Re: [Haskell-cafe] ANNOUNCE: pqueue-mtl, stateful-mtl

2009/2/16 Josef Svenningsson :

On Mon, Feb 16, 2009 at 2:30 AM, wren ng thornton wrote:

...

Louis Wasserman wrote:

...

I follow. The primary issue, I'm sort of wildly inferring, is that use of STT -- despite being pretty much a State monad on the inside -- allows access to things like mutable references?

That's exactly the problem. The essential reason for ST's existence are STRefs which allow mutability.

I'd like to point out one other thing that ST provides, which is often forgotten. It provides *polymorphic* references. That is, we can create new references of any type.

So ST is a really magical beast. Not only does it provide mutation, it also provides mutable references. And these are two different features. Now, everyone agrees that mutation is not something that you can implement in a functional language, so ST cannot be implemented in Haskell for that reason. It has to be given as a primitive. But what about polymorphic references? Can they be implemented in Haskell? The Claessen conjecture (after Koen Claessen) is that they cannot be implemented in Haskell. See the following email for more details: http://www.haskell.org/pipermail/haskell/2001-September/007922.html

One could try and separate mutation and polymorphic references and give them as two different primitives and implement ST on top of that. But I haven't seen anyone actually trying that (or needing it for that matter).

Actually, I was interested in making a state holding polymorphic references in the state monad, so that the state could be passed around. I made an attempt, and if my memory serves me right, it worked like this : It was based on Dynamics, with an IntMap indexed, indirectly, by TypeRep, yielding, for each type, a new IntMap, providing references for any value of that type. In fact, I think the next stpe would be to have some TH to generate specific state monads to hold references on specific types. Cheers, Thu