GH-282, GH-283: Restrict triple terms to have only IRI as a constant subject in VALUES and expressions

[afs]

Replaces #329
Closes #283
Closes #282

This PR breaks test basic-tripleterm-subject.rq.

==== Positive syntax test
====          1         2         3         4
====  12345789_123456789_123456789_123456789_
 1 -- PREFIX : <http://example.com/ns#>
 2 -- 
 3 -- SELECT * {
 4 --     VALUES ?x {
 5 --         <<(:s :p :o )>>
 6 --         <<( <<(:s :p :o )>> :q :z )>>
 7 --     }
 8 --     VALUES (?y ?z) { (<<(:s :p :o )>> 123 )
 9 --                       (123 <<(:s :p :o )>> ) }
10 -- }
====
  64 [F] basic-tripleterm-subject.rq (basic-tripleterm-subject.rq)

---

Preview | Diff

[afs]

I am neutral on this change.

SPARQL (all versions) are what is now called symmetric RDF. There must be some way to write literal in the subject position of triple terms.

I do not support keeping TRIPLE as identical to <<( )>>. See #284.

I do not support changing <<( )>> in graph patterns. It is inconsistent with triples in BGPs (where they are property paths).

[Tpt]

My rational for the change:

Allowing the following queries

SELECT ?t WHERE { VALUES ?s { <<( <<(:s :p :o )>> :q :z )>> } }

SELECT (<<( <<(:s :p :o )>> :q :z )>> AS ?t) WHERE {}

means that:

1. either they return the solution { ?s -> <<( <<(:s :p :o )>> :q :z )>>. This mean that, in conformant SPARQL implementations, SELECT queries can now return solutions that are not anymore composed of only valid RDF terms but of a generalized flavor of them allowing triple terms and literals as subject of triple terms.
2. either it return an empty solution/no solution. In this case the <<( <<(:s :p :o )>> :q :z )>> term gets "lost" somewhere in the evaluation and it's confusing to the user.

No variant sounds good to me. I would prefer that the SPARQL 1.2 standard still allow SELECT queries to return only valid RDF terms. This does not prevent implementations supporting generalized RDF to support and return such terms by extending the SPARQL grammar.

[hartig]

I am supporting this change.

[afs]

w3c/rdf-tests#265 removes the test basic-tripleterm-subject.rq and adds new negative syntax tests.