Modular Stacking-based Context-Sensitive Program Analysis

This paper presents a systematic approach to scaling stackingbased context-sensitive program analysis yielded by weighted pushdown systems, and demonstrates its effectiveness by applications to Java pointsto analysis. Our approach relies on the view offormulating stacking-based analysis as AGPs (abstract grammar problem), and reducing the analysis problem to fixed-point calculation over the equation system encoded by grammar productions. To characterize points-to analysis that does not assume an existing program control flow, we propose Contracted AGP, and a symbolic sliding-window based algorithm for it. The algorithm aims at reducing both time and memory costs, and its correctness is proved in terms of chaotic iteration. Finally, we instantiated the algorithm within Japot, a context-sensitive points-to analyzer for Java. Experiments show two-fold benefits of our approach in practice, with (i) scaling Japot to Dacapo benchmark suite, and (ii) a speed-up of 3x faster in average, given an adjustable memory budget.