Automated Large Program Repair based on Big Code

The task of automatic program repair is to automatically localize and generate the correct patches for the bugs. A prominent approach is to produce a space of candidate patches, then find and validate candidates on test case sets. However, searching for the correct candidates is really challenging, since the search space is dominated by incorrect patches and its size is huge. This paper presents several methods to improve the automated program repair system Prophet, called Prophet+. Our approach contributes three improvements over Prophet: 1) extract twelve relations of statements and blocks for Bi-gram model using Big code, 2) prune the search space, 3) develop an algorithm to re-rank candidate patches in the search space. The experimental results show that our proposed system enhances the performance of Prophet, recognized as the state-of-the-art system, significantly. Specifically, for the top 1, our system generates the correct patches for 17 over 69 bugs while the number achieved by Prophet is 15.