A manual inspection of Defects4J bugs and its implications for automatic program repair

Published in Science China Information Sciences, 2019

Jiajun Jiang, Yingfei Xiong, Xin Xia

Brief Intro.

Automatic program repair techniques, which target to automatically generate correct patches for real-world defects, have gained a lot of attention in the last decade. Though many different techniques and tools have been proposed and developed, the ability of current techniques is still limited with low recall.

To better understand how far can we go on automatic program repair (APR), and identify useful insights and concrete directions for future research, we conducted a manual inspection on 50 real-word bugs from Defects4J benchmark, from which we summarized a set of fault localization and patch generation strategies that can guide future research directly.

[Paper] [Resources]