Incremental Deductive Verification for Relational Model Transformations

Abstract

In contract-based development of model transformations, continuous deductive verification may help the transformation developer in early bug detection. However, because of the execution performance of current verification systems, re-verifying from scratch after a change has been made would introduce impractical delays. We address this problem by proposing an incremental verification approach for the ATL model-transformation language. Our approach is based on decomposing each OCL contract into sub-goals, and caching the sub-goal verification results. At each change we exploit the semantics of relational model transformation to determine whether a cached verification result may be impacted. Consequently, less postconditions/sub-goals need to be re-verified. When a change forces the re-verification of a postcondition, we use the cached verification results of sub-goals to construct a simplified version of the postcondition to verify. We prove the soundness of our approach and show its effectiveness by mutation analysis. Our case study presents an approximate 50% reuse of verification results for postconditions, and 70% reuse of verification results for sub-goals. The user perceives about 56% reduction of verification time for postconditions, and 51% for sub-goals.