Question

IntHistogram counts words and displays them count-descending:

object IntHistogram {

  def main(args: Array[String]) {
    val wordsToCount = "foo foo bar foo bar wtf foo bar".split(" ")
    val histogram = new IntHistogram
    for (word <- wordsToCount) histogram(word) += 1
    println(histogram)
    /* 
       (foo,4)
       (bar,3)
       (wtf,1)
    */
  }

}

class IntHistogram extends collection.mutable.HashMap[String,Int] {
  override def default(key:String) = 0
  def descendingPairs = toList.sortBy(_._2).reverse
  override def toString() = descendingPairs.mkString("\n")
}

I needed a DoubleHistogram, and I resorted to copy-and-paste because I couldn’t figure out how to define a generic “Histogram[NumberSuperClassOfIntAndDouble]” trait:

class DoubleHistogram extends collection.mutable.HashMap[String,Double] {
  override def default(key:String) = 0
  def descendingPairs = toList.sortBy(_._2).reverse
  override def toString() = descendingPairs.mkString("\n")
}

Can a smart/knowledgeable person show me how to define such a supertrait, so I can avoid the ugly copy-and-paste boilerplate?

Thanks in advance,

PT

P.S. I really want Histogram to be a trait, so I can mix the Histogram behavior into any numerically-valued Map. In reality there are lots of behaviors I need, besides a descending toString method; I simplified it for this question. Here Num is the fictional numeric superclass of Int and Double:

trait Histogram[Num] extends collection.Map[String,Num] {
  override def default(key:String) = 0
  def descendingPairs = toList.sortBy(_._2).reverse
  override def toString() = descendingPairs.mkString("\n")
}

I tried using Numeric, Number, import Ordering.Implicits._, all kinds of stuff… to no avail.

Answer 1

This seems to work just fine with Numeric. Not that you can’t make Histogram a trait because of the context bound on N, but IntHistogram and DoubleHistogram can be traits.

object IntHistogram {

  def main(args: Array[String]) {
    val wordsToCount = "foo foo bar foo bar wtf foo bar".split(" ")
    val histogram = new IntHistogram {}
    for (word <- wordsToCount) histogram(word) += 1
    println(histogram)
    /* 
       (foo,4)
       (bar,3)
       (wtf,1)
    */
  }

}

abstract class Histogram[N:Numeric] extends collection.mutable.HashMap[String,N] {
  override def default(key:String) = implicitly[Numeric[N]].zero
  def descendingPairs = toList.sortBy(_._2).reverse
  override def toString = descendingPairs.mkString("\n")
}

trait IntHistogram extends Histogram[Int]
trait DoubleHistogram extends Histogram[Double]

If you really, really need Histogram to be a trait, you can do it this way:

trait Histogram[N] extends collection.mutable.HashMap[String,N] {
  implicit val n:Numeric[N]
  override def default(key:String) = n.zero
  def descendingPairs = toList.sortBy(_._2).reverse
  override def toString = descendingPairs.mkString("\n")
}

But then you have to instantiate it like this in main:

val histogram = new Histogram[Int] { val n = implicitly[Numeric[Int]] }