Typeclasses and higher-kinded types in Wipple

Typeclasses are a powerful construct used in functional programming languages like Haskell...

Wipple


Typeclasses are a powerful construct used in functional programming languages like Haskell. Essentially, they allow you to overload functions so that they can operate on different types. Let’s take a look at Wipple’s show function:

show :: Show A => A -> ()

This means that show accepts any value where Show is defined for the value’s type. Well what is Show? It’s a trait, Wipple’s name for typeclasses. To create a trait, just provide a type definition:

A => Show A :: A -> Text

Note: This syntax is lightweight but ambiguous in some edge cases. If you want to add a type annotation on a value like f x :: A, you should wrap the value in parentheses: (f x) :: A.

In Wipple, a trait is a construct whose value depends on its type. We can provide multiple definitions for Show to define it for various types!

Show Text : it
Show Number : n -> ...
Show Boolean : b -> if b "True" "False"

Another good example of traits in action is the Equal trait, which can be defined to support equality on types:

A => Equal A :: A -> A -> Boolean

[operator 5] = : Equal  -- = :: Equal A => A -> A -> Boolean

Now let’s take a look at the | (map) operator. What do you think its type is?

[operator 7] | : functor -> f -> Functor f functor

First, what’s a functor??? Without getting too technical, a functor is basically a container that supports the map operation on its contents. For example, we can use | with Lists, Maybes, and Results:

'(1 2 3) | increment = '(2 3 4)

Some 1 | increment = Some 2
None | increment = None

Success 1 | increment = Success 2
Error "oh no" | increment = Error "oh no"

So, the job of the | operator is to take a container holding some value(s) A, and apply a function A -> B to the value(s) so that we end up with the same container holding the new value(s) B. To express this in the type system, we need a new feature — higher-kinded types! It looks like this:

(F _) => Functor F :: A => B => (A -> B) -> F A -> F B

Notice that instead of providing a type variable like A, we provide a “type function” F that takes an input _. We can then provide the input A to build a container F A, and so on. For example, we can define Functor for Maybe A, for any A:

A => Functor (Maybe A) : f -> x? -> when x? {
    Some x -> Some (f x)
    None -> None
}

Thus, the type of | is:

| :: Functor (F _) => A => B => F A -> (A -> B) -> F B

For now, only traits with a single parameter are supported. (This restriction is also present in vanilla Haskell because it causes ambiguity.) But that shouldn’t prevent very many use cases!

The new type function syntax

You may have noticed that I reworked the syntax for what used to be for functions. In the new syntax, you declare a single type variable with an optional set of traits for which the type is defined. This syntax is a little more concise while still making clear the difference between implicit and explicit type checking.

A => ...  -- introduce type variable A
Show A => ...  -- introduce type variable A where Show A is defined
(Show Equal) A => ...  -- introduce type variable A where Show A and Equal A are defined
(F _) => ...  -- introduce higher-kinded type F
Functor (F _) => ... -- introduce higher-kinded type F where Functor (F _) is defined

Conclusion

All in all, I’m really excited to start working on implementing this in the Wipple compiler. Once I have a working, type safe | operator, I can start work on pattern matching and then the standard library!