An Empirical Study On Spectrum-Based Fault Localization Techniques For Multiple Faults

Faults location, an indispensable issue when software failure occurs, is widely recognized as one of the most costly activities in software engineering. In order to reduce costs and raise efficiency, plenty of automatic fault localization techniques have been raised to assist developers in identifying faults.Studies have shown that spectrum-based fault localization (SBFL) is a simple, and efficient technique to locate faults, so higher efficiency can be obtained when testing and debugging. Dozens of risk evaluation formulas (REF) have been proposed for SBFL. However, one of the most important foundations of all these REFs is the assumption of single fault. Some experimental studies have indicated that SBFL works efficiently for single fault, while its performance for multiple faults is still unclear.Aiming at the above shortcomes, an empirical study on SBFL for multiple faults is conducted to figure out the impact of multiple faults and single fault. Also a new method to adjust REF parameters’ weights is proposed to find out the best REF in the study.Our study shows that 1) SBFL performs much worse for multiple faults than single fault,2) the over-reliance on failing test cases will damage the performance of SBFL for multiple faults, while it is the opposite for the test cases that pass,3) AMPLE2 with adjusted parameters’ weights gains the optimal fault localization result.The results could be used to apply SBFL to multiple bugs in practice.