The 'impossible' happened: type-inference succeeds somehow in code which isn't type-safe

While trying to figure out how Haskell handles OCaml's value restriction, I created some code which I thought shouldn't type-check.

It did, though, and makes GHCi panic on execution, or makes GHC fail during compilation.

Minor changes to the code makes it no longer type-check, as expected.

Here's the code, including some comments which show when compilation does fail as expected:

{-# LANGUAGE ScopedTypeVariables #-}

import Data.IORef

main :: IO ()
main = do
    (r :: IORef (t -> t)) <- newIORef id
    -- r <- newIORef i -- => Type-check error

    writeIORef r (\v -> if v then False else True)

    c <- readIORef r

    print $ c True

    print $ c 1234

    -- print $ c Nothing -- => Type-check error
    -- print $ c (1 :: Int) -- => Type-check error

When replacing the "print $ c 1234" line with one of the last 2 lines, type-checking fails. When removing the explicit type-annotation on 'r', type-checking fails when "print $ c 1234" is left in place.

Here's the GHCi and GHC output:

Prelude> :load demo1.hs
[1 of 1] Compiling Main             ( demo1.hs, interpreted )
Ok, modules loaded: Main.
*Main> main
ghc: panic! (the 'impossible' happened)
  (GHC version 7.0.4 for x86_64-unknown-linux):
	nameModule $dNum{v ann}

Please report this as a GHC bug:  http://www.haskell.org/ghc/reportabug

$ ghc --make demo1.hs
[1 of 1] Compiling Main             ( demo1.hs, demo1.o )
ghc: panic! (the 'impossible' happened)
  (GHC version 7.0.4 for x86_64-unknown-linux):
	initC: srt_lbl

Please report this as a GHC bug:  http://www.haskell.org/ghc/reportabug

Trac metadata

Trac field	Value
Version	7.0.4
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler (Type checker)
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture