馃敟 Pure & PureC #307
馃敟 Pure & PureC #307
Conversation
There was a problem hiding this comment.
This PR is the result of three observations by @patrickt:
-
Every monad
mcould be an algebra forLift m. -
The
Algebrainstance forIdentityobsoletesPureC. -
Lift Identityis an equivalent effect toPure.
I鈥檝e been working on this thing for 14 months straight and @patrickt still manages to show me new things about it 馃挅
| instance Algebra Pure Identity where | ||
| alg v = case v of {} | ||
| instance Algebra (Lift Identity) Identity where | ||
| alg = join . unLift |
There was a problem hiding this comment.
When I wrote the Algebra instance for IO, above, @patrickt asked if it wouldn鈥檛 hold for any Monad m that you could write an Algebra (Lift m) m instance鈥攁nd indeed it would!
| -- | A carrier for pure effects, used to kick off a stack of effects with 'run'. | ||
| -- | ||
| -- @since 1.0.0.0 | ||
| module Control.Carrier.Pure |
There was a problem hiding this comment.
When I added the Algebra instance for Identity, @patrickt asked if we should eliminate PureC since it鈥檚 now redundant. At the time I felt like it would be better to keep our own definition around and define our own instances, but upon reflection this only makes interoperation with other libraries harder since it鈥檚 probable that our users are using run instead of runIdentity as the terminus of pure effect stacks. By eliminating PureC, and redefining run as a synonym for runIdentity, we not only reduce our surface area, but also operate seamlessly with anything else defining instances for Identity.
| @@ -1,11 +1,12 @@ | |||
| {- | Provides a mechanism to kick off the evaluation of an effect stack that takes place in a monadic context. | |||
|
|
|||
| 'Lift' effects are always the last effect in a given effect stack. These stacks are invoked with 'Control.Effect.Lift.runM'. The 'Control.Effect.Pure.Pure' effect is equivalent to @Lift Identity@. | |||
| 'Lift' effects are always the last effect in a given effect stack. These stacks are invoked with 'Control.Carrier.Lift.runM' or 'Control.Algebra.run'. | |||
There was a problem hiding this comment.
-
I noticed the final sentence (added by @patrickt in Document the effects and their carriers.聽#251) yesterday, and it blew my mind. I don鈥檛 know how I鈥檇 never realized this before, but yes,
Lift Identityis an equivalent terminal effect for pure code.sendM (Identity x)isn鈥檛 the most readable synonym forpure x, but by contrast,Puredoesn鈥檛 offer any operations at all鈥攎aking it kind of a degenerate edge case. -
I love that the first sentence is now true in two ways: if
Liftoccurs, it鈥檚 the final effect; andLiftnow occurs in all (ok, most, since Base instances聽#206) stacks!
|
As the Beatles sang,
|
| @@ -55,9 +53,13 @@ class (Effect sig, Monad m) => Algebra sig m | m -> sig where | |||
| -- | Construct a value in the carrier for an effect signature (typically a sum of a handled effect and any remaining effects). | |||
| alg :: sig m a -> m a | |||
|
|
|||
| instance Algebra Pure PureC where | |||
| alg v = case v of {} | |||
There was a problem hiding this comment.
I think we can ditch the EmptyCase extension in this file now.
| -- @since 1.0.0.0 | ||
| run :: Identity a -> a | ||
| run = runIdentity | ||
| {-# INLINE run #-} |
There was a problem hiding this comment.
Should we deprecate this? I鈥檓 honestly not sure.
No, we shouldn鈥檛. This is good, as per your comment below.
| runR run = Run (fmap Identity . run) | ||
|
|
||
| -- | Handlers with curried input state (e.g. 'Control.Carrier.Reader.ReaderC', 'Control.Carrier.State.Strict.StateC'). | ||
| runC :: (forall a . s -> m a -> PureC (f a)) -> Run ((,) s) f m | ||
| runC :: (forall a . s -> m a -> Identity (f a)) -> Run ((,) s) f m | ||
| runC run = Run (uncurry run) | ||
|
|
||
|
|
There was a problem hiding this comment.
There鈥檚 some underlying thing going on here that鈥檚 intensely beautiful but that I don鈥檛 have the vocabulary to describe.
| -- | A carrier for pure effects, used to kick off a stack of effects with 'run'. | ||
| -- | ||
| -- @since 1.0.0.0 | ||
| module Control.Carrier.Pure |
This PR:
PureC.Pure.