PyType.is_subclass does not work with two dynamic PyTypes

[Eric-Arellano]

Hello! During Pants's migration to PyO3, I'm stuck on how to get &PyType.is_subclass() working.

This is what I would like to do:

#[pyfunction]
fn check_subclass(typ1: &PyType, typ2: &PyType) -> PyResult<bool> {
  typ1.is_subclass(typ2)
}

If I understand correctly, this is not possible because the API is is_subclass<T>(&self), and it doesn't seem possible to say this due to how Generics work in Rust:

typ1.is_subclass::<typ2>()

For now, I'm thinking I can work around this by writing our own unsafe implementation, but I'd love to be able to contribute a more flexible API to PyO3. I'm thinking is_subclass(&self, T: ToPyObject) -> bool.

What do you think? How would we handle deprecations? (Note that rust-cpython calls it is_subtype_of, maybe that's an option)

Also I'm curious what you think about it returning bool instead of PyResult<bool>?

[birkenfeld]

Interesting! The current API is fine for some cases, but for cases such as yours we should also offer one that takes the type as parameter. Same for is_instance on PyAny.

For the moment, the best you can do is to call typ1.__subclasscheck__(typ2) using call.

[Eric-Arellano]

Thank you for that workaround! Much better than unsafe code.

[birkenfeld]

Ok, I've started a PR at #1985 to fix this for the next version.

[birkenfeld]

Also I'm curious what you think about it returning bool instead of PyResult<bool>?

Since the subclass/instance checks can run arbitrary Python code (by overriding __subclasscheck__/__instancecheck__ in the metaclass), we kind of have to keep the possibility for exceptions open.

If you're fine with swallowing exceptions, you can just use unwrap_or(false).