Type parameters and inheritance in Scala

Question

Is there a simple way to return a concrete type in an override method? And what about creating an instance of a concrete implementation? And calling chained methods implemented in the concrete class, so they return a correct type, too? I have a solution (based on https://stackoverflow.com/a/14905650) but I feel these things should be simpler.

There are many similar questions, but everyone's case is a little different, so here is another example (shortened from https://github.com/valdanylchuk/saiml/tree/master/src/main/scala/saiml/ga). When replying, if possible, please check if the whole code block compiles with your suggested change, because there are subtle cascading effects. I could not make this work with the "curiously recurring template pattern", for example (not that I find it nicer).

import scala.reflect.ClassTag
import scala.util.Random

abstract class Individual(val genome: String) {
  type Self
  def this() = this("")  // please override with a random constructor
  def crossover(that: Individual): Self
}
class HelloGenetic(override val genome: String) extends Individual {
  type Self = HelloGenetic
  def this() = this(Random.alphanumeric.take("Hello, World!".length).mkString)
  override def crossover(that: Individual): HelloGenetic = {
    val newGenome = this.genome.substring(0, 6) + that.genome.substring(6)
    new HelloGenetic(newGenome)
  }
}
class Population[A <: Individual {type Self = A} :ClassTag]( val size: Int,
    tournamentSize: Int, givenIndividuals: Option[Vector[A]] = None) {
  val individuals: Vector[A] = givenIndividuals getOrElse
    Vector.tabulate(size)(_ => implicitly[ClassTag[A]].runtimeClass.newInstance.asInstanceOf[A])
  def tournamentSelect(): A = individuals.head  // not really, skipped
  def evolve: Population[A] = {
    val nextGen = (0 until size).map { _ =>
      val parent1: A = tournamentSelect()
      val parent2: A = tournamentSelect()
      val child: A = parent1.crossover(parent2)
      child
    }.toVector
    new Population(size, tournamentSize, Some(nextGen))
  }
}
class Genetic[A <: Individual {type Self = A} :ClassTag](populationSize: Int, tournamentSize: Int) {
  def optimize(maxGen: Int, maxMillis: Long): Individual = {
    val first = new Population[A](populationSize, tournamentSize)
    val optPop = (0 until maxGen).foldLeft(first) { (pop, _) => pop.evolve }
    optPop.individuals.head
  }
}

Alexey Romanov · Accepted Answer · 2016-07-11 09:42:24Z

1

The CRTP version is

abstract class Individual[A <: Individual[A]](val genome: String) {
  def this() = this("")  // please override with a random constructor

  def crossover(that: A): A
}
class HelloGenetic(override val genome: String) extends Individual[HelloGenetic] {
  def this() = this(Random.alphanumeric.take("Hello, World!".length).mkString)
  override def crossover(that: HelloGenetic): HelloGenetic = {
    val newGenome = this.genome.substring(0, 6) + that.genome.substring(6)
    new HelloGenetic(newGenome)
  }
}
class Population[A <: Individual[A] :ClassTag]( val size: Int,
    tournamentSize: Int, givenIndividuals: Option[Vector[A]] = None) {
  val individuals: Vector[A] = givenIndividuals getOrElse
    Vector.tabulate(size)(_ => implicitly[ClassTag[A]].runtimeClass.newInstance.asInstanceOf[A])
  def tournamentSelect(): A = individuals.head  // not really, skipped

  def evolve: Population[A] = {
    val nextGen = (0 until size).map { _ =>
      val parent1: A = tournamentSelect()
      val parent2: A = tournamentSelect()
      val child: A = parent1.crossover(parent2)
      child
    }.toVector
    new Population(size, tournamentSize, Some(nextGen))
  }
}
class Genetic[A <: Individual[A] :ClassTag](populationSize: Int, tournamentSize: Int) {
  def optimize(maxGen: Int, maxMillis: Long): Individual[A] = {
    val first = new Population[A](populationSize, tournamentSize)
    val optPop = (0 until maxGen).foldLeft(first) { (pop, _) => pop.evolve }
    optPop.individuals.head
  }
}

which compiles. For creating the instances, I'd suggest just passing functions:

class Population[A <: Individual[A]](val size: Int,
    tournamentSize: Int, makeIndividual: () => A, givenIndividuals: Option[Vector[A]] = None) {
  val individuals: Vector[A] = givenIndividuals getOrElse
    Vector.fill(size)(makeIndividual())
  ...
}

If you want to pass them implicitly, you can easily do so:

trait IndividualFactory[A] {
  def apply(): A
}

class HelloGenetic ... // remove def this() 
object HelloGenetic {
  implicit val factory: IndividualFactory[HelloGenetic] = new IndividualFactory[HelloGenetic] {
    def apply() = new HelloGenetic(Random.alphanumeric.take("Hello, World!".length).mkString)
  }
}
class Population[A <: Individual[A]](val size: Int,
    tournamentSize: Int, givenIndividuals: Option[Vector[A]] = None)
    (implicit factory: IndividualFactory[A]) {
  val individuals: Vector[A] = givenIndividuals getOrElse
    Vector.fill(size)(factory())
  ...
}

answered Jul 11, 2016 at 9:42

Alexey Romanov

172k38 gold badges329 silver badges509 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

Valentyn Danylchuk Over a year ago

Thank you! The CRTP version does look a bit cleaner. For creating instances, I will try to play around passing the function. Not sure if I can make it work without requiring the library user to implement a companion object as well as the HelloGenetic class. I want to keep the interface simple, even at the cost of reflection in my internal implementation. I will try this on the full project and see if there are more suggestions before marking this as the accepted answer, although it definitely qualifies.

Valentyn Danylchuk Over a year ago

Yes, the CRTP works perfectly for all my cases in this project. I was not consistent enough when I tried it earlier.

Collectives™ on Stack Overflow

Type parameters and inheritance in Scala

1 Answer 1

2 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related