PITMuS: A Tool for Automated Bug Dataset Generation via Source-Level Mutant Reconstruction

Abstract

Mutation testing provides a scalable mechanism for generating controlled faults for software testing and empirical software engineering. In Java, PIT is a widely used mutation testing tool that creates large numbers of mutants for evaluating test suites. However, PIT creates mutants at the bytecode level and reports mutation metadata rather than the corresponding source edits. Consequently, these mutants are difficult to inspect, replay, and reuse as structured bug artifacts. We present PITMuS, a tool that reconstructs PIT mutants at the source level and automatically generates reusable mutation-based bug datasets. Given standard PIT reports and the original Java source, and using compiled bytecode when available, PITMuS localizes the affected source statement, applies the corresponding edit, and associates the result with its enclosing method, documentation, and mutation metadata. The tool also supports source-level mutant injection for inspection, replay, and downstream experimentation. We evaluate PITMuS on eight open-source Java systems. From 69,229 mutations reported by PIT, PITMuS produces 69,198 source-level original-mutant method pairs across 1,913 source files, yielding source-level original-mutant pairs for 99.96% of the reported mutation records. These results show that bytecode-level mutation reports can be converted at high coverage into inspectable, context-rich source artifacts for software testing, program repair, and learning-based software engineering.