using numpydoc resulting in autosummary import error.

[sirno]

When not explicitly declaring a method section in the class docstring numpydoc will create a autosummary table containing all public methods as an entry.
Also when only the methods names are contained in the section, i.e. method instead of method() numpydoc will also create an autosummary:

.. autosummary::

    method1
    method2
    <...>

This behaviour would be amazing, but it doesn't work unfortunately because autosummary will be unable to import the methods and thus create a table not containing any descriptions.
This may be an issue with autosummary, either on sphinx side or because the table is not generated properly. It may also be that the way I documented things is unintended. Anyways, here is a small example producing exactly the issue I am having. I'd be glad about any advice, thanks.

Minimal example:
example.zip

[rgommers]

It may also be that the way I documented things is unintended.

I don't think it's unintended, but what you're doing does look different from what numpy, scipy & co are doing (which is to have a function listing per submodule, so then you'd be in the right currentmodule already probably.

Browsing your example, you do seem to need the Child subclass to demonstrate the problem. Your problem description above doesn't mention this only happening for subclasses. I have the feeling your problem summary is a bit too general.

[sirno]

Oh yes, indeed it is a bit general, sorry.

The problem only occurs when using .. autoclass:: and not when using .. automodule::. I didn't notice.

Also, I don't need the childclass to demonstrate the problem, it was just the way I produced the problem in the first place. When you remove the summary string (and the brackets) of aMethod in the methods section of Base, import will fail:

/example/docs/src/lib.classes/Base.rst:15: WARNING: failed to import aMethod

It does not fail before, because aMethod() (with brackets) can not be imported either way, so numpydoc generates a table directly from the docstring.

[rgommers]

gh-81 merged, so closing. Thanks @sirno