Why can't scalac optimize tail recursion in certain scenarios?

Question

Why doesn't scalac (the Scala compiler) optimize tail recursion?

Code and compiler invocations that demonstrates this:

> cat foo.scala 
class Foo {
 def ifak(n: Int, acc: Int):Int = {  
   if (n == 1) acc  
   else ifak(n-1, n*acc)  
 }
}

> scalac foo.scala
> jd-gui Foo.class
import scala.ScalaObject;

public class Foo
  implements ScalaObject
{
  public int ifak(int n, int acc)
  {
    return ((n == 1) ? acc : 
      ifak(n - 1, n * acc));
  }
}

Mckelvey

note that JVM-level tailcall optimisation is contributed for java 7 see wikis.sun.com/display/mlvm/TailCalls – Preliminaries

Answer 1

Methods that can be overridden can NOT be tail recursive. Try this:

class Foo {
  private def ifak(n: Int, acc: Int): Int = {  
    if (n == 1) acc  
    else ifak(n-1, n*acc)  
  }
}

Gompers

+1 What's the rationale? ( I mean I can imagine it , but I would like to know ) – Cockloft

@OscarRyz: see #4786002 – Aquinas

Answer 2

Try this:

class Foo {
  def ifak(n: Int, acc: Int):Int = {
    if (n == 1) acc
    else ifak(n-1, n*acc)
  }
}

class Bar extends Foo {
  override def ifak(n: Int, acc: Int): Int = {
    println("Bar!")
    super.ifak(n, acc)
  }
}

val foobar = new Bar
foobar.ifak(5, 1)

Notice that ifak may be recursive, but it may not as well. Mark the class or the method final, and it shall probably made tail-recursive.

Radcliffe

Answer 3

Inner functions are also eligible for TCO.

Ferity