Building docs fails with the latest version of sphinx

[datapythonista]

It seems like one week ago sphinx 1.7.0 was released, and the version of numpydoc included with pandas is not compatible with it (it uses the sphinx.util.compat which has been deprecated).

Anaconda already includes sphinx 1.7.0, which is used by default. As in requirements_dev.txt sphinx is specified without version, after following the instructions in the contributing documentation, building the documentation fails.

I assume that numpydoc has been modified, if included in the repository instead of being added as a dependency. So, may be a quick solution is to specify sphinx==1.6.7 in requirements_dev.txt.

Happy to contribute a pull request, if you can clarify what needs to be done.

[TomAugspurger]

See #19687 and #18147

We're currently blocked by an IPython release to fix an issue with newer versions of the IPython directive. I thought we already pinned sphinx in our requirements_dev, but apparently not. We want to pin to sphinx 1.5.*, since 1.6 has a regression slowing down the API docs.

Alternatively, we could backport @jorisvandenbossche's fix to IPython and vendor that? I think pinning is best for now.

[jcontesti]

When I install the build dependencies with pip, I get the following error:

python -m pip install -r ci/requirements_dev.txt
Invalid requirement: 'sphinx=1.5*'
= is not a valid operator. Did you mean == ?

Simply changing sphinx=1.5* with sphinx==1.5.* solves the issue.

Let me know if a can contribute with a PR.

[TomAugspurger]

Sorry about that @jcontesti. I think #19687 is already addressing this.