matplotlib

Table Of Contents

Previous topic

The Matplotlib Developers’ Guide

Next topic

Working with matplotlib source code

This Page

Coding guide

Committing changes

When committing changes to matplotlib, there are a few things to bear in mind.

  • if your changes are non-trivial, please make an entry in the CHANGELOG
  • if you change the API, please document it in doc/api/api_changes.rst, and consider posting to matplotlib-devel
  • Are your changes python2.6 compatible? We support python2.6 and later
  • Can you pass examples/tests/backend_driver.py? This is our poor man’s unit test.
  • Can you add a test to lib/matplotlib/tests to test your changes?
  • If you have altered extension code, do you pass unit/memleak_hawaii3.py?
  • if you have added new files or directories, or reorganized existing ones, are the new files included in the match patterns in MANIFEST.in. This file determines what goes into the source distribution of the mpl build.
  • Keep the maintenance branches and master in sync where it makes sense.

Style guide

Importing and name spaces

For numpy, use:

import numpy as np
a = np.array([1,2,3])

For masked arrays, use:

import numpy.ma as ma

For matplotlib main module, use:

import matplotlib as mpl
mpl.rcParams['xtick.major.pad'] = 6

For matplotlib modules (or any other modules), use:

import matplotlib.cbook as cbook

if cbook.iterable(z):
    pass

We prefer this over the equivalent from matplotlib import cbook because the latter is ambiguous as to whether cbook is a module or a function. The former makes it explicit that you are importing a module or package. There are some modules with names that match commonly used local variable names, eg matplotlib.lines or matplotlib.colors. To avoid the clash, use the prefix ‘m’ with the import some.thing as mthing syntax, eg:

import matplotlib.lines as mlines
import matplotlib.transforms as transforms   # OK
import matplotlib.transforms as mtransforms  # OK, if you want to disambiguate
import matplotlib.transforms as mtrans       # OK, if you want to abbreviate

Naming, spacing, and formatting conventions

In general, we want to hew as closely as possible to the standard coding guidelines for python written by Guido in PEP 0008, though we do not do this throughout.

  • functions and class methods: lower or lower_underscore_separated
  • attributes and variables: lower or lowerUpper
  • classes: Upper or MixedCase

Prefer the shortest names that are still readable.

Configure your editor to use spaces, not hard tabs. The standard indentation unit is always four spaces; if there is a file with tabs or a different number of spaces it is a bug – please fix it. To detect and fix these and other whitespace errors (see below), use reindent.py as a command-line script. Unless you are sure your editor always does the right thing, please use reindent.py before committing your changes in git.

Keep docstrings uniformly indented as in the example below, with nothing to the left of the triple quotes. The matplotlib.cbook.dedent() function is needed to remove excess indentation only if something will be interpolated into the docstring, again as in the example below.

Limit line length to 80 characters. If a logical line needs to be longer, use parentheses to break it; do not use an escaped newline. It may be preferable to use a temporary variable to replace a single long line with two shorter and more readable lines.

Please do not commit lines with trailing white space, as it causes noise in git diffs. Tell your editor to strip whitespace from line ends when saving a file. If you are an emacs user, the following in your .emacs will cause emacs to strip trailing white space upon saving for python, C and C++:

; and similarly for c++-mode-hook and c-mode-hook
(add-hook 'python-mode-hook
          (lambda ()
          (add-hook 'write-file-functions 'delete-trailing-whitespace)))

for older versions of emacs (emacs<22) you need to do:

(add-hook 'python-mode-hook
          (lambda ()
          (add-hook 'local-write-file-hooks 'delete-trailing-whitespace)))

Keyword argument processing

Matplotlib makes extensive use of **kwargs for pass-through customizations from one function to another. A typical example is in matplotlib.pylab.text(). The definition of the pylab text function is a simple pass-through to matplotlib.axes.Axes.text():

# in pylab.py
def text(*args, **kwargs):
    ret =  gca().text(*args, **kwargs)
    draw_if_interactive()
    return ret

text() in simplified form looks like this, i.e., it just passes all args and kwargs on to matplotlib.text.Text.__init__():

# in axes.py
def text(self, x, y, s, fontdict=None, withdash=False, **kwargs):
    t = Text(x=x, y=y, text=s, **kwargs)

and __init__() (again with liberties for illustration) just passes them on to the matplotlib.artist.Artist.update() method:

# in text.py
def __init__(self, x=0, y=0, text='', **kwargs):
    Artist.__init__(self)
    self.update(kwargs)

update does the work looking for methods named like set_property if property is a keyword argument. I.e., no one looks at the keywords, they just get passed through the API to the artist constructor which looks for suitably named methods and calls them with the value.

As a general rule, the use of **kwargs should be reserved for pass-through keyword arguments, as in the example above. If all the keyword args are to be used in the function, and not passed on, use the key/value keyword args in the function definition rather than the **kwargs idiom.

In some cases, you may want to consume some keys in the local function, and let others pass through. You can pop the ones to be used locally and pass on the rest. For example, in plot(), scalex and scaley are local arguments and the rest are passed on as Line2D() keyword arguments:

# in axes.py
def plot(self, *args, **kwargs):
    scalex = kwargs.pop('scalex', True)
    scaley = kwargs.pop('scaley', True)
    if not self._hold: self.cla()
    lines = []
    for line in self._get_lines(*args, **kwargs):
        self.add_line(line)
        lines.append(line)

Note: there is a use case when kwargs are meant to be used locally in the function (not passed on), but you still need the **kwargs idiom. That is when you want to use *args to allow variable numbers of non-keyword args. In this case, python will not allow you to use named keyword args after the *args usage, so you will be forced to use **kwargs. An example is matplotlib.contour.ContourLabeler.clabel():

# in contour.py
def clabel(self, *args, **kwargs):
    fontsize = kwargs.get('fontsize', None)
    inline = kwargs.get('inline', 1)
    self.fmt = kwargs.get('fmt', '%1.3f')
    colors = kwargs.get('colors', None)
    if len(args) == 0:
        levels = self.levels
        indices = range(len(self.levels))
    elif len(args) == 1:
       ...etc...

Documentation and docstrings

Matplotlib uses artist introspection of docstrings to support properties. All properties that you want to support through setp and getp should have a set_property and get_property method in the Artist class. Yes, this is not ideal given python properties or enthought traits, but it is a historical legacy for now. The setter methods use the docstring with the ACCEPTS token to indicate the type of argument the method accepts. Eg. in matplotlib.lines.Line2D:

# in lines.py
def set_linestyle(self, linestyle):
    """
    Set the linestyle of the line

    ACCEPTS: [ '-' | '--' | '-.' | ':' | 'steps' | 'None' | ' ' | '' ]
    """

Since matplotlib uses a lot of pass-through kwargs, eg. in every function that creates a line (plot(), semilogx(), semilogy(), etc...), it can be difficult for the new user to know which kwargs are supported. Matplotlib uses a docstring interpolation scheme to support documentation of every function that takes a **kwargs. The requirements are:

  1. single point of configuration so changes to the properties don’t require multiple docstring edits.
  2. as automated as possible so that as properties change, the docs are updated automagically.

The functions matplotlib.artist.kwdocd and matplotlib.artist.kwdoc() to facilitate this. They combine python string interpolation in the docstring with the matplotlib artist introspection facility that underlies setp and getp. The kwdocd is a single dictionary that maps class name to a docstring of kwargs. Here is an example from matplotlib.lines:

# in lines.py
artist.kwdocd['Line2D'] = artist.kwdoc(Line2D)

Then in any function accepting Line2D pass-through kwargs, eg. matplotlib.axes.Axes.plot():

# in axes.py
def plot(self, *args, **kwargs):
    """
    Some stuff omitted

    The kwargs are Line2D properties:
    %(Line2D)s

    kwargs scalex and scaley, if defined, are passed on
    to autoscale_view to determine whether the x and y axes are
    autoscaled; default True.  See Axes.autoscale_view for more
    information
    """
    pass
plot.__doc__ = cbook.dedent(plot.__doc__) % artist.kwdocd

Note there is a problem for Artist __init__ methods, eg. matplotlib.patches.Patch.__init__(), which supports Patch kwargs, since the artist inspector cannot work until the class is fully defined and we can’t modify the Patch.__init__.__doc__ docstring outside the class definition. There are some some manual hacks in this case, violating the “single entry point” requirement above – see the artist.kwdocd['Patch'] setting in matplotlib.patches.

Developing a new backend

If you are working on a custom backend, the backend setting in matplotlibrc (Customizing matplotlib) supports an external backend via the module directive. if my_backend.py is a matplotlib backend in your PYTHONPATH, you can set use it on one of several ways

  • in matplotlibrc:

    backend : module://my_backend
  • with the use directive is your script:

    import matplotlib
    matplotlib.use('module://my_backend')
    
  • from the command shell with the -d flag:

    > python simple_plot.py -d module://my_backend

Writing examples

We have hundreds of examples in subdirectories of matplotlib/examples, and these are automatically generated when the website is built to show up both in the examples and gallery sections of the website. Many people find these examples from the website, and do not have ready access to the file:examples directory in which they reside. Thus any example data that is required for the example should be added to the sample_data git repository. Then in your example code you can load it into a file handle with:

import matplotlib.cbook as cbook
fh = cbook.get_sample_data('mydata.dat')

The file will be fetched from the git repo using urllib and updated when the revision number changes.

If you prefer just to get the full path to the file instead of a file object:

import matplotlib.cbook as cbook
datafile = cbook.get_sample_data('mydata.dat', asfileobj=False)
print 'datafile', datafile

Writing a new pyplot function

A large portion of the pyplot interface is automatically generated by the boilerplate.py script (in the root of the source tree). To add or remove a plotting method from pyplot, edit the appropriate list in boilerplate.py and then run the script which will update the content in lib/matplotlib/pyplot.py. Both the changes in boilerplate.py and lib/matplotlib/pyplot.py should be checked into the repository.

Testing

Matplotlib has a testing infrastructure based on nose, making it easy to write new tests. The tests are in matplotlib.tests, and customizations to the nose testing infrastructure are in matplotlib.testing. (There is other old testing cruft around, please ignore it while we consolidate our testing to these locations.)

Requirements

The following software is required to run the tests:

Running the tests

Running the tests is simple. Make sure you have nose installed and run the script tests.py in the root directory of the distribution. The script can take any of the usual nosetest arguments, such as

-v increase verbosity
-d detailed error messages
--with-coverage enable collecting coverage information

To run a single test from the command line, you can provide a dot-separated path to the module followed by the function separated by a colon, eg. (this is assuming the test is installed):

python tests.py matplotlib.tests.test_simplification:test_clipping

An alternative implementation that does not look at command line arguments works from within Python:

import matplotlib
matplotlib.test()

Running tests by any means other than matplotlib.test() does not load the nose “knownfailureif” (Known failing tests) plugin, causing known-failing tests to fail for real.

Writing a simple test

Many elements of Matplotlib can be tested using standard tests. For example, here is a test from matplotlib.tests.test_basic:

from nose.tools import assert_equal

def test_simple():
    '''very simple example test'''
    assert_equal(1+1,2)

Nose determines which functions are tests by searching for functions beginning with “test” in their name.

Writing an image comparison test

Writing an image based test is only slightly more difficult than a simple test. The main consideration is that you must specify the “baseline”, or expected, images in the image_comparison() decorator. For example, this test generates a single image and automatically tests it:

import numpy as np
import matplotlib
from matplotlib.testing.decorators import image_comparison
import matplotlib.pyplot as plt

@image_comparison(baseline_images=['spines_axes_positions'])
def test_spines_axes_positions():
    # SF bug 2852168
    fig = plt.figure()
    x = np.linspace(0,2*np.pi,100)
    y = 2*np.sin(x)
    ax = fig.add_subplot(1,1,1)
    ax.set_title('centered spines')
    ax.plot(x,y)
    ax.spines['right'].set_position(('axes',0.1))
    ax.yaxis.set_ticks_position('right')
    ax.spines['top'].set_position(('axes',0.25))
    ax.xaxis.set_ticks_position('top')
    ax.spines['left'].set_color('none')
    ax.spines['bottom'].set_color('none')

The first time this test is run, there will be no baseline image to compare against, so the test will fail. Copy the output images (in this case result_images/test_category/spines_axes_positions.*) to the baseline_images tree in the source directory (in this case lib/matplotlib/tests/baseline_images/test_category) and put them under source code revision control (with git add). When rerunning the tests, they should now pass.

There are two optional keyword arguments to the image_comparison decorator:

  • extensions: If you only wish to test some of the image formats (rather than the default png, svg and pdf formats), pass a list of the extensions to test.
  • tol: This is the image matching tolerance, the default 1e-3. If some variation is expected in the image between runs, this value may be adjusted.

Known failing tests

If you’re writing a test, you may mark it as a known failing test with the knownfailureif() decorator. This allows the test to be added to the test suite and run on the buildbots without causing undue alarm. For example, although the following test will fail, it is an expected failure:

from nose.tools import assert_equal
from matplotlib.testing.decorators import knownfailureif

@knownfailureif(True)
def test_simple_fail():
    '''very simple example test that should fail'''
    assert_equal(1+1,3)

Note that the first argument to the knownfailureif() decorator is a fail condition, which can be a value such as True, False, or ‘indeterminate’, or may be a dynamically evaluated expression.

Creating a new module in matplotlib.tests

Let’s say you’ve added a new module named matplotlib.tests.test_whizbang_features. To add this module to the list of default tests, append its name to default_test_modules in lib/matplotlib/__init__.py.

Using tox

Tox is a tool for running tests against multiple Python environments, including multiple versions of Python (e.g.: 2.6, 2.7, 3.2, etc.) and even different Python implementations altogether (e.g.: CPython, PyPy, Jython, etc.)

Testing all 4 versions of Python (2.6, 2.7, 3.1, and 3.2) requires having four versions of Python installed on your system and on the PATH. Depending on your operating system, you may want to use your package manager (such as apt-get, yum or MacPorts) to do this, or use pythonbrew.

tox makes it easy to determine if your working copy introduced any regressions before submitting a pull request. Here’s how to use it:

$ pip install tox
$ tox

You can also run tox on a subset of environments:

$ tox -e py26,py27

Tox processes everything serially so it can take a long time to test several environments. To speed it up, you might try using a new, parallelized version of tox called detox. Give this a try:

$ pip install -U -i http://pypi.testrun.org detox
$ detox

Tox is configured using a file called tox.ini. You may need to edit this file if you want to add new environments to test (e.g.: py33) or if you want to tweak the dependencies or the way the tests are run. For more info on the tox.ini file, see the Tox Configuration Specification.

Using Travis CI

Travis CI is a hosted CI system “in the cloud”.

Travis is configured to receive notifications of new commits to GitHub repos (via GitHub “service hooks”) and to run builds or tests when it sees these new commits. It looks for a YAML file called .travis.yml in the root of the repository to see how to test the project.

Travis CI is already enabled for the main matplotlib GitHub repository – for example, see its Travis page.

If you want to enable Travis CI for your personal matplotlib GitHub repo, simply enable the repo to use Travis CI in either the Travis CI UI or the GitHub UI (Admin | Service Hooks). For details, see the Travis CI Getting Started page.

Once this is configured, you can see the Travis CI results at http://travis-ci.org/#!/your_GitHub_user_name/matplotlib – here’s an example.

Licenses

Matplotlib only uses BSD compatible code. If you bring in code from another project make sure it has a PSF, BSD, MIT or compatible license (see the Open Source Initiative licenses page for details on individual licenses). If it doesn’t, you may consider contacting the author and asking them to relicense it. GPL and LGPL code are not acceptable in the main code base, though we are considering an alternative way of distributing L/GPL code through an separate channel, possibly a toolkit. If you include code, make sure you include a copy of that code’s license in the license directory if the code’s license requires you to distribute the license with it. Non-BSD compatible licenses are acceptable in matplotlib toolkits (eg basemap), but make sure you clearly state the licenses you are using.

Why BSD compatible?

The two dominant license variants in the wild are GPL-style and BSD-style. There are countless other licenses that place specific restrictions on code reuse, but there is an important difference to be considered in the GPL and BSD variants. The best known and perhaps most widely used license is the GPL, which in addition to granting you full rights to the source code including redistribution, carries with it an extra obligation. If you use GPL code in your own code, or link with it, your product must be released under a GPL compatible license. I.e., you are required to give the source code to other people and give them the right to redistribute it as well. Many of the most famous and widely used open source projects are released under the GPL, including linux, gcc, emacs and sage.

The second major class are the BSD-style licenses (which includes MIT and the python PSF license). These basically allow you to do whatever you want with the code: ignore it, include it in your own open source project, include it in your proprietary product, sell it, whatever. python itself is released under a BSD compatible license, in the sense that, quoting from the PSF license page:

There is no GPL-like "copyleft" restriction. Distributing
binary-only versions of Python, modified or not, is allowed. There
is no requirement to release any of your source code. You can also
write extension modules for Python and provide them only in binary
form.

Famous projects released under a BSD-style license in the permissive sense of the last paragraph are the BSD operating system, python and TeX.

There are several reasons why early matplotlib developers selected a BSD compatible license. matplotlib is a python extension, and we choose a license that was based on the python license (BSD compatible). Also, we wanted to attract as many users and developers as possible, and many software companies will not use GPL code in software they plan to distribute, even those that are highly committed to open source development, such as enthought, out of legitimate concern that use of the GPL will “infect” their code base by its viral nature. In effect, they want to retain the right to release some proprietary code. Companies and institutions who use matplotlib often make significant contributions, because they have the resources to get a job done, even a boring one. Two of the matplotlib backends (FLTK and WX) were contributed by private companies. The final reason behind the licensing choice is compatibility with the other python extensions for scientific computing: ipython, numpy, scipy, the enthought tool suite and python itself are all distributed under BSD compatible licenses.