Re: [GHC] #13985: GHC 8.0 regression: ‘k’ is not in scope during type checking, but it passed the renamer

[GHC]

#13985: GHC 8.0 regression: ‘k’ is not in scope during type checking, but it passed
the renamer
-------------------------------------+-------------------------------------
        Reporter:  RyanGlScott       |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler (Type    |              Version:  8.0.1
  checker)                           |
      Resolution:                    |             Keywords:  TypeFamilies
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Compile-time      |  Unknown/Multiple
  crash or panic                     |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:  #13738            |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by RyanGlScott):

 I tried looking at this, but I'm not nearly smart enough to sort out the
 chicken-and-egg problem that I uncovered. I //think// the solution would
 be to call
 [http://git.haskell.org/ghc.git/blob/d774b4e2de4f07d2432b67010305fede7aeefc78:/compiler/typecheck/TcHsType.hs#l2135
 reportFloatingKvs] somewhere in the code that typechecks data family
 instances (e.g.,
 [http://git.haskell.org/ghc.git/blob/d774b4e2de4f07d2432b67010305fede7aeefc78:/compiler/typecheck/TcInstDcls.hs#l622
 tcDataFamInstDecl] seems like a good candidate). However, there's a
 serious problem with that: `reportFloatingKvs` needs tyvars to do its
 thing, but `tcDataFamInstDecl` starts out only having access to type
 patterns, not tyvars.

 To obtain tyvars, `tcDataFamInstDecl` calls `tcFamTyPats`, which in turn
 calls `kcDataDefn` to kind-check the type patterns. But therein lies the
 problem: `kcDataDefn` kind-checks each of the data constructors! By the
 time we do this, it's too late, since `k` wasn't rejected yet, so when we
 hit `k` when kind-checking a data constructor's existential type
 variables, GHC can't find it and throws an internal error.

 That is, `reportFloatingKvs` needs tyvars that `tcFamTyPats` produces, but
 `tcFamTyPats` can't produce the tyvars because it hits a kind variable
 which should have been rejected earlier by `reportFloatingKvs`. Ugh.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13985#comment:1>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler