[GHC] #13611: Segfault due to levity polymorphism of mkWeak#

[GHC]

#13611: Segfault due to levity polymorphism of mkWeak#
-------------------------------------+-------------------------------------
        Reporter:  nomeata           |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.1
      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 nomeata):

 > This is tantamount to unsafePerformIO, is it not?

 Not quite, I am not producing a `RealWorld#` token out of thin air.

 Normal user code calls `mkWeak#` via `mkWeak` in `GHC.Weak`, which is not
 levity polymorphic. But there is one use of `mkWeak#` in the libraries
 that seems to use a different `TypeRep`:
 {{{
 addForeignPtrConcFinalizer_ :: ForeignPtrContents -> IO () -> IO ()
 addForeignPtrConcFinalizer_ (PlainForeignPtr r) finalizer = do
   noFinalizers <- insertHaskellFinalizer r finalizer
   if noFinalizers
      then IO $ \s ->
               case r of { IORef (STRef r#) ->
               case mkWeak# r# () (unIO $ foreignPtrFinalizer r) s of {
                 (# s1, _ #) -> (# s1, () #) }}
      else return ()
 }}}
 The `r#` here is of type `MutVar# s a`, which is `UnliftedRep`; there are
 similar calls in the modules for for `MVar` and `IORef`.

 So this is a use case where we want levity polymorphism that allows any
 *boxed* type, whether lifted or not.

 Or, the easy way out, is to have two copies of the `mkWeak#` primop, one
 for `LiftedRep` and one for `UnliftedRep`. They could use the same code
 and info-pointer.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13611#comment:4>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler