BDD-based software verification

Abstract

In software model checking, most successful symbolic approaches use predicates as representation of the state space, and SMT solvers for computations on the state space; BDDs are often used as auxiliary data structure. Although BDDs are applied with great success in hardware verification, BDD representations of software state spaces were not yet thoroughly investigated, mainly because not all operations that are needed in software verification are efficiently supported by BDDs. We evaluate the use of a pure BDD representation of integer values, and focus on a particular class of programs: event-condition-action (ECA) programs with limited operations. A symbolic representation using BDDs seems appropriate for ECA programs under certain conditions. We configure a program analysis based on BDDs and experimentally compare it to four approaches to verify reachability properties of ECA programs: an explicit-value analysis, a symbolic bounded-loops analysis, a predicate-abstraction analysis, and a predicate-impact analysis. The results show that BDDs are efficient for a restricted class of programs, which yields the insight that BDDs could be used selectively for variables that are restricted to certain program operations (according to the variable's domain type), even in general software model checking. We show that even a naive portfolio approach, in which after a pre-analysis either a BDD-based analysis or a predicate-impact analysis is performed, outperforms all above-mentioned analyses.