Thanks for getting to the bottom of this. I think it would be a fair bit of work to make WCSBase picklable, and I'd rather not bother unless absolutely necessary. Maybe we should wait to see what the response is to the Sphinx bug first -- I'm not sure why one would need to pickle an entire class just to retain its docstrings...

On further investigation, it looks like it only needs to pickle the "type" WCSBase, not a WCSBase instance, which should be an easier problem, since it has no state. In fact, Python should be able to handle that automatically, I would have thought. It turns out that by getting Sphinx to use pickle rather than cPickle, you get a better error message out:

  File "/usr/lib64/python2.7/pickle.py", line 748, in save_global
    (obj, module, name))
pickle.PicklingError: Can't pickle <type 'astropy.wcs.WCSBase'>: it's not found as astropy.wcs.WCSBase

And from that, it's clear what's going on. astropy.wcs.WCSBase wasn't actually accessible by that path. The attached PR fixes that. I'm comfortable with this as a workaround for Sphinx 1.2b1, since I think it actually makes the documentation match the code better.

Getting through this, I did uncover another issue hiding beneath it, but I'll file a new issue for that.

Ahhh, that explains why I was having so much trouble finding WCSBase in the first place. Alright, looks good to me, then.

It's still a bit confusing why this behavior changed from 1.1.3 to 1.2, but it's not really a big deal.