Wow, this looks like a bug to me. If it's not a bug, then it's horribly  unintuitive.
I extended your example in an effort to figure out what was going on. Apparently weak pointers loath live IORefs:

import Data.IORef
import Data.Maybe
import System.Mem
import System.Mem.Weak

import Control.Monad

data A = A String
data B = B String (IORef Int)
showA (A s) = s
showB (B s _) = s

main = do
 -- works as expected:
 ref <- return $ A "A"
 ptr <- mkWeak ref 21 Nothing
 performGC
 print . isNothing =<< deRefWeak ptr
 print (showA ref)
 
 -- why doesn't this work?
 ref <- liftM (B "B") $ newIORef 42
 ptr <- mkWeak ref 21 Nothing
 performGC
 print . isNothing =<< deRefWeak ptr
 print (showB ref)
 
 -- this works, wtf???
 ref <- liftM (B "B") $ return undefined
 ptr <- mkWeak ref 21 Nothing
 performGC
 print . isNothing =<< deRefWeak ptr
 print (showB ref)


I don't think this is the expected behavior. The docs on Weak pointers don't mention anything like this. I suspect something in the GC is getting confused by the IORef somehow.

- Job



2009/11/2 Patai Gergely <patai_gergely@fastmail.fm>
> Could mkWeakPair do what you want?
> http://haskell.org/ghc/docs/latest/html/libraries/base/System-Mem-Weak.html#v:mkWeakPair
No, it's just a convenience function that doesn't help much, because the
value already refers to the IORef anyway.

Here's a minimal example to illustrate the problem:

import Data.IORef
import Data.Maybe
import System.Mem
import System.Mem.Weak

main = do
 ref <- newIORef 42
 ptr <- mkWeak ref 21 Nothing
 performGC
 print . isNothing =<< deRefWeak ptr
 print =<< readIORef ref

Depending on whether you compile with optimisations, the weak reference
might be reported dead, even though the IORef is alive and kicking.
Switching to mkWeakPair (or just mentioning ref in the value somehow)
doesn't affect that.

> Or are you trying to do something else?
The goal is to create mutable objects whose update codes are tracked by
the main program, but they can be thrown out when all the other
references to the objects are lost. Creating weak pointers with MutVar#s
seems to do the trick, but I'm not confident if it is a solution I can
trust...

Gergely

--
http://www.fastmail.fm - A fast, anti-spam email service.

_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe