Coding guide

Version control

svn checkouts

Checking out everything in the trunk (matplotlib and toolkits):

svn co https://matplotlib.svn.sourceforge.net/svnroot/matplotlib/trunk \
matplotlib --username=youruser --password=yourpass

Checking out the main source:

svn co https://matplotlib.svn.sourceforge.net/svnroot/matplotlib/trunk/\
matplotlib mpl --username=youruser --password=yourpass

Branch checkouts, eg the maintenance branch:

svn co https://matplotlib.svn.sourceforge.net/svnroot/matplotlib/branches/\
v0_91_maint mpl91 --username=youruser --password=yourpass

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 API_CHANGES, and consider posting to matplotlib-devel

  • Are your changes python2.4 compatible? We still support 2.4, so avoid features new to 2.5

  • Can you pass examples/tests/backend_driver.py? This is our poor man’s unit test.

  • If you have altered extension code, do you pass unit/memleak_hawaii.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 branch and trunk in sync where it makes sense. If there is a bug on both that needs fixing, use svnmerge.py to keep them in sync. The basic procedure is:

    • install svnmerge.py in your PATH:

      > wget http://svn.collab.net/repos/svn/trunk/contrib/client-side/\
        svnmerge/svnmerge.py
    • get a svn copy of the maintenance branch and the trunk (see above)

    • Michael advises making the change on the branch and committing it. Make sure you svn upped on the trunk and have no local modifications, and then from the svn trunk do:

      > svnmerge.py merge

      If you wish to merge only specific revisions (in an unusual situation), do:

      > svnmerge.py merge -rNNN1-NNN2

      where the NNN are the revision numbers. Ranges are also acceptable.

      The merge may have found some conflicts (code that must be manually resolved). Correct those conflicts, build matplotlib and test your choices. If you have resolved any conflicts, you can let svn clean up the conflict files for you:

      > svn -R resolved .

      svnmerge.py automatically creates a file containing the commit messages, so you are ready to make the commit:

      > svn commit -F svnmerge-commit-message.txt

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 checking changes into svn.

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 svn 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.

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 different 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 sagemath, linux, gcc and emacs.

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 two primary reasons why early matplotlib developers selected a BSD compatible license. 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 in general, who use matplotlib often make significant contributions, since they have the resources to get a job done, even a boring one, if they need it in their code. Two of the matplotlib backends (FLTK and WX) were contributed by private companies.

The other reason is licensing 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.