Fine-Grained Fix Localization with new embeddings of code
Bug localization, i.e., the process of finding code locations that are suspicious with respect to a program failure, can be tedious and costly for the software industry. It is thus an important research problem that has significant practical relevance. In this project, our ambition is to develop novel approaches to precisely identify fix locations (i.e. which code to fix) when a program crash. Concretely, we target fine-grained localization (e.g., statement level when within the code) to accelerate debugging. Our goal in this project is to devise a retrieval engine that will exploit the mapping between various information sources (i.e., crash report, bug report, execution traces of test cases, etc..) and source code by using new embeddings to retrieve fine-grained code fix locations.
