Create __annotations__ for functions

[konstin]

In python 3, functions and methods can have an __annotations__ attribute, which is automatically initialized with type hints:

>>> def foo(a: str, b: int) -> str:
...    return a * b
>>> help(foo)
Help on function foo in module __main__:
foo(a:str, b:int) -> str
>>> foo.__annotations__
{'a': <class 'str'>, 'b': <class 'int'>, 'return': <class 'str'>}

Those annotations are used for ide type hints and can be used as basis for generating .pyi files. (iirc there is even a tool that can do that automatically)

Implementation

In the proc macros, we get the function argument types, but only as ast, not as resolved typed.
So first, implement a function we can call on all possible rust types for arguments (rust native, python native and custom) that returns the corresponding type object. Then generate code that builds a dict with the annotations using that function. During runtime, add that dict with PyFunction_SetAnnotations. Make sure to take of tyke possible cyclic dependencies (afaik this is done by using the string name instead the type object when there would otherwise be a cycle).

This could possibly even be extended to modules and class, s. datamodel.

[mehcode]

From what I understand PyFunction_SetAnnotations is totally illegal for extension modules. I hope I'm wrong though.

[ijl]

I tried PyFunction_SetAnnotations and indeed it failed. This is comprehensive: https://stackoverflow.com/a/50763175

[konstin]

Thanks for the link @ijl. With that I found the corresponding cpython bug: https://bugs.python.org/issue3208.