Hybrid approach combines runtime and historical analysis to predict JavaScript code change impacts

When developers modify JavaScript code, understanding what else might break as a result is critical for testing and maintenance. A new research framework called Caprese demonstrates that the best approach to predicting these cascading effects combines two complementary analysis strategies rather than relying on just one.

The challenge with JavaScript#

JavaScript's dynamic features—callbacks, event handlers, asynchronous operations, and mutable shared state—make it unusually difficult to trace the true dependencies within an application. Unlike statically typed languages, the connections between different code sections can be opaque, making it hard to confidently predict what a change will affect downstream.

Two competing strategies#

Researchers evaluated two established techniques for mapping change impact. History-based analysis mines repositories to find patterns in how code sections have changed together in the past, revealing which parts tend to break when others are modified. Dynamic dependency analysis observes actual program execution to capture real runtime interactions and data flows.

Each method has tradeoffs: dynamic analysis captures what genuinely happens during testing but may miss relationships not exercised by the test suite, while historical analysis unearths evolutionary patterns but can suffer from noise and spurious co-change correlations.

Key finding: complementary strengths#

Testing the framework across ten open-source JavaScript applications revealed that the two approaches identify largely different candidates for inspection. With broader testing budgets, overlap between the two methods was only about 22%, suggesting they're capturing distinct signals about code dependencies.

Dynamic analysis generally delivered better precision—fewer false positives—while history-based techniques surfaced additional relevant files that dynamic analysis missed. This asymmetry points to a clear path forward: combining both signals yields more complete coverage than either alone.

Practical implications#

For software teams, the results argue against betting entirely on either runtime monitoring or historical co-change analysis. The hybrid approach implemented in Caprese shows that JavaScript change impact recommendation improves substantially when developers can draw on both evolutionary patterns and observed execution behavior to guide which code sections warrant testing and review.