Alleviating the Impact of Coincidental Correctness on the Effectiveness of SFL by Clustering Test Cases

Abstract

Spectrum-based fault localization techniques leverage coverage information to identify the faulty elements of the program via passed and failed runs. However, the effectiveness of these techniques can be affected adversely by coincidental correctness, which occurs when faulty elements are executed, but the program produces the correct output. This paper proposes a clustering-based strategy to improve the effectiveness of spectrum-based fault localization. The basis of this strategy is that test cases in the same cluster have similar behaviors. Our experimental results show that, the percentage of clusters that contain coincidentally correct test cases in clusters which do not contain failed test cases, is usually smaller than the percentage of coincidentally correct test cases in passed test cases. By clustering test cases and reconstructing the coverage matrix, our extensive experiments demonstrated that the fault-localization accuracy of Spectrum-based fault localization techniques can be effectively improved.