Introduction
Version Control Systems (VCS), particularly Distributed Version Control Systems (DVCS) like Git, are fundamental to professional software development. Consequently, educators frequently integrate these tools into computer science curricula to foster collaboration and prepare students for industry standards. However, a significant gap often exists between professional best practices—characterized by frequent, granular, and well-documented commits—and student usage patterns. This study investigates whether students leverage the iterative benefits of Git or merely treat it as a mandatory "submission system" for final code delivery.
Methodology
The researchers conducted a multi-faceted analysis of Git repositories from three courses at the University of Jyväskylä:
- Introduction to Software Engineering (SE): 72 undergraduate students (33 groups) with mandatory VCS training.
- Functional Programming (FP): 36 master's students (13 groups).
- Service Oriented Architectures & Cloud Computing (SOA&CC): 36 master's students (9 groups).
The analysis employed several techniques:
- Commit Activity Analysis: Gaussian kernel density estimation was used to visualize commit frequency over the semester and within specific lab sessions.
- Message Classification: Commits were manually categorized as useful (descriptive), trivial (e.g., "fixed"), or nonsense.
- Work Distribution: The "dominant user" phenomenon was measured by calculating the percentage of commits attributed to the most active student in each group.
- Repository Hygiene: Automated and manual inspections identified "garbage" files, such as binaries and IDE metadata, that should ideally be excluded via
.gitignore.
Results
Temporal Activity Patterns
Analysis revealed that student activity is heavily influenced by deadlines and lab schedules rather than continuous iterative development. In courses lacking strict VCS training, commit activity peaked sharply around submission deadlines, showing a periodic "bursty" pattern rather than a steady flow of work.
Furthermore, within individual lab sessions, students often delayed commits until the very end of the period. In the Functional Programming course, for instance, a significant activity peak occurred after the lab ended, as students rushed to store their work for the day.
Commit Quality and Repository Hygiene
The quality of version control usage was often subpar. Between 15% and 33% of commit messages were classified as trivial or nonsense. In the SOA&CC course, repositories were frequently bloated with non-versionable files, indicating that students viewed the repository as bulk storage for project folders rather than a source code manager.
The "Dominant User" Phenomenon
A striking finding was the lack of collaborative distribution. In many groups, a single "dominant user" was responsible for 60% to 100% of all commits. This suggests that while the group submitted the project via Git, the practical learning of the tool was often concentrated in one individual.
Discussion
The study highlights several barriers to effective VCS adoption in education. The "face-saving" barrier—where students avoid committing messy, unfinished code—likely discourages the granular commits seen in industry. Additionally, physical lab constraints, such as having only one computer per group, naturally lead to a single student handling all repository interactions.
Interestingly, the 2nd-year SE course, which included explicit VCS training, exhibited a smoother commit curve compared to advanced master's courses. This underscores the necessity of continuous, explicit instruction on how and why to use Git, regardless of the students' academic level.
Conclusion
To move beyond the "submission system" mentality, educators should integrate VCS training earlier and more explicitly. Future improvements could include reconfiguring lab environments to ensure all students have individual access to the repository and emphasizing the value of the commit history as a learning tool rather than just a final delivery mechanism.
Summary updated