Re: [GHC] #10867: Primop types should not mention ()

14 Sep 2015

      #10867: Primop types should not mention ()
-------------------------------------+-------------------------------------
        Reporter:  simonpj           |                   Owner:
            Type:  bug               |                  Status:  new
        Priority:  normal            |               Milestone:
       Component:  Compiler          |                 Version:  7.10.2
      Resolution:                    |                Keywords:
Operating System:  Unknown/Multiple  |            Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |               Test Case:
      Blocked By:                    |                Blocking:
 Related Tickets:                    |  Differential Revisions:
-------------------------------------+-------------------------------------
Changes (by simonpj):

 * cc: core-libraries-committee (added)

Comment:

 Simon, can you help a little more?  I'm also copying the core libraries
 committee,
 since I propose an API change for weak-pointer primpops.

 The type `Weak#` and primop `mkWeak#`, are declared thus:
 {{{
 primtype Weak# b

 primop  MkWeakOp "mkWeak#" GenPrimOp
    o -> b -> c -> State# RealWorld -> (# State# RealWorld, Weak# b #)
 }}}
 I think `c` is the finalizer.  This type suggests that you can put any
 old value in for the finalizer, which seems unlikely.  '''In fact it can
 easily lead to a seg-fault (with no use of `unsafeCoerce` by building
 a weak pointer whose "finaliser" with the wrong type.'''

 Then the primop `finalizeWeak#` is declared thus:
 {{{
 primop  FinalizeWeakOp "finalizeWeak#" GenPrimOp
    Weak# a -> State# RealWorld
    -> (# State# RealWorld
        , Int#
        , State# RealWorld -> (# State# RealWorld, () #)
        #)
 }}}
 I believe the following:

  * The third argument to `mkWeak#` is returned as the third component of
 the result to `finalizeWeak#`, but is otherwise totally un-examined by the
 RTS.

 Is that true?  In that case it would be far, far better to declare them
 like this:
 {{{
 primtype Weak# value finalizer

 primop  MkWeakOp "mkWeak#" GenPrimOp
    key -> value -> finalizer
    -> State# RealWorld -> (# State# RealWorld, Weak# payload finalizer #)
 }}}
 So `Weak#` gets two type parameters, the second recording the type of
 the "finaliser".  All the RTS promises to do is to cough up the
 "finalizer"
 when the `key` dies.

 Now in `GHC.Weak` we can declare `Weak` thus:
 {{{
 data Weak v = Weak (Weak# v) (IO ())
 }}}
 That is `Weak` specialises the "any only value can be a finaliser" API
 offered by `Weak#` to "the finaliser is a vlaue of type `IO ()`" API
 for `Weak`.

 Doing this would eliminate an unpleasant potential source of seg-faults;
 and it would eliminate one of the two unpleasant uses of `()` in
 `primpops.txt.pp`.

--
Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/10867#comment:10
GHC http://www.haskell.org/ghc/
The Glasgow Haskell Compiler