Improving Random

[Alexey Kuleshevich]

Hey Ben,

> Two weeks, perhaps four?

Already 3 weeks has passed since the proposal has been submitted. If Carter said "Give me two weeks and I'll have a release ready", I personally would not believe it, but at least it could have been considered a concrete deadline. Instead we got "I am taking my time with it".

> Class PRNG g => MonadRandom m g | m-> g where

What Carter suggested, I believe, would only complicate things. FunctionalDependencies in definition means that for every generator you need to create your own monad, which will have to also be a monad transformer for some generators, so you'd have to worry about all these instance and the user would have to worry about yet another monad transformer. What we currently have in the proposal does not require any special new data types or new monads being created. Existing stateful generators can just create an instance for any monad they suppose to work in, notable ones are `ST`, `IO` and of course all monads that have `PrimMonad` instance.

This is what we have now:

class Monad m => MonadRandom m g where

which means stateful generators can create:

instance (s ~ PrimState m, PrimMonad m) => MonadRandom (MWC.Gen s) m where -- mwc-random package

instance (s ~ PrimState m, PrimMonad m) => MonadRandom (PCG.Gen s) m where -- pcg-random package

instance (s ~ PrimState m, PrimMonad m) => MonadRandom (SFMT.Gen s) m where -- sfmt package

instance MonadRandom MTGen IO where -- mersene-random package


and every pure generator that has `RandomGen` instance can be used with any monad that has `MonadState` instance:

instance (RandomGen g, MonadState g m) => MonadRandom (StateGenM g) m where


These instances cover all of pure and stateful non-crypto RNGs available in Haskell, which can be used with any monad transformer (through PrimMonad and MonadState instances). There are also wrappers around IORef and STRef for pure generators that can be used in IO and ST respectfully, but that is a minor detail.


Hopefully I now addressed that comment for you, because Carter does not seem to be interested in this, since he locked PR I opened and it is now a one directional communication.

Alexey.

‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐
On Wednesday, June 3, 2020 9:31 PM, Ben Franksen <ben.franksen at online.de> wrote:

> Am 03.06.20 um 15:55 schrieb dominic at steinitz.org:
>
> > We invited anyone interested including Carter to participate in the
> > design back in February. He did not do so. How much longer do you
> > think we should give him?
>
> Two weeks, perhaps four?
>
> > No serious work has been done on it since September 2014 and numerous
> > promises to do something have never materialised.
>
> I can't possibly comment on this.
>
> > Is there any part of Carter’s critique that in your view has not been
> > addressed by replies in this thread?
>
> He wrote
>
> > The better choice is
> > Class PRNG g => MonadRandom m g | m-> g where ...
> > And then their example instance should be something like
> > Newtype MWCT m a = ... newtype wrapper around StateT m a threading an MWC
> > indexed by the state token of the underlying PrimMonad
> > (Or newtype MWCT s m a = .... stateT thing)
> > Instance (PrimMonad m) => MonadRandom (MWCT m) (MCWGen (PrimState m))
> > where ...
>
> which hasn't been addressed fully. However, this would be difficult
> since the above comment is missing too many details.
>
> This is why I wanted to entice Carter to share his
> thoughts/misgivings/counter-proposals in greater detail on the PR.
>
> > We should be welcoming contributions and improve libraries when the
> > opportunities arise and not wait another 5 years for perfection
> > (which may never arrive).
>
> I can understand that you don't want this to stall indefinitely, so
> perhaps you need to give him a (comfortable) deadline? Just a thought.
>
> Cheers
> Ben
>
> Libraries mailing list
> Libraries at haskell.org
> http://mail.haskell.org/cgi-bin/mailman/listinfo/libraries