Support #[pyfunction] inside #[pymodule]

[davidhewitt]

It struck me as strange that #[pyfn] exists at all when it should be possible to use #[pyfunction] inside pymodule.

This PR adds exactly that. Maybe this is a road to deprecating / removing #[pyfn] so as to make the overall API simpler? (Just one #[pyfunction] attribute instead of two.)

TODO:

• Update documentation

[programmerjake]

one reason to have separate #[pyfn] and #[pyfunction] attributes is for when you want a function that's not automatically added to a Python module.

[kngwyu]

#[pymodule]
fn pyfunction_module(_py: Python, module: &PyModule) -> PyResult<()> {
    #[pyfunction]
    fn foobar() -> usize {
        101
    }

    Ok(())
}

Hmm... 🤔
It's confusing for me that pyfunction automatically adds itself to module.
I think we need alternative syntax if we want to do this kind of automatic module declaration.
E.g.,

#[pymodule] 
mod pyfunction_module {
    #[pyfunction]
    fn foobar() -> usize {
        101
    }
}

[davidhewitt]

one reason to have separate #[pyfn] and #[pyfunction] attributes is for when you want a function that's not automatically added to a Python module.

Is there any case where you would write a #[pyfunction] inside a #[pymodule] but not want it automatically added to a module?

This PR doesn't change the fact that #[pyfunction]s outside of a #[pymodule] still need to be manually added.

[davidhewitt]

I think we need alternative syntax if we want to do this kind of automatic module declaration.

I can add mod syntax if that's preferable.

However, my original motivation is that we don't seem to need two differently-named attributes #[pyfn] and #[pyfunction].

I wasn't planning on changing anything besides making it possible to use #[pyfunction] in a functionally-equivalent way to #[pyfn], so that we just need one attribute.

[davidhewitt]

@kngwyu I see what you mean by automatic adding being confusing. I realise now this PR is probably a breaking change, as this is already valid code, but does not add foobar to the module:

#[pymodule]
fn pyfunction_module(_py: Python, module: &PyModule) -> PyResult<()> {
    #[pyfunction]
    fn foobar() -> usize {
        101
    }

    Ok(())
}

I think the solution is instead to add an attribute #[add_to_module(module)], so that instead of:

 #[pyfn(module, "foobar")]
 fn foobar() -> usize {
      101
 }

It would be possible to write:

#[pyfunction]
#[add_to_module(module)]
fn foobar() -> usize {
    101
}

Does this seem better? I've also clarified my rationale for doing this at all on #694.

[kngwyu]

I see what you mean by automatic adding being confusing. I realise now this PR is probably a breaking change, as this is already valid code, but does not add foobar to the module:

I mean that few users expect that the generated code uses unrelated local variable module.
Not about the breaking change.

Does this seem better?

Hmm... 🤔
I'm sorry but I feel diffult to understand why it's better.

[davidhewitt]

Ok 👍

Let's pull back on this idea for now. I still think it would be worth having just #[pyfunction] so that users don't have to learn two different attribute syntaxes for almost exactly the same thing. I'll leave that in #694 .