DeepMerge: Learning to Merge Programs

In collaborative software development, program merging is the mechanism to integrate changes from multiple programmers. Merge algorithms in modern version control systems report a conflict when interfere textually. conflicts require manual intervention and frequently stall continuous integration pipelines. Prior work found that, although costly, large majority of resolutions involve re-arranging text without writing any new code. Inspired by this observation we propose the xmlns:xlink="http://www.w3.org/1999/xlink">first data-driven approach resolve merge with machine learning model. We realize our approach tool DeepMerge that uses novel combination (i) an edit-aware embedding inputs (ii) variation pointer networks, construct input segments. also algorithm localize resolved file employ it curate ground-truth dataset comprising 8,719 non-trivial JavaScript programs. Our evaluation shows on held out test set, can predict correct for 37% merges, compared only 4% state-of-the-art semistructured technique. Furthermore, subset merges upto 3 lines (comprising 24% total dataset), 78% accuracy.