[GHC] #13959: substTyVar's definition is highly dubious

[GHC]

#13959: substTyVar's definition is highly dubious
-------------------------------------+-------------------------------------
        Reporter:  RyanGlScott       |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.0.1
      Resolution:                    |             Keywords:
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):

 * comment:2: Sorry, I shouldn't have used this example, since it's not
 actually tripping up the `ASSERT`. In fact, it does close over the kinds
 of the type variables, which you can see by looking at the
 [http://git.haskell.org/ghc.git/blob/f656fba19d0cefe05643ddea35d080ea332a6584:/compiler/types/TyCoRep.hs#l1897
 definition of zipTvSubst]:

 {{{#!hs
 zipTvSubst :: [TyVar] -> [Type] -> TCvSubst
 zipTvSubst tvs tys
   | debugIsOn
   , not (all isTyVar tvs) || neLength tvs tys
   = pprTrace "zipTvSubst" (ppr tvs $$ ppr tys) emptyTCvSubst
   | otherwise
   = mkTvSubst (mkInScopeSet (tyCoVarsOfTypes tys)) tenv
   where
     tenv = zipTyEnv tvs tys
 }}}

   It uses `tyCoVarsOfTypes` to get the tyvars for the in-scope set, so
 they will include the necessary kind variables. Phew.

 * comment:3. Aha! What I should have done is split `substTyVar` into a
 checked version and an unchecked version, and use the unchecked version in
 `subst_ty` (`subst_ty` is already using `substTyVarBndrUnchecked`
 [http://git.haskell.org/ghc.git/blob/f656fba19d0cefe05643ddea35d080ea332a6584:/compiler/types/TyCoRep.hs#l2185
 here], so this doesn't make things any less palatable than before).

   I'll post an update after I've ran `./validate` with this change.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13959#comment:5>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler