Potential problem with tests/sparql/pre-binding/unsupported-sparql-006.ttl

[ashleysommer]

Test in question: https://github.com/w3c/data-shapes/blob/gh-pages/data-shapes-test-suite/tests/sparql/pre-binding/unsupported-sparql-006.ttl

I'm currently trying to get all of the unsupported-sparql-xxx tests to pass in pySHACL.

This test expects the SHACL implementation to throw a Validation Failure due to the use of BIND (true AS ?value) pre-bound variable reassignment in the ASK Validator SPARQL query.

However in our pySHACL implementation, this test is generating a different error, due to the rule that validators of type SPARQLAskValidator cannot be used on SHACL PropertyShapes.

Some facts I've pulled from the ttl file:

• In the test graph, the constraint ex:LanguageConstraintComponentUsingASK applies to both of the blank-node Shapes _:b41651 and _:b75747 because they both have the mandatory parameter ex:lang.
• Both SHACL Shapes _:b41651 and _:b75747 are PropertyShapes.
• The constraint ex:LanguageConstraintComponentUsingASK defines its validator using the sh:propertyValidator predicate.
• The validator ex:hasLang is an instance of SPARQLAskValidator and contains an sh:ask predicate.

In the Spec:

Section 6.2.3 Validator rules: Rule 2: For property shapes, use one of the values of sh:propertyValidator, if present.

Section 6.2.3.1 SELECT-based Validators The values of sh:propertyValidator must be SELECT-based validators

Based on those rules, the pySHACL implementation is throwing an internal ConstrainLoad error, rather than a formal SHACL Validation Failure. Is that correct? Is the test wrong? If not, what are we doing wrong?

Thanks

[HolgerKnublauch]

Thanks for your report, Ashley. This test case does indeed have a glitch. The issue is that the use of an ASK validator as value of sh:propertyValidator is unsupported by SHACL, which technically means that the shapes graph is invalid which means that the result of validation is undefined. Therefore, the test case cannot assume that a Failure gets reported.

While I am sure there is a formal process to update these tests, I went ahead and fixed it on github: https://github.com/w3c/data-shapes/commit/a5582ac559bfa0c24fb03e651ec880fe966f6444 switching to sh:validator.

I have verified that the TopBraid test result from the report is still accurate (it still produces the failure).

Could you kindly double-check whether you agree that this fixes the issue with the test? I would then announce this to the SHACL CG so that those who have reported test results on this in the past can verify that they still produce the same results.

[ashleysommer]

Hello Holger, Yes, that change fixes the problem with that test case, and that test case is now passing for me. Go ahead and announce the change.

• Ashley

[HolgerKnublauch]

Thanks, I have announced this and other recent changes: https://lists.w3.org/Archives/Public/public-shacl/2018Sep/0013.html