Research On The Spectrum-based Fault Localization For Software Systems

Spectrum-based Fault Localization (SBFL) is a kind of technique using programinstrumentation to collect program dynamic execution information to help programmersquickly locate faulty program statements or functions.But SBFL is still not being widely used in the industry. We consider the reason asfollows:(1) The fault localization accuracy of this technique is still not satisfactory.(2)SBFL requires users to manually prepare large amount of input data.(3) It can only beused with internal testing. Aiming at addressing those issues, we propose three methodsand this paper is organized as follows:(1) For the first issue we propose an interactive fault localization approach whichincorporates simple user feedback. It aims to expose truly faulty elements earlier, butdoes not require additional effort than existing methods. Our experimentsshow that ourmethod achieves significant improvements on fault localization accuracy.(2) We propose a novel test case selection strategy based on DiversityMaximization Speedup. It maximizes the effectiveness of a fault localization techniqueand only expects developers to label a much smaller number of test cases along theprioritized order. The experimental results show that our method can help existing faultlocalization techniques to achieve comparable accuracy with significantly fewer labeledtest cases than existing best techniques.(3) We also propose a fault localization techniqueenabling fault localization to beused for debugging user-reported crashes. This method combines stack traces in crashreports and static information from program source code to generate a approximateprogram spectra. Besides, we also propose several heuristics to further improve thelocalization accuracy for diagnosing user-reported crashes. The results show that theperformance of the crash fault localization method is promising.