Inconsistency between reflect Typed and

[nicolasstucki]

Compiler version

3.0.0-RC3

Issue

Both TypeTests will may match the same tpd.Typed tree

• TypedTypeTest.unapply
• UnapplyTypeTest.unapply

This makes the type hierarchy unsound.

UnapplyTypeTest.unapply only matches tpd.Typed for historical reasons. Back when trees were separated from patterns we needed to handle this specially.

Solution

UnapplyTypeTest.unapply only match tpd.Unapply

[Snowflake8]

I think this is what I encountered (quoted from gitter) (or this since I noticed a problem with the Unapply and a mix with #12222 ). I can provide the code if needed

Hello I've been been working in metaprogramming with scala.reflect and have stumbled across what could be a bug. Basically I have this fuction

def treePrint(tree: Tree): Unit = {
            tree match {
                case body : Term => {
                        body match 
                           case typed: Typed =>
                              // Point A
                              // Code for Typed
                        //other Terms
                       }
                // case Unapply(_, _, _) => 
                         // Point C
                case typed : Typed =>
                        // Point B
                        // This should never run
                // other Tree things
            }
        }

When I run the function point B is sometimes reached even though Typed should always go to Point A (since Typed is a subtype of Term). Even stranger, if I uncomment the Unapply case, instead of point B getting reached, it is point C instead.
I can provide a failing input if necessary

[nicolasstucki]

@Snowflake8 does sound like it would be caused by this bug. Could you create a minimal self-contained code example that fails with your use case. This way we can ensure that everything is properly fixed.

[Snowflake8]

@nicolasstucki You were too quick
Here was the code for legacy purpose
https://github.com/Snowflake8/Scala_Typed_bug

[nicolasstucki]

@Snowflake8 no problem, I will add that extra test or reference it in one of the related issues if it hits another bug.