A type-and-effect system for asynchronous, typed events

Abstract

Implicit concurrency between handlers is important for event driven systems because it helps simultaneously promote modularity and scalability. Knowing the side-effect of the handlers is critical in these systems to avoid concurrency hazards such as data races. As event systems are dynamic because statically known and unknown handlers can register at almost any time during program execution, static effect analyses must reconcile over competing goals such as precision, soundness and modularity. We recently developed asynchronous, typed events, a system that can analyze the effects of the handlers at runtime. This mechanism utilizes runtime information to enable precise effect computation and greatly improves concurrency between handlers. In this paper, we present the formal underpinnings of asynchronous, typed events, and examine the concurrency safety properties it provides. The technical innovations of our system include a novel effect system to soundly approximate the dynamism introduced by runtime handlers registration, a static analysis to precompute the effects and a dynamic analysis that uses the precomputed effects to improve concurrency. Our design simplifies modular concurrency reasoning and avoids concurrency hazards.