We present the results of an empirical study evaluating the precisionof subset-based points-to analysis with several variations of contextsensitivity on Java benchmarks of significant size. We compare theuse of call site strings as the context abstraction, object sensitivity,and the BDD-based context-sensitive algorithm proposed by Zhu andCalman, and by Whaley and Lam. Our study includes analyses that context-sensitivelyspecialize only pointer variables, as well as ones that also specializethe heap abstraction. We measure both characteristics of the points-tosets themselves, as well as effects on the precision of client analyses.To guide development of efficient analysis implementations, we measurethe number of contexts, the number of distinct contexts, and thenumber of distinct points-to sets that arise with each context sensitivityvariation. To evaluate precision, we measure the size of the callgraph in terms of methods and edges, the number of devirtualizablecall sites, and the number of casts statically provable to be safe.
CITATION STYLE
Lhoták, O., & Hendren, L. (2006). Context-sensitive pointsLhoták, O., & Hendren, L. (2006). Context-sensitive points-to analysis: is it worth it? Compiler Construction, 47–64. https://doi.org/10.1007/11688839-to analysis: is it worth it? Compiler Construction, (March 2006), 47–64. Retrieved from http://link.springer.com/chapter/10.1007/11688839_5
Mendeley helps you to discover research relevant for your work.