Evaluating template-based automatic program repair in industry

Abstract: Automatic Program Repair (APR) is a field that has gained much attention in recent years. The idea of automatically fixing bugs could save time and money for companies. Template Based Automatic Program Repair is an area within APR that uses fix templates for generating patches and a test suite for evaluating them. However, there exists many various tools and datasets, and the concept has not widely been evaluated at companies or tried in production. Critique of current research is that the bug datasets are gathered from only a few projects, are sparsely updated and are not representative of real-world projects. This thesis evaluates, TBar, Template based automatic program repair tool for Java on a large open-source bug dataset Bears and a small company dataset. Further, TBar is modified to kBar to be used for experiments. The results show that kBar presents Plausible patches to 35% (19/54) of the selected bugs and 13% (7/54) of them is Correct. Finally, a prototype were implemented at Saab, waiting for the developers to submit the first real-world bug to fix.