pyo3 and inspect.get_annotations

[dean-shaff]

I would like to be able to get type information about my Rust structs at runtime.

#[pyclass]
#[pyo3(text_signature = "(start, end)")]
pub struct Foo {
    #[pyo3(get)]
    pub start: i64,
    #[pyo3(get)]
    pub end: i64,
}

in Python land:

>>> import inspect
>>> from my_module import Foo 
>>> inspect.get_annotations(Foo)
{}

Ideally, I would like to see something like

>>> inspect.get_annotations(Foo)
{"start": <class 'int'>, "end": <class 'int'>}

Looking at the get_annotations source code I see that for classes, we're calling cls.__dict__["__annotations__"]. I've tried setting Span's __dict__ attribute (I know this is a bit of a hack!), but I get a RuntimeError at import time. How might I go about supporting this behaviour?

[konstin]

This looks like the same problem as #510. To summarize, python unfortunately simply doesn't allow you to set __annotations__ from native code.

[dean-shaff]

I implemented a work around for this.

#[pyclass]
#[pyo3(text_signature = "(start, end)")]
pub struct Foo {
    #[pyo3(get)]
    pub start: i64,
    #[pyo3(get)]
    pub end: i64,
}

#[pymethods]
impl Foo {
    #[classattr]
    pub fn __annotations__(py: Python<'_>) -> PyResult<Py<PyAny>> 
    {
        let builtins = PyModule::import(py, "builtins")?;
        let t = builtins.getattr(intern!(py, "int"))?;
        let dict = PyDict::new(py);
        dict.set_item("start", t)?; 
        dict.set_item("end", t)?;
        Ok(dict.into()) 
    }
}

Now in Python land

>>> import inspect
>>> from my_module import Foo 
>>> inspect.get_annotations(Foo)
{"start": <class 'int'>, "end": <class 'int'>}

[konstin]

oh, i never even considered cpython would just accept a class attribute!

for you code, i think you can skip the import by using PyInt directly.

[kngwyu]

Interesting. Automatically generating this class attr would be an OK approach if we can't find any alternative.