sys.argv support: Consider call PySys_SetArgv on python initialization.

[XUWeijiang]

Currently if I use sys.argv in python code, it would complain: 'module' object has no attribute 'argv'
This is because python's argv is not initialized.

I hope PyO3 can provide an option to initialize argv in its initialization.

Repro Example

use pyo3::prelude::*;


fn main_(py: Python) -> PyResult<()> {
    let sys = py.import("sys")?;
    let argv: Vec<String> = sys.get("argv")?.extract()?;
    println!("{:?}", argv);
    Ok(())
}

fn main() -> Result<(), ()> {
    Python::with_gil(|py| {
        main_(py).map_err(|e| {
            e.print_and_set_sys_last_vars(py);
        })
    })
}

Output:

AttributeError: module 'sys' has no attribute 'argv'
Error: ()

[davidhewitt]

I wonder if we should be using the python initialization configuration from Python 3.8 onwards? I'll have a play with this this morning.

[davidhewitt]

@XUWeijiang I took a look this morning and I can't reproduce this issue. On my system the snippet you provide above works fine, and sys.argv exists.

Can you please provide detail on your OS and Python version?

[XUWeijiang]

@davidhewitt I am using ubuntu 18.04 on wsl, and python3.6.

[davidhewitt]

So I can confirm that on Python 3.6 this issue reproduces on my system.

On Python 3.7 and up sys.argv exists but is empty, so the original reported issue no longer applies.

However, I'm unsure whether pyo3 should be setting sys.argv unconditionally. Perhaps we can add an API to do this.

[davidhewitt]

Note for the future - the folks in pybind11 look like they might merge something similar to this: pybind/pybind11#2341

The design of their implementation would take arguments passed from c++ main, which isn't quite how Rust does it. There's also a question of whether we would ever want something like this for the auto-initialize feature.

Additionally, it's clearly less of a problem once we move away from supporting Python 3.6, because after that point accessing sys.argv stops being an AttributeError.