Advanced

In this section different “Advanced” topics are covered, things you normally dont need to worry about when you use pymunk but might be of interest if you want a better understanding of pymunk for example to extend it.

First off, pymunk is a pythonic wrapper around the C-library Chipmunk.

To wrap Chipmunk pymunk uses ctypes. In the bottom an autogenerated layer, and then a handmade pythonic layer on top to make it nice to use from Python programs.

Why ctypes?

The reasons for ctypes instead of [your favorite wrapping solution] can be summarized as

  • You only need to write pure python code when wrapping. This is good for several reasons. I can not really code in c. Sure, I can read it and write easy things, but Im not a good c coder. What I do know quite well is python. I imagine that the same is true for most people using pymunk, after all its a python library. :) Hopefully this means that users of pymunk can look at how stuff is actually done very easily, and for example add a missing chipmunk method/property on their own in their own code without much problem, and without being required to compile/build anything.
  • ctypes is included in the standard library. Anyone with python has it already, no dependencies on 3rd party libraries, and some guarantee that it will stick around for a long time.
  • The only thing required to run pymunk is python and a c compiler (in those cases a prebuilt version of chipmunk is not included). This should maximize the multiplatformness of pymunk, only thing that would even better would be a pure python library (which might be a bad idea for other reasons, mainly speed).
  • Not much magic going on. Working with ctypes is quite straight forward. Sure, pymunk uses a generator which is a bit of a pain, but at least its possible to sidestep it if required, which Ive done in some cases. Ive also got a share amount of problems when stuff didnt work as expected, but I imagine it would have been even worse with other solutions. At least its only the c library and python, and not some 3rd party in between.
  • Non api-breaking fixes in chipmunk doesnt affect pymunk. If a bugfix, some optimization or whatever is done in chipmunk that doesnt affect the API, then its enough with a recompile of chipmunk with the new code to benefit from the fix. Easy for everyone.
  • Ctypes can run on other python implementations than cpython. Right now pypy feels the most promising and it is be able to run ctypes just fine.

As I see it, the main benefit another solution could give would be speed. However, there are a couple of arguments why I dont find this as important as the benefits of ctypes

  • You are writing your game in python in the first place, if you really required top performance than maybe rewrite the whole thing in c would be better anyway? Or make a optimized binding to chipmunk.

    For example, if you really need excellent performance then one possible optimization would be to write the drawing code in c as well, and have that interact with chipmunk directly. That way it can be made more performant than any generic wrapping solution as it would skip the whole layer.

  • The bottleneck in a full game/application is somewhere else than in the physics wrapping in many cases. If your game has AI, logic and so on in python, then the wrapper overhead added by ctypes is not so bad in comparison.

  • Pypy. ctypes on pypy has the potential to be very quick. However, right now with pypy-1.9 the speed of pymunk is actually a bit slower on pypy than on cpython. Hopefully this will improve in the future.

Note that pymunk has been around since late 2007 which means not all wrapping options that exist today did exist or was not stable/complete enough for use by pymunk in the beginning. There are more options available today, and using ctypes is not set in stone. If a better alternative comes around then pymunk might switch given the improvements are big enough.

Code Layout

Most of pymunk should be quite straight forward.

Except for the documented API pymunk has a couple of interesting parts. Low level bindings to Chipmunk, a custom library load function, a custom documentation generation extension and a customized setup.py file to allow compilation of Chipmunk.

The low level chipmunk bindings are located in the two files _chipmunk.py and _chipmunk_ffi.py. In order to locate and load the compiled chipmunk library file pymunk uses a custom load_library function in libload.py

docs/src/ext/autoexample.py
A Sphinx extension that scans a directory and extracts the toplevel docstring. Used to autogenerate the examples documentation.
pymunk/_chipmunk.py
This file contains autogenerated low level ctypes binding created by the generate_bindings script.
pymunk/_chipmkunk_ffi.py
This file contains manual bindings not automatically generated.
pymunk/libload.py
This file contains the custom ctypes library load fucntion that is used by the rest of pymunk to load the Chipmunk library file.
setup.py
Except for the standard setup stuff this file also contain the custom build commands to build Chipmunk from source.
tests/*
Collection of (unit) tests. Does not cover all cases, but most core things are there. The tests require a working chipmunk library file.
tools/*
Collection of helper scripts that can be used to various development tasks such as generating documentation.

Tests

There are a number of unit tests included in the tests folder. Not exactly all the code is tested, but most of it (at the time of writing its about 85% of the core parts).

There is a helper script in the tools folder to easily run the tests:

> cd tools
> python run_tests.py

Working with non-wrapped parts of Chipmunk

In case you need to use something that exist in Chipmunk but currently is not included in pymunk the easiest method is to add it manually. All

Note: If you only want one or two new functions its probably easier to just add them manually to _chipmunk.py. See the ctypes documentation for instructions on what the function definitons/structs/whatever should look like.

For example, lets assume that the is_sleeping property of a body was not wrapped by pymunk. The Chipmunk method to get this property is named cpBodyIsSleeping.

First we need some imports:

from ctypes import *
from ._chipmunk import cpBody, chipmunk_lib, function_pointer

Then the actual ctypes wrapping:

cpBodyIsSleeping = (function_pointer(cpBool, POINTER(cpBody))).in_dll(chipmunk_lib, '_cpBodyIsSleeping')

Then to make it easy to use we want to create a python method that looks nice:

def is_sleeping(body):
    return cpffi.cpBodyIsSleeping(body._body)

Now we are ready with the mapping and ready to use our new method.

Full example:

from ctypes import *
from ._chipmunk import cpBody, chipmunk_lib, function_pointer

cpBodyIsSleeping = (function_pointer(cpBool, POINTER(cpBody))).in_dll(chipmunk_lib, '_cpBodyIsSleeping')

def is_sleeping(body):
    return cpffi.cpBodyIsSleeping(body._body)

import pymunk
body = pymunk.Body(1,2)
print is_sleeping(body)

Regenerate bindings to Chipmunk

You need the ctypes code generator. It is part of the ctypeslib package. You will also need GCC_XML. See the ctypes wiki for instructions on how to set it up: http://starship.python.net/crew/theller/wiki/CodeGenerator

I have found that ctypeslib and gcc_xml are easiest to get to work under Linux. Even if you normally work under Windows I suggest you put up a virtual machine with a linux dist to make things easier.

When ctypeslib (h2xml and xml2py) and gcc_xml are installed you can use the helper file to regenerate the bindings. It is located in the tools folder:

> cd tools
> python generate_bindings.py

You have now created a _chipmunk.py file with generated bindings. (use –help to display options, you will most probably want to change the include path and possibly the lib path)