validating custom datatypes in SHACL

[onderwijsarchitectuur]

Hoi,

SHACL works with build-in datatypes, the same als XSD and SPARQL. I found references that SHACL also works with custom datatypes defined as rdfs:Datatype. This is useful when the datatype is used in more places. One way to define such a thing is:

schema:Postcode a rdfs:Datatype; owl:onDatatype xsd:string; owl:withRestrictions ( [xsd:maxLength 6] [xsd:minLength 6] ) . Testing with https://shacl-play.sparna.fr/play/ I find that the SHACL-API ignores the facets in those datatypes. How can I make sure that SHACL validates custom datatypes as well?

Thanks Gerald

[HolgerKnublauch]

In SHACL I would suggest to write this as

ex:MyClass-myProperty
    a sh:PropertyShape ;
    sh:path ex:myProperty ;
    sh:node schema:Postcode .

schema:Postcode
    a sh:NodeShape ;
    sh:datatype xsd:string ;
    sh:minLength 6 ;
    sh:maxLength 6 .        

I would generally avoid using custom datatypes, because most tools don't understand them. Even in SPARQL you cannot do something like

BIND (STRLEN("12345"^^schema:Postcode) AS ?length)

and you cannot do mathematical functions for typed derived from numeric types.

[HolgerKnublauch]

If despite this, you really really want to continue using custom datatypes, you could write a SPARQL-based constraint that examines the xsd:minLength and xsd:maxLength and walks through all literals that may use that datatype.

[tfrancart]

If despite this, you really really want to continue using custom datatypes, you could write a SPARQL-based constraint that examines the xsd:minLength and xsd:maxLength and walks through all literals that may use that datatype.

A generalization of this approach would be to associate a regex to the rdfs:Datatype, and then writes a SPARQL-based constraint that test whether literals that use this datatype conforms to the regex. Seems like a generic-enough requirement to be included in DASH ?

[HolgerKnublauch]

If the RDF model already contains xsd:minLength etc then I don't see why going through a regex would be beneficial. We need to query what's in the model.

And the problem is to detect all literals of a given datatype. This requires iterating over all triples in the graph! That's why I would model this explicitly, e.g. using sh:node.

[onderwijsarchitectuur]

Thanks for the info

[HolgerKnublauch]

I assume this can be closed...