Chapter 7. Purely functional parallelism


Because modern computers have multiple cores per CPU, and often multiple CPUs, it’s more important than ever to design programs in such a way that they can take advantage of this parallel processing power. But the interaction of programs that run with parallelism is complex, and the traditional mechanism for communication among execution threads—shared mutable memory—is notoriously difficult to reason about. This can all too easily result in programs that have race conditions and deadlocks, aren’t readily testable, and don’t scale well.

In this chapter, we’ll build a purely functional library for creating parallel and asynchronous computations. We’ll rein in the complexity inherent in parallel programs by describing them using only pure functions. This will let us use the substitution model to simplify our reasoning and hopefully make working with concurrent computations both easy and enjoyable.

7.1. Choosing data types and functions

7.2. Picking a representation

7.3. Refining the API

7.4. The algebra of an API

7.5. Refining combinators to their most general form

7.6. Summary