Dichotomy method in testing-based fault localization
Sun Ji-rong, Jiancheng Ni, Baolin Li · 2007
In practice, testing-based fault localization (TBFL), which uses test information to locate faults, has become a research focus in recent years. Dichotomy method is presented to perform on TBFL. First we optimize the test information itself from three aspects: searching scope localization using slice technique, redundant test case removal, and test suite reduction with nearest series. Secondly, the diagnosis matrix is set up according to the optimized test information and each code in failed slice is prioritized accordingly. Thirdly, the dichotomy method is iteratively applied to an interactive process for seeking the bug: the searching scope is cut in two by the checking point cp, which is of highest priority in searching scope; If cp is wrong, the bug is found; else we will ignore the codes before/after it according to the result of cp. Finally, we conduct three studies with Siemens suite of 132 program mutants. Our method scores 0.85 on average, which means we only need to check less than 15% of the program before finding out the bug.