Ranking-Based Approaches for Localizing Faults

摘要: A fault is the root cause of program failures where a behaves differently from intended behavior. Finding or localizing faults often laborious (especially so for complex programs), yet it an important task in software lifecycle. An automated technique that can accurately and quickly identify faulty code greatly needed to alleviate costs debugging. Many localization techniques assume are localizable, i.e., each manifests only single few lines close one another. To verify this assumption, we study how spread across elements. We find most localizable within methods, while around 30% manifest line code. Spectrum-based approach lightweight analyzes execution traces highlight top-most suspicious elements (i.e., statement, blocks, etc.) inspection by developers. Our be categorized into spectrum-based approach. localizes measuring strength relationship between element occurrence failure. Various association measures proposed domains statistics data mining quantify two variables interest. However, their effectiveness not well studied. investigate 40 single-bug multiple-bug programs. Some achieve smaller percentage inspected on average than well-known techniques, namely Ochiai Tarantula, number comparable Tarantula. Different have different buggy propose called Fusion Localizer leverage differences boost faults. combines scores ranking information produced existing spectrumbased particular, measures, Ochiai, inexpensively rank using fusion methods been studied domain retrieval. evaluation demonstrates our significantly improve state-of-the-art techniques. The above approaches localize potential traces. However at times, full available Code clones pieces similar code) shown useful detecting bugs because inconsistent changes among clone group may indicate bugs. clone-based bug detection suffer excessive false positives. ranks anomaly reports contain earlier list as compared original list. By actively incrementally incorporating user feedback iteratively refine classification model reorder reports, successfully reduce positive rate. In summary, dissertation has empirically demonstrated need novel ranking-based faults, which advances previous state-of-the-art.