Towards fixing the Py2 bug in #163

[jnothman]

The error in #163 seems to come from a combination of changes between 0.7 and 0.8:

• our reimplementation of autosummary logic in listing attributes seems to mix unicode objects (from the currently processed dosctring) with native str due to pydoc.getdoc explicitly encoding its result (https://github.com/python/cpython/blob/2.7/Lib/pydoc.py#L85).
• the Jinja templating of docstrings might be forcing it to be unicode...?

I'm not sure yet how to fix this completely. SphinxDocString currently provides __str__ and not __unicode__. But the __str__ implementation can currently return a unicode in py2, resulting in automatic casting / encoding.

[jnothman]

Any Py2/3 experts hanging around?

[larsoner]

Given that everyone is moving to Python3, maybe we should close this and move on