Practical Demand-Driven Program Analysis with Recursion

Demand-Driven Program Analysis (DDPA) is a context-sensitive control-flow analysis for higher-order functional programming languages in the same space as k-CFA and related analyses. DDPA is provably sound and polynomial, but that alone does not guarantee its applicability to real-world programs. This project studies DDPA’s practicality in two dimensions: expressiveness (the analysis closely approximates the run-time) and performance (the analysis is fast to compute). To address expressiveness, we extended the analysis to directly support recursion, eliminating the need for encodings. To address performance, we developed a higher-level programming language that compiles to DDPA’s core language, ported over benchmarks that exercise all its features, and measured the reference implementation’s performance, comparing to other state-ofthe-art program analyses. DDPA is competitive and, in some cases, faster than the alternatives.