Research On Mutation Execution Strategy For Mutation Based Fault Localization

How to quickly and accurately locate the faults in programs has become an important issue in academic research. Mutation Based Fault Localization (MBFL) is a kind of new approach, which analyzes the similar behaviors between faulty program and mutants with artificial faults. This kind of approach provides the suspiciousness values for illustrating the probabilities of statements being errors at the same time to assist in finding faulty statements. MBFL shows a high precision on fault localization but requires a large execution cost, since the test suite needs to be executed on lots of mutants. This disadvantage limits MBFL to practical application in software fault localization.Aiming at the execution cost of MBFL, this paper focus on the optimization of mutants and test cases executions, and presents associated Mutants Execution Optimizing (TEO) strategy and Test cases Execution Optimizing (TEO) strategy. On this basis, a Dynamic Mutation Execution Strategy (DMES) is proposed and the framework of MBFL with DMES is displayed. Furthermore, this paper proves that MBFL with or without DMES has the same localization precision in any suspiciousness formulas. This paper also discusses the minimum mutation execution cost of MBFL theoretically, and gives a formal description with respect to the minimum cost problem. And then a quick solving approach with low complexity is presented especially for the minimum execution cost of MBFL.To evaluate the effectiveness of DMES, the empirical studies are conducted on 127 faulty programs. We compare the fault localization precision and mutation execution cost among spectrum-based approach, original MBFL, MBFL with DMES and theoretically optimal MBFL, namely minimum execution cost of MBFL in theory. The empirical results show that DMES can reduce the mutation execution cost of MBFL and shorten the distance with theoretically optimal MBFL, under the case of keeping fault localization precision unchanged. As a result, the efficiency of MBFL can be remarkabley increased. Moreover, additional execution cost caused by DMES is so little that can be ignored. In additional, the experiments also synthetically analyze the fault localization precision and mutation execution cost of MBFL with DMES in different suspiciousness formulas. The results provide support for popularizing MBFL in actual software fault localization.