Spark Scala Dataset Type Hierarchy

Question

Trying to enforce classes that extend W to have a method get that returns a Dataset of a subclass of a WR.

abstract class WR

case class TGWR(
          a: String,
          b: String
        ) extends WR

abstract class W {

  def get[T <: WR](): Dataset[T]

}


class TGW(sparkSession: SparkSession) extends W {

  override def get[TGWR](): Dataset[TGWR] = {
    import sparkSession.implicits._

    Seq(TGWR("dd","dd").toDF().as[TGWR]
  }

}

Compilation error:

Unable to find encoder for type stored in a Dataset.  Primitive types (Int, String, etc) and Product types (case classes) are supported by importing spark.implicits._  Support for serializing other types will be added in future releases.

If I change the get function to following:

  def get(): Dataset[TGWR]

and

  override def get(): Dataset[TGWR] = {...

it compiles - therefore I suspect a problem due to inheritance/type hierarchy.

Answer 1

Forget my comment, I re-read your question and noticed a simple problem.

Here override def get[TGWR] you are not saying that this class produces instances of TGWR, but you are creating a new type parameter of name TGWR, that will shadow your real type.
I fixed it with the following code:

import org.apache.spark.sql.{SparkSession, Dataset}

abstract class WR extends Product with Serializable

final case class TGWR(a: String, b: String) extends WR

abstract class W[T <: WR] {
  def get(): Dataset[T]
}

final class TGW(spark: SparkSession) extends W[TGWR] {
  override def get(): Dataset[TGWR] = {
    import spark.implicits._
    Seq(TGWR("dd","dd")).toDF().as[TGWR]
  }
}

That you can use right this:

val spark = SparkSession.builder.master("local[*]").getOrCreate()
(new TGW(spark)).get()
// res1: org.apache.spark.sql.Dataset[TGWR] = [a: string, b: string]
res1.show()
// +---+---+
// |  a|  b|
// +---+---+
// | dd| dd|
// +---+---+

Hope this is what you are looking for.
Do not doubt to ask for clarification.