[GHC] #15116: GHC internal error when GADT return type mentions its own constructor name

[GHC]

#15116: GHC internal error when GADT return type mentions its own constructor name
-------------------------------------+-------------------------------------
        Reporter:  RyanGlScott       |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:  8.6.1
       Component:  Compiler (Type    |              Version:  8.5
  checker)                           |             Keywords:  GADTs,
      Resolution:                    |  TypeInType
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Compile-time      |  Unknown/Multiple
  crash or panic                     |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by RyanGlScott):

 Ah, I see what is going on. This is due to the following change in
 `kcLTyClDecl`:

 {{{#!diff
  kcLTyClDecl :: LTyClDecl GhcRn -> TcM ()
    -- See Note [Kind checking for type and class decls]
  kcLTyClDecl (L loc decl)
 +  | hsDeclHasCusk decl
 +  = traceTc "kcTyClDecl skipped due to cusk" (ppr tc_name)
 +  | otherwise
    = setSrcSpan loc $
      tcAddDeclCtxt decl $
 -    do { traceTc "kcTyClDecl {" (ppr (tyClDeclLName decl))
 +    do { traceTc "kcTyClDecl {" (ppr tc_name)
         ; kcTyClDecl decl
 -       ; traceTc "kcTyClDecl done }" (ppr (tyClDeclLName decl)) }
 +       ; traceTc "kcTyClDecl done }" (ppr tc_name) }
 +  where
 +    tc_name = tyClDeclLName decl
 }}}

 Now, if the declaration has a CUSK, then `kcTyClDecl` is skipped entirely.
 As a consequence, subsequent checks for each data constructor
 (`kcConDecl`) are also skipped. But `kcConDecl` was what was throwing the
 error message for the original program, so skipping it leads to
 pandemonium later down the line.

 How should we fix this? I can think of at least a couple of possibilities:

 1. Never skip `kcTyClDecl` if the declaration has constructors.
 2. Do the `kcConDecl` checks separately from `kcTyClDecl`.

 Thoughts?

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15116#comment:2>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler