An Empirical Study of Architectural Changes in Code Commits

The maintenance of software architecture is challenged by fast-delivery code changes since developers are rarely aware of the architectural impacts of their code changes. To ease the burdens of architects, in this work, we proposed a light-weight framework to identify changes in architectures from code commits automatically. The framework identifies architectural changes without heavy architecture recovery techniques. Instead, it only takes a code commit as input. The framework, on the one hand, can be integrated into prevalent continuous integration systems to monitor architectural changes. On the other hand, it can be plugged into code review systems to help developers realize the architectural changes they introduce. Based on the framework, we further conducted a large-scale empirical study on 368,847 commits of 16 Apache open projects to study architectural changes. Our study reveals several new findings regarding the frequency of architectural change commits, the common and risky intents under which developers introduce architectural changes, and the correlations of architectural changes with lines of code and number of modified source files in commits. Our findings provide practical implications for software contributors and shed light on potential research directions on architecture maintenance.