The Death of Finalizers

[Simon Marlow]

> > It has a different type:
> > 
> >   atomicModifyIORef :: IORef a -> (a -> (a,b)) -> IO b
> >   modifyIORef :: IORef a -> (a -> a) -> IO ()
> 
> It would break backwards compatability, but
> 
> 	modifyIORef_ :: IORef a -> (a -> IO a) -> IO ()
> 	modifyIORef :: IORef a -> (a -> IO (a, b)) -> IO b
> 
> (both specified atomic) would be consistent with the MVar names.

Did you really mean those types, and not the types I quoted?  That
implies unsafeInterleaveIO if they have to be atomic.

Cheers,
	Simon