Dynamic mutation scheduling: highly parallel, efficient evaluation of mutations for rust programs through program splitting

Mutation analysis has been considered prohibitively expensive for large software projects due to its high computational cost. Even with efficient mutant schemata approaches, the evaluation of the constituent mutants still takes considerable time. Classical mutation analysis systems introduce large overheads by evaluating mutants as separate processes. State-of-the-art Static Mutation Batching uses threads to evaluate batches of multiple mutations at the same time; however, it suffers from a “waiting tail” problem causing idling, often hindering it in practice. Our novel Dynamic Mutation Scheduling approach addresses these limitations and solves the online mutation scheduling problem: maximizing parallelism by safely scheduling new, compatible mutations and their tests as capacity becomes available, resulting in reduced mutation evaluation runtimes. Mutations are compatible with regard to parallel evaluation if they appear in distinct parts of the program. We implement our approach by extending mutest-rs, the state-of-the-art mutation analysis tool for the Rust programming language. We perform an extensive empirical evaluation of our novel Dynamic Mutation Scheduling approach on 25 top subject Rust programs, comparing it to a basic thread-based mutation evaluation approach, and to state-of-the-art Static Mutation Batching. We find that Dynamic Mutation Scheduling effectively reduces wasted time and mutation evaluation runtimes over Static Mutation Batching, by up to $75.6 \%$, while also reducing exhaustive mutation analysis runtimes, by up to $25.2 \%$.