Scala Kleisli 在 IntelliJ 中抛出错误

Question

尝试在 Scala 中为虚构的部分类型实现 Kleisli 类别(阅读 Bartosz Milewski 的“程序员的类别理论”，这是第 4 章的练习)

object Kleisli {
  type Partial[A, B] = A => Option[B]

  implicit class KleisliOps[A, B](f1: Partial[A, B]) {

    def >=>[C](f2: Partial[B, C]): Partial[A, C] =
      (x: A) =>
        for {
          y <- f1(x)
          z <- f2(y)
        } yield z

    def identity(f: Partial[A, B]): Partial[A, B] = x => f(x)

  }

  val safeRecip: Partial[Double, Double] = {
    case 0d => None
    case x => Some(1d / x)
  }

  val safeRoot: Partial[Double, Double] = {
    case x if x < 0 => None
    case x => Some(Math.sqrt(x))
  }
  
  val safeRootRecip: Partial[Double, Double] = safeRoot.>=>(safeRecip) 
  
  safeRootRecip(1d)
  safeRootRecip(10d)
  safeRootRecip(0d)
}

IDE (IntelliJ) 没有显示任何错误，但是当我运行这段代码时，我得到:

Error:(27, 57) value >=> is not a member of $line5.$read.$iw.$iw.Kleisli.Partial[Double,Double]
val safeRootRecip: Partial[Double, Double] = safeRoot.>=>(safeRecip)

在隐式类之外定义 >=>=> 效果很好。可能是什么原因？

Answer 1

@sinanspd 是对的。在 Dotty 中，代码似乎可以编译: https://scastie.scala-lang.org/n17APWgMQkWqy93ct2cghw

手动解决

val safeRootRecip: Partial[Double, Double] = KleisliOps(safeRoot).>=>(safeRecip)

编译但编译器本身没有发现这个转换

Information: KleisliOps{<null>} is not a valid implicit value 
  for App.safeRoot.type => ?{def >=> : ?} because:
type mismatch;
 found   : App.safeRoot.type (with underlying type App.Partial[Double,Double])
 required: App.Partial[A,Double]
    (which expands to)  A => Option[Double]
  val safeRootRecip: Partial[Double, Double] = safeRoot.>=>(safeRecip)

似乎没有推断类型参数 A。

(顺便说一下，Martin Odersky 在这里解释了为什么语言中存在隐式转换会使类型推断变得更糟:https://contributors.scala-lang.org/t/can-we-wean-scala-off-implicit-conversions/4388)

尝试使 Partial 与 B 协变，并且(特别是)与 A 协变(类似于 A = > Option[B] 与 B 协变，与 A 协变)

type Partial[-A, +B] = A => Option[B]

然后代码似乎可以编译。

另一种解决方法是用类型类 (MyTransform) 和隐式转换 (X => Y, KleisliOps) 替换隐式转换 ( myConversion) 根据此类型类定义(sometimes 这有助于隐式转换)

trait MyTransform[X, Y] {
  def transform(x: X): Y
}
implicit def myConversion[X, Y](x: X)(implicit mt: MyTransform[X, Y]): Y = 
  mt.transform(x)

type Partial[A, B] = A => Option[B]

implicit def partialToKleisliOps[A, B]: MyTransform[Partial[A, B], KleisliOps[A, B]] = 
  f1 => new KleisliOps(f1)
class KleisliOps[A, B](f1: Partial[A, B]) {    
  def >=>[C](f2: Partial[B, C]): Partial[A, C] =
    (x: A) =>
      for {
        y <- f1(x)
        z <- f2(y)
      } yield z

  def identity(f: Partial[A, B]): Partial[A, B] = x => f(x)
}