Software repositories such as Git are significant sources of metadata about software projects, containing information such as modified files, change authors, and often commentary describing the change. An emerging approach to support software change impact analysis is to exploit this metadata to determine which files are linked by co-committal, i.e., when two files are frequently updated together within the same Git commit. Such information can serve as an indicator for identifying potential change-impact sets in future development activities. The aim of this study is to determine whether co-committal is a reliable indicator of links between software artifacts stored in Git and, if so, whether these links persist as the artifacts evolve—thereby offering a potentially valuable dimension for change impact analysis. To investigate this, we mined the metadata of five large Git repositories comprising over 14K commits and extracted co-change sets from the resulting data. The results show that: (1) co-committal links between artifacts vary widely in both strength and frequency, with these variations strongly influenced by the development style and activity levels of the contributing developers, and (2) although co-committal can serve as an indicator of evolutionary coupling in certain scenarios, its usefulness depends on project-specific development practices and observable patterns of developer behavior.
Price et al. (Sun,) studied this question.