owlgen: no class needed for terms defined on other vocabularies

[Silvanoc]

Important NOTE: this MR is a PoC or demo not ready for merging for the time being. I'm lack the expertise to asses if it's breaking something else.

This MR tries to address the issues mentioned in #1281.

If a class or slot are to be inherited from terms defined in external vocabularies (for example, https://rds.posccaesar.org/ontology/lis14/rdl/ in my case), a declaration is needed in the schema. That declaration provokes unexpected "exactMatch" entries instead of the expected "subClassOf"/"subPropertyOf".

Example

Following schema

id: https://example.org/my-schema
prefixes:
  linkml: https://w3id.org/linkml/
  rdl: https://rds.posccaesar.org/ontology/lis14/rdl/
imports:
  - linkml:types
default_range: string

classes:
  MySubclass:
    is_a: FunctionalObject
  FunctionalObject:
    class_uri: rdl:FunctionalObject

slots:
  mySubslot:
    is_a: hasFeature
  hasFeature:
    slot_uri: rdl:hasFeature

Results in an OWL file with following entries:

[...]

<https://example.org/my-schema/MySubclass> a owl:Class,
        linkml:ClassDefinition ;
    rdfs:label "MySubclass" ;
    rdfs:subClassOf <https://example.org/my-schema/FunctionalObject> .

<https://example.org/my-schema/mySubslot> a owl:ObjectProperty,
        linkml:SlotDefinition ;
    rdfs:label "mySubslot" ;
    rdfs:range linkml:String ;
    rdfs:subPropertyOf rdl:hasFeature .

[...]

<https://example.org/my-schema/FunctionalObject> a owl:Class,
        linkml:ClassDefinition ;
    rdfs:label "FunctionalObject" ;
    skos:exactMatch rdl:FunctionalObject .

rdl:hasFeature a owl:ObjectProperty,
        linkml:SlotDefinition ;
    rdfs:label "hasFeature" ;
    rdfs:range linkml:String ;
    skos:exactMatch rdl:hasFeature .

[...]

Notice that

1. FunctionalObject gets a wrong URI, since it should be rdl:FunctionalObject. That's what the first commit tries to fix.
2. <https://example.org/my-schema/mySubslot> is already being defined as a rdfs:subPropertyOf rdl:hasFeature, so no need to have the rdl:hasFeature skos:exactMatch rdl:hasFeature at the bottom. That's what the second commit tries to fix.

With the patch the output is what I'd expect:

[...]

<https://example.org/my-schema/MySubclass> a owl:Class,
        linkml:ClassDefinition ;
    rdfs:label "MySubclass" ;
    rdfs:subClassOf rdl:FunctionalObject .

<https://example.org/my-schema/mySubslot> a owl:ObjectProperty,
        linkml:SlotDefinition ;
    rdfs:label "mySubslot" ;
    rdfs:range linkml:String ;
    rdfs:subPropertyOf rdl:hasFeature .

[...]

without neither rdl:Functional skos:exactMatch rdl:Functional nor rdl:hasFeature skos:exactMatch rdl:hasFeature.

[cmungall]

Hi @Silvanoc - apologies we let this one slip. Generally things that are not passing github actions fall behind in our PR review priorities, but I would be very keen to get your changes in, let us know if you are able to do another pass to get tests to pass, or if you want one of us to take this over

[Silvanoc]

Hi @Silvanoc - apologies we let this one slip. Generally things that are not passing github actions fall behind in our PR review priorities, but I would be very keen to get your changes in, let us know if you are able to do another pass to get tests to pass, or if you want one of us to take this over

No, problem. It's understandable that PRs not passing tests fell down in your priority list. But I couldn't find a better way to provide some non-ready code as a discussion basis. That's why added the initial prominent note 🙂

I've tried to document in the description and commit messages what they're trying to accomplish. I lack deep knowledge about the LinkML internals to be sure that I'm fixing it "the right way". Therefore I'd appreciate if you could have a look at the description, commit messages and drafted patch and consider taking over.

[Silvanoc]

What wonders me is that changes in the OWL generator breaks test in multiple, apparently unrelated, places (starting with test_base/test_context.py).

Am I really breaking all those tests? Or simply triggering something else (like caches being erased) that is bringing the tests to new "unexplored" paths that make them fail?

[Silvanoc]

I've decided to split this PR in two, since one of the commits is fixing what I'm sure is a bug. That goes to PR #1375. I'd expect the pipelines to succeed on that PR.

The other commit is fixing what I consider a bug, but it's discutible. I'll keep this single commit in this PR.

This way we have more focussed discussions.