Investigating the Impact of Test Cases on the Performance of Automated Program Repair

Abstract

Several techniques have been available in research studies on automated program repair. In test-based repair techniques among them, the preparation of test cases is key to the success of repair. Nevertheless, there is no sufficient knowledge based on the investigation of such test cases and programs to be fixed so far. We designed an experiment framework for investigation of how the characteristics of prepared test cases affect the number of generated patches, their correctness, and the consuming time of their generation. This paper shows results of an experiment with programs containing five kinds of well-known bugs. The results demonstrate that either the increasing or decreasing of the test cases improves the performance of patch generation, depending on the kinds of bugs.