[Haskell-cafe] MonadMemory class

[Derek Elkins]

On Fri, 2008-03-28 at 00:02 +0200, Ariel J. Birnbaum wrote:
> Two questions came to mind while thinking of memory references in Haskell:
> 
> 1. Is there a "standard" equivalent of the following:
> 
> class (Monad m) => MonadMemory m r | m -> r where
>     new :: a -> m (r a)
>     read :: r a -> m a
>     write :: r a -> a -> m ()
> 

This has come up a few times and been written by many Haskellers
(usually called MonadRef.)  The problem is the functional dependencies.
Any choice you make is either too limiting or too annoying to use and it
was never agreed which would be best.


> What kind of axioms should an instance of this class satisfy?

There are quite a few obvious ones, but I'm not sure it's particularly
necessary.

> 
> 2. How would a "pure" instance of this class look like (obvious 
> unsafePerformIO-based solutions aside)? Is it even possible in pure Haskell?

The issue here is a type one, not a semantics one.  I believe the only
way to purely get such an interface (in current Haskell) is to use
unsafeCoerce.  Other than that, it's a straightforward state monad.