Error when compiling to ML backend about effekt HOT 2 OPEN

I am sorry, but with the MLton backend you are working on the bleeding edge. This is indeed a known issue that we still need to address. ML does not have support for rank-2 types, which are required by our translation to system-f. We are working on #229 to avoid this.

The easiest workaround is avoiding the interaction of mutable state, recursive functions (while) and calling the continuation under the recursive function for now. This is not really satisfying and we are working on better solutions.

EDIT: I believe the issue was actually not caused by mutable state and recursive functions, but by the fact that Choice contained both effect operations choose and fail. They are naturally called at different answer types (int and unit, the result of the operations). MLton will try to monomorphize the evidence passed for Choice but it cannot, because it would need to work for int as well as unit (this problem is dealt with in #229 -- but some other problems resurfaced there).

I slightly massaged your program manually to avoid the rank-2 problem and it should work now:

effect Choose {
  def choose(n: Int): Int
}
effect Fail {
  def fail(): Unit
}
effect Choice = {Choose,Fail}

effect Yield {
  def yield(a: Int, b: Int, c: Int): Unit
}

def triple(n: Int, s: Int): Unit / {Choice, Yield} = {
  val i = do choose(n)
  val j = do choose(i - 1)
  val k = do choose(j - 1)
  if (i + j + k == s) {
    do yield(i, j, k)
  } else {
    do fail()
  }
}

def yieldTriples(n: Int, s: Int): Unit / {Yield} = {
  try {
    triple(n, s)
  } with Choose {
    def choose(n) = {
      var i = 1;
      while (i < n) {
        resume(i);
        i = i + 1
      }
    }
  } with Fail {
    def fail() = ()
  }
}

def countTriples(n: Int, s: Int): Int = {
  var cnt = 0;
  try {
    yieldTriples(n, s)
  } with Yield {
    def yield(i: Int, j: Int, k: Int) = {
      cnt = cnt + 1;
      resume(())
    }
  };
  cnt
}

def main() = println(countTriples(8, 6))

from effekt.

BTW, if you are benchmarking this; I would be interested to learn about your findings :)

from effekt.