Improving Mutation-Based Fault Localization via Mutant Categorization

Abstract

Fault localization is one of the most important activities in software debugging. Among various fault localization techniques, mutation-based fault localization (MBFL) has been commonly studied with its promising performance. However, MBFL should be improved further by incorporating more useful program information. In this paper, we propose MuCatFL, a novel and lightweight technique for better MBFL via mutant categorization. In details, after executing the original test suite against all generated mutants, we categorize the mutants into two groups, positive mutants and negative mutants, to rank the tied program elements. We evaluate MuCatFL by performing an extensive study on 395 real software faults from the widely used benchmark Defects4J. The experimental results show that MuCatFL can significantly outperform MBFL techniques (e.g., localizing 138 faults within the Top-1 position on method-level, 43.75% more than traditional Metallaxis technique). We also investigate that only positive mutants can contribute to the effectiveness of MBFL. Our findings can also provide guidance for the strategies to reduce the execution cost of MBFL.