Consider the following implementaion sketch:
sealed trait Type
object Type {
case object Type1 extends Type
case object Type2 extends Type
}
sealed trait Data{
type T <: Type
}
object Data {
type Aux[TT] = Data{
type T = TT
}
case class Data1(i: Int){
type T = Type1.type
} extends Data
case class Data2(s: String){
type T = Type2.type
} extends Data
}
case class Main(
//lots of other fields
data: Data.Aux[T] forSome { type T <: Type}
)
// This method is supposed to accept the only Main's
// that have data of type Data.Aux[Type2.type]
def handleMainType2(
main: Main
): Unit = ???
The problem:
Having the case class
containing a field of an existential type is it possible to implement a method that would accept the only branch of the existential type.
Maybe shapeless could be helpful here?
CodePudding user response:
First of all, Data.Aux[T] forSome { type T <: Type}
can be written as Data.Aux[_]
and is just Data
implicitly[(Data.Aux[T] forSome { type T <: Type}) =:= Data] // compiles
implicitly[Data =:= (Data.Aux[T] forSome { type T <: Type})] // compiles
If you put println
inside handleMainType2
import scala.reflect.runtime.universe.{Type, TypeTag, typeOf}
def getType[A: TypeTag](a: A): Type = typeOf[A]
def handleMainType2(main: Main): Unit =
println(getType(main.data) "=" showRaw(getType(main.data)))
then
handleMainType2(Main(Data1(1)))
handleMainType2(Main(Data2("a")))
will print one of the following
(depending on how you define the type of Main
parameter: Data.Aux[T] forSome { type T <: Type}
, Data.Aux[_]
or Data
)
App.Data{type T = T}=RefinedType(List(TypeRef(ThisType(App), App.Data, List())), Scope(TypeName("T")))
App.Data{type T = _$1}=RefinedType(List(TypeRef(ThisType(App), App.Data, List())), Scope(TypeName("T")))
App.Data=TypeRef(ThisType(App), App.Data, List())
both times. So inside method handleMainType2
main.data
has type just Data
and Data1
/Data2
are indistinguishable by type. What is distinguishable is runtime class:
def handleMainType2(main: Main): Unit =
println(main.data.getClass)
//class App$Data$Data1
//class App$Data$Data2
So you can define
def handleMainType2(main: Main): Unit =
assert(main.data.getClass.isAssignableFrom(classOf[Data2]))
with runtime behavior.
If you want compile-time behavior then you can try to make handleMainType2
a macro and use runtime reflection in the macro
// in a different subproject
import scala.language.experimental.macros
import scala.reflect.macros.blackbox
def handleMainType2(main: Main): Unit = macro handleMainType2Impl
def handleMainType2Impl(c: blackbox.Context)(main: c.Tree): c.Tree = {
import c.universe._
val clazz = c.eval(c.Expr[Main](c.untypecheck(main))).data.getClass
if (!clazz.isAssignableFrom(classOf[Data2]))
c.abort(c.enclosingPosition, s"${clazz.getName} <:!< Data2")
else q"()"
}
handleMainType2(Main(Data1(1))) // doesn't compile
handleMainType2(Main(Data2("a"))) // compiles
You can even make the macro implicit if you prefer not to make handleMainType2
a macro itself.
trait IsData2[D <: Data with Singleton]
object IsData2 {
implicit def mkIsData2[D <: Data with Singleton]: IsData2[D] = macro mkIsData2Impl[D]
def mkIsData2Impl[D <: Data with Singleton : c.WeakTypeTag](c: whitebox.Context): c.Tree = {
import c.universe._
val clazz = c.eval(c.Expr[ValueOf[D]](c.untypecheck(
c.inferImplicitValue(weakTypeOf[ValueOf[D]], silent = false)
))).value.getClass
if(!clazz.isAssignableFrom(classOf[Data2])) {
c.abort(c.enclosingPosition, s"${weakTypeOf[D]} <:!< Data2")
} else q"new IsData2[${weakTypeOf[D]}] {}"
}
object App {
val m1: Main = Main(Data1(1))
val m2: Main = Main(Data2("a"))
}
def handleMainType2(main: Main)(implicit ev: IsData2[main.data.type]) = ()
handleMainType2(App.m1) // doesn't compile
handleMainType2(App.m2) // compiles
Notice that handleMainType2(Main(Data2("a")))
or even
val m2: Main = Main(Data2("a"))
handleMainType2(m2)
will not work.