The hidden challenges of merging: a large-scale empirical study across merge tools.

Abstract

Software development has shifted from centralized to collaborative approaches, with Ver sion Control Systems (VCSs), such as Git, facilitating integration through merge operations. However, merge conflicts that arise during code integration may delay projects and reduce productivity. This study investigates factors influencing merge conflicts by analyzing merge outcomes from three tools: Git, jFSTMerge, and IntelliMerge. Git is a widely used text based tool, whereas jFSTMerge and IntelliMerge are state-of-the-art Java merge tools de signed to handle more complex scenarios. We analyzed over 507,000 merge scenarios from 530 Java projects on GitHub, considering five factors: the number of evolutionary commits (the commits with changes that lead to a merge conflict), the number of developers, the num ber of refactorings, the number of refactoring regions, and the number of refactorings in the left and right merge parents. While refactoring aims to improve code quality, it may also intensify these conflicts. We found evidence that nearly 43% of Git, 32% of jFSTMerge, and 37% of IntelliMerge conflicting scenarios may be related to refactorings. Additionally, Git’s conflicting scenarios involving refactorings were also associated with longer resolution times. Refactorings show a strong association with merge conflicts across all tools. After controlling for change-magnitude variables, such as code churn and the number of modified files, refactorings remain strongly associated with merge conflicts. The presence of refac torings also increased the likelihood of a conflict to about 58% in our Git dataset. Among the 62 refactoring types identified, a consistent subset (33.8%), including renaming, moving classes, and extracting methods, frequently appears in conflicts across all tools. We also clas sify refactorings by tactic as pure or floss, finding that 65% of conflicting scenarios involve floss refactorings. Additionally, conflicts involving floss refactorings tend to require slightly longer resolution times. Overall, our results suggest that refactoring is not only a practice aimed at improving code quality, but also a factor associated with the dynamics of the merg ing process. These findings highlight the importance of carefully planning and coordinating refactorings. Developers may use these observations to define merging guidelines, while researchers may build on them to design or improve merge tools better equipped to handle refactoring-related changes.