Deriving pyclass/pymethods seems to break when inside of a macro-by-example.

[tritoke]

Bug Description

When placed inside of a macro code that compiles outside of the macro no longer compiles. The bug seems to be specifically around returning an instance of another pyclass from a method on another.

Steps to Reproduce

maturin new repro --bindings pyo3
cd repro
cargo add sha3
cat >src/lib.rs <<'EOF'
∙ use pyo3::prelude::*;

macro_rules! impl_repro {
    ($name_a:ident, $name_b:ident) => {
        #[pyclass]
        struct $name_a {}

        #[pymethods]
        impl $name_a {
            fn finalize(&mut self) -> $name_b {
                $name_b {}
            }
        }

        #[pyclass]
        struct $name_b {}
    };
}

impl_repro!(MyClass1, MyClass2);
EOF
maturin develop --release

Backtrace

error[E0425]: cannot find value `py` in this scope
  --> src/lib.rs:20:23
   |
20 | impl_repro!(MyClass1, MyClass2);
   |                       ^^^^^^^^ not found in this scope

error[E0425]: cannot find value `ret` in this scope
  --> src/lib.rs:20:23
   |
20 | impl_repro!(MyClass1, MyClass2);
   |                       ^^^^^^^^ not found in this scope

### Your operating system and version

Arch Linux (kernel version 6.9.9)

### Your Python version (`python --version`)

Python 3.12.4

### Your Rust version (`rustc --version`)

 rustc 1.80.0 (051478957 2024-07-21)

### Your PyO3 version

0.20.0

### How did you install python? Did you use a virtualenv?

python is installed via pacman
maturin is installed in a venv:
python -m venv .env
source .env/bin/activate
pip install maturin

### Additional Info

The original context for this is effectively attempting to get around the no-generics for pyclasses by templating out an implementation for a set of XOF functions where each hasher has a "Shaker" which absorbs data in and a "Sponge" which produces data, and you must finalize a Shaker to get a Sponge so I want to have finalize reset the state of the Shaker and return a Sponge object.

[Icxolu]

I can't reproduce that on 0.20 (or 0.21), but I do see it on 0.22. (Did you perhaps mistype the version?)

The 0.22 case that I could reproduce is a tricky one 😅. It seems like this was introduced by #4220, which improved some error messages by changing the Spans of our generated code. Together with macro_rules hygiene these now seem to get different syntax contexts attached, which makes them different from the compilers perspective in this specific context. PR fixing will follow shortly.

[tritoke]

Ah yeah i did it was 0.22, I will use 0.21 for now, thanks for finding the bug :D