Context-Sensitive Dynamic Partial Order Reduction

Dynamic Partial Order Reduction (DPOR) is a powerful technique used in verification and testing to reduce the number of equivalent executions explored. Two executions are equivalent if they can be obtained from each other by swapping adjacent, non-conflicting (independent) execution steps. Existing DPOR algorithms rely on a notion of independence that is context-insensitive, i.e., the execution steps must be independent in all contexts. In practice, independence is often proved by just checking no execution step writes on a shared variable. We present context-sensitive DPOR, an extension of DPOR that uses context-sensitive independence, where two steps might be independent only in the particular context explored. We show theoretically and experimentally how context-sensitive DPOR can achieve exponential gains.