Church-encoded State carrier #363
Conversation
| {-# INLINE execState #-} | ||
|
|
||
| -- | @since 1.1.0.0 | ||
| newtype StateC s m a = StateC { runStateC :: forall r . (a -> s -> m r) -> s -> m r } |
There was a problem hiding this comment.
I debated exporting a CPS handler for this à la runNonDet &c., but:
-
following the extant examples would suggest naming it
runState, but that name already has a well-understood meaning forState, so I’d have to come up with a new naming scheme; -
it’s not immediately clear whether it would be more useful as
(a -> s -> m r) -> s -> StateC s m a -> m ror(a -> s -> m r) -> StateC s m a -> s -> m r; and -
I didn’t really want to have to document it.
I export the constructor (tho not the selector), so users are always free to define their own.
There was a problem hiding this comment.
I wouldn’t be averse to exporting the constructor, fwiw.
There was a problem hiding this comment.
I do export the constructor, just not the selector.
test/State.hs
Outdated
| import qualified Control.Carrier.State.Lazy as LazyStateC | ||
| import qualified Control.Carrier.State.Strict as StrictStateC | ||
| import qualified Control.Carrier.State.Church as C.Church | ||
| import qualified Control.Carrier.State.Lazy as C.Lazy | ||
| import qualified Control.Carrier.State.Strict as C.Strict | ||
| import Control.Effect.State | ||
| import qualified Control.Monad.Trans.RWS.Lazy as LazyRWST | ||
| import qualified Control.Monad.Trans.RWS.Strict as StrictRWST | ||
| import qualified Control.Monad.Trans.State.Lazy as LazyStateT | ||
| import qualified Control.Monad.Trans.State.Strict as StrictStateT | ||
| import qualified Control.Monad.Trans.RWS.Lazy as RWST.Lazy | ||
| import qualified Control.Monad.Trans.RWS.Strict as RWST.Strict | ||
| import qualified Control.Monad.Trans.State.Lazy as T.Lazy | ||
| import qualified Control.Monad.Trans.State.Strict as T.Strict |
There was a problem hiding this comment.
I renamed all of these because they were overcomplicated and we’re probably going to end up extending them at least a little anyway.
| instance MonadFix m => MonadFix (StateC s m) where | ||
| mfix f = StateC $ \ k s -> mfix (runState s . f . snd) >>= uncurry (flip k) |
There was a problem hiding this comment.
I’m quite pleased with how this turned out.
src/Control/Carrier/State/Church.hs
Outdated
| alg hom = \case | ||
| L (Get k) -> StateC $ \ k' s -> runState s (hom (k s)) >>= uncurry (flip k') | ||
| L (Put s k) -> StateC $ \ k' _ -> runState s (hom k) >>= uncurry (flip k') | ||
| R other -> StateC $ \ k s -> alg id (thread (s, ()) (uncurry runState . fmap hom) other) >>= uncurry (flip k) |
There was a problem hiding this comment.
This turned out pretty nicely, but I’m not completely convinced that this is the best way to be threading the state through. I plan to revisit that in/after #361.
|
I think this is a fine addition, though I think we should document something about its performance. Do we expect it to be better or worse than the |
|
@patrickt: I don’t have a strong intuition, and in fact I don’t even think there’s going to be a cut and dried answer to that. My guess is that it’d depend pretty crucially on the carriers surrounding it. My goal is to try to provide some more church-encoded carriers and measure how they fare against the classical varieties in larger carrier compositions in e.g. Starlight or |
|
@robrix That’s fine by me. We should probably mention somewhere that the choice of the “best” carrier for state is contextual. |
|
Yeah, I might go a bit further and recommend |
This PR defines a Church-encoded
Statecarrier, equivalent toCodensity ((->) s).I mostly wanted to see how the
MonadFixinstance would turn out, since I’m sort of combining this and theCutCinstance forfused-effects-parser.Algebra, GADT effects, &c.Control.Effect.State.