我试图在Haskell中复制这段Idris代码,该代码通过类型强制执行正确的动作排序:
data DoorState = DoorClosed | DoorOpen
data DoorCmd : Type ->
DoorState ->
DoorState ->
Type where
Open : DoorCmd () DoorClosed DoorOpen
Close : DoorCmd () DoorOpen DoorClosed
RingBell : DoorCmd () DoorClosed DoorClosed
Pure : ty -> DoorCmd ty state state
(>>=) : DoorCmd a state1 state2 ->
(a -> DoorCmd b state2 state3) ->
DoorCmd b state1 state3
由于(>>=)
运算符的重载,可以编写如下的monadic代码:
do Ring
Open
Close
但编译器拒绝不正确的转换,如:
do Ring
Open
Ring
Open
我试图在下面的Haskell片段中遵循这种模式:
data DoorState = Closed | Opened
data DoorCommand (begin :: DoorState) (end :: DoorState) a where
Open :: DoorCommand 'Closed 'Opened ()
Close :: DoorCommand 'Opened 'Closed ()
Ring :: DoorCommand 'Closed 'Closed ()
Pure :: x -> DoorCommand b e x
Bind :: DoorCommand b e x -> (x -> DoorCommand e f y) -> DoorCommand b f y
instance Functor (DoorCommand b e) where
f `fmap` c = Bind c (\ x -> Pure (f x))
-- instance Applicative (DoorCommand b e) where
-- pure = Pure
-- f <*> x = Bind f (\ f' -> Bind x (\ x' -> Pure (f' x')))
-- instance Monad (DoorCommand b e) where
-- return = Pure
-- (>>=) = Bind
但当然失败了:Applicative
和Monad
实例无法正确定义,因为它们需要两个不同的实例才能正确排序操作。构造函数Bind
可用于强制执行正确的排序,但我无法使用&#34;更好的&#34;做-符号。
我如何编写此代码以便能够使用do-notation,例如防止Command
s的无效序列?
答案 0 :(得分:8)
你正在寻找的确是Atkey的parameterised monad,现在更常被称为索引monad 。
class IFunctor f where
imap :: (a -> b) -> f i j a -> f i j b
class IFunctor m => IMonad m where
ireturn :: a -> m i i a
(>>>=) :: m i j a -> (a -> m j k b) -> m i k b
IMonad
是一类类似monad的事物m :: k -> k -> * -> *
,用于描述属于k
种类型的有向图的路径。 >>>=
绑定计算,该计算将类型级状态从i
转换为j
到将j
从k
转移到i
的计算,返回更大的计算从k
到ireturn
。 IMonad
允许您将纯值提升为不会改变类型级别状态的monadic计算。
我将使用索引的免费monad 来捕获此类请求 - 响应操作的结构,主要是因为我不想弄清楚如何为自己的类型编写data IFree f i j a where
IReturn :: a -> IFree f i i a
IFree :: f i j (IFree f j k a) -> IFree f i k a
instance IFunctor f => IFunctor (IFree f) where
imap f (IReturn x) = IReturn (f x)
imap f (IFree ff) = IFree $ imap (imap f) ff
instance IFunctor f => IMonad (IFree f) where
ireturn = IReturn
IReturn x >>>= f = f x
IFree ff >>>= f = IFree $ imap (>>>= f) ff
实例:
Door
我们可以从以下仿函数中免费构建您的data DoorState = Opened | Closed
data DoorF i j next where
Open :: next -> DoorF Closed Opened next
Close :: next -> DoorF Opened Closed next
Ring :: next -> DoorF Closed Closed next
instance IFunctor DoorF where
imap f (Open x) = Open (f x)
imap f (Close x) = Close (f x)
imap f (Ring x) = Ring (f x)
type Door = IFree DoorF
open :: Door Closed Opened ()
open = IFree (Open (IReturn ()))
close :: Door Opened Closed ()
close = IFree (Close (IReturn ()))
ring :: Door Closed Closed ()
ring = IFree (Ring (IReturn ()))
monad:
open
你可以close
一扇门,导致当前关闭的门打开,ring
当前打开的门,或RebindableSyntax
门保持关闭,大概是因为这个房子的居住者并不想见到你。
最后,Monad
语言扩展意味着我们可以使用自己的自定义IMonad
替换标准(>>=) = (>>>=)
m >> n = m >>>= const n
return = ireturn
fail = undefined
door :: Door Open Open ()
door = do
close
ring
open
类。
Open
但是我注意到你并没有真正使用monad的绑定结构。您的所有构建基块Close
,Ring
或data Path g i j where
Nil :: Path g i i
Cons :: g i j -> Path g j k -> Path g i k
都不会返回任何值。所以我认为你真正需要的是以下更简单的类型对齐列表类型:
Path :: (k -> k -> *) -> k -> k -> *
在操作上,k
就像一个链表,但它有一些额外的类型级结构,再次描述通过有节点在g
中的有向图的路径。列表的元素是边Nil
。 i
表示您始终可以找到从节点Cons
到自身的路径,i
提醒我们千里之行始于一步:如果您有一个优势{ {1}}到j
以及从j
到k
的路径,您可以将它们组合在一起,形成从i
到k
的路径。它被称为类型对齐列表,因为一个元素的结束类型必须与下一个元素的起始类型匹配。
在Curry-Howard Street的另一边,如果g
是二元逻辑关系,那么Path g
构造其自反传递闭包。或者,Path g
分类为图g
的免费类别中的态射类型。在自由类别中编写态射只是(翻转)附加类型对齐列表。
instance Category (Path g) where
id = Nil
xs . Nil = xs
xs . Cons y ys = Cons y (xs . ys)
然后我们可以用Door
:
Path
data DoorAction i j where
Open :: DoorAction Closed Opened
Close :: DoorAction Opened Closed
Ring :: DoorAction Closed Closed
type Door = Path DoorAction
open :: Door Closed Opened
open = Cons Open Nil
close :: Door Opened Closed
close = Cons Close Nil
ring :: Door Closed Closed
ring = Cons Ring Nil
door :: Door Open Open
door = open . ring . close
你没有得到do
符号(虽然我认为 RebindableSyntax
确实允许你重载列表文字),但用(.)
构建计算看起来像纯函数的排序,我认为这对你正在做的事情来说是一个相当不错的类比。对我而言,它需要额外的智力 - 一种稀有而珍贵的自然资源 - 才能使用索引monad。当一个更简单的结构可以做到时,避免monad的复杂性会更好。