Cheap (But Functional) Threads

This article demonstrates how a powerful and expressive abstraction from concurrency theory plays a dual rôle as a programming tool for concurrent applications and as a foundation for their verification. This abstraction—monads of resumptions expressed using monad transformers—is cheap: it is easy to understand, easy to implement, and easy to reason about. We illustrate the expressiveness of the resumption monad with the construction of an exemplary multitasking operating system kernel with process forking, preemption, message passing, and synchronization constructs in the pure functional programming language Haskell. Because resumption computations are stream-like structures, reasoning about this kernel may be posed as reasoning about streams, a problem which is well-understood. And, as an example, this article illustrates how a liveness property— fairness—is proven and especially how the resumption-monadic structure promotes such verifications. 1 “Cheap” Concurrency This paper presents a low overhead approach to multi-threaded concurrent computation. We demonstrate how a rich variety of concurrent behaviors—including synchronization, message passing, the forking and suspension of threads among others—may be expressed succinctly in a non-strict functional language with no changes to the host language implementation. The necessary machinery—monads of resumptions—is so expressive, in fact, that the kernel described here requires fewer than fifty lines of Haskell 98 code. And, because this machinery may be generalized as monad transformers, the functionality described here may be reused and refined easily. Many techniques and structures have emigrated from programming language theory to programming practice (e.g., types, CPS, etc.), and this article advocates that resumption monads make the journey as well. This work is not intended to be theoretical; on the contrary, its purpose is to demonstrate how a natural (but, perhaps, under-appreciated) computational model of concurrency is used to construct and verify concurrent applications. The approach taken here is informal and should be accessible to anyone familiar with monads in Haskell. And while the constructions 1 All the code presented in this paper is available from the author.