Skeletons and the Anatomy of Monads

Monads are used heavily in Haskell for supporting computational effects, and the language offers excellent support for defining monadic computations. Unfortunately, defining a monad remains a difficult challenge. There are no libraries that a programmer can use to define a monad that is not a composition of existing monad transformers; therefore every such effort must start from scratch despite that all monads share the same structure and need to satisfy the same minimum set of properties. I propose a monadic programming framework called skeletons to simplify the task of defining monads and monad transformers. Skeletons provide a modular term representation that can be used for all monadic computations, and observer functions that implement the basic semantics of monad operators and guarantee their compliance with monad laws. Using the skeleton framework frees programmers from the semantic boilerplate in defining monads so that they can focus on the more important and more interesting task of implementing computational effects.