Hybrid partial order reduction with under-approximate dynamic points-to and determinacy information

Abstract

Verification techniques for concurrent systems are often based on systematic state space traversal. An important piece of such techniques is partial order reduction (POR). Many algorithms of POR have been already developed, each having specific advantages and drawbacks. For example, fully dynamic POR is very precise but it has to check every pair of visible actions to detect all interferences. Approaches involving static analysis can exploit knowledge about future behavior of program threads, but they have limited precision. We present a new hybrid POR algorithm that builds upon (i) dynamic POR and (ii) hybrid field access analysis that combines static analysis with data taken on-the-fly from dynamic program states. The key feature of our algorithm is usage of under-approximate dynamic points-to and determinacy information, which is gradually refined during a run of the state space traversal procedure. Knowledge of dynamic points-to sets for local variables improves precision of the field access analysis. Our experimental results show that the proposed hybrid POR achieves better performance than existing techniques on selected benchmarks, and it enables fast detection of concurrency errors.