[MaxFreedomPollard]

If a struct field's type is a renamed struct or enum, the Kotlin backend generates code that does not compile. For example, with a module-level rename:

#[diplomat::bridge]
#[diplomat::attr(kotlin, rename = "Renamed{0}")]
mod ffi {
    pub struct Inner {
        pub x: i32,
    }

    pub enum Color {
        Red,
        Green,
    }

    pub struct Outer {
        pub inner: Inner,
        pub color: Color,
    }
}

the classes come out as RenamedInner, RenamedInnerNative and RenamedColor, but RenamedOuter.kt still uses the Rust names in two places:

internal var inner: InnerNative = RenamedInnerNative();
// ...
val color: RenamedColor = Color.fromNative(nativeStruct.color)

The cause is two helpers in tool/src/kotlin/formatter.rs that read .name directly. #1069 moved the rest of the backend over to fmt_type_name, but these two were left:

• fmt_struct_field_type_native, Type::Struct arm: this gives the field's JNA type. It also names the Option...Native wrapper for DiplomatOption<Inner> fields and, through gen_kt_to_c_for_type, for Option<Inner> method parameters. Those break the same way, as OptionInnerNative.
• fmt_struct_field_native_to_kt, Type::Enum arm: this generates the enum's fromNative conversion.

The neighbouring helpers fmt_field_default and fmt_struct_field_type_kt already call fmt_type_name. That is why the declared type and the initializer above disagree.

The fix calls fmt_type_name in both arms. Types without a rename produce the same output as before. Nothing in feature_tests or example has a struct field or DiplomatOption parameter of a renamed type, so the checked-in bindings do not change.

test_struct_field_renamed_types in the formatter tests builds a struct with a renamed struct field, a renamed enum field, and DiplomatOption fields of each. It checks the strings both helpers return.

This has not been compiled or run locally, because I don't have a Rust toolchain on this machine. I checked the change and the expected strings by reading the formatter, Struct.kt.jinja and Option.kt.jinja, and laid out the new code by hand to match rustfmt. CI will be its first build.

🤖 Generated with Claude Code

[Manishearth]

Hm, did none of our existing feature_tests hit this already? We have a lot of renaming tests; do we need more?

On balance I'd probably prefer a feature_test test (i.e. code in feature_tests that exercises this) instead of an in-code test. We try to keep those to insta tests most of the time.