autodoc add_module_names equivalent for arguments

[RuRo]

The add_module_names = False configuration seems to only affect the class/function/attribute header names.
The type hints are still always rendered as fully qualified names.

mypackage/mymodule.py:

class MyClass:
    """Whatever 1."""
    pass


def foo(arg: MyClass):
    """Whatever 2."""
    pass

conf.py:

# ...
add_module_names = False
# ...

index.rst:

mypackage.mymodule module
=========================

.. automodule:: mypackage.mymodule
   :members:
   :undoc-members:
   :show-inheritance:

Expected documentation:

foo(arg: MyClass)
    Whatever 2.

Actual documentation:

foo(arg: mypackage.mymodule.MyClass)
    Whatever 2.

Describe the solution you'd like

I would be OK with any of the following:

add_module_names = False # now affects type annotations too
# or
add_type_module_names = False # new sphinx config option (name up for debate)
# or
autodoc_add_module_names = False # new autodoc config option (name up for debate)

Describe alternatives you've considered

There's a StackOverflow post which suggests using the autodoc_docstring_signature option to manually specify the function signature. This is not really a viable solution in my opinion.

[tk0miya]

+1 for adding a new confval only for autodoc. It would be nice if we can give better name to it. I feel "add_module_names" is a bit ambiguous and difficult to understand its behavior from the name.

[RuRo]

To be clear, the add_module_names confval already exists. It's just that it doesn't affect parameter types for some reason.

[tk0miya]

Sorry for confuse. I know that. I thought it's better to separate the confval for autodoc and python-domain.

[tk0miya]

Just FYI: Sphinx-4.0 provides a new configuration python_use_unqualified_type_names. It suppresses the module name if hyperlinks can be resolved.

[felix-hilden]

So does this mean that the proposed add_module_names configuration is completely unnecessary?

[RuRo]

My understanding is that python_use_unqualified_type_names only works with resolved refs and only for the python domain. Although, I would personally consider this issue fixed if python_use_unqualified_type_names would also work for unresolved refs.

[felix-hilden]

Oh that's a fair point! Yep.

[RuRo]

Also, not sure if this deserves a separate issue/feature request, but one more place where sphinx currently produces fully qualified names is in the package/module headings generated with apidoc. So for example, if you have

foo/__init__.py
foo/bar/__init__.py
foo/bar/baz.py

you will get the documentation for

foo package
 - foo.bar package
    - foo.bar.baz module

instead of

foo package
 - bar package
    - baz module

The FQN version is kind of redundant, since the parent packages of bar and baz are already obvious from the ToC. And with longer package/module names, this can lead to really ugly wrapping in the sidebar ToC.