Testing Static Type Analysis Scalability in a Large Object-Oriented Language Codebase – An Experiment with Loupe
Shopify Engineering introduces an experiment on static type analysis scalability within large object-oriented codebases using their prototype tool, Loupe. They create two programs: one with classes and another without to analyze the impact of adding complexity through objects. The program with classes has 175K reachable functions (205K total), 3 million instructions, and megamorphic method lookups up to 144 classes; it takes 2.5 seconds on a single core for analysis completion. Comparatively, the program without classes contains only 200K functions and finishes in 1.3 seconds on a single core. Although these numbers are specific to their setup, they indicate promising scalability results that could potentially work with large human-crafted software after further optimization or selective implementation of advanced features like object sensitivity or increased call site sensitivity without significantly impacting runtime.
The authors suggest exploring related topics such as control flow analysis (CFA) and points-to analysis for deeper understanding in the vast field of program analysis. They also share resources, including a lecture by Ondřej Lhoták that provides an overview of this domain.\
Complete Article after the Jump: Here!