Scala 2.13.0-RC1 - method without override modifier rejected when overridden both earlier and later in trait linearisation

[Sciss]

I think this is a bug:

trait Basic {
  def numFrames: Long
}

trait Readable extends Basic {
  def numFrames: Long = ??? // doesn't matter for bug report
}
    
trait Writable extends Basic {
  override final def numFrames: Long = ??? // doesn't matter for bug report
}

trait Bidi extends Readable with Writable  // writable beats readable
  
trait WritableFileLike extends Writable

trait BidiFileLike extends WritableFileLike with Bidi // compiles up to 2.13.0-M5, not RC1

Note that in the linearisation order, Writable is last and thus should "eliminate" its first occurrence in WritableFileLike, which I guess triggers the error in 2.13.0-RC1:

[error] final override def numFrames: Long (defined in trait Writable)
[error]   with def numFrames: Long (defined in trait Readable)
[error] trait BidiFileLike extends WritableFileLike with Bidi
[error]       ^

[smarter]

Note that in the linearisation order, Writable is last and thus should "eliminate" its first occurrence in WritableFileLike

No, for linearization order the first occurence is the important one, see https://www.scala-lang.org/files/archive/spec/2.12/05-classes-and-objects.html#class-linearization. I think 2.13.0-RC1 is doing the right thing.

[Sciss]

I can actually remove the final qualifier, the problem stays. If I add override also to the Readable trait, it compiles. So not sure, if it was "wrong" interpretation of the spec before 2.13.0-RC1 or not (and thus, whether it's a bug or not).

[hrhino]

Isn't this the opposite of #11136? (Not that I'm biased...)

[som-snytt]

The PR to the earlier ticket is explanatory (though it also has a long tail of follow-up).

What do you call a duplicate ticket that is the opposite? A mirror ticket?

I'm not sure which ticket I admire more for their creative variants of IO-related trait names.

[Sciss]

I don't understand - is this resolved?

[som-snytt]

@Sciss See smarter's comment -- the current behavior is correct, the old behavior was wrong, and was always wrong. Now I have to check my "stackable traits" answer on stackoverflow from 2012.