ST6RI-924 Problems with TypeAdapter and AssignmentActionUsageAdapter

[seidewitz]

This PR fixes bugs in TypeAdapter and AssignmentUsageAdapter.

TypeAdapter

• Previously, TypeAdapter::removeUnnecessaryImplicitGeneralTypes could throw an NullPointerException if implicitEClassGenerals was null for a certain specialization EClass (which can happen during processing of an ill-formed model).
• Added a check on that implicitEClassGenerals is non-null before processing it further.

AssignmentUsageAdapter

• Previously, AssignmentActionUsageAdapter::addTargetRedefinition called FeatureAdapter::addFeatureWriteTypes to satisfy redefinition constraints on target::StartingAt and target::startingAt::accessedFeature nested in an AssignmentActionUsage. However, placing this within AssignmentActionUsageAdapter means that the transformation is not done if Feature::computeImplicitGeneralTypes is called directly on these features, which can happen during Xtext editor reconciliation of text changes. The result can be spurious validation errors.

• Moved the implementation satisfying these constraints to FeatureAdapter, called from addComputedRedefinitions.

  Note. This change results in the KerML-level FeatureAdapter depending on SysML AssignmentActionUsage. However, in general, "on-demand" calls to computeImplicitGeneralTypes can only be properly handled if the satisfaction of all semantic constrains effecting a metaclass are implemented in the adapter for that metaclass.

[himi]

I encountered this issue during Xtext editing, but I do not remember any concrete steps to reproduce that NullPointerException. Do you know anything about it? So far, it works quite well without NPE :).

[seidewitz]

I encountered this issue during Xtext editing, but I do not remember any concrete steps to reproduce that NullPointerException. Do you know anything about it? So far, it works quite well without NPE :).

I have not been able to reliably reproduce the exception, so I have no been able to track down its cause, which does bother me. I am not even sure how it is possible for implicitEClassGenerals to be null at this point, since it is the result of accessing a Map using a known member of the keySet. But it does happen, and there is an easy way to prevent it just by checking for null.

[himi]

Thank you. I tried many patterns and could not get any NPEs so far. In my experience, analysis and subject edit has had caused an NPE often, but I could not reproduce such problems in this version.