LATERAL: Unexpectedly unbound variable.

[Aklakan]

Version

5.3.0-SNAPSHOT

What happened?

I am not sure whether this is according to SEP7 or a bug in Jena:

I am trying to inject a value using lateral into a BIND in a subquery. The following works:

SELECT * {
  BIND('x' AS ?xIn)
  LATERAL {
    { SELECT ?xIn ?xOut { BIND(?xIn AS ?xOut) } } # The algebraic no-op SELECT * gives the same result.
  }
}

The result is:

?xIn	?xOut
x	x

However, if in the lateral block there is a Join, then it breaks:

SELECT * {
  BIND('x' AS ?xIn)
  LATERAL {
    BIND("foo" AS ?foo)
    { SELECT ?xIn ?xOut { BIND(?xIn AS ?xOut) } } # SELECT * gives the same result.
  }
}

In the result, ?xOut is (for me) unexpectedly unbound.

?xIn	?foo	?xOut
x	foo

I think this is a bug in Jena (but perhaps I am misunderstanding something about the scoping rules).

Relevant output and stacktrace

No response

Are you interested in making a pull request?

Maybe

[afs]

This could be related to a recent report about substitution interacting with SELECT * that I've been investigating in a non-LATERAL context.

PS - not related directly
The other issue #2799 is similar but different.

[afs]

The (table unit) needs the binding injected but the algorithm does not consider that. Actually any (table) i.e. VALUES.

(lateral
  (extend ((?xIn "x"))
    (table unit))
  (join
    (extend ((?foo "foo"))
      (table unit))
    (project (?xIn ?xOut)
      (extend ((?xOut ?xIn))
        (table unit)))))

Elsewhere (the SimpleEvcaluator "reference" query evaluator) does exploit a #2799 related-execution to come up with the right answer for the wrong reasons.

(table unit) is an empty BGP. However, the VALUES which is a (table ...) is looking like a SEP-0007 bug.

Thank you for the report!