Would the Patch Be Quickly Merged?

Abstract

Code review is one of the most time-consuming and costly activities in modern software development. For the code submissions that can not be accepted by reviewers, developers need to re-modify the code again. Developers desire to minimize the time-cost that spends in the code review process. In some cases, a submission might be submitted many times and still not be accepted. The number of review times has serious implications for defect repairs and the progress of development. Therefore, a few recent studies focused on discussing factors that effect submission acceptance, while these prior studies did not try to predict submission acceptance or the number of review times. In this paper, we propose a novel method to predict the time-cost in code review before a submission is accepted. Our approach uses a number of features, including review meta-features, code modifying features and code coupling features, to better reflect code changes and review process. To examine the benefits of our method, we perform experiments on two large open source projects, namely Eclipse and OpenDaylight. Our results show that the proposed approach in the problem of predicting submission acceptance achieves an accuracy of 79.72%, 80.03% for Eclipse and OpenDaylight, respectively. For the prediction of review times ranges, our method achieves an accuracy of 66.42% and 60.42% for Eclipse and OpenDaylight, respectively.