andyscott/mercator

Automatic creation of typeclass evidence for monad-like types

Mercator

Mercator is a macro for automatically constructing evidence that a known type may be used in a for-comprehension, for abstracting over monad-like types with no impact on performance.

This allows us to write code against generic type constructors which can assume that they are usable in a for-comprehension, without having the evidence that they are until the application point when the type constructor is known, at which point, Mercator will construct it.

Usage

It is not possible to write code such as,

// does not compile
def increment[F[_]](xs: F[Int]) = for(x <- xs) yield x + 1

because the compiler is not able to enforce the constraint that the type constructor F[_] provides the methods map and flatMap (with the correct signatures) which are necessary for the for-comprehension to compile.

With Mercator, it is possible to demand an implicit instance of Monadic[F] to enforce this constraint. Mercator will automatically instantiate such an instance at the use-site for any type which has the required methods, like so,

import mercator._
def increment[F[_]: Monadic](xs: F[Int]) = for(x <- xs) yield x + 1

The methods flatMap and map will be provided to the instance of F[_] as extension methods, using an implicit value class in the mercator package. This incurs no allocations at runtime, and the performance overhead should be zero or negligible.

Point

An instance of Monadic[F] will generate an implementation of point, which constructs a new instance of the type from a single value. This implementation assumes the existence of an apply method on the type's companion object, and that applying the value to it will produce a result of the correct type.

If this is not the case, Mercator will try to find a unique subtype of F[_] whose companion object has an apply method taking a single value and returning the correct type. In the case of Either or Scalaz's \/, this will do the right thing.