协变类型E出现在类型A => returnerror.Error [E，B]值f的相反位置

Question

I have this definition:

sealed trait Error[+E, +A] {
  final def map[B](f: A => B): Error[E, B] = {
    this match {
      case Ok(a) => Ok(f(a))
      case Fail(e) => Fail(e)
    }
  }
  final def flatMap[B](f: A => Error[E, B]): Error[E, B] = {
    this match {
      case Ok(a) => f(a)
      case Fail(e) => Fail(e)
    }
  }
}

object Error {
  final case class Ok[A](a: A) extends Error[Nothing, A]
  final case class Fail[E](e: E) extends Error[E, Nothing]
}

This won't compile with this error:

Error.scala:12: covariant type E occurs in contravariant position in type A => returnerror.Error[E,B] of value f
  final def flatMap[B](f: A => Error[E, B]): Error[E, B] = {
                       ^
one error found

I don't get why. Can you, please, explain why this doesn't compile? Thanks

Answer 1

I will simplify the example a bit in order to make it easier to read.

Imagine you have trait like this and let's assume that such declaration is legal:

trait Error[+E]  {
  def someMethod (f: String => Error[E])  // actually illegal
}

And implementations

class ErrorString extends  Error[String]  {

  def someMethod (f: String => Error[String]) = {
    val someVar: Error[String] = f("Somestring")
  }

}

class ErrorInt extends  Error[Int]  {
    ...
} 

As E is covariant we can treat Error[String] as subtype of Error[Any] , so we can write the follwing

val anyError: Error[Any] = new ErrorString()

And then pass String => Error[Int] function as a parameter to someMethod

val intErrorReturnFunction: String => Error[Any] = (k) => new ErrorInt
anyError.someMethod(intErrorReturnFunction) // actually illegal

But as anyError is still have type Error[String] in a background means that we are trying to pass String => Error[Int] function to method with String => Error[String] parameter.

So at the end we would have something like this

 val someVar: Error[String] = f("Somestring"):Error[Int]