Demand-Driven Refinement of Points-to Analysis

Abstract

We present DynaSens, a demand-driven approach to points-to analysis that uses slicing to automatically adjust the analysis' context-sensitivity. Within a points-to analysis, heap-carried data flows are composed of loads and stores, and these heap-carried dependences are difficult to resolve. Having observed the limitations of existing techniques, we propose a slicing analysis based on a demand-driven approach to resolve such dependences. Given a points-to query, a collection of relevant program elements is identified by the slicing analysis and handled context-sensitively by the points-to analysis. We compare the precision and cost of our points-to analysis against two state-of-the-art uniformly context-sensitive analyses that achieve the best trade between cost and precision to date. Evaluation results shows the points-to analysis refined by the slicing analysis achieves higher precision in most tests than the uniformly context-sensitive analyses, which are many times more costly.