[GHC] #15852: Bad axiom produced for polykinded data family

[GHC]

#15852: Bad axiom produced for polykinded data family
-------------------------------------+-------------------------------------
        Reporter:  goldfire          |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.7
      Resolution:                    |             Keywords:  TypeFamilies,
                                     |  TypeInType
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by RyanGlScott):

 Oh dear. Yes, I //was// barking up the wrong tree—thank you for setting me
 straight.

 Now that I dig closer into `ppr_co_axiom_branch` (which is the real
 culprit), I now understand what went wrong. When implementing
 `etaExpandFamInstLHS`, I left
 [http://git.haskell.org/ghc.git/blob/d30352add1da67dd0346613853cd423c7becbaeb:/compiler/types/Type.hs#l3067
 this comment]:

 {{{#!hs
 -- @Note [Eta reduction for data families]@ in "FamInstEnv". Because the
 -- right-hand side is where the eta-reduced variables are obtained from,
 it
 -- is not returned from this function (as there is never a need to modify
 it).
 }}}

 In retrospect, I have no idea what I was smoking when I wrote this
 comment, because it's completely wrong. The right-hand side //is// eta-
 reduced, and that's what we're seeing in `-ddump-tc`. (Moreover, we aren't
 obtaining the eta-reduced variables from the RHS at all. They're actually
 being taken from the `tyConTyVars` of the representation `TyCon`.)

 This suggests a straightforward fix: just eta expand the RHS as well, and
 return that from `etaExpandFamInstLHS` (or really, it just should be
 called `etaExpandFamInst`). I'll give that a shot.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15852#comment:5>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler