ddarius at hotpop.com
Thu Nov 13 22:59:08 EST 2003
On Thu, 13 Nov 2003 23:58:29 +0000
Ross Paterson <ross at soi.city.ac.uk> wrote:
> [switching to libraries]
Yes, I did think of adding a comment to that effect.
> On Thu, Nov 13, 2003 at 04:29:49PM -0500, Derek Elkins wrote:
The class declaration I missed in my cut & paste:
class MonadRef m r | m -> r where
newRef :: a -> m (r a)
readRef :: r a -> m a
writeRef :: r a -> a -> m ()
> > instance MonadRef IO IORef where
> > newRef = newIORef
> > readRef = readIORef
> > writeRef = writeIORef
> > instance MonadRef (Lazy.ST s) (STRef s) where
> > newRef = Lazy.strictToLazyST . newSTRef
> > readRef = Lazy.strictToLazyST . readSTRef
> > writeRef = (Lazy.strictToLazyST .) . writeSTRef
> > instance MonadRef (Strict.ST s) (STRef s) where
> > newRef = newSTRef
> > readRef = readSTRef
> > writeRef = writeSTRef
> For an alternative design, see
Here's the earlier RefMonad emails:
Well, you have a point. MonadRef is broken, at least as I have it.
Unless both Lazy and Strict ST's didn't exist at the time, I don't know
how Simon Peyton Jones came up with his fundeps, but those are the
problem. An r -> m fundep breaks right off the bat with STRef
and Lazy/Strict.ST. With an m -> r fundep, as I have it, it's also
broken right off the bat via IO and MVar/IORef*.
So basically, with the MonadRef interface, the only
(potentially) reasonable interface is with no fundeps, though this has
it's nuisances too.
Andrew Bromage brought up one problem with the MonadST interface, namely
that it requires you to use STRefs. (Which obviously means it doesn't
work with MVars as well.) Another problem with the MonadST interface(or
at least I don't see what to do about it) is the following:
I'm using MonadRef to abstract away from the implementation of a
backtrackable reference type. Though it isn't now, it would be
perfectly reasonable to make it too an instance of MonadRef like so,
instance MonadRef m r => MonadRef (LP m) r where
newRef = newLPRef -- etc.
newLPRef/readLPRef just lift the underlying functions, but writeLPRef
handles the restoring when backtracking. Now if we write some function
f :: MonadRef m r => r a -> m () (e.g. \r -> writeRef r 10) it will
behave correctly even when used in the LP monad. I don't see how to do
this with the MonadST interface.
In a nutshell, MonadST seems to solve a different problem than MonadRef.
Altogether, I don't (yet) see any crippling problem with MonadRef with
no fundeps, but this has turned out to be a lot trickier than I
expected. As several people have their MonadRef/MRef/RefMonad/MonadST
classes floating around, I certainly had no illusion that I was the
first to suggest it, it looks like a good thing to add to the libraries.
Though MonadRef with no fundeps seems less than ideal.
* Andrew Bromage just pointed this one out. I'd thought up milder
cases, but this one is devastating.
More information about the Libraries