我正在尝试在Idris中编写一个zip函数,该函数将任意多个相同长度(len
)的向量组合成一个HList
s的向量。
也就是说,我正在尝试概括以下功能:
module Zip
import Data.Vect
%default total
zip2 : (Vect len a, Vect len b) -> Vect len (a, b)
zip2 ([], []) = []
zip2 ((x :: xs), (y :: ys)) = (x, y) :: zip2 (xs, ys)
我使用向量定义自己的HList
(“异构列表”):
data HList : Vect n Type -> Type where
Nil : HList []
(::) : (x : a) -> (xs : HList as) -> HList (a :: as)
这里是zip2
函数的变体,它使用以下HList
:
zip2H : HList [Vect len a, Vect len b] -> Vect len (HList [a, b])
zip2H [[], []] = []
zip2H [(x :: xs), (y :: ys)] = [x, y] :: zip2H [xs, ys]
到目前为止,很好。
现在是一般情况。
任意多个要压缩的矢量的类型签名要复杂得多,但我相信我做对了。
n
是要压缩的向量数量。 len
是每个向量的长度:
vects : (len : Nat) -> Vect n Type -> Vect n Type
vects len as = map (\type => Vect len type) as
-- Example:
-- `vects len [a, b] = [Vect len a, Vect len b]`
-- You cannot pattern-match on types in Idris, so you cannot get an `a` from an `Vect len a`. Instead, I go the other way around in `zip` and pass my `a`s implicitly.
zip : {types : Vect (S n) Type} -> {len : Nat} -> HList (vects len types) -> Vect len (HList types)
现在我的问题是:我什至不能写zip
定义的左手边。类型检查器不断抱怨。
一个例子:
zip {n = Z} [xs] = ?zip_rhs1
zip xs = ?zip_rhs2
When checking left hand side of Zip.zip:
When checking an application of Zip.zip:
Type mismatch between
HList [a] (Type of [xs])
and
HList (Data.Vect.Vect n implementation of Prelude.Functor.Functor, method map (\type =>
Vect len
type)
types) (Expected type)
Specifically:
Type mismatch between
[a]
and
Data.Vect.Vect n implementation of Prelude.Functor.Functor, method map (\type =>
Vect len
type)
types
我想念什么?我是否以错误的方式使用了隐式参数?我需要写一些证明吗?有没有更好的方法来构造函数类型签名?
(我的Idris版本是1.3.1-git:a93d8c9
。)
编辑:使用HTNW的代码,我仍然会得到基本相同的错误:
module Zip
import Data.Vect
%default total
data HList : Vect n Type -> Type where
Nil : HList []
(::) : (x : a) -> (xs : HList as) -> HList (a :: as)
vects : (len : Nat) -> Vect n Type -> Vect n Type
vects len as = map (\type => Vect len type) as
multiUnCons : {len : Nat} -> {types : Vect n Type} ->
HList (vects (S len) types) -> (HList types, HList (vects len types))
multiUnCons {types = []} [] = ([], [])
multiUnCons {types = t :: ts} ((x :: xs) :: xss) with (multiUnCons xss)
| (ys, yss) = (x :: ys, xs :: yss)
zip : {types : Vect n Type} -> {len : Nat} ->
HList (vects len types) -> Vect len (HList types)
zip {len = Z} _ = []
zip {len = S n} xss with (multiUnCons xss)
| (ys, yss) = ys :: zip yss
testVectors : HList [Vect 3 Nat, Vect 3 Char]
testVectors = [[1, 2, 3], ['a', 'b', 'c']]
*zip> :t Zip.zip testVectors
(input):1:4-23:When checking an application of function Zip.zip:
Type mismatch between
HList [Vect 3 Nat, Vect 3 Char] (Type of testVectors)
and
HList (vects len types) (Expected type)
Specifically:
Type mismatch between
[Vect 3 Nat, Vect 3 Char]
and
Data.Vect.Vect n implementation of Prelude.Functor.Functor, method map (\type =>
Vect len
type)
types
解决方案: zip
需要更多信息:
*zip> the (Vect 3 (HList [Nat, Char])) (zip testVectors)
[[1, 'a'], [2, 'b'], [3, 'c']] : Vect 3 (HList [Nat, Char])
*zip> zip {types=[Nat, Char]} testVectors
[[1, 'a'], [2, 'b'], [3, 'c']] : Vect 3 (HList [Nat, Char])
答案 0 :(得分:0)
您也必须在types
上进行匹配。通过在types
上进行匹配,您还可以揭示关于vects len types
的信息,这使您可以进一步在HList (vects len types)
参数上进行匹配。另外,S n
上的types
长度要求是不必要的,也没有必要。最后,我认为您实际上需要首先在len
上递归,然后在types
上递归。最好将types
上的递归编写为其他函数:
multiUnCons : {len : Nat} -> {types : Vect n Type} ->
HList (vects (S len) types) -> (HList types, HList (vects len types))
multiUnCons {types = []} [] = ([], [])
multiUnCons {types = t :: ts} ((x :: xs) :: xss) with (multiUnCons xss)
| (ys, yss) = (x :: ys, xs :: yss)
zip
本身很简单:
zip : {types : Vect n Type} -> {len : Nat} ->
HList (vects len types) -> Vect len (HList types)
zip {len = Z} _ = []
zip {len = S n} xss with (multiUnCons xss)
| (ys, yss) = ys :: zip yss