Bridge method clash with value class

[travisbrown]

minimized code

trait TC[A] {
  def apply(a: A): Unit
}

class Foo[A](val value: A) extends AnyVal
object Foo {
  val tc = new TC[Foo[String]] {
    def apply(a: Foo[String]): Unit = ()
  }
}

Compilation output

-- Error: TC.scala:8:8 ---------------------------------------------------------
8 |    def apply(a: Foo[String]): Unit = ()
  |        ^
  |bridge generated for member method apply(a: Foo[String]): Unit in anonymous class Object with TC {...}
  |which overrides method apply(a: A): Unit in trait TC
  |clashes with definition of the member itself; both have erased type (a: Object): Unit."
1 error found

expectation

Another minimization from the Cats tests. Scala 2 is fine with this.

[odersky]

There's a long history about this bug which is really a shortcoming of the value class encoding. Given that value classes are not pursued further, and the bug would likely cost a person several weeks to fix, I am not sure we will ever fix it. I do hope there is a workaround.

[travisbrown]

@odersky We can easily work around this, thanks.

[odersky]

@travisbrown Good to know, thanks!

[bishabosha]

minimise to this

case class Box(a: Any) extends AnyVal

or perhaps this to avoid case:

class Box(a: Any) extends AnyVal
object Box extends (Any => Box)
  def apply(a: Any): Box = new Box(a)

[Jasper-M]

@bishabosha The code that Travis posted compiles fine in Scala 2, while yours will error in Scala 2 as well.

I think the bug (or just difference from scala 2, don't know...) here is that dotc thinks that Foo[String] or Foo[Int] erase to Object instead of String or Int respectively.

[smarter]

I think the bug (or just difference from scala 2, don't know...) here is that dotc thinks that Foo[String] or Foo[Int] erase to Object instead of String or Int respectively.

Yes, exactly. I'm working on a fix.

[larsrh]

This is marked as fixed in M3, but there's another situation where this happens: https://scastie.scala-lang.org/g5dE8K7wT5aCPYTYFETtLA

trait Monad[F[_]] {
  def flatMap[A, B](fa: F[A])(f: A => F[B]): F[B]
}

final case class Dual[A](getDual: A) extends AnyVal

object Dual {
  implicit val dualInstances: Monad[Dual] = new Monad[Dual] {
    def flatMap[A,B](fa: Dual[A])(f: A => Dual[B]): Dual[B] = f(fa.getDual)
  }
}

Compiler:

bridge generated for member method flatMap[A, B](fa: Dual[A])(f: A => Dual[B]): Dual[B] in anonymous class Object with Monad {...}
which overrides method flatMap[A, B](fa: F[A])(f: A => F[B]): F[B] in trait Monad
clashes with definition of the member itself; both have erased type (fa: Object, f: Function1): Object.

[smarter]

Hmm, interesting, if I replace the anonymous instantiation by:

object Foo extends Monad[Dual] {
  def flatMap[A,B](fa: Dual[A])(f: A => Dual[B]): Dual[B] = f(fa.getDual)
}

Then Scala 2 emits the same error as Dotty. It looks like Scala 2 has some extra magic to deal with the special-case of value class bridge clashes in anonymous classes where it can get away with renaming one of the method to avoid the clash: scala/scala#3428, feel free to open an issue about this, but I don't know if we have the bandwidth to port this in time for 3.0.

[larsrh]

I have to admit I don't understand at all what's going on, and what a possible workaround could be.

Having said that, I found this problem while minimizing another problem found by @fmsbeekmans while porting semigroups to Dotty. A similar instantiation produced the following message:

[error]    |bad parameter reference io.chrisdavenport.semigroups.Dual#A at elimOpaque
[error]    |the parameter is type A in class Dual but the prefix io.chrisdavenport.semigroups.Dual
[error]    |does not define any corresponding arguments.

... and I haven't yet minimized it in a satisfactory way, nor can I tell whether or not this is an internal error. The error goes away when removing extends AnyVal.

[smarter]

That bad parameter reference is definitely a bug, but I don't think it's related to the bridge issue. I've gone ahead and opened #11264 to keep track of the bridge issue.