Do we have idiom for lifting a state monad into pair of states?
Consider you have developed library routines that act on (State s1 a). For some reason, you need more than one state simultaneously. Let's say two side by side for simple example, that is (State (s1,s2) a). To use library functions on one state monad in a two state monad, we need to wrapper that lifts actions of (State s1 a) to (State (s1,s2) a). It is not difficult to write a lifter for such purposes as below. It is kind of like doing liftA in Applicative libarary, but instead of the last argument 'a' but on the fist argument 's' of (State s a). This seems like an idiom that can often come up. So, I tried some searching in some applicative related libraries and monad transformer related libraries but haven't found this idiom yet. If you had a need for idioms like below, what do you call it? Or, is there a way more general way I've missed that makes this a very special case of it.
import Control.Monad.State import Control.Monad.Error import Control.Applicative
-- lift an action over a state into a pair of states -- do1st :: State s1 a -> State (s1,s2) a do1st m1 = do (s1, s2) <- get let (a, s1') = runState m1 s1 put (s1',s2) return a
do2nd :: State s2 a -> State (s1,s2) a do2nd m2 = do (s1, s2) <- get let (a, s2') = runState m2 s2 put (s1,s2') return a
-- lift an action over a state with error -- into a pair of states with error -- do1 :: Error e => ErrorT e (State s1) a -> ErrorT e (State (s1,s2)) a do1 m1 = do (s1, s2) <- lift get let (ma,s1') = (runState . runErrorT) m1 s1 case ma of Left e -> throwError e Right a -> do lift $ put (s1',s2) return a
do2 :: Error e => ErrorT e (State s2) a -> ErrorT e (State (s1,s2)) a do2 m2 = do (s1, s2) <- lift get let (ma,s2') = (runState . runErrorT) m2 s2 case ma of Left e -> throwError e Right a -> do lift $ put (s1,s2') return a
It's called 'zoom' in various lens libraries. On 26/03/15 10:28, Ki Yung Ahn wrote:
Consider you have developed library routines that act on (State s1 a). For some reason, you need more than one state simultaneously. Let's say two side by side for simple example, that is (State (s1,s2) a). To use library functions on one state monad in a two state monad, we need to wrapper that lifts actions of (State s1 a) to (State (s1,s2) a).
It is not difficult to write a lifter for such purposes as below. It is kind of like doing liftA in Applicative libarary, but instead of the last argument 'a' but on the fist argument 's' of (State s a). This seems like an idiom that can often come up. So, I tried some searching in some applicative related libraries and monad transformer related libraries but haven't found this idiom yet.
If you had a need for idioms like below, what do you call it? Or, is there a way more general way I've missed that makes this a very special case of it.
import Control.Monad.State import Control.Monad.Error import Control.Applicative
-- lift an action over a state into a pair of states -- do1st :: State s1 a -> State (s1,s2) a do1st m1 = do (s1, s2) <- get let (a, s1') = runState m1 s1 put (s1',s2) return a
do2nd :: State s2 a -> State (s1,s2) a do2nd m2 = do (s1, s2) <- get let (a, s2') = runState m2 s2 put (s1,s2') return a
-- lift an action over a state with error -- into a pair of states with error -- do1 :: Error e => ErrorT e (State s1) a -> ErrorT e (State (s1,s2)) a do1 m1 = do (s1, s2) <- lift get let (ma,s1') = (runState . runErrorT) m1 s1 case ma of Left e -> throwError e Right a -> do lift $ put (s1',s2) return a
do2 :: Error e => ErrorT e (State s2) a -> ErrorT e (State (s1,s2)) a do2 m2 = do (s1, s2) <- lift get let (ma,s2') = (runState . runErrorT) m2 s2 case ma of Left e -> throwError e Right a -> do lift $ put (s1,s2') return a
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
Thanks, yes, it should of course been in Lens!! *Main Control.Lens> :t zoom _1 :: State s1 a -> State (s1,s2) a zoom _1 :: State s1 a -> State (s1,s2) a :: State s1 a -> State (s1, s2) a *Main Control.Lens> :t zoom _2 :: State s2 a -> State (s1,s2) a zoom _2 :: State s2 a -> State (s1,s2) a :: State s2 a -> State (s1, s2) a It works beautifully :) 2015년 03월 26일 01:32에 Roman Cheplyaka 이(가) 쓴 글:
It's called 'zoom' in various lens libraries.
On 26/03/15 10:28, Ki Yung Ahn wrote:
Consider you have developed library routines that act on (State s1 a). For some reason, you need more than one state simultaneously. Let's say two side by side for simple example, that is (State (s1,s2) a). To use library functions on one state monad in a two state monad, we need to wrapper that lifts actions of (State s1 a) to (State (s1,s2) a).
It is not difficult to write a lifter for such purposes as below. It is kind of like doing liftA in Applicative libarary, but instead of the last argument 'a' but on the fist argument 's' of (State s a). This seems like an idiom that can often come up. So, I tried some searching in some applicative related libraries and monad transformer related libraries but haven't found this idiom yet.
If you had a need for idioms like below, what do you call it? Or, is there a way more general way I've missed that makes this a very special case of it.
import Control.Monad.State import Control.Monad.Error import Control.Applicative
-- lift an action over a state into a pair of states -- do1st :: State s1 a -> State (s1,s2) a do1st m1 = do (s1, s2) <- get let (a, s1') = runState m1 s1 put (s1',s2) return a
do2nd :: State s2 a -> State (s1,s2) a do2nd m2 = do (s1, s2) <- get let (a, s2') = runState m2 s2 put (s1,s2') return a
-- lift an action over a state with error -- into a pair of states with error -- do1 :: Error e => ErrorT e (State s1) a -> ErrorT e (State (s1,s2)) a do1 m1 = do (s1, s2) <- lift get let (ma,s1') = (runState . runErrorT) m1 s1 case ma of Left e -> throwError e Right a -> do lift $ put (s1',s2) return a
do2 :: Error e => ErrorT e (State s2) a -> ErrorT e (State (s1,s2)) a do2 m2 = do (s1, s2) <- lift get let (ma,s2') = (runState . runErrorT) m2 s2 case ma of Left e -> throwError e Right a -> do lift $ put (s1,s2') return a
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
One solution is to use lens package.
The lens package provides an interesting combinator, zoom, in
Control.Lens.Zoom[1]:
zoom https://hackage.haskell.org/package/lens-4.8/docs/Control-Lens-Zoom.html#v:z...
:: Monad https://hackage.haskell.org/package/base-4.7.0.2/docs/Control-Monad.html#t:M...
m => Lens'
https://hackage.haskell.org/package/lens-4.8/docs/Control-Lens-Type.html#t:L...
s t -> StateT
https://hackage.haskell.org/package/mtl-2.1.3.1/docs/Control-Monad-State-Str...
t m a -> StateT
https://hackage.haskell.org/package/mtl-2.1.3.1/docs/Control-Monad-State-Str...
s m a
Lens' s a is the type of accessors to `a`. When it comes to tuples, you can
use
_1 :: Lens' (a, b) a
_2 :: Lens' (a, b) b
So do1st, do2nd are zoom _1, zoom _2 respectively.
zoom also can be used for monad transformer stacks; zoom _1 may have the
type of do1.
[1] https://hackage.haskell.org/package/lens-4.8/docs/Control-Lens-Zoom.html
2015-03-26 17:28 GMT+09:00 Ki Yung Ahn
Consider you have developed library routines that act on (State s1 a). For some reason, you need more than one state simultaneously. Let's say two side by side for simple example, that is (State (s1,s2) a). To use library functions on one state monad in a two state monad, we need to wrapper that lifts actions of (State s1 a) to (State (s1,s2) a).
It is not difficult to write a lifter for such purposes as below. It is kind of like doing liftA in Applicative libarary, but instead of the last argument 'a' but on the fist argument 's' of (State s a). This seems like an idiom that can often come up. So, I tried some searching in some applicative related libraries and monad transformer related libraries but haven't found this idiom yet.
If you had a need for idioms like below, what do you call it? Or, is there a way more general way I've missed that makes this a very special case of it.
import Control.Monad.State import Control.Monad.Error import Control.Applicative
-- lift an action over a state into a pair of states -- do1st :: State s1 a -> State (s1,s2) a do1st m1 = do (s1, s2) <- get let (a, s1') = runState m1 s1 put (s1',s2) return a
do2nd :: State s2 a -> State (s1,s2) a do2nd m2 = do (s1, s2) <- get let (a, s2') = runState m2 s2 put (s1,s2') return a
-- lift an action over a state with error -- into a pair of states with error -- do1 :: Error e => ErrorT e (State s1) a -> ErrorT e (State (s1,s2)) a do1 m1 = do (s1, s2) <- lift get let (ma,s1') = (runState . runErrorT) m1 s1 case ma of Left e -> throwError e Right a -> do lift $ put (s1',s2) return a
do2 :: Error e => ErrorT e (State s2) a -> ErrorT e (State (s1,s2)) a do2 m2 = do (s1, s2) <- lift get let (ma,s2') = (runState . runErrorT) m2 s2 case ma of Left e -> throwError e Right a -> do lift $ put (s1,s2') return a
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
Let me grab this opportunity to advertise the multistate package [1]. The underlying assumption is that the types of the states in your stack are distinct. In that case, type inference can give you something like "automatic zooming" for a MultiState that contains arbitrary heterogenous lists (tuples). Generous usage of the Monad(Multi)State type class would thereby eliminate the need for do1st and do2nd. See the example in the package. The package does not (yet!) work well with existing transformers, for example one might think of running a StateT on one of the states in a MultiStateT. This is not a problem when using MultiStateT exclusively, but of course bad for interoperability. I am open for specific requests in that direction. In the last few days I have been working on adding a MultiRWST, but this is not completely finished yet (and I am not sure yet how to make the interface consistent; the whole run/eval/exec distinction seems unintuitive, especially when you have multiple states..) Lennart [1] https://hackage.haskell.org/package/multistate On 26/03/15 09:28, Ki Yung Ahn wrote:
Consider you have developed library routines that act on (State s1 a). For some reason, you need more than one state simultaneously. Let's say two side by side for simple example, that is (State (s1,s2) a). To use library functions on one state monad in a two state monad, we need to wrapper that lifts actions of (State s1 a) to (State (s1,s2) a).
It is not difficult to write a lifter for such purposes as below. It is kind of like doing liftA in Applicative libarary, but instead of the last argument 'a' but on the fist argument 's' of (State s a). This seems like an idiom that can often come up. So, I tried some searching in some applicative related libraries and monad transformer related libraries but haven't found this idiom yet.
If you had a need for idioms like below, what do you call it? Or, is there a way more general way I've missed that makes this a very special case of it.
import Control.Monad.State import Control.Monad.Error import Control.Applicative
-- lift an action over a state into a pair of states -- do1st :: State s1 a -> State (s1,s2) a do1st m1 = do (s1, s2) <- get let (a, s1') = runState m1 s1 put (s1',s2) return a
do2nd :: State s2 a -> State (s1,s2) a do2nd m2 = do (s1, s2) <- get let (a, s2') = runState m2 s2 put (s1,s2') return a
-- lift an action over a state with error -- into a pair of states with error -- do1 :: Error e => ErrorT e (State s1) a -> ErrorT e (State (s1,s2)) a do1 m1 = do (s1, s2) <- lift get let (ma,s1') = (runState . runErrorT) m1 s1 case ma of Left e -> throwError e Right a -> do lift $ put (s1',s2) return a
do2 :: Error e => ErrorT e (State s2) a -> ErrorT e (State (s1,s2)) a do2 m2 = do (s1, s2) <- lift get let (ma,s2') = (runState . runErrorT) m2 s2 case ma of Left e -> throwError e Right a -> do lift $ put (s1,s2') return a
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
monad-classes can do this, and more: http://bit.ly/1xBZr3I On 26/03/15 17:32, lennart spitzner wrote:
Let me grab this opportunity to advertise the multistate package [1].
The underlying assumption is that the types of the states in your stack are distinct. In that case, type inference can give you something like "automatic zooming" for a MultiState that contains arbitrary heterogenous lists (tuples). Generous usage of the Monad(Multi)State type class would thereby eliminate the need for do1st and do2nd. See the example in the package.
The package does not (yet!) work well with existing transformers, for example one might think of running a StateT on one of the states in a MultiStateT. This is not a problem when using MultiStateT exclusively, but of course bad for interoperability. I am open for specific requests in that direction.
In the last few days I have been working on adding a MultiRWST, but this is not completely finished yet (and I am not sure yet how to make the interface consistent; the whole run/eval/exec distinction seems unintuitive, especially when you have multiple states..)
Lennart
[1] https://hackage.haskell.org/package/multistate
On 26/03/15 09:28, Ki Yung Ahn wrote:
Consider you have developed library routines that act on (State s1 a). For some reason, you need more than one state simultaneously. Let's say two side by side for simple example, that is (State (s1,s2) a). To use library functions on one state monad in a two state monad, we need to wrapper that lifts actions of (State s1 a) to (State (s1,s2) a).
It is not difficult to write a lifter for such purposes as below. It is kind of like doing liftA in Applicative libarary, but instead of the last argument 'a' but on the fist argument 's' of (State s a). This seems like an idiom that can often come up. So, I tried some searching in some applicative related libraries and monad transformer related libraries but haven't found this idiom yet.
If you had a need for idioms like below, what do you call it? Or, is there a way more general way I've missed that makes this a very special case of it.
import Control.Monad.State import Control.Monad.Error import Control.Applicative
-- lift an action over a state into a pair of states -- do1st :: State s1 a -> State (s1,s2) a do1st m1 = do (s1, s2) <- get let (a, s1') = runState m1 s1 put (s1',s2) return a
do2nd :: State s2 a -> State (s1,s2) a do2nd m2 = do (s1, s2) <- get let (a, s2') = runState m2 s2 put (s1,s2') return a
-- lift an action over a state with error -- into a pair of states with error -- do1 :: Error e => ErrorT e (State s1) a -> ErrorT e (State (s1,s2)) a do1 m1 = do (s1, s2) <- lift get let (ma,s1') = (runState . runErrorT) m1 s1 case ma of Left e -> throwError e Right a -> do lift $ put (s1',s2) return a
do2 :: Error e => ErrorT e (State s2) a -> ErrorT e (State (s1,s2)) a do2 m2 = do (s1, s2) <- lift get let (ma,s2') = (runState . runErrorT) m2 s2 case ma of Left e -> throwError e Right a -> do lift $ put (s1,s2') return a
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
participants (4)
-
Fumiaki Kinoshita -
Ki Yung Ahn -
lennart spitzner -
Roman Cheplyaka