How to avoid stack overflow when using scalaz's free monad?

Question

I had previously thought that part of the goal of the implementation was to avoid this very problem, so maybe I'm doing something obviously dumb?

Here is some code:

    // Stack overflow
import scalaz._

sealed trait Command[T]
case class Wait(ms: Long) extends Command[Unit]

case object Evaluator extends (Command ~> Id.Id) {
  override def apply[T](cmd: Command[T]) = cmd match {
    case Wait(t)  => Thread.sleep(t)
  }
}

object Api {
  def sleep(ms: Long): Free.FreeC[Command, Unit] = Free.liftFC(Wait(ms))
}

val sleep: Free.FreeC[Command, Unit] =
  Api.sleep(1).flatMap { _ => sleep }

Free.runFC(sleep)(Evaluator)

Note: I realize this is silly :) In practice, my command class has many commands, and I have a command which does this same loop...basically, poll some state, if true abort, if false, keep waiting.

I want to avoid the stack overflow that this causes... I THOUGHT this was already trampolined, but I guess I need to manually do it again? Is there a clean way to do it within the free monad way of thinking?

Update:

Thinking further on this, I think the issue isn't the sleep Free Monad but rather the Id.Id monad taht we bind into on evaluation... so I tried something like:

case object Evaluator2 extends (Command ~> ({ type t[x] = Free[Id.Id, x] })#t) {
  override def apply[T](cmd: Command[T]) = cmd match {
    case Wait(t)  => Thread.sleep(t); Free.liftF[Id.Id, Unit](())
  }
}

Free.runFC[Command, ({ type t[x] = Free[Id.Id, x] })#t, Unit](sleep)(Evaluator2)(Free.freeMonad[Id.Id])

But the problem with this is that it will only evaluate one step. Ideally I would like runFC to block until some condition is satisfied (or in this case, to loop forever until I kill it, but without a stack overflow)

Answer 1

The Id monad is not trampolined. You end up in an infinite mutual recursion between the bind method for the Id monad and the foldMap method for the free monad. Use Trampoline or Task instead of Id.