Static-only objects are included in wildcards expansions

[pawamoy]

Describe the bug
Currently, expanding a wildcard will include members that are only "statically" present.
More specifically: members that are defined or imported under an if TYPE_CHECKING or if False condition.
This is problematic because this can lead to cyclic aliases.

To Reproduce
Steps to reproduce the behavior:

# package/module_a.py
from typing import TYPE_CHECKING

if TYPE_CHECKING:  # always false
    from package.module_b import CONST

# package/__init__.py
from .module_a import *

At runtime, package would definitely not have CONST in its members.

Expected behavior
These "static-only" objects should not be included in wildcards expansions.

Additional context
mkdocstrings/mkdocstrings#382
Maybe an other way to fix that would be to better handle duplicate members.
But that depends on the order we add expanded wildcard's members...

[tristanlatr]

This is how I check for the TYPE checking blocks: https://github.com/tristanlatr/pydocspec/blob/main/pydocspec/astroidutils.py#L359-L374

Objects have a field is_type_guarded wich is set to True is the context is under a type guard if condition.

Then the wildcard import rely on __all__ or the public names returnde with this function: https://github.com/tristanlatr/pydocspec/blob/main/pydocspec/processor/mod_attr.py#L75-L87

[pawamoy]

Thanks @tristanlatr! I was indeed thinking about adding a field on objects so that we know they are type-guarded or not.

[pawamoy]

Released in 0.17.0!