Documenting IPython

When contributing code to IPython, you should strive for clarity and consistency, without falling prey to a style straitjacket. Basically, ‘document everything, try to be consistent, do what makes sense.’

By and large we follow existing Python practices in major projects like Python itself or NumPy, this document provides some additional detail for IPython.

Standalone documentation

All standalone documentation should be written in plain text (.txt) files using reStructuredText [reStructuredText] for markup and formatting. All such documentation should be placed in the directory docs/source of the IPython source tree. Or, when appropriate, a suitably named subdirectory should be used. The documentation in this location will serve as the main source for IPython documentation.

The actual HTML and PDF docs are built using the Sphinx [Sphinx] documentation generation tool. Once you have Sphinx installed, you can build the html docs yourself by doing:

$ cd ipython-mybranch/docs
$ make html

Our usage of Sphinx follows that of matplotlib [Matplotlib] closely. We are using a number of Sphinx tools and extensions written by the matplotlib team and will mostly follow their conventions, which are nicely spelled out in their documentation guide [MatplotlibDocGuide]. What follows is thus a abridged version of the matplotlib documentation guide, taken with permission from the matplotlib team.

If you are reading this in a web browser, you can click on the “Show Source” link to see the original reStricturedText for the following examples.

A bit of Python code:

for i in range(10):
    print i,
print "A big number:",2**34

An interactive Python session:

>>> from IPython.utils.path import get_ipython_dir
>>> get_ipython_dir()
'/home/fperez/.config/ipython'

An IPython session:

In [7]: import IPython

In [8]: print "This IPython is version:",IPython.__version__
This IPython is version: 0.9.1

In [9]: 2+4
Out[9]: 6

A bit of shell code:

cd /tmp
echo "My home directory is: $HOME"
ls

Docstring format

Good docstrings are very important. Unfortunately, Python itself only provides a rather loose standard for docstrings [PEP257], and there is no universally accepted convention for all the different parts of a complete docstring. However, the NumPy project has established a very reasonable standard, and has developed some tools to support the smooth inclusion of such docstrings in Sphinx-generated manuals. Rather than inventing yet another pseudo-standard, IPython will be henceforth documented using the NumPy conventions; we carry copies of some of the NumPy support tools to remain self-contained, but share back upstream with NumPy any improvements or fixes we may make to the tools.

The NumPy documentation guidelines [NumPyDocGuide] contain detailed information on this standard, and for a quick overview, the NumPy example docstring [NumPyExampleDocstring] is a useful read.

For user-facing APIs, we try to be fairly strict about following the above standards (even though they mean more verbose and detailed docstrings). Wherever you can reasonably expect people to do introspection with:

In [1]: some_function?

the docstring should follow the NumPy style and be fairly detailed.

For purely internal methods that are only likely to be read by others extending IPython itself we are a bit more relaxed, especially for small/short methods and functions whose intent is reasonably obvious. We still expect docstrings to be written, but they can be simpler. For very short functions with a single-line docstring you can use something like:

def add(a, b):
   """The sum of two numbers.
   """
   code

and for longer multiline strings:

def add(a, b):
   """The sum of two numbers.

   Here is the rest of the docs.
   """
   code

Here are two additional PEPs of interest regarding documentation of code. While both of these were rejected, the ideas therein form much of the basis of docutils (the machinery to process reStructuredText):

• Docstring Processing System Framework
• Docutils Design Specification

Note

In the past IPython used epydoc so currently many docstrings still use epydoc conventions. We will update them as we go, but all new code should be documented using the NumPy standard.

Building and uploading

The built docs are stored in a separate repository. Through some github magic, they’re automatically exposed as a website. It works like this:

• You will need to have sphinx and latex installed. In Ubuntu, install texlive-latex-recommended texlive-latex-extra texlive-fonts-recommended. Install the latest version of sphinx from PyPI (pip install sphinx).
• Ensure that the development version of IPython is the first in your system path. You can either use a virtualenv, or modify your PYTHONPATH.
• Switch into the docs directory, and run make gh-pages. This will build your updated docs as html and pdf, then automatically check out the latest version of the docs repository, copy the built docs into it, and commit your changes.
• Open the built docs in a web browser, and check that they’re as expected.
• (When building the docs for a new tagged release, you will have to add its link to index.rst, then run python build_index.py to update index.html. Commit the change.)
• Upload the docs with git push. This only works if you have write access to the docs repository.
• If you are building a version that is not the current dev branch, nor a tagged release, then you must run gh-pages.py directly with python gh-pages.py <version>, and not with make gh-pages.

[reStructuredText]

reStructuredText. http://docutils.sourceforge.net/rst.html

[Sphinx]

Sphinx. http://sphinx.pocoo.org/

[MatplotlibDocGuide]

http://matplotlib.sourceforge.net/devel/documenting_mpl.html

[PEP257]

PEP 257. http://www.python.org/peps/pep-0257.html

[NumPyDocGuide]

NumPy documentation guide. https://github.com/numpy/numpy/blob/master/doc/HOWTO_DOCUMENT.rst.txt

[NumPyExampleDocstring]

NumPy example docstring. https://github.com/numpy/numpy/blob/master/doc/HOWTO_BUILD_DOCS.rst.txt