No fallback to __rmod__ magic method when __mod__ is not implemented

[lycantropos]

🐛 Bug Reports

🌍 Environment

• Your operating system and version: macOS 11.6 (20G165)
• Your python version: 3.9
• How did you install python (brew install python)? Did you use a virtualenv?: yes
• Your Rust version (rustc --version): rustc 1.53.0 (53cb7b09b 2021-06-17)
• Your PyO3 version: 0.14.4
• Have you tried using latest PyO3 main (replace version = "0.x.y" with git = "https://github.com/PyO3/pyo3")?: yes, the same

💥 Reproducing

This repository has MCVE with README.

[lycantropos]

I've managed to solve this locally by changing

pyo3/src/class/number.rs

Line 617 in 9d846c3

py_binary_num_func!(mod_, PyNumberModProtocol, T::__mod__);

to

py_binary_fallback_num_func!(
    mod_rmod,
    T,
    PyNumberModProtocol::__mod__,
    PyNumberRModProtocol::__rmod__
);
py_binary_num_func!(mod_, PyNumberModProtocol, T::__mod__);
py_binary_reversed_num_func!(rmod, PyNumberRModProtocol, T::__rmod__);

and

pyo3/pyo3-macros-backend/src/defs.rs

Line 531 in d544c07

SlotDef::new(&["__mod__"], "Py_nb_remainder", "mod_"),

to

        SlotDef::new(&["__mod__", "__rmod__"], "Py_nb_remainder", "mod_rmod"),
        SlotDef::new(&["__mod__"], "Py_nb_remainder", "mod_"),
        SlotDef::new(&["__rmod__"], "Py_nb_remainder", "rmod"),

if this is enough and there are no other potential places to fix -- I can open PR.

[davidhewitt]

A PR would be very welcome!

Note that in #1884 we're merging #[pyproto] into #[pymethods], so hopefully this already works on main if my implementation was correct:

#[pyclass]
struct A;

#[pymethods]
impl A {
    #[new]
    fn new() -> A {
        A {}
    }

    fn __mod__(&self, other: &PyAny) -> PyResult<PyObject> {
        println!("A.__mod__ is called");
        Ok(other.py().NotImplemented())
    }

    fn __rmod__(&self, other: &PyAny) -> PyResult<PyObject> {
        println!("A.__rmod__ is called");
        Ok(other.py().NotImplemented())
    }
}