Fix regression in #81

[jschueller]

It seems #81 introduced some kind of bug:

I have the following rst file documenting MyClass object from ottemplate module:

User manual
===========

Reference
---------

.. currentmodule:: ottemplate

.. autosummary::
    :toctree: _generated/
    :template: class.rst_t

    MyClass

with class.rst_t:

{{ objname }}
{{ underline }}

.. currentmodule:: {{ module }}

.. autoclass:: {{ objname }}

   {% block methods %}
   .. automethod:: __init__
   {% endblock %}

results in the following error:

following exception was raised:
Traceback (most recent call last):
  File "/home/schueller/.local/lib/python3.5/site-packages/sphinx/ext/autodoc.py", line 551, in import_object
    __import__(self.modname)
ImportError: No module named 'ottemplate.ottemplate.MyClass'; 'ottemplate.ottemplate' is not a package

I use sphinx latest stable 1.5.5 with python 3.5.2.

What do you think @sirno ?

[sirno]

Hey,

could it be that the double use of .. currentmodule:: causes the issue? your template
already sets the module, but then again this should also fail when someone wants to
autodoc multiple classes in the same .rst file, which would be an issue.

idk, back then i just knew i needed to include that line and searched for some way to
put it there. if you want i can take a closer look tomorrow.

[jschueller]

Hi, thanks for your quick reply

I tried to get rid of my currentmodule but it doesnt work either.

It seems it's more complicated than that ; MyClass lies in ottemplate/ottemplate.py but in ottemplate/__init__.py I do from .ottemplate import * so from ottemplate import MyClass is valid.
So, the code location (ottemplate.ottemplate.MyClass) is different from the import location (ottemplate.MyClass). Then your code actually hardcodes the code location whereas it should use the import location.

I think we have no easy way of doing that ; I propose to revert to avoid regressions in lots of python codebases, what do you think ?

[jschueller]

ping @sirno

[sirno]

hey, sorry i was quite busy and am without computer the next couple of days.

however, it shouldnt be a problem that you packaged like this since ottemplate.ottemplate.MyClass should still be directly importable from the absolute path.

either way. i think reverting is a good plan. i can easily fix my own stuff by manually adding the line
and it will probably will cause more headaches for many people once there is a release.

so if possible somebody should merge this pr before the next release.

[jschueller]

hello maintainers (@rgommers @pv, @stefanv ?), could you review this one ?

[jschueller]

ping

[stefanv]

This was merged by @rgommers. Also see gh-86.

[jschueller]

@stefanv #86 was meant to fix a single test introduced after #81, this basically reverts #81 as it introduced a severe regression

[stefanv]

@jschueller Understood; just tying together all relevant PRs.

[rgommers]

Looks like this makes sense to revert, thanks @jschueller, all