我正在尝试像这样写一个固定大小的矢量:
{-# LANGUAGE GADTs, DataKinds, KindSignatures, TypeOperators #-}
import GHC.TypeLits
data NVector (n :: Nat) a where
Nil :: NVector 0 a
Cons :: a -> NVector n a -> NVector (n + 1) a
instance Eq a => Eq (NVector n a) where
Nil == Nil = True
(Cons x xs) == (Cons y ys) = x == y && xs == ys
但无法使用此消息进行编译:
Could not deduce (n2 ~ n1)
from the context (Eq a)
bound by the instance declaration at prog.hs:8:10-33
or from (n ~ (n1 + 1))
bound by a pattern with constructor
Cons :: forall a (n :: Nat). a -> NVector n a -> NVector (n + 1) a,
in an equation for `=='
at prog.hs:10:6-14
or from (n ~ (n2 + 1))
bound by a pattern with constructor
Cons :: forall a (n :: Nat). a -> NVector n a -> NVector (n + 1) a,
in an equation for `=='
at prog.hs:10:21-29
但如果我手动引入类型级自然,它会成功编译
{-# LANGUAGE GADTs, DataKinds, KindSignatures, TypeOperators, TypeFamilies #-}
data Nat = Z | S Nat
infixl 6 :+
type family (n :: Nat) :+ (m :: Nat) :: Nat
type instance Z :+ m = m
type instance (S n) :+ m = S (n :+ m)
data NVector (n :: Nat) a where
Nil :: NVector Z a
Cons :: a -> NVector n a -> NVector (S n) a
instance (Eq a) => Eq (NVector n a) where
Nil == Nil = True
(Cons x xs) == (Cons y ys) = x == y && xs == ys
ghc版本7.8.3
答案 0 :(得分:7)
ghc
无法(尚未?)从n ~ n'
推断出类型相等(n+1) ~ (n'+1)
从S n ~ S n'
中推断它是没有问题的。 Append for type-level numbered lists with TypeLits有一个解释和一个可能的出路(即同时拥有Peano风格的自然,仍然可以使用像5
这样的文字)
但是,如果您将Nvector
的定义更改为
data NVector (n :: Nat) a where
Nil :: NVector 0 a
Cons :: a -> NVector (n -1) a -> NVector n a
它必须从n-1 ~ n'-1
中推断n ~ n'
,这是一个更容易的推论!这编译,并仍然产生一个正确的类型,例如, Cons () Nil
:
*Main> :t Cons () Nil
Cons () Nil :: NVector 1 ()
请注意,这是无用的,因为我们仍然无法定义
append :: NVector n a -> NVector m a -> NVector (n + m) a -- won't work
2014年10月ghc
{{1}}说:
Iavor Diatchki正致力于在GHC的约束求解器中使用现成的SMT求解器。目前,主要关注点是使用类型级自然数[...]
改进对推理的支持
所以你的例子可能适用于ghc 7.10或7.12!