[GHC] #12503: Template Haskell regression: GHC erroneously thinks a type variable is also a kind

[GHC]

#12503: Template Haskell regression: GHC erroneously thinks a type variable is also
a kind
-------------------------------------+-------------------------------------
        Reporter:  RyanGlScott       |                Owner:  goldfire
            Type:  bug               |               Status:  new
        Priority:  high              |            Milestone:  8.2.1
       Component:  Template Haskell  |              Version:  8.0.1
      Resolution:                    |             Keywords:  TypeInType
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  GHC rejects       |  Unknown/Multiple
  valid program                      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by RyanGlScott):

 I _think_ I know what's happening here. The kind equalities patch changed
 `FreeKiTyVars` from this:

 {{{#!hs
 type FreeKiTyVars = ([RdrName], [RdrName])
 }}}

 to this:

 {{{#!hs
 data FreeKiTyVars = FKTV { fktv_kis    :: [Located RdrName]
                          , _fktv_k_set :: OccSet  -- for efficiency,
                                                   -- only used internally
                          , fktv_tys    :: [Located RdrName]
                          , _fktv_t_set :: OccSet
                          , fktv_all    :: [Located RdrName] }
 }}}

 Now `FreeKiTyVars` is using two `OccSets`. (One each for the kind and type
 variables extracted so far? That's what I'm guessing, since `Note [Kind
 and type-variable binders]` is out of date.)

 The problem is that two exact `RdrName`s reified from Template Haskell can
 have the same underlying `Unique` but with different `Name`s (see `Note
 [Local bindings with Exact Names]`). That is what is shown in comment:7 —
 two exact `RdrName`s (`k`) have the same `Unique` (`vESeYM`), but
 different `Name`s (`a1lx` and `a1ly`).

 But as far as fixing this goes, I am a bit clueless. I suppose we could
 add `NameSet`s to `FreeKiTyVars` as well that track only `Exact`
 `RdrName`s, and if `lookupLocalRdrEnv` finds a hit for an `RdrName`, we
 could use the corresponding `NameSet`. Otherwise, fall back to the
 `OccSet`.

 Thoughts, Richard?

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/12503#comment:8>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler