ST6RI-923 Problem with name resolution when using Semantic metadata

[seidewitz]

This PR fixes a bug that could cause an incorrect name resolution when using semantic metadata.

Background
The following SysML model should parse, with p1::a redefining P::a:

part def P {
    attribute a;
}
part p : P;

metadata def M :> Metaobjects::SemanticMetadata {
    :>> baseType = p meta SysML::Usage;
}

part def Q {
    attribute a = 1;
    part p1 {
        metadata m : M;
        attribute :>> a = 2;
    }
}

Previously, this worked in the Eclipse Xtext editor, but in Jupyter the model gave the errors:

ERROR:Cannot override a binding feature value (1.sysml line : 16 column : 29)
WARNING:Duplicate of inherited member name 'a' from P (1.sysml line : 16 column : 13)

In Jupyter, name resolution happens during top-down transformation of the model. During the transformation of p1, the reference to the semantic metadata M must be resolved in order to get the implied base type for p1. This name resolution begins in the scope of p1, resultin in the redefined feature namea being resolved early, in order to check the (effective) name of the redefining feature. But, at this point, inheritance from the implied base type is blocked, in order to avoid circular resolution of M. So the reference to a in p1 resolves to Q::a.

In Eclipse, the Xtext reconciler proxy resolution works bottom up, resolving the proxy for M in the declaration of m before processing m as semantic metadata during the transformation of p1. When a is resolved, this triggers the adding of the implied subsetting of p1 by P, so the reference to a resolves to P::a (as it should).

Changes

1. Created an Xpect test ParsingTests_SemanticMetadata.kerml.xt. Since the underlying problem is in the KerML core, it is a KerML Xpect test similar to the above SysML model. Name resolution in Xpect is similar to Jupyter, so the following model had errors previously, but does not have errors after the change below is made.

   classifier P {
       feature a;
   }
   feature p : P;
   
   metaclass M :> Metaobjects::SemanticMetadata {
       :>> baseType = p meta KerML::Feature;
   }
   
   classifier Q {
       feature a = 1;
       feature p1 {
           metadata m : M;
           feature :>> a = 2;
       }
   }
   

2. Add the call metadataFeature.getMetaclass() in the method TypeAdapter::getBaseTypes in the loop over the owned metadata features of a typ ethat compute the implied base types from those metadata features that are semantic metadata. The call is added such that any proxy resolution (requiring name resolution) for the metaclass will happen before the computation of the base type.

[himi]

I'm trying to test your example on Jupyter. Is there any way to evaluate KerML on Jupyter SysML kernel?

[seidewitz]

I'm trying to test your example on Jupyter. Is there any way to evaluate KerML on Jupyter SysML kernel?

No, the Jupyter deployment cannot parse KerML.

I wrote the PR description poorly. The bug was originally identified in a SysML model. However, since the underlying cause was in the KerML Core implementation, the Xpect test for the resolution is a KerML test. The KerML model shown in the PR description is from the Xpect test.

I have updated the PR description to make this clearer.

[himi]

Ed, thank you. I found one typo in your example in Line 17 (Metadataobjcts should be Metaobjects) and

part def P {
    attribute a;
}
part p : P;

metadata def M :> Metaobjects::SemanticMetadata {
    :>> baseType = p meta SysML::Usage;
}

part def Q {
    attribute a = 1;
    part p1 {
        metadata m : M;
        attribute :>> a = 2;
    }
}

was evaluated successfully with

PartDefinition P (735d88fe-7971-4a91-8ea4-91393f8de22e)
PartUsage p (de414c8d-bfa7-4275-972c-232d83846b3c)
MetadataDefinition M (c67eec8d-8221-4bb8-9bbf-89548910c411)
PartDefinition Q (017f6523-1350-478c-a7ff-9524e6a742b2)

[seidewitz]

I found one typo in your example in Line 17 (Metadataobjcts should be Metaobjects)

Oops. That was a manual change after I copied the validated text from the editor. I've fixed it now!