Add option to relax to recursively relax

[cmungall]

Example:

Prefix: : <http://x.org/>

Ontology: <http://x.org>

ObjectProperty: hasPart
ObjectProperty: inheresIn
        
Class: P1
  EquivalentTo: hasPart some (Q and inheresIn some E)

Class: P2
  EquivalentTo: Foo and (hasPart some (Q and inheresIn some E))

Class: Q
Class: E
Class: Foo

Example of current relax (filtering equivalence just for illustration purposes):

robot relax -i nested-relax-test.omn remove --axioms "equivalent"

gives:

Class: <http://x.org/P1>
    SubClassOf: 
        <http://x.org/hasPart> some 
            (<http://x.org/Q>
             and (<http://x.org/inheresIn> some <http://x.org/E>))
    
    
Class: <http://x.org/P2>
    SubClassOf: 
        <http://x.org/Foo>,
        <http://x.org/hasPart> some 
            (<http://x.org/Q>
             and (<http://x.org/inheresIn> some <http://x.org/E>))

in both cases the inner expressions are intersections. The goal of relax is to weaken to axioms that use SubClassOf and restrictions only.

Desired:

Class: <http://x.org/P1>
    SubClassOf: 
        <http://x.org/hasPart> some <http://x.org/Q>,
        <http://x.org/hasPart> some <http://x.org/inheresIn> some <http://x.org/E>
    
    
Class: <http://x.org/P2>
    SubClassOf: 
        <http://x.org/Foo>,
        <http://x.org/hasPart> some <http://x.org/Q>
          <http://x.org/hasPart> some <http://x.org/inheresIn> some <http://x.org/E>

side note: it could be argued that end goals are better served by using robot/materialize or relation-graph with suitable blank-node hopping property chains, this may be true, but I think there is a logical coherence in having relax operate recursively.

1. X=Y => X sub Y
2. X sub A and ... => X sub A
3. X sub (R1 some R2 some ... Rn some (A and ...)) => X sub (R1 some R2 some ... Rn some (A)). [for any n >= 1]

[matentzn]

I think that makes sense and is easy to implement. Your 1/2 are already in relax. 3 as a rewrite weakens the semantics of the logical statements somewhat compared to the old relax. In particular in conjunction with remove --axioms "equivalent" this DL query wont work anymore unless R is functional: R some (A and ...B). Since this is at the heart of the idea of relax, its fine, I just wanted to mention it for completeness..

Once we are all agreed I will make a PR. The question is now if the current relax behaviour constitute a bug (i.e. it should have been recursive all along) or whether this is an addition feature, in which case we would have to add a parameter --recursive to relax. I guess some argument can be made though that the default of this should be "true".

[cmungall]

I think that makes sense and is easy to implement. Your 1/2 are already in relax. 3 as a rewrite weakens the semantics of the logical statements somewhat compared to the old relax. In particular in conjunction with remove --axioms "equivalent" this DL query wont work anymore unless R is functional: R some (A and ...B). Since this is at the heart of the idea of relax, its fine, I just wanted to mention it for completeness..

I think that is fine. Another way to think about this is the target profile excluding intersectionOf, so queries with intersectionOf would be incomplete.

Once we are all agreed I will make a PR. The question is now if the current relax behaviour constitute a bug (i.e. it should have been recursive all along) or whether this is an addition feature, in which case we would have to add a parameter --recursive to relax. I guess some argument can be made though that the default of this should be "true".

I think it is most in keeping with current robot practice to add a new parameter with default false.

[minor: fixing a typo in my original comment, I originally had n<= 1!]

[cmungall]

Although not a priority, for completeness, the --recursive option should also recurse over the min 1=>some rule too

i.e. my rule 1, the "somes" should be treated as matching EITHER 'some' OR 'min 1'

i.e

Class: P3
  EquivalentTo: Foo and (hasPart min 1 (hasPart min 1 E))

should relax to

Class: P3
  SubClassOf: Foo,
  hasPart some (hasPart some E)

(yes this is an odd example since hasPart is usually transitive but this is just for illustration...)