Hybrid change impact analysis using patterns of change in software history

Change impact analysis is essential for predicting the impact of change in complex and dynamic applications. Balancing the precision and completeness of the analysis is challenging, especially for a highly-dynamic programming language such as JavaScript. Existing impact analyses rely on techniques such as dynamic analysis and mining software repositories. Despite the benefits, dynamic analysis misses dependencies that are not exercised, and mining software repositories may yield imprecise results. In this paper, we propose a novel function-level hybrid impact analysis for JavaScript that utilizes the advantages of both techniques. For identifying the potential impact of a change set, our approach first extracts recurring patterns of change in software history that involve the change set. Next, it infers a dependency graph of the software using dynamic analysis. Integrating the findings of analyzing the patterns of change in software history with the dynamic code dependencies, our proposed approach provides a ranked impact set. We implement our approach in a tool named Caprese, which we then evaluate using 10 open-source Node.js applications. The results show that both dynamic analysis and frequent pattern detection units contribute uniquely to the approach. We also demonstrate that Caprese outperforms a state-of-the-art impact analysis tool by achieving 59% to 84% higher precision and 41% to 64% higher recall, on average.