Reducing Static Analysis Unsoundness with Approximate Interpretation
Explore a 20-minute video presentation from PLDI 2024 that introduces a novel dynamic analysis technique called approximate interpretation for improving static program analysis of JavaScript. Learn how this approach addresses challenges posed by dynamic property accesses in JavaScript, leading to more comprehensive call graph construction. Discover how the technique infers likely determinate facts about dynamic property accesses, particularly in complex library API initialization code, and utilizes this information to enhance static analysis. Examine the results of experiments conducted on 141 real-world Node.js-based JavaScript applications and libraries, showcasing significant improvements in call graph construction, including increased call edges, reachable functions, and resolved call sites. Gain insights into how approximate interpretation improves analysis recall from 75.9% to 88.1% for 36 JavaScript projects with available dynamic call graphs, while maintaining precision.