问题描述
这个问题实际上是由于尝试将几个数学组实现为类型而引起的.
This problem actually emerged from attempt to implement few mathematical groups as types.
循环组没有问题(在其他地方定义的Data.Group
实例):
Cyclic groups have no problem (instance of Data.Group
defined elsewhere):
newtype Cyclic (n :: Nat) = Cyclic {cIndex :: Integer} deriving (Eq, Ord)
cyclic :: forall n. KnownNat n => Integer -> Cyclic n
cyclic x = Cyclic $ x `mod` toInteger (natVal (Proxy :: Proxy n))
但是对称组在定义某些实例时存在一些问题(通过阶乘数系统实现):
But symmetric groups have some problem on defining some instances (implementation via factorial number system):
infixr 6 :.
data Symmetric (n :: Nat) where
S1 :: Symmetric 1
(:.) :: (KnownNat n, 2 <= n) => Cyclic n -> Symmetric (n-1) -> Symmetric n
instance {-# OVERLAPPING #-} Enum (Symmetric 1) where
toEnum _ = S1
fromEnum S1 = 0
instance (KnownNat n, 2 <= n) => Enum (Symmetric n) where
toEnum n = let
(q,r) = divMod n (1 + fromEnum (maxBound :: Symmetric (n-1)))
in toEnum q :. toEnum r
fromEnum (x :. y) = fromInteger (cIndex x) * (1 + fromEnum (maxBound `asTypeOf` y)) + fromEnum y
instance {-# OVERLAPPING #-} Bounded (Symmetric 1) where
minBound = S1
maxBound = S1
instance (KnownNat n, 2 <= n) => Bounded (Symmetric n) where
minBound = minBound :. minBound
maxBound = maxBound :. maxBound
ghci的错误消息(仅简短地):
Error message from ghci (only briefly):
Overlapping instances for Enum (Symmetric (n - 1))
Overlapping instances for Bounded (Symmetric (n - 1))
那么GHC如何知道n-1
是否等于1?我还想知道是否可以在不使用FlexibleInstances
的情况下编写解决方案.
So how can GHC know whether n-1
equals to 1 or not? I'd also like to know whether the solution can be written without FlexibleInstances
.
推荐答案
添加Bounded (Symmetric (n-1))
和Enum (Symmetric (n-1))
作为约束,因为要完全解决这些约束将需要知道n的确切值.
Add Bounded (Symmetric (n-1))
and Enum (Symmetric (n-1))
as constraints, because fully resolving those constraints would require knowing the exact value of n.
instance (KnownNat n, 2 <= n, Bounded (Symmetric (n-1)), Enum (Symmetric (n-1))) =>
Enum (Symmetric n) where
...
instance (KnownNat n, 2 <= n, Bounded (Symmetric (n-1))) =>
Bounded (Symmetric n) where
...
要避免使用FlexibleInstances
(这对IMO来说不值得,FlexibleInstances
是一个良性扩展),请使用Peano数字data Nat = Z | S Nat
代替GHC的原始表示形式.首先用Bounded (Symmetric (S (S n')))
替换实例头Bounded (Symmetric n)
(这扮演约束2 <= n
的角色),然后使用辅助类(可能更多)分解实例,以满足实例头的标准要求.可能看起来像这样:
To avoid FlexibleInstances
(which is not worth it IMO, FlexibleInstances
is a benign extension), use Peano numbers data Nat = Z | S Nat
instead of GHC's primitive representation. First replace the instance head Bounded (Symmetric n)
with Bounded (Symmetric (S (S n')))
(this plays the role of the constraint 2 <= n
), and then break up the instance with an auxiliary class (possibly more) to satisfy the standard requirement on instance heads. It might look like this:
instance Bounded_Symmetric n => Bounded (Symmetric n) where ...
instance Bounded_Symmetric O where ...
instance Bounded_Symmetric n => Bounded_Symmetric (S n) where ...
这篇关于通过Nat-kind重叠实例的文章就介绍到这了,希望我们推荐的答案对大家有所帮助,也希望大家多多支持!