Retracing the Semantics of CSP

Abstract

CSP was originally introduced as a parallel programming language in which sequential imperative processes execute concurrently and communicate by synchronized input and output. The influence of CSP and the closely related process algebra TCSP is widespread. Over the years CSP has been equipped with a series of denotational semantic models, involving notions such as communication traces, failure sets, and divergence traces, suitable for compositional reasoning about safety properties and deadlock analysis. We revisit these notions (and review some of the underlying philosophy) with the benefit of hindsight, and we introduce a semantic framework based on action traces that permits a unified account of shared memory parallelism, asynchronous communication, and synchronous communication. The framework also allows a relatively straightforward account of (a weak form of) fairness, so that we obtain models suitable for compositional reasoning about liveness properties as well as about safety properties and deadlock. We show how to incorporate race detection into this semantic framework, leading to models more independent of hardware assumptions about the granularity of atomic actions.