support dependent argument types for constructor parameter lists (in the same way as for methods)

[scabug]

scala> import scala.tools.nsc._
import scala.tools.nsc._

scala> def mkReifier(global: Global)(typer: global.analyzer.Typer) = typer
mkReifier: (global: scala.tools.nsc.Global)(typer: global.analyzer.Typer)global.analyzer.Typer

scala> class Reifier(global: Global)(typer: global.analyzer.Typer) { }
<console>:10: error: not found: value global
      class Reifier(global: Global)(typer: global.analyzer.Typer) { }
                                           ^

[scabug]

Imported From: https://issues.scala-lang.org/browse/SI-5712?orig=1
Reporter: @paulp
Affected Versions: 2.10.0, 2.11.6
See #5700

[scabug]

@retronym said:
Just happened across this looking into #5070, it struck me that it might need to be expanded for dependent constructor types.

override def isImmediatelyDependent = (sym ne NoSymbol) && (sym.owner.isMethod && sym.isValueParameter)

[scabug]

@adriaanm said (edited on Jul 26, 2012 9:37:36 AM UTC):
I wonder why we're using an outer namer for parameter accessors.
Commenting out these lines fixes the test case.

    def namerOf(sym: Symbol): Namer = {
      val usePrimary = sym.isTerm && (
//           (sym.isParamAccessor)
//        ||
        (sym.isParameter && sym.owner.isPrimaryConstructor)
      )

      if (usePrimary) createPrimaryConstructorParameterNamer
      else innerNamer
    }

Here's a self-contained test case that also makes sure the constructor's type parameters stay in scope.

class Global {
  class Typer[T]
}

class Reifier[T](global: Global)(typer: global.Typer[T])

object Test {
  def mkReifier[T](global: Global)(typer: global.Typer[T]) = typer
}

[scabug]

@adriaanm said (edited on Jul 26, 2012 10:20:19 AM UTC):
Lukas explains: we shouldn't see fields in the constructor signature, but we should see the parameter accessors.

[scabug]

@gkossakowski said:
Unassigning and rescheduling to M7 as previous deadline was missed.

[scabug]

Eduardo Pareja Tobes (eparejatobes) said:
just curious, why is this no longer considered a bug but an improvement?

[scabug]

@adriaanm said:
Dependent method types were an improvement as well. This is an improvement to support dependent constructor types, which were never implemented. So, technically, they can't be buggy :-)

[scabug]

Eduardo Pareja Tobes (eparejatobes) said:
OK thanks :)

something about the decreased priority and the unassigned state?

[scabug]

@adriaanm said:
Exactly.

[scabug]

Alexander Bulaev (alexbool) said:
Any news on this?

[scabug]

Andrei Pozolotin (andrei.pozolotin) said:
Please implement?

[scabug]

Nathan M (nemccarthy) said:
+1

[scabug]

Owen Healy (ellbur) said:
This would be really useful. Scala's type system is amazing in that it makes dependent types usable by mere humans, but gaps like this partially undo that.

[scabug]

@manojo said:
+1, any timeline on this?

[scabug]

Ruben (lqbweb) said:
+1

[scabug]

@milessabin said (edited on Jun 20, 2016 5:28:08 PM UTC):
@adriaanm left a pretty strong hint above ... why not take a stab at it and claim the glory?

[scabug]

Hamish Dickson (hamish.dickson-at-gmail.com) said:
I'm taking a look a this under scala/scala#5238

[soronpo]

Another simple example which fails to compile in scalac, but compiles fine in dotty:

trait Impl {
  type Out
}
object Impl {
  implicit def ev : Impl = new Impl {
    type Out = Int
  }
}

class Foo[T]
class Bar[T](implicit val impl: Impl) extends Foo[impl.Out] //ERROR=   not found: value impl

Workaround: use Aux pattern

[Sciss]

Cross-building against Scala 2.13 and Dotty would be greatly simplified if this was fixed for Scala 2.13.

[milessabin]

I hadn't realized this was still assigned to me. I agree it would be extremely helpful, but unfortunately I don't have time to work on it.

[som-snytt]

Still DNC. Maybe Miles didn't have time to work on it.

[milessabin]

I didn't.