int-index originally spotted this bug at https://github.com/goldfirere/singletons/issues/150\#issuecomment-305909199. To reproduce, compile this file with GHC 8.0.1, 8.0.2, 8.2.1, or HEAD:

{-# LANGUAGE DataKinds #-}
{-# LANGUAGE ExistentialQuantification #-}
{-# LANGUAGE PolyKinds #-}
{-# LANGUAGE TemplateHaskell #-}
module Bug where

import Language.Haskell.TH

$(do TyConI (DataD _ _ [KindedTV a1 _] _ _ _) <- reify ''Maybe
     [f,a2] <- mapM newName ["f","a"]
     return [ SigD f (ForallT [KindedTV a2 (AppT (ConT ''Maybe) (VarT a1))]
                   [] (ConT ''Int))
            , ValD (VarP f) (NormalB (LitE (IntegerL 42))) []
            ])

GHCi, version 8.2.0.20170522: http://www.haskell.org/ghc/  :? for help
Loaded GHCi configuration from /home/rgscott/.ghci
[1 of 1] Compiling Bug              ( Bug.hs, interpreted )

Bug.hs:9:3: error:
    • GHC internal error: ‘a_11’ is not in scope during type checking, but it passed the renamer
      tcl_env of environment: []
    • In the first argument of ‘Maybe’, namely ‘a_11’
      In the kind ‘Maybe a_11’
      In the type signature: f :: forall (a_a4Qz :: Maybe a_11). Int
  |
9 | $(do TyConI (DataD _ _ [KindedTV a1 _] _ _ _) <- reify ''Maybe
  |   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^...

The root cause of the issue seems to be that the name a (which we picked for newName) happens to clash with the type variable we reified from Maybe (since data Maybe a = ...). If we pick a different name:

     [f,a2] <- mapM newName ["f","albatross"]

Then it will compile.

This is a regression from GHC 7.10.3, as it compiles in that version (with a slight change to accommodate the API differences in DataD between 7.10.3 and 8.0):

{-# LANGUAGE DataKinds #-}
{-# LANGUAGE ExistentialQuantification #-}
{-# LANGUAGE PolyKinds #-}
{-# LANGUAGE TemplateHaskell #-}
module Works where

import Language.Haskell.TH

$(do TyConI (DataD _ _ [KindedTV a1 _] _ _) <- reify ''Maybe
     [f,a2] <- mapM newName ["f","a"]
     return [ SigD f (ForallT [KindedTV a2 (AppT (ConT ''Maybe) (VarT a1))]
                   [] (ConT ''Int))
            , ValD (VarP f) (NormalB (LitE (IntegerL 42))) []
            ])

$ /opt/ghc/7.10.3/bin/ghci Works.hs -ddump-splices
GHCi, version 7.10.3: http://www.haskell.org/ghc/  :? for help
[1 of 1] Compiling Works            ( Works.hs, interpreted )
Works.hs:(9,3)-(14,13): Splicing declarations
    do { TyConI (DataD _ _ [KindedTV a1_a3jt _] _ _) <- reify ''Maybe;
         [f_a3nh, a2_a3ni] <- mapM newName ["f", "a"];
         return
           [SigD
              f_a3nh
              (ForallT
                 [KindedTV a2_a3ni (AppT (ConT ''Maybe) (VarT a1_a3jt))]
                 []
                 (ConT ''Int)),
            ValD (VarP f_a3nh) (NormalB (LitE (IntegerL 42))) []] }
  ======>
    f_a4vc :: forall (a_a4vd :: Maybe a_a4v3). Int
    f_a4vc = 42
Ok, modules loaded: Works.