Second Validation for Unacceptable Fixes in Automatic Program Repair

抄録

We propose a method for improving a generate-and-validate (G&V)-based automatic program repair (APR) system. Fixes are first generated and then output if they pass a test suite validation. However, there is a chance that the output fixes are unacceptable for developers. To reduce the number of unacceptable fixes, we propose a method called second validation that validates the behavior of the fixed program. The results of our experiment showed that 81.3% of the unacceptable fixes were filtered.