A Study On The Impact Of Reusing Redundant Patches On Automatic Program Repair

Automatic program repair is one of the hottest topics in software engineering in recent years, but it still needs a lot of efforts to improve ability. Prophet is an efficient patch generation system that works with a set of successful human patches to rank the candidate patches in order of likely correctness [2]. However, when inspecting the Prophet's human patches manually, we find there are about 65 redundant patches (777 in total). In order to confirm whether these redundant patches affect the conclusion, we remove them and relaunch the experiment. Moreover, we try to add redundant patches to the original dataset randomly. Both the results show that in terms of Prophet and our experimental settings, the redundant patches do not have a significant Impact on the first correct patches ranking results.