An Empirical Study of Bug Isolation on the Effectiveness of Multiple Fault Localization

Bug isolation is the main approach to multi-fault localization, where failed test cases are divided into groups, and each group failed test cases are used to localize a single fault combined with all passed test cases. Ideally, all failed test cases within a single group execute the same faulty statements. However, misgrouping usually occurs due to the clustering algorithms may not able to divide failed test cases accurately. This paper focuses on the impact of fault localization by the accuracy of the clustering algorithm. A large quantitative empirical study is conducted on 12786 version programs with multiple faults, in which the misgrouping are simulated with different accuracy by a controlled experiment. The results indicate that the effect of fault localization will become worse as the accuracy of clustering decreases.