Debugging consumes a considerable amount of time in software engineering, but it is rarely automated. In this paper, we focus on improving existing fault localization techniques. Spectrum-based fault localization (SFL) and slicing-hitting-set-computation (SHSC) are two techniques based on program execution traces. Both techniques come with small computational overhead and aid programmers to faster identify possible locations of faults. However, they have disadvantages: SHSC results in an undesirable high ranking of statements which are executed in many test cases, such as constructors. SFL operates on block level. Therefore, it cannot provide fine-grained results. We combine SHSC with SFL in order to eliminate these disadvantages. Our objective is to improve the ranking of faulty statements so that they allow for better fault localization than when using the previously mentioned methods separately. We show empirically that the resulting approach reduces the number of statements a programmer needs to check manually. In particular, we gain improvements of about 50% percent for SHSC and 25% for SFL.
展开▼
