Beyond Isolated Changes: A Context-aware and Dependency-enhanced Code Change Detection Method

Code change detection is essential for understanding software evolution. However, existing techniques mainly concentrate on modifications at the entity level (like methods and statements) while often overlooking changes in dependencies and failing to group related changes that are scattered throughout the code. To tackle these issues, this paper introduces ChangeDelta, a code change detection method that is both context-aware and dependency-enhanced. ChangeDelta utilizes static analysis and rule-driven techniques to identify changes not only in code entities but also in their dependencies, including both intra- and inter-procedural dependencies. This approach enhances the semantic understanding of changes across different scopes. Additionally, we present a context-aware change association mechanism that aggregates individual changes, highlighting their logical relationships and the intent behind their implementation. Our evaluations on two Java benchmarks demonstrate that ChangeDelta outperforms leading tools such as RefactoringMiner and CodeShovel, achieving 95.2% precision and 98.0% recall in change detection. The results also show that ChangeDelta effectively associates related scattered changes. Moreover, our analysis of real-world commit history indicates that dependency changes occur twice as frequently as entity changes, underscoring their importance in the detection process. Overall, our work will help developers in navigating numerous discrete code modifications, providing clarity on the motivations and purposes behind these changes.