Online Adaptable Bug Localization for Rapidly Evolving Software

Bug localization aims to reduce debugging time by recommending program elements that are relevant for a specific bug report. To date, researchers have primarily addressed this problem by applying different information retrieval techniques that leverage similarities between a given bug report and source code. However, with modern software development trending towards increased speed of software change and continuous delivery to the user, the current generation of bug localization techniques, which cannot quickly adapt to the latest version of the software, is becoming inadequate. In this paper, we propose a technique for online bug localization, which enables rapidly updatable bug localization models. More specifically, we propose a streaming bug localization technique, based on an ensemble of online topic models, that is able to adapt to both specific (with explicit code mentions) and more abstract bug reports. By using changesets (diffs) as the input instead of a snapshot of the source code, the model naturally integrates defect prediction and co-change information into its prediction. Initial results indicate that the proposed approach improves bug localization performance for 42 out of 56 evaluation projects, with an average MAP improvement of 5.9%.