Enhancing the Efficiency of Automated Program Repair via Greybox Analysis

In this paper, we pay attention to the efficiency of automated program repair (APR). Recently, an efficient patch scheduling algorithm, Casino, has been proposed to improve APR efficiency. Inspired by fuzzing, Casino adaptively chooses the next patch candidate to evaluate based on the results of previous evaluations. However, we observe that Casino utilizes only the test results, treating the patched program as a black box. Inspired by greybox fuzzing, we propose a novel patch-scheduling algorithm, Gresino, which leverages the internal state of the program to further enhance APR efficiency. Specifically, Gresino monitors the hit counts of branches observed during the execution of the program and uses them to guide the search for a valid patch. Our experimental evaluation on the Defects4J benchmark and eight APR tools demonstrates the efficacy of our approach.