Type mapper, visitor, iterator, finder all in one.

The interface is wrapped into a monad, that allows, by choosing a proper monad implement all the above morphisms and iterators. If you're afraid of the monads, then calm down and continue reading, there is a way to use this interface without any monads.

Each syntactical element t:T of the type system is represented with three methods:

• enter_T t
• map_T t
• leave_T t

The map_T t method first calls enter_T t, the applies a deep mapping of the t to t' and finally calls leave_T t'.

Override enter_T if an element shouldn't be morphed. The combination of enter_t, leave_T allows to perform different visiting strategies. If mapping is needed then a map_T method should be overridden. A usual pattern would be:

class my_mapper = object(self)
  inherit base as super

  method map_T t =
    super#map_T t >>| self#my_transformation

  method private my_transformation t = t
end

All method calls are bound with monadic operations. This makes it possible to parametrize visitor with different computation strategies (depending on a binding strategy of a monad).

If the monad is a null monad, where bind is a reverse application, and returns is an identity, then we have a normal execution and the whole visitor degrade to a regular visitor/mapper without any monads. Such visitor is instantiated as C.Type.Mapper.base.

The other two useful monads, are State and Search. The visitor in these monads is instantiated as C.Type.Mapper.State.base and C.Type.Mapper.Finder.base correspondingly. The former is useful to implement a regular visitor, that will fold an abitrary value over the type structure, or to implement a mapper, that can also have a state. The latter, is useful for implementing a search with a shortcut, i.e., when the searching is terminated as soon as the target is found.