Fusion fault localizers

摘要: Many spectrum-based fault localization techniques have been proposed to measure how likely each program element is the root cause of a failure. For various bugs, best technique localize bugs may differ due characteristics buggy programs and their spectra. In this paper, we leverage diversity existing better using data fusion methods. Our approach consists three steps: score normalization, selection, fusion. We investigate two normalization methods, selection five methods resulting in twenty variants Fusion Localizer. bug specific which set be fused are adaptively selected for based on its Also, it requires no training data, i.e., execution traces past programs.We evaluate our common benchmark dataset consisting real from medium large programs. evaluation demonstrates that can significantly improve effectiveness state-of-the-art techniques. Compared these techniques, Localizer statistically reduce amount code inspected find all bugs. increase proportion localized when developers only inspect top 10% most suspicious elements by more than number successfully up 10 blocks 20%.