GitLab

Today, I've been bitten again by the fact that I can write

{-# LANGUAGE GeneralisedNewtypeDeriving #-}
{-# LANGUAGE RoleAnnotations #-}

module Lib where

import Data.Coerce

class C a where
  emp :: a

instance C [a] where
  emp = []

newtype MyList a = MyList [a]
  deriving C
type role MyList representational -- just to be sure this isn't an issue

But I can't write

empty :: [a]
empty = []

myEmpty :: MyList a
myEmpty = coerce empty

It works if I insert the necessary -XTypeApplications. But that is so ugly! I really want it to be inferred for me, just as the deriving strategy does (and hence GHC already knows what to do):

==================== Derived instances ====================
Derived class instances:
  instance Lib.C (Lib.MyList a) where
    Lib.emp :: Lib.MyList a_ago
    Lib.emp
      = GHC.Prim.coerce
          @([a_ago]) @(Lib.MyList a_ago) (Lib.emp @([a_ago]))

This came up while I wanted to have

newtype Prio = Prio Int
newtype PrioMap a = PrioMap (IntMap a)

insertPrioMap :: Prio -> a -> PrioMap a -> PrioMap a
insertPrioMap = coerce IntMap.insert

Which also doesn't work without further elaboration.