使用多态ADT类型检查的错误Scala代码

在下面的示例中,我正在编写IO monad的堆栈安全解释器.代码在多态代数数据类型(FlatMap [A??,B])上进行模式匹配.代码中的错误是k1,然后是k2;这两个函数不能组合,因为k1返回的不同类型(IO [B])比k2期望的(B).代码仍以某种方式进行类型检查,这显然是一个类型检查错误,因为在运行时,自动取消装箱时会出现ClassCastException(就像我在Java中使用不安全的强制转换一样).也没有发出编译器警告.

代码(也见于gist)：

object IOMonadExercise extends App {

  sealed trait IO[A]    
  case class Return[A](value: A) extends IO[A]    
  case class Suspend[A](f: () => A) extends IO[A]    
  case class FlatMap[A,B](io: IO[A],cont: A => IO[B]) extends IO[B]

  object IO {
    def apply[A](a: => A): IO[A] = Suspend(() => a)
  }

  object Interpreter {
    def run[A](io: IO[A]): A = {
      io match {
        case Return(a) => a
        case Suspend(f) => f()

        case FlatMap(Return(a),cont) => run(cont(a))
        case FlatMap(Suspend(f),cont) => run(cont(f()))

        // this case compiles for whatever reason but shouldn't type check (k1 returns IO[B] and k2 expects just B)
        // accordingly,there is a ClassCastException in the runtime
        case FlatMap(FlatMap(io1,k1),k2) => run(FlatMap(io1,k1 andThen k2))

        // this case is the one that actually works
//        case FlatMap(FlatMap(io1,k2) => run(flatten(io1,k1,k2))
      }
    }

    def flatten[A,B,C](io: IO[A],k1: A => IO[B],k2: B => IO[C]): FlatMap[A,C] = {
      FlatMap(io,a => FlatMap(k1(a),k2))
    }
  }


  def sum(i: Int): IO[Int] = {
    Stream.range(0,i).foldLeft(IO(0))((io,i) => FlatMap(io,(s: Int) => IO(s + i)))
  }

  val n = 100000
  val sumNIO: IO[Int] = sum(n)
  val sumN: Int = Interpreter.run(sumNIO)
  println(s"sum of 1..$n by IO loop  : $sumN")
  println(s"sum of 1..$n by math expr: ${n * (n - 1) / 2}")
  assert(sumN == n * (n - 1) / 2)
}

到底是怎么回事？这是编译器错误吗？或者这是类型推断的已知限制？或者有解释吗？

我已经在Scala 2.11.8和2.12.0上进行了测试,行为似乎是相同的：代码编译时没有警告.