[GHC] #15591: Inconsistent kind variable binder visibility between associated and non-associated type families

[GHC]

#15591: Inconsistent kind variable binder visibility between associated and non-
associated type families
-------------------------------------+-------------------------------------
        Reporter:  RyanGlScott       |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:  8.6.1
       Component:  Compiler          |              Version:  8.5
      Resolution:                    |             Keywords:
                                     |  TypeApplications, TypeFamilies
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):

 Replying to [comment:5 goldfire]:
 > As for the patch: I would expect that the solution to this problem would
 be in the type-checker, not the renamer. As it stands, I'm pretty sure
 that the renamer identifies the `a` in these associated type definitions
 with the same unique as the `a` in the class.

 Well, it does identify the `a` in `T2` to be the same `a` as in `C`. But
 after it does so, it immediately filters it out! That's problematic,
 because by the time you reach `kcLHsQTyVars`, the `hsq_implicit` field of
 the `LHsQTyVars` argument only contains `f`—`a` is no longer within reach.
 (You can verify this for yourself by compiling this program with `-ddump-
 tc-trace` and searching for `kcLHsQTyVars: cusk`.)

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15591#comment:9>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler