destroy the self-loop, so that inE/bothE/outE all fail to emit anything associated with the self-loop? Should it transform the self-loop into a single-directional edge (which seems like one interpretation for what it does now, in which case it might be correct)? I noticed that the name of my original branch is nonsense ("DSP-1397"). I don't think that can be changed after a PR is created. So, I'm opening a pair of new PRs: one for tp31 and one for master.
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. ---