On Computing Instructions to Repair Failed Model Refinements

A model refinement step is the process of removing underspecification from a model by applying syntactic changes such that the transformed model's semantics is subsumed by the semantics of the original model. Performing a refinement step is error-prone and thus needs automated and meaningful support for repair in case an intended refinement step yields an incorrect result. This paper introduces sufficient conditions on a modeling language that enable fully automatic calculation of syntactic changes, which transform one model to a refinement of another model. In contrast to previous work, this paper's approach is independent of a concrete modeling language, computes shortest syntactic changes to maintain the developer's intention behind the model as much as possible, and does not assume availability of powerful model composition operators. The method relies on partitioning the syntactic change operations applicable to each model in equivalence classes and on excluding syntactic changes that are not part of shortest changes leading to a refining model. This paper contains formal proofs for the modeling language independent results and shows the method's applicability and usefulness by instantiating the framework with three modeling languages. The results provide a language independent and fully automated method to repair refinement steps under intuitive assumptions as well as language independent foundational insights concerning the relation between syntactic changes and the impact of their application on a model's semantics.