Exploiting fault localisation for efficient program repair:2020 Genetic and Evolutionary Computation Conference, GECCO 2020

Search-based program repair generates variants of a defective program to find its repair. This could reduce the time and effort necessary for the manual software development and maintenance. However, applying even a limited set of mutations on a small piece of code (that repairs only trivial defects) generates a huge number of possible program variants (also called a search space). The reduction of the search space, while preserving the number and quality of repairs, would make these tools more efficient and practical. We present an end-to-end repair tool for Java programs. It localises lines of source code that introduced a defect into the history of the program's development and applies a set of mutations targeting only these lines. In the reduced search space, our tool repaired defects covered by failing tests in an open-source Java program. © 2020 Owner/Author