Understanding the Impact of Branch Edit Features for the Automatic Prediction of Merge Conflict Resolutions.

Abstract

Developers regularly have to resolve merge conflicts, i.e., two conflicting sets of changes to the same files in different branches, which can be tedious and error-prone. To resolve conflicts, developers typically: keep the local version (KL) or the remote version (KR) of the code. They also sometimes manually edit both versions into a single one (ME). However, most existing techniques only support merging the local and remote versions (the ME strategy). We recently proposed RPRedictor, a machine learning-based approach to support developers in choosing how to resolve a conflict (by KL, KR, or ME), by predicting their resolution strategy. In its original design, RPRedictor uses a set of Evolution History Features ( s) that capture: the magnitude of the changes in conflict, their evolution, and the experience of the developers involved. In this paper, we proposed and evaluated a new set of Branch Edit Features ( s), that capture the fine-grained edits that were performed on each branch of the conflict. We learned multiple lessons. First, s provided lower effectiveness (F-score) than the original s. Second, combining s with s still did not improve the effectiveness of s, it provided the same f-score. Third, the feature set that provided highest effectiveness in our experiments was the combination of with a subset of s that captures the number of insertions performed in the local branch, but this combination only improved s by 3 pp. f-score. Finally, our experiments also share the lesson that some feature sets provided higher C-score (i.e., the safety of the technique’s mistakes) as a trade-off for lower f-scores. This may be valued by developers and we believe that it should be studied in the future.