Simple utility for Spark for Coproducts functionality emulated as TupleX[(None, None, Some(Thing3), ... None)] where a coproduct [Thing1, Thing2, Thing3... ThingN] is desired.
Here is a simple use case. Given the following entities
case class Person(...)
case class Robot(...)
You are you are attempting to create:
PersonOrRobot = Person | Robot
This can be emulated as:
(Option[Person], Option[Robot]) // where one variant will always be None
In order to achieve this without major boilerplate pains, this library can be used in the following way:
import import com.github.ctl.trivialspark.TupleSumFactories.TupleSumFactory2 // Factories exist for Tuple1-22
object Model {
val factory = TupleSumFactory[Person, Robot]
type PersonOrRobot = factory.TupleSum // a type alias for (Option[Person], Option[Robot])
}
object UserCode {
val people: Dataset[Person] = ...
val robots: Dataset[Robot] = ...
val peopleOrRobots1: Dataset[PersonOrRobot] = people.map(factory.construct1(_)) // use 'construct1' to create (Some[Person], None)
val peopleOrRobots2: Dataset[PersonOrRobot] = robots.map(factory.construct2(_)) // use 'construct2' to create (None, Some[Robot])
// A union of this dataset can now be taken
val union: Datset[PersonOrRobot] = peopleOrRobots1 union peopleOrRobots2
// ...
}This library then creates extensions methods for easily unpacking the coproduct-tuple
import com.github.ctl.trivialspark.TupleSumExtensions._
object UserCode {
// ...
val output: Dataset[SomethingElse] = union.map { personOrRobot =>
personOrRobot.fold {
(p: Person) => SomethingElse(p.name, p.age)
(r: Robot) => SomethingElse(r.id, r.yearsService)
}
}
}This allows easy mapping to some other data-type.
