[ME-0004] Generics Traits

[illusory0x0]

No description provided.

[HCHogan]

I think there’s an important conceptual split that will help the design land cleanly:

Generic traits (multiparam typeclassess/rank-1 polymorphic methods) ≠ Higher-Kinded Types (HKT).

Functor, Monad, ... examples require quantifying over type constructors (Self :: Type -> Type), i.e. Fω/HKT, not just “traits with generic parameters.”

We can split the proposal into to independent capabilities:

Track 1 -- Generic Traits

// Actually fundep `Self Rhs -> Output` is needed to avoid ambiguity
trait HAdd[Rhs, Output] {
  add(Self, Rhs) -> Output;
}

No HKT required.

Track 2 -- Abstraction over type constructors (HKT)

// Self :: Type -> Type
trait Functor {
  fn[A,B] map(Self[A], (A) -> B raise?) -> Self[B] raise?
}

In this way we’ll need:

• Kind checking & partial application rules
• Clear inference story (be explicit; HM+HKT inference is hard)

[illusory0x0]

@HCHogan you are right.

I think this proposal should be divided into three proposals.

• Polymorphic methods

• Parameterized classes

• HKT allow parameterized other Kind such as (Type) -> Type, (Type,Type) -> Type, not only for Type,

The current Self keyword actually supports Kind poly, when writing the MoonBit method.

Related issues:

moonbitlang/core#2433 (comment)