GitLab

servant-0.19 fails to build on GHC HEAD. Here is a minimized version of the issue:

{-# LANGUAGE AllowAmbiguousTypes      #-}
{-# LANGUAGE DataKinds                #-}
{-# LANGUAGE Haskell2010              #-}
{-# LANGUAGE PolyKinds                #-}
{-# LANGUAGE StandaloneKindSignatures #-}
{-# LANGUAGE TypeFamilies             #-}
{-# LANGUAGE TypeOperators            #-}
{-# LANGUAGE UndecidableInstances     #-}

module Bug where

import Data.Kind
import GHC.TypeLits

-- | No instance exists for @tycls (expr :> ...)@ because
-- @expr@ is not recognised.
type NoInstanceForSub (tycls :: k) (expr :: k') =
  Text "There is no instance for " :<>: ShowType tycls
  :<>: Text " (" :<>: ShowType expr :<>: Text " :> ...)"

-- | No instance exists for @expr@.
type NoInstanceFor (expr :: k) =
  Text "There is no instance for " :<>: ShowType expr

-- | No instance exists for @tycls (expr :> ...)@ because @expr@ is not fully saturated.
type PartialApplication (tycls :: k) (expr :: k') =
  NoInstanceForSub tycls expr
  :$$: ShowType expr :<>: Text " expects " :<>: ShowType (Arity expr) :<>: Text " more arguments"

-- The arity of a combinator, i.e. the number of required arguments.
type Arity (ty :: k) = Arity' k

type Arity' :: k -> Nat
type family Arity' (ty :: k) :: Nat where
  Arity' (_ -> ty) = 1 + Arity' ty
  Arity' _ = 0

data (path :: k) :> (a :: *)
class HasLink endpoint where
instance TypeError (PartialApplication HasLink arr) => HasLink ((arr :: a -> b) :> sub)

This compiles with GHC 9.0.2 and 9.2.1, but fails to compile with HEAD:

$ ~/Software/ghc-9.3.20220216/bin/ghc Bug.hs
[1 of 1] Compiling Bug              ( Bug.hs, Bug.o )

Bug.hs:40:10: error:
    • There is no instance for HasLink (arr :> ...)
      arr expects 1 + Arity' b more arguments
    • In the ambiguity check for an instance declaration
      In the instance declaration for ‘HasLink ((arr :: a -> b) :> sub)’
   |
40 | instance TypeError (PartialApplication HasLink arr) => HasLink ((arr :: a -> b) :> sub)
   |          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

I'm unclear if this change is intended behavior or not, but since I couldn't find any mention of TypeError's behavior changing in the 9.4 release notes, I decided to open an issue.