Possible regression in the old and the new pattern matcher (SI-3968) and possibly fixed bugs (SI-3481, SI-3556, SI-2080)

Hi,

I found the following issue:

array matching showing covariance
https://issues.scala-lang.org/browse/SI-3968

object Test {
  def f1(x: Any) = x match { 
    case _: Array[AnyRef] => 1
    case _: Array[String] => 2
    case _                => 3
  }
  def f2(x: Any) =
    if (x.isInstanceOf[Array[AnyRef]]) 1
    else if (x.isInstanceOf[Array[String]]) 2
    else 3
  
  def main(args: Array[String]): Unit = {
    println(f1(args))
    println(f2(args))
  }
}

The following prints "1 1" but should print "2 2".

Well, now it prints “3 3”.

Is this a regression?


It seems like the following bugs might have been fixed, at least I can verify that the compiler rejects the unsound code of these tickets:

Unsound type projection
https://issues.scala-lang.org/browse/SI-3481

method's return type can't be inferred, Nothing is used (unsoundly)
https://issues.scala-lang.org/browse/SI-3556

It should not be possible to override concrete member types
https://issues.scala-lang.org/browse/SI-2080

Can someone verify that?

Thanks and bye,

Simon