haskell / haskell/vector

Possible Changes to Bundle.Chunk?

未关闭
#91 4 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看
design-shed enhancement question
主要语言
Haskell
星标
400
派生
145
PR 合并指标
30 天内没有已合并 PR

描述

As it is, the Chunk type in Data.Vector.Fusion.Bundle.Monadic is much less useful than it could be.Here are some ways, ranked by desirability and compatibility-breaking, on how it could be better.

1) Add a Monoid instance and utility functions for Chunk.
This is the simplest change, and one that breaks the fewest programs. The most general utility functions would probably be toChunk :: (Vector v a) => v a -> Chunk v a and fromChunk :: (Vector v a) => Chunk v a -> v a, and possibly singleton :: (Vector v a) => a -> Chunk v a. This would let Chunks be used like ByteString Builders, as a low-overhead, O(1) concatenation option.

The definition for Monoid would be:

``` haskell
import qualified Data.Vector.Generic.Mutable as M
instance Monoid (Chunk v a) where
mempty = Chunk 0 (const $ return ())
mappend (Chunk na wa) (Chunk nb wb) = Chunk (na + nb) (\v -> wa (M.take na v) >> wb (M.drop na v))
```

2) Change the type of Chunk.
This would involve a Yoneda transform on Chunk's type:

``` haskell
data Chunk v a = Chunk Int (forall m r. (PrimMonad m, Vector v r) => Mutable v (PrimState m) r -> (a -> r) -> m ())
```

This would turn Chunk into a Functor, meaning the Functor instance for Bundle could be a lot simpler and involve a lot less copying. Also, this prevents badly-behaved Chunks from reading and mutating the MVector they are given.

3) Change the kind of Chunk.
If you generalize, then you can make Chunks that can construct any type of Vector:

``` haskell
data Chunk a = Chunk Int (forall m v r. (PrimMonad m, Vector v r) => Mutable v (PrimState m) r -> (a -> r) -> m ())
```

In addition to making Chunks able to produce any kind of Vector, this also admits an Applicative instance for Chunk!

``` haskell
import qualified Data.Vector.Generic.Mutable as M
import qualified Data.Vector.Mutable as MV
instance Applicative Chunk where
pure a = Chunk 1 (\v c -> M.write v 0 (c a))
(Chunk nf wf) <*> (Chunk na wa) = Chunk (nf * na) $ \v c -> do
vf <- MV.new nf
wf vf id
let loop n = if n >= nf then return () else do
f <- MV.read vf n
wa (M.slice (n * na) na v) (c . f)
loop (n + 1)
loop 0
(Chunk na _) *> (Chunk nb wb) = Chunk (na * nb) $ \v c -> do
let loop n = if n >= na then return () else wb (M.slice (n * nb) nb) c >> loop (n + 1)
loop 0
```

This also makes it an Alternative, with empty and <|> defined as mempty and mappend, respectively.

If these changes are made, then it would probably be a good idea to move Chunk from its current location and give it a dedicated import, like Data.Vector.Chunk, for use in building Vectors cheaply.

So is it possible to make these changes and make Chunk more than just a utility type for Bundle?

贡献指南

这个仓库没有索引到贡献指南

评估

这个 Issue 还没有评估数据。

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。