Augmented Bug Localization Using Past Bug Information

Traditional bug localization techniques involve a developer using his or her knowledge of the software system to locate bugs in source code. Various automated techniques simulate knowledge of the system using source code retrieval models such as latent semantic indexing (LSI) and latent Dirichlet allocation (LDA). While these methods do an adequate job, they do not make use of another wealth of information stored in the form of past bug reports. In this paper, I present an extension to the LSI model for bug localization in which the information stored in past bug reports augments the LSI model of bug localization. I describe the details of implementing this process along with the novel patch cartographer tool that is necessary for its execution. Presented along with this description is a pair of case studies verifying the effectiveness of the patch cartographer and process respectively. Results show that the patch cartographer indeed correctly identifies affected methods from a patch file. Additionally, the study of the augmented process shows significant improvement in performance compared to LSI alone.