GitLab

From #18982 (closed):

{-# LANGUAGE GADTs #-}

module Lib where

data GADT a where
  GADT :: Int -> GADT Int

h :: GADT a -> Int
h (GADT n) = n
{-# NOINLINE h #-}

produces

Lib.$wh [InlPrag=NOINLINE]
  :: forall a. (a GHC.Prim.~# Int) -> GHC.Prim.Int# -> GHC.Prim.Int#
Lib.$wh
  = \ (@ a_sxO)
      (ww_sxS [Dmd=<L,A>] :: a_sxO GHC.Prim.~# Int)
      (ww1_sxW :: GHC.Prim.Int#) ->
      ww1_sxW

-- RHS size: {terms: 14, types: 18, coercions: 1, joins: 0/0}
h [InlPrag=NOUSERINLINE[0]] :: forall a. GADT a -> Int
h = \ (@ a_sxO) (w_sxP :: GADT a_sxO) ->
      case w_sxP of { GADT ww1_sxS [Dmd=<L,A>] ww2_sxT ->
      case ww2_sxT of { GHC.Types.I# ww4_sxW ->
      case Lib.$wh @ a_sxO @~ (ww1_sxS :: a_sxO GHC.Prim.~# Int) ww4_sxW
      of ww5_sy1
      { __DEFAULT ->
      GHC.Types.I# ww5_sy1
      }
      }
      }

The coercion argument to the worker $wh is absent, but still present in the simplified code. It would be good if we can change that, as it also makes the type argument @a dead. Then we end up with a much simpler worker and call site.

Not that it would impact runtime performance other than maybe through inlining decisions.

I think the problem has to do with "what kind of absent dummy should we pick for the coercion type?".