I'm seeing a strange behavior in klib? deserializa...

# compiler

[Zac Sweers]

I'm seeing a strange behavior in klib? deserialization in 2.3.20 dev builds with native targets where annotations are seemingly deserialized incorrectly. Working on trying to narrow this down, but here's a dump of what I've found so far: https://github.com/ZacSweers/metro/issues/1556#issuecomment-3693475872

In short, I'm seeing annotations on some FIR-generated function parameters replaced with class annotations from an unrelated FIR-generated class 🤔

Any pointers on where to look for this would be helpful as I'm not super familiar with what goes on in the klib serialization layer

[Zac Sweers]

Does seem to be specific to klib serialization, I see it in the final klib .knm file headers too

[Zac Sweers]

Tracing this through IrFileSerializer in the debugger, I don't see any issue. It appears to serialize the annotation correctly there

[Zac Sweers]

Further debugging - if I trace through FirMemberDeserializer, I do see it parse the right annotation (!)

[Zac Sweers]

but somewhere between then and IR transformation, that annotation is then changed to the other one in the FIR declaration instance

[Zac Sweers]

Filed this: KT-83427