Cerebro: Static Subsuming Mutant Selection

Mutation testing research has indicated that a major part of its application cost is due to the large number low utility mutants it introduces. Although previous identified this issue, no study proposed any effective solution problem. Thus, remains unclear how mutate and test given piece code in best effort way, i.e., achieving good trade-off between invested effectiveness. To achieve this, we propose Cerebro , machine learning approach xmlns:xlink="http://www.w3.org/1999/xlink">statically selects subsuming mutants, set resides on top subsumption hierarchy, based mutants’ surrounding context. We evaluate using 48 10 programs written C Java, respectively, demonstrate preserves mutation benefits while limiting cost, reduces all factors such as equivalent mutant executions, requiring analysis. strong inter-project prediction ability, which significantly higher than two baseline methods, supervised features by state-of-the-art, random selection. More importantly, our results show ’s selected lead tests are respectively capable killing 2 times killed baselines when selecting same mutants. At time, cost-related factors, selects, average, 68% fewer 90% executions baselines.