Doctor of Philosophy (PhD)
Computer Science and Engineering
Software systems are often shipped with defects. When a bug is reported, developers use the information available in the associated report to locate source code fragments that need to be modified to fix the bug. However, as software systems evolve in size and complexity, bug localization can become a tedious and time-consuming process. Contemporary bug localization tools utilize Information Retrieval (IR) methods for automated support to minimize the manual effort. IR methods exploit the textual content of bug reports to capture and rank relevant buggy source files. However, for an IR-based bug localization tool to be useful, it must achieve adequate retrieval accuracy. Lower precision and recall can leave developers with large amounts of incorrect information to wade through. Motivated by these observations, in this dissertation, we propose a new paradigm of information-theoretic IR methods to support bug localization tasks in software systems. These methods exploit the co-occurrence patterns of code terms in
software systems to reveal latent semantic information that other methods often fail to capture. We further investigate the impact of combining various IR methods on the retrieval accuracy of bug localization engines. The main assumption is that different IR methods, targeting different dimensions of similarity between software artifacts, can enhance the confidence in each other's results. Furthermore, we propose a novel approach for enhancing the performance of IR-enabled bug localization methods in the context of Open-Source Software (OSS). The proposed approach exploits knowledge from previously resolved bugs to help localize new bugs. Our analysis uses multiple datasets generated for multiple open-source and closed source projects. Our results show that a) information-theoretic IR methods can significantly outperform classical IR methods in bug localization tasks, b) optimized IR-hybrids can significantly outperform individual IR methods, and near-optimal global configurations can be determined for different combinations of IR methods, and c) information extracted from previously resolved bug reports can significantly enhance the accuracy of IR-enabled bug localization methods in OSS.
Khatiwada, Saket, "Efficient Information Retrieval for Software Bug Localization" (2022). LSU Doctoral Dissertations. 5769.