Effective fault localization via mutation analysis

When programs fail, developers face the problem of identifying the code fragments responsible for this failure. To this end, fault localization techniques try to identify suspicious program places (program statements) by observing the spectrum of the failing and passing test executions. These statements are then pointed out to assist the debugging activity. This paper considers mutation-based fault localization and suggests the use of a sufficient mutant set to locate effectively the faulty statements. Experimentation reveals that mutation-based fault localization is significantly more effective than current state-of-the-art fault localization techniques. Additionally, the results show that the proposed approach is capable of reducing the overheads of mutation analysis. In particular the number of mutants to be considered is reduced to 20% with only a limited loss on the method's effectiveness.