Prioritizing pull requests

Abstract

Previous work showed that in the pull-based development model integrators face challenges with regard to prioritizing work in the face of multiple concurrent pull requests. We identified the manual prioritization heuristics applied by integrators and extracted features from these heuristics. The features are used to train a machine learning model, which is capable of predicting a pull request’s importance. The importance is then used to create a prioritized order of the pull requests. Our main contribution is the design and initial implementation of a prototype service, called PRioritizer, which automatically prioritizes pull requests. The service works like a priority inbox for pull requests, recommending the top pull requests the project owner should focus on. It keeps the pull request list up-to-date when pull requests are merged or closed. In addition, the service provides functionality that GitHub is currently lacking. We implemented pairwise pull request conflict detection and several new filter and sorting options e.g. the pull request’s size. A preliminary user study showed that the PRioritizer service, although it is positively evaluated as a whole, needs to give users more insight into how the priority ranking is established to make it really useful.