[adaussy]

Similarly to changes made in PRs #815 and #818, this PR moves functionality for completing the abstract syntax of a FeatureChainExpression from its transformation adapter to its parse post-processor.

Background

A FeatureChainExpression has the form src.f.g.h, where src is an expression and f.g.h is a feature chain that is applied to the result of src. Semantically, an expression of this form is equivalently to

expr : ControlFunctions::'.' subsets Performances::evaluations {
    feature redefines ControlFunctions::'.'::source = src {
        feature redefines ControlFunctions::'.'::source::target
            redefines tgt;
    }
    feature tgt chains f.g.h;
    return subsets source.tgt;
}

(see KerML, 8.4.4.9.6 Operator Expressions/Control Functions). Previously, this structure was created for a FeatureChainExpression in roughly three steps:

1. Parse the textual into the outer expression containing an empty owned feature bound to the src expression and a feature for the tgt feature chain.
2. During transformation, create (via FeatureChainExpressionAdapter.addAdditionalMembers) a nested feature under the expression’s source parameter when none exists.
3. Add all the implied specializations (as shown above).

The problem is that creation of the sourceTarget feature created in step 2 can also be triggered indirectly by a derived-property query, if the addAdditionalMembers mechanism is to be used, which is to be avoided.

Objectives

• Preserve feature-chain semantics.
• Make structural preparation explicit.
• Keep derived queries free of structural mutations.

Change

• The computation of the sourceTarget feature (step 2 above) has been moved from FeatureChainExpressionAdapter to FeatureChainExpressionParserPostProcessing.

[seidewitz]

I have added two more commits. Tests still pass. Please review.

[adaussy]

On a small problem on headers.

[adaussy]

Why did you remove this copyright ?