Scalable Parametric Static Analysis

Parametric static analysis allows choosing a parameter value tobalance the precision and cost of the instantiated analysis. Wepropose an efficient approach to either find a cheapest parame-ter value to prove a given query or show that no such parame-ter value exists. Our approach is based on refinement, as in CE-GAR (counterexample-guided abstraction refinement), but appliesa novel meta-analysis to abstract counterexample traces to effi-ciently find parameter values that are incapable of proving thequery. We formalize our approach in a generic framework andapply it to two parametric analyses: a thread-escape analysis anda type-state analysis. The thread-escape analysis is implementedand applied to eight Java benchmarks comprising 2.5 MLOC. Ourexperiments show that our approach is effective in practice: forour four largest benchmarks, searching 2 parameter values foreach of 10K queries on average per benchmark, it finds a cheapestone for proving 46% queries and shows that none exists for 37%queries, in one minute per query on average.