GADT with subtype cause non exhaustive match

[peter-jerry-ye]

Compiler version

3.2.2

Minimized code

scastie

enum T[U] {
  case G[A <: Int](i: A) extends T[A]

  // match may not be exhaustive.
  //
  // It would fail on pattern case: T.G(_)
  def f: Int = this match
    case G(i) => ???
}

Output

Warning on match being not exhaustive

Expectation

It should compile

[dwijnand]

This is due to how type inference instantiates things and GADT constraints. The information is in the GADT constraints, but GADT constraints are thrown away after Typer/PostTyper. So the proper solution is to keep GADT constraints (using AssumeInfo trees). In general we've tried to do without GADT constraints in other ways, like casting. So I tried to fix this by tweaking how types are instantiated, to keep the bound, in #18297. That fixes the case, but other tests break - and I've had to fight against and for type inference changes a lot, so I decided to drop that approach. This will/would be fixed if we introduce AssumeInfo trees.