Abstract

Automatically-synthesized random data generators are an appealing option when using property-based testing. There exists a variety of techniques that extract static information from the codebase to produce random test cases. Unfortunately, such techniques cannot enforce the complex invariants often needed to test properties with sparse preconditions.

Coverage-guided, property-based testing (CGPT) tackles this limitation by enhancing synthesized generators with structure-preserving mutations guided by execution traces. Albeit effective, CGPT relies largely on randomness and exhibits poor scheduling, which can prevent bugs from being found.

We present MUTAGEN, a CGPT framework that tackles such limitations by generating mutants exhaustively. Our tool incorporates heuristics that help to minimize scalability issues as well as cover the search space in a principled manner. Our evaluation shows that MUTAGEN not only outperforms existing CGPT tools but also finds previously unknown bugs in real-world software.