Parameters of types

You can define, for example, a type T of kind Constraint -> *. Then you can apply T to (Read a, Show a), for example, resulting in the type T (Read a, Show a) of kind *.

Instances of classes

Note that there is a common use of the term “instance” in the Haskell community that does not match the terminology used in the Haskell Report. Something like instance C T where […] is not an instance, and it also does not define an instance. Instead, it makes the type T and instance of the class C. Types are instances of classes. So when I said that contexts can be instances of classes, I meant that you can have, for example, an instance declaration instance C (Read a, Show a) where […], meaning that C must have kind Constraint -> Constraint.

Classes, partially applied classes, and contexts can be parameters of types and instances of classes.

Does it mean that “contexts can be parameters of instances of classes”? Or does it mean that “contexts can be instances of classes”? Either way I’d appreciate some elaboration.

The Monad class given (but not invented) by me uses not only the ConstraintKinds, but also the TypeFamilies extension. ConstraintKinds is very new, and also TypeFamilies is not very old. So I suppose that this generalized Monad class will not make it into base very soon. We would have to ask the folks on the GHC or the library mailing list to know for sure.

That said, relying on extensions for base does not seem to be a problem as such, as base is already beyond standard Haskell. For example, Control.Monad.ST uses Rank2Types.

The example you give of generalizing the Monad type class is fantastic, as it seems to have no downsides, since any previously allowable instance can use the existing instance definition. Apart from the dependence on ConstraintKinds, is there any negative to this Monad definition? Is there any movement to thus generalize it? If not, why not? A reluctance to rely on an extension for a base library?