Monadic input-output with OCaml. Why? There are many possibilities to implement monadic IO with OCaml. It's very annoying to copy and paste types and implementations over the projects, but the real reason was the "diamond problem": two libraries are using almost same implementation, but different types. How? There will be type signatures and implementations for monadic input-output using Lwt and using direct OCaml input-output (since OCaml type system implements "monad with errors", where monad that carries 'a is the 'a itself, and errors are propagated by exceptions). Details! Module IO_Types contains signatures: - IO_Monad -- just return, bind, bind_rev - IO_MonadError -- IO_Monad + error, catch - IO_Common (functions' signatures that are common to Lwt, Direct IO and possible future IO libraries) There are no need for signatures of IO_Lwt and IO_Direct, so there are no its signatures. Module implementations: - IO_Pure -- no Lwt, no -threads, no exceptions handling - IO_PureError -- IO_Pure + exceptions - IO_Lwt -- IO via Lwt - IO_Direct -- IO via OCaml's Pervasives + Unix + Sys + Thread modules IO_Lwt and IO_Direct must match all the signatures contained in IO_Types. Why "bind_rev"? "bind" function has signature "('a -> m 'b) -> m 'a -> m 'b". Note, that ">>=" is not "bind", and we should name this function anyhow. Let's name it "bind_rev". But for performance reasons we can't always do "let bind_rev m f = bind f m" , so it's up to you, how the bind and bind_rev will be implemented: one using another or independently.