"Multiple top-level property shapes for the same triple" when validating with SHACL

[tfrancart]

Contact Details (Optional)

thomas.francart@sparna.fr

What happened?

I read https://github.com/diggsweden/interoperable-specifications/blob/main/docs/ap.md#multiple-top-level-property-shapes-for-the-same-triple and I don't think it can work this way.

Consider the given SHACL example

ex:ns1 a sh:NodeShape ;
  sh:label "Person" ;
  sh:property ex:ps1, ex:ps2 .
ex:ps1 a sh:PropertyShape ;
  sh:label "Hobbies" ;
  sh:path foaf:topic_interests ;
  sh:nodeKind sh:URI ;
  sh:pattern "^http://example.com/hobbies/.*$";
ex:ps2 a sh:PropertyShape ;
  sh:label "Professional interests in computer science" ;
  sh:path foaf:topic_interests ;
  sh:nodeKind sh:URI ;
  sh:pattern "^http://example.com/computer_science/.*$";

and consider the following 2 triples:

ex:Thomas foaf:topic_interests <http://example.com/hobbies/Cooking> .
ex:Thomas foaf:topic_interests <http://example.com/computer_science/SHACL> .

When validated with the above SHACL spec, the first triple will produce a Violation for ex:ps2, while the second triple will produce a Violation for ex:ps1. Put differently, any triple using a property defined with multiple different structures this way will generate a violation for all other structures that its own.
Or maybe SHACL-INSPEC does not care about its application profiles being used for validating data ?

I think a possible way to encode multiple structurally different usage of the same property is through sh:qualifiedValueShape:

ex:ns1 a sh:NodeShape ;
  sh:label "Person" ;
  sh:property ex:ps1, ex:ps2 .
ex:ps1 a sh:PropertyShape ;
  sh:label "Hobbies" ;
  sh:path foaf:topic_interests ;
  sh:qualifiedValueShape [
    sh:nodeKind sh:URI ;
    sh:pattern "^http://example.com/hobbies/.*$";
  ];
  
ex:ps2 a sh:PropertyShape ;
  sh:label "Professional interests in computer science" ;
  sh:path foaf:topic_interests ;
  sh:qualifiedValueShape [
    sh:nodeKind sh:URI ;
    sh:pattern "^http://example.com/computer_science/.*$";
  ] ;

Or using an sh:or to refer to those different property shapes could maybe be another option

Steps To Reproduce

see above

What did you expect?

I expect to be able to use a SHACL-INSPEC specification, containing multiple top-level property shapes for the same property to correctly validate RDF graphs.

Version (Optional)

No response

Screenshots (Optional)

No response

Relevant log output (Optional)

[lokal-profil]

Expanding this with some data that can be easily plugged into e.g. https://shacl.org/playground/

The test data with two cases

@prefix ex: <http://example.org/ns#> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .

ex:ShouldPass a ex:Person;
  foaf:topic_interests <http://example.com/hobbies/Cooking> ;
  foaf:topic_interests <http://example.com/computer_science/SHACL> .

ex:ShouldFail a ex:Person;
  foaf:topic_interests <http://example.com/interests/Cooking> ;
  foaf:topic_interests <http://example.com/computer_science/SHACL> .

This Shapes graph fails to validate either case (as explained in this issue)

@prefix ex: <http://example.org/ns#> .
@prefix sh: <http://www.w3.org/ns/shacl#> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .

ex:ns1 a sh:NodeShape ;
  sh:targetClass ex:Person ;
  sh:label "Person" ;
  sh:property ex:ps1, ex:ps2 .

ex:ps1 a sh:PropertyShape ;
  sh:label "Hobbies" ;
  sh:path foaf:topic_interests ;
  sh:nodeKind sh:IRI ;
  sh:pattern "^http://example.com/hobbies/.*$" .

ex:ps2 a sh:PropertyShape ;
  sh:label "Professional interests in computer science" ;
  sh:path foaf:topic_interests ;
  sh:nodeKind sh:IRI ;
  sh:pattern "^http://example.com/computer_science/.*$" .

Whereas this one does not trigger (it needs a sh:qualifiedMinCount and/or sh:qualifiedMaxCount condition as well to be activated, information we wouldn't have for 0..* cardinality fields). Even adding e.g. sh:qualifiedMaxCount 1 this doesn't catch the failing case since the ValueShapes are only concerned with counting the number of successful validations.

@prefix ex: <http://example.org/ns#> .
@prefix sh: <http://www.w3.org/ns/shacl#> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .

ex:ns1 a sh:NodeShape ;
  sh:targetClass ex:Person ;
  sh:label "Person" ;
  sh:property ex:ps1, ex:ps2 .

ex:ps1 a sh:PropertyShape ;
  sh:label "Hobbies" ;
  sh:path foaf:topic_interests ;
  sh:qualifiedValueShape [
    sh:nodeKind sh:IRI ;
    sh:pattern "^http://example.com/hobbies/.*$";
  ] .

ex:ps2 a sh:PropertyShape ;
  sh:label "Professional interests in computer science" ;
  sh:path foaf:topic_interests ;
  sh:qualifiedValueShape [
    sh:nodeKind sh:IRI ;
    sh:pattern "^http://example.com/computer_science/.*$";
  ] .

This last one correctly distinguishes the Failing from the Passing case but it no longer retains the simple 1 field -> 1 PropertyShape connection.

@prefix ex: <http://example.org/ns#> .
@prefix sh: <http://www.w3.org/ns/shacl#> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .

ex:ns1 a sh:NodeShape ;
  sh:targetClass ex:Person ;
  sh:label "Person" ;
  sh:property ex:ps1 .

ex:ps1 a sh:PropertyShape ;
  sh:path foaf:topic_interests ;
  sh:or (
    [
      sh:label "Hobbies" ;
      sh:nodeKind sh:IRI ;
      sh:pattern "^http://example.com/hobbies/.*$"
    ]
    [
      sh:label "Professional interests in computer science" ;
      sh:nodeKind sh:IRI ;
      sh:pattern "^http://example.com/computer_science/.*$"
    ]) .

[tfrancart]

Even adding e.g. sh:qualifiedMaxCount 1 this doesn't catch the failing case since the ValueShapes are only concerned with counting the number of successful validations.

I don't understand this sentence.
Here is test that shows that using sh:qualifiedMaxCount 1 works as expected:

To be more realistic let's pretend that values from computer_science are limited to one maximum value, while values from hobbies are not.
Then we have this test case:

@prefix ex: <http://example.org/ns#> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .

ex:ShouldPass a ex:Person;
  foaf:topic_interests <http://example.com/hobbies/Cooking> ;
  foaf:topic_interests <http://example.com/computer_science/SHACL> .

ex:ShouldPassToo a ex:Person;
  foaf:topic_interests <http://example.com/hobbies/Cooking>, <http://example.com/hobbies/Thinking> ;
  foaf:topic_interests <http://example.com/computer_science/SHACL> .

ex:ShouldFail a ex:Person;
  foaf:topic_interests <http://example.com/interests/Cooking> ;
  foaf:topic_interests <http://example.com/computer_science/SHACL>, <http://example.com/computer_science/RDF> .

And this SHACL with sh:qualifiedMaxCount 1 :

@prefix ex: <http://example.org/ns#> .
@prefix sh: <http://www.w3.org/ns/shacl#> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .

ex:ns1 a sh:NodeShape ;
  sh:targetClass ex:Person ;
  sh:label "Person" ;
  sh:property ex:ps1, ex:ps2 .

ex:ps1 a sh:PropertyShape ;
  sh:label "Hobbies" ;
  sh:path foaf:topic_interests ;
  sh:qualifiedValueShape [
    sh:nodeKind sh:IRI ;
    sh:pattern "^http://example.com/hobbies/.*$";
  ] .

ex:ps2 a sh:PropertyShape ;
  sh:label "Professional interests in computer science" ;
  sh:path foaf:topic_interests ;
  # maximum one such value
  sh:qualifiedMaxCount 1 ;
  sh:qualifiedValueShape [
    sh:nodeKind sh:IRI ;
    sh:pattern "^http://example.com/computer_science/.*$";
  ] .

And when I test in https://shacl-play.sparna.fr/play/validate, then the ShouldFail resource is reported as invalid, as expected:

[lokal-profil]

Apologies for being unclear in my phrasing.

What I meant was that the initial intention was to restrict foaf:topic_interests to a closed set of options (only http://example.com/hobbies/* or http://example.com/computer_science/*). The sh:qualifiedValueShape construction does not do this (which is why http://example.com/interests/Cooking passes).

The sh:or construct does enforce the closed nature of the options.

[tfrancart]

I don' t think that the sh:or option allows you to distinguish different cardinalities on each alternative.

sh:or could also be expressed on the NodeShape level as well, which would be more inline with 1 property shape = 1 property in the AP. something like (untested) :

@prefix ex: <http://example.org/ns#> .
@prefix sh: <http://www.w3.org/ns/shacl#> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .

ex:ns1 a sh:NodeShape ;
  sh:targetClass ex:Person ;
  sh:label "Person" ;
  sh:or ( [ sh:property ex:ps1 ] [ sh:property ex:ps2 ] )  .

ex:ps1 a sh:PropertyShape ;
  sh:path foaf:topic_interests ;
  sh:label "Hobbies" ;
   sh:nodeKind sh:IRI ;
  sh:pattern "^http://example.com/hobbies/.*$" ;
.

ex:ps2 a sh:PropertyShape ;
  sh:path foaf:topic_interests ;
  sh:label "Professional interests in computer science" ;
  sh:nodeKind sh:IRI ;
  sh:pattern "^http://example.com/computer_science/.*$"
.

But again, we cannot express different cardinalities on each option. sh:minCount and sh:maxCount are gloabl.

A closing axiom could be added on top of the qualifiedValueShape option to enforce the closing
this would provide a way to have different cardinalities for the 2 options, as well as the "closing" of the property.
example, again, untested:

@prefix ex: <http://example.org/ns#> .
@prefix sh: <http://www.w3.org/ns/shacl#> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .

ex:ns1 a sh:NodeShape ;
  sh:targetClass ex:Person ;
  sh:label "Person" ;
  sh:property ex:ps1, ex:ps2, ex:psCLOSE .

ex:ps1 a sh:PropertyShape ;
  sh:label "Hobbies" ;
  sh:path foaf:topic_interests ;
  sh:qualifiedValueShape [
    sh:nodeKind sh:IRI ;
    sh:pattern "^http://example.com/hobbies/.*$";
  ] .

ex:ps2 a sh:PropertyShape ;
  sh:label "Professional interests in computer science" ;
  sh:path foaf:topic_interests ;
  # maximum one such value
  sh:qualifiedMaxCount 1 ;
  sh:qualifiedValueShape [
    sh:nodeKind sh:IRI ;
    sh:pattern "^http://example.com/computer_science/.*$";
  ] .

ex:psCLOSE a sh:PropertyShape ;
  sh:path foaf:topic_interests ;
  sh:description "Enforces the fact that foaf:topic_interests must be either one of these patterns" ;
  sh:or (
    [
      sh:pattern "^http://example.com/hobbies/.*$"
    ]
    [
      sh:pattern "^http://example.com/computer_science/.*$"
    ]) .