GHC HEAD regression: piResultTys1

The following program typechecks on GHC 8.0.2 through 8.6.3, but panics on HEAD:

{-# LANGUAGE GADTs      #-}
{-# LANGUAGE TypeInType #-}
module Bug where

import Data.Kind

data T f :: f Type -> Type where
  MkT :: T f a

$ ~/Software/ghc4/inplace/bin/ghc-stage2 Bug.hs
[1 of 1] Compiling Bug              ( Bug.hs, Bug.o )
ghc-stage2: panic! (the 'impossible' happened)
  (GHC version 8.7.20190219 for x86_64-unknown-linux):
        piResultTys1
  k_a10k[tau:1]
  [*]
  Call stack:
      CallStack (from HasCallStack):
        callStackDoc, called at compiler/utils/Outputable.hs:1159:37 in ghc:Outputable
        pprPanic, called at compiler/types/Type.hs:1063:5 in ghc:Type

changed weight to 10

added Tbug Trac import TypeInType typechecker labels

changed the description

This regression was introduced in commit 68278382 (Make a smart mkAppTyM).

Interestingly this is a consequence of the broken-ness described in #16344 (closed). Fixing the latter fixes this ticket too.

In conversation, Simon and I uncovered an unstated invariant: the tyVarKind of occurrences of a bound tyvar must share the tyVarKind of the binding site. In this example, that is not the case, because the kind of the tyvar contains a metavariable; it is zonked at an occurrence but not at the binding site.

The fix is easy: do an extra zonk in kcLHsQTyVars_NonCusk, but fixing #16344 (closed) will also make this problem melt away.

As expected, this issue was fixed in !436 (merged) / commit 1f5cc9dc. I'll prepare a regression test.

marked this issue as related to #16344 (closed)

mentioned in merge request !525 (merged)

!525 (merged)

mentioned in commit 0d4eae05

mentioned in commit 6336388a

closed via commit 4db9bdd9

changed milestone to %8.10.1

added compiler crash label

mentioned in commit 4db9bdd9

added Phighest label