Unreachable case warning (E121) in inline generic function

[alexishorner]

Compiler version

3.4.0

Minimized code

sealed abstract class Parent // The warning goes away if `sealed` is removed

class A extends Parent
class B extends Parent

inline def matchAs[T <: Parent](p: Parent): Unit = p match // The type bound is necessary
    case p: T =>
    case _ => println("unreachable case reached") // Warning here: "Unreachable case except for null [...]"
end matchAs

@main def main: Unit =
    matchAs[A](new B)
end main

Output

Compilation Output

[warn] -- [E121] Pattern Match Warning: [...]/inline_poc/src/main/scala/poc.scala:9:9 
[warn] 9 |    case _ => println("unreachable case reached")
[warn]   |         ^
[warn]   |Unreachable case except for null (if this is intentional, consider writing case null => instead).
[warn] one warning found

Run Output

unreachable case reached

Expectation

No warning should be issued, since the case is reachable.

Notes

The following conditions are required to trigger the warning

• Parent must be sealed
• T must be upper bounded

The warning stays even when p is declared inline.

[alexishorner]

The warning is still issued with the latest version 3.5.0-RC1-bin-SNAPSHOT built from e9fd9e1.

[mbovel]

Thanks for reporting @alexishorner 🙂

@dwijnand do you think that could be appropriate for a future spree? And do you think this is a duplicate of #20395?

[dwijnand]

Yes, it looks the same, but not, doesn't seem spree-sized.

[dwijnand]

The problem occurs because when to verify if the case is unreachable we check if the space of values it represents is a subspace of the previous cases (of which there's only one here):

  ==> isSubspace(Typ(_: NodeId), Typ(_: T))?

In doing so we decompose NodeId into Hi and Hello. But then we also decompose T into Hi and Hello, which is a widening, and it all cancels out and thus believes it to be a subspace thus unreachable. I'm assuming that RHS decomposition is important for other cases, so we'd have to see how to fix this.