Code merging using transformations and member identity

Abstract

Conventionally, merging code files is performed using generic line-based merging algorithms (e.g., diff3) that are unaware of the syntax and semantics of the programming language, outputting conflicts that could be avoided. Structured and semistructured merging techniques are capable of reducing conflicts, but they still suffer from false positives (conflicts that could be avoided) and false negatives (conflicts that go undetected). We propose a merging technique that combines semistructured and transformation-based strategies, where conflict detection is aware of semantic aspects of the programming language. We extract transformations of two branches and apply a merging process that analyzes incompatible transformations, avoiding false positives and false negatives that occur in existing approaches. We developed Jaid, a prototype merging tool for Java based on the assumption that structural code elements evolve with attached UUIDs (representing identity). We performed an early experiment with 63 merge scenarios from two open-source projects to test the technique and assess its feasibility.