Precise semantic history slicing through dynamic delta refinement

Semantic history slicing solves the problem of extracting changes related to a particular high-level functionality from the software version histories. State-of-the-art techniques combine static program analysis and dynamic execution tracing to infer an over-approximated set of changes that can preserve the functional behaviors captured by a test suite. However, due to the conservative nature of such techniques, the sliced histories may contain irrelevant changes. In this paper, we propose a divide-and-conquer-style partitioning approach enhanced by dynamic delta refinement to produce minimal semantic history slices. We utilize deltas in dynamic invariants generated from successive test executions to learn significance of changes with respect to the target functionality. Empirical results indicate that these measurements accurately rank changes according to their relevance to the desired test behaviors and thus partition history slices in an efficient and effective manner.