Characterizing Commits in Open-Source Software

Mining software repositories has been the basis of many studies on software engineering. Many of these works rely on commits’ data extracted since commit is the basic unit of information about activities performed on the projects. However, not knowing the characteristics of commits may introduce biases and threats in studies that consider commits’ data. This work presents an empirical study to characterize commits in terms of four aspects: the size of commits in the total number of files; the size of commits in the number of source-code files, the size of commits by category; and the time interval of commits performed by contributors. We analyzed 1M commits from the 24 most popular and active Java-based projects hosted on GitHub. The main findings of this work show that: the size of commits follows a heavy-tailed distribution; most commits involve one to 10 files; most commits affect one to four source-code files; the commits involving hundreds of files not only refer to merge or management activities; the distribution of the time intervals is approximately a Normal distribution, i.e., the distribution tends to be symmetric, and the mean is representative; in the average, a developer proceed a commit every eight hours. The results of this study should be considered by researchers in empirical works to avoid biases when analyzing commits’ data. Besides, the results provide information that practitioners may apply to improve the management and the planning of software activities.