Regression: 2.13.x higher kinded match requires type parameter bounds

[soronpo]

When trying to match on a internal class of a higher kinded class, an error is generated for missing type arguments (works fine in 2.12.x and in dotty).
type arguments [_] do not conform to class Foo's type parameter bounds [B[I <: Int] <: Test.Bar[I]]

object Test {
  class Bar[I <: Int]
  class Foo[B[I <: Int] <: Bar[I]] {
    trait SomeThing
  }

  val x = new Bar[Int]

  x match {
    case y : Foo[_]#SomeThing => //error here
    case _ => 
  }  
}

I'll try to bisect this in the coming days and report back.

[joroKr21]

We talked about this on Gitter, the relevant PR is scala/scala#7260
Nothing changed behaviorally, but phases were reordered and now refchecks happens before patmat. Foo[_]#SomeThing is translated to a TypeTreeWithDeferredRefCheck to check the bounds during refchecks. But since patmat used to run before refchecks I guess those bounds were never checked before.

[joroKr21]

We might say that the bug is why the bounds don't match, but this bug has probably been always there, it just surfaced now because of the phase reordering.

[joroKr21]

We might say that the bug is why the bounds don't match, but this bug has probably been always there, it just surfaced now because of the phase reordering.

This is just #8039

[SethTisue]

@joroKr21 so we should close this as a duplicate, then?

[joroKr21]

I don't know - this is a regression after all. But I think a principled solution would also have to fix #8039

[som-snytt]

Ignoring a type constructor requires rocket science

The rocket scientists say it's perfectly natural to just blow up and crash.