Generate SPARQL for nodegroups with cycles

[cuddihyge]

No description provided.

[cuddihyge]

@glguy No real fault of yours, but the fact that the query generator is now non-deterministic sent me on a couple-days wild goose chase. It turns out the infamous LM string query arbitrarily returned in 0.5 seconds or 45 seconds depending upon which random route the query generator happened to choose. I added a sort by ID as a tie-breaker in NodeGroup.getNextHeadNode() just to make it (more?) deterministic.

I'm hoping to experiment with the Jena query optimizer now, so I have it arbitrarily sorting in the direction that gives me the long-running version.

[cuddihyge]

@glguy reports that simpler cycles with MINUS are implemented. This might be a good stopping point.
@cuddihyge and others need to test

[cuddihyge]

@glguy is working on a "final" set of tests
...and put a nodegroup with cycles into a report

[cuddihyge]

I did a merge and test and found this example that doesn't seem to work.
The MINUS is not generated.

I believe it should be MINUS { ?Link chain:nextLink ?Link_1 . }

[cuddihyge]

Or more likely, this one is ambiguous. ??

[cuddihyge]

@glguy
The sparql generator gets confused inside generateSparqlSubgraphClausesNodeItem().

• It is processing ?Link_0's links
• It recurses into do ?Link_1's links
• It sees ?Link_1's incoming "minus nextlink" but it is only set up to presume a tree, so it only looks for "reverse minus nextlink" and ignores the forward "minus"
• If it were not presuming a tree, it could process the link properly and stop because ?Link is already done

But will this actually work?

• will ?Link always be the head node?
• is it arbitrary luck or actually enforced that the algorithm went to ?Link_0 first?
• if it had done ?Link_1 first, how would it have known to stop, and not put ?Link_0 inside the MINUS {} too?
• is there any real hint that the minus clause is only supposed to contain ?Link_1

[cuddihyge]

Storing these excellent examples from @galguy here in the issue:
This example looks for components that are impacted requirements that shouldn’t be governing their behavior

Extract correctly linked test results.

Finding linked requirements from the same baseline

Looking for bad cases of someone reviewing their own requirements