[GHC] #7662: Improve GC of mutable objects

[GHC]

#7662: Improve GC of mutable objects
-------------------------------------+-------------------------------------
        Reporter:  ezyang            |                Owner:
            Type:  feature request   |               Status:  new
        Priority:  normal            |            Milestone:  ⊥
       Component:  Runtime System    |              Version:  7.7
      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 rwbarton):

 I think this has been discussed elsewhere too, but I don't understand how
 the above could be consistent with Haskell's semantics. What if I define
 {{{#!hs
 upperize :: SomeType -> SomeType
 upperize s = s { y = toUpper (y s) }
 }}}
 Obviously this needs to allocate a new `SomeType` closure on the heap, but
 if the `IORef Int` is completely unpacked into a mutable field of
 `SomeType` closures, then the new closure can't share `x` with the old
 closure. In general just constructing a `SomeType` value would have the
 observable effect of creating a new mutable cell and so would have to live
 in IO.

 In short, ML's records with mutable fields have identity, and so are quite
 different than just sticking an `IORef` inside a record in Haskell. In
 Haskell the `IORef` itself has identity (and thus is created and accessed
 through IO actions) but the record `SomeType` does not.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/7662#comment:10>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler