overloads without implementations (in stub files) are not detected

[tlambert03]

Describe the bug
Methods and functions that are decorated with @overload but do not have an implementation (as is standard in a stub file) are not detected:

To Reproduce

# mystub.pyi
from typing import overload

class M:
    @overload
    def f(self, x: int) -> int: ...
    @overload
    def f(self, x: str) -> str: ...

class M2:
    @overload
    def f(self, x: int) -> int: ...
    @overload
    def f(self, x: str) -> str: ...
    def f(self, x: str | int) -> str | int: ...  # one wouldn't normally do this in a stub

In [3]: griffe.load('mystub._test.M').members
Out[3]: {}

In [4]: griffe.load('mystub._test.M2').members
Out[4]: {'f': <Function('f', 14, 14)>}

Expected behavior
Expected f method to be detected.

System (please complete the following information):

• griffe version: 0.24.1
• Python version: 3.10
• OS: mac

Will probably be relevant for mkdocstrings/mkdocstrings#308 as well

[pawamoy]

Thanks for the report!
It may be a bug in the code that merges stubs data into actual module data, but I can't see it 🤔
Do you have a public repo I could use to debug?

[tlambert03]

the repo i found it on was that same (compiled) pymmcore repo. so

pip install pymmcore

def test_overloads():
    import griffe
    core = griffe.load('pymmcore.CMMCore')
    
    # defineConfig is overloaded
    assert 'defineConfig' in core.members

    # this is a non-overloaded function that works already
    assert 'addSearchPath' in core.members

and the corresponding stub is at https://github.com/micro-manager/pymmcore/blob/main/pymmcore/__init__.pyi

[pawamoy]

Ah, found the issue. It's because we merge stubs before wildcard imports were expanded.

[pawamoy]

Another issue is that ./pymmcore is found first, rather than the installed version which is the only one containing the compiled module _pymmcore_swig.cpython-311-x86_64-linux-gnu. So the compiled module is not found, not inspected, and we can't expand wildcards, and merge stubs.

An argument in favor of the src-layout I guess. I'll try with it, to see if it's better.

[pawamoy]

OK, we need #115 first.

[pawamoy]

Fixed in 0ed9c36