How to: Make your Sphinx documentation compile with ReadTheDocs when you’re using Numpy and Scipy
Sphinx is a great tool for documenting Python programs (and lots of other things – I recently saw a lecturer who had done all of his lecture notes using Sphinx!) and I’ve used it for my new project (which will be announced on this blog in the next few days). Now that the project is near release, I wanted to get the documentation onto ReadTheDocs to make it nice and easily accessible (and so that it can be easily built every time I commit to GitHub).
The theory is that you just point ReadTheDocs to your GitHub repository, it finds the Sphinx conf.py file, and all is fine. However, if you use any module outside of the standard library, and you’re using the Sphinx autodoc module, then it will fail to compile the documentation. This is because the Python code that you are documenting needs to be able to be imported for autodoc to work, and if you are trying to import a module that doesn’t exist by default on a Python install then an error will be produced.
The ReadTheDocs FAQ says that you can setup a pip_requirements file to install any modules that are needed for your code, but this won’t work for any modules that include C code. This is understandable – as ReadTheDocs don’t want any random C code executing on their server – but it means that trying to build the docs for any code that uses numpy, scipy or matplotlib (or many other modules) will fail.
The FAQ suggests how to solve this – using a so-called ‘mock’. This is an object that pretends to be one of these modules, so that it can be imported, but doesn’t actually do anything. This doesn’t matter as it is not normally necessary to actually run the code to produce the docs, just to be able to import it. However, the code that is provided by ReadTheDocs doesn’t work for any modules that you import using the * operator – for example, from matplotlib import *. After asking a StackOverflow question, I found how to fix this for the code that ReadTheDocs provide, but a comment suggested a far easier way to do this, simply add code like the following to the top of your conf.py file:
import mock MOCK_MODULES = ['numpy', 'scipy', 'matplotlib', 'matplotlib.pyplot', 'scipy.interpolate'] for mod_name in MOCK_MODULES: sys.modules[mod_name] = mock.Mock()
In the MOCK_MODULES list put the names of all of the modules that you import. It is important to list submodules (such as matplotlib.pyplot) as well as the main modules. After committing your changes and pushing to GitHub, you should find that your docs compile properly.