SH_targetClass does not handle transitive rdfs:subClassOf

[gtfierro]

Hi @ashleysommer! Hope all is well with you

I found what I believe is a similar issue to #87, this time around the handling of sh:targetClass. Below is a relatively minimal repro:

from pyshacl import validate

mixed_file_text = """
@prefix sh: <http://www.w3.org/ns/shacl#> .
@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
@prefix owl: <http://www.w3.org/2002/07/owl#> .
@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .
@prefix : <urn:ex#> .

:Class0 a owl:Class .

:Class1 a owl:Class ;
    rdfs:subClassOf :Class0 .

:Class2 a owl:Class ;
    rdfs:subClassOf :Class1 .

:Class3 a owl:Class ;
    rdfs:subClassOf :Class2 .

:prop  a   owl:DatatypeProperty .

:shape a sh:NodeShape ;
    sh:targetClass :Class0 ;
    sh:property [
        sh:path :prop ;
        sh:hasValue "test" ;
        sh:minCount 1 ;
    ] .

:s2 a :Class2 ;
    :prop "fail" .

:s3 a :Class3 ;
    :prop "fail" .
"""

def test_fail():
    res1 = validate(mixed_file_text, data_graph_format='turtle', shacl_graph_format='turtle', debug=True)
    conforms, _, _ = res1
    assert not conforms

per this file, I would expect the validation to fail because :s2 and :s3 do not have the required "test" value on the property :prop. However, in the current handling of sh:targetClass, they are not targeted by the :shape NodeShape and thus do not fail the test. I expect this because of the SPARQL definition given in the SHACL standard.

I think the fix is fairly straightforward and I will submit a PR soon

[ashleysommer]

Hi @gtfierro
Thanks for opening this Issue and PR.

I believe, similar to the case in #87, I limited this to just two graph lookups (ex:focus rdf:type ex:target and ex:focus rdfs:subcClassOf ex:target) for performance reasons, and because that logic passes all SHACL test suite tests. We also assumed that the vast majority of users would use inference="rdfs" to perform graph expansion prior to validating, so all of the class memberships would be filled in anyway, so no further traversal would be necessary.

But you're right, the shacl spec does specify:

the set of SHACL instances of c in a data graph DG is a target

Where "SHACL instances" is the same definition we discussed in #87, so all subClassOf relationships need to be walked.

[ashleysommer]

Fixed by #97