Java/C++/C#: Add nodes predicate to PathGraph.

[aschackmull]

This ensures that we show the toString and thus the accesspath in all contexts, in particular lgtm.com.

[hvitved]

This means we should be able to get rid of the direct source-to-sink step (https://github.com/aschackmull/ql/blob/a898f5781e9bff1b5dce747d18a8c37ccb47aada/cpp/ql/src/semmle/code/cpp/dataflow/internal/DataFlowImpl.qll#L1662), right?

[yh-semmle]

Expected test output still needs updating.

[aschackmull]

I've included a commit to fix the toString of an empty accesspath, which was a postponed change from #1797.

[hvitved]

See comment above.

[aschackmull]

This means we should be able to get rid of the direct source-to-sink step

Possibly, yes. But I think that's best left to a separate PR. This is about getting the right toString to lgtm.com.

[hvitved]

Possibly, yes. But I think that's best left to a separate PR. This is about getting the right toString to lgtm.com.

OK, I would have thought the two would go together; the sink-to-source step was basically a workaround needed because of the missing nodes predicate.

[aschackmull]

Language-Tests/Java is btw. not expected to succeed here, but it is green on the internal PR.