Efficient module-level dynamic analysis for dynamic languages with module recontextualization

Efficient module-level dynamic analysis for dynamic languages with module recontextualization

연구 분야: Analysis

학회: ESEC/FSE 2021: Proceedings of the 29th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering

Dynamic program analysis is a long-standing technique for obtaining information about program execution. We present module recontextualization, a new dynamic analysis approach that targets modern dynamic languages such as JavaScript and Racket, enabled by the fact that they feature a module-import mechanism that loads code at runtime as a string. This approach uses lightweight load-time code transformations that operate on the string representation of the module, as well as the context to which it is about to be bound, to insert developer-provided, analysis-specific code into the module before it is loaded. This code implements the dynamic analysis, enabling this approach to capture all interactions around the module in unmodified production language runtime environments. We implement this approach in two systems targeting the JavaScript and Racket ecosystems. Our evaluation shows that this approach can deliver order-of-magnitude performance improvements over state-of-the-art dynamic analysis systems while supporting a range of analyses, implemented on average in about 100 lines of code.