Incorrect inexhaustivity warning when leafs are defined in an unrelated sealed trait

[kitlangton]

Compiler version

3.1.1

Minimized code

Here is a Scastie.

The exhaustivity check fails when instances of Schema are defined in a trait.

sealed trait Schema[A]

object Schema extends RecordInstances

sealed trait RecordInstances:
  case class Field[A]() extends Schema[A]
  case object Thing extends Schema[Int]

import Schema._

// Match not exhaustive error! (with fatal warnings :P)
def handle[A](schema: Schema[A]) =
  schema match
    case Field() => println("field")
    case Thing   => println("thing")

Expectation

While it is a bit odd to define these instances in a trait, I would still expect this to correctly check exhaustivity. Scala 2 works as expected here.

[kitlangton]

On second thought. Maybe this is more sound. I don't know why the code I'm working on was originally organized like this—I think I can merely remove these intermediate traits. Feel free to close this if you agree! But I'll leave it open in case this is interesting. It causes another similar compiler error here (https://scastie.scala-lang.org/2ME3I1HIQuyUAe67T2jziw)

[som-snytt]

For example,

@main def test() = println {
  val r = new RecordInstances {}
  val s = r.Thing
  handle(s)
}

[smarter]

There is an exhaustiveness issue here since RecordInstances isn't sealed (but the error message wording is a bit confusing: it says RecordInstances.Field() when it should say something like _: RecordInstances#Field). However, if RecordInstances is sealed the error doesn't go away when it should (but it's a pretty weird thing to do indeed).

[kitlangton]

@smarter @som-snytt Apologies. It actually still fails: https://scastie.scala-lang.org/WwDjucNPS4ekVwn2977OKA
Forgot to seal both in my example (updating my issue link now)

Oh. You already realized that 😆 Misunderstood your comment whoops :)

[som-snytt]

Also Scala 2 warns. It would fail on the following inputs: Field(), Thing, _

[smarter]

It causes another similar compiler error here (scastie.scala-lang.org/2ME3I1HIQuyUAe67T2jziw)

That looks like an unrelated bug (missing substitution of RecordInstances.this into Schema), could you open a separate issue for it?

[kitlangton]

Also Scala 2 warns. It would fail on the following inputs: Field(), Thing, _

Oh yeah... Don't know why that didn't trigger fatal warnings for me in Scala 2. The solution for me was to replace these traits with // RECORD INSTANCES comment headers instead 😛. I think someone just wanted the organizational niceties of traits to organize a very large sealed trait file.