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 library built around the C-library Chipmunk2D, which provides almost all of the base functionality around the physics simulation such as collision detection, impulse solving etc. Bascially it runs the simulation, and Pymunk calls it with input, and receives the result.

To wrap Chipmunk Pymunk uses CFFI in API mode. On top of the CFFI wrapping is a handmade pythonic layer to make it nice to use from Python code.

Impulse Solver Algorithm

Pymunk in itself only performs a minimum amount of physics calculation, instead those are handled by the underlying C-library Chipmunk 2D. Chipmunk2D (and therefor Pymunk) uses standard Euler to perform the integration

Scott/slembcke (the creator of Chipmunk2D), wrote this to describe the method used on the Chipmunk2D Forum:

Chipmunk works like this:

  1. Integrate the positions of everything and finds colliding pairs.

  2. Pre-calculate a number of properties of the contacts and joints. (mass properties, bounce velocities, etc)

  3. Integrate the velocities of everything.

  4. Run a number of solver iterations, to fix velocity constraints.

Case 1, a box sitting on the ground:

  1. The box is at rest, it’s velocity is (very near) zero, it’s position doesn’t change. Generate a contact with the ground.

  2. Precalculate the contact properties, if elasticity is set, the “bounce” velocity is calculated now (as 0).

  3. Integrate the velocity, gravity makes the object accelerate downwards.

  4. Solve the contact. Velocity should converge back to 0. If elasticity was set, it will be handled correctly without resorting to threshold velocities.

#1 is really the most important part. If you were to update the velocity before the postition, it would cause the box to move itself into a position where it would intersect the ground. While Chipmunk has to solve these overlaps anyway, avoiding them seems desirable. Another very useful property is that when cpSpaceStep() returns, all of the collision detection data structures are up to date. No need to reindex them twice in a single frame if you want to make queries.

Collision Detection Algorithm

Just as the impulse solver, the collision detection is also handled by the underlying C-library Chipmunk2D.

Chipmunk uses GJK/EPA to find collisions between the tricky cases (e.g. polygons, segment shapes). There is a blog post here with more details.

Why CFFI?

This is a straight copy from the github issue tracking the CFFI upgrade. https://github.com/viblo/pymunk/issues/99

CFFI have a number of advantages but also a downsides.

Advantages (compared to ctypes):

  • Its an active project. The developers and users are active, there are new releases being made and its possible to ask and get answers within a day on the CFFI mailing list.

  • Its said to be the way forward for Pypy, with promise of better performance compares to ctypes.

  • A little easier than ctypes to wrap things since you can just copy-paste the c headers.

Disadvatages (compared to ctypes):

  • ctypes is part of the CPython standard library, CFFI is not. That means that it will be more difficult to install Pymunk if it uses CFFI, since a copy-paste install is no longer possible in an easy way.

For me I see the 1st advantage as the main point. I have had great difficulties with strange segfaults with 64bit pythons on windows, and also sometimes on 32bit python, and support for 64bit python on both windows and linux is something I really want. Hopefully those problems will be easier to handle with CFFI since it has an active community.

Then comes the 3rd advantage, that its a bit easier to wrap the c code. For ctypes I have a automatic wrapping script that does most of the low level wrapping, but its not supported, very difficult to set up (I only managed inside a VM with linux) and quite annoying. CFFI would be a clear improvement.

For the disadvantage of ctypes I think it will be acceptable, even if not ideal. Many python packages have to be installed in some way (like pygame), and nowadays with pip its very easy to do. So I hope that it will be ok.

See the next section on why ctypes was used initially.

Why ctypes? (OBSOLETE)

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 documentation generation extension and a customized setup.py file to allow compilation of Chipmunk.

docs/src/ext/autoexample.py

A Sphinx extension that scans a directory and extracts the toplevel docstring. Used to autogenerate the examples documentation.

pymunk/_callbacks.py

Callbacks cannot be specified on a class, so they are all gathered here.

pymunk/_chipmunk_cffi.py

This file only contains a call to _cffi_backend .py, and exists mostly as a wrapper to be able to switch between abi and api mode of Cffi. This is currently not in use in the released code, but is used during experimentation.

pymunk/pymunk_extension_build.py

The low level Chipmunk bindings are located in this file. Contains configuration for how to build Chipmunk2D by Cffi into a pyd file.

setup.py

Except for the standard setup stuff this file also contain the custom build commands to build Chipmunk from source, using a build_ext extension.

pymunk/examples/*

Collection of examples of usages of Pymunk. Tries to showcase most common features of Pymunk.

pymunk/tests/*

Collection of (unit) tests. Does not cover all cases, but most core things are there. The tests require a working chipmunk library file.

pymunk_cffi/*

Custom (c source and header) extensions to Chipmunk2D used by Pymunk. Will be used when Cffi builds the pyd extension.

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 pymunk.tests package (pymunk/tests). Not exactly all the code is tested, but most of it (at the time of writing its about 85% of the core parts).

The tests can be run by calling the module

> python -m pymunk.tests

Its possible to control which tests to run, by specifying a filtering argument. The matching is as broad as possible, so UnitTest matches all the unit tests, test_arbiter all tests in test_arbiter.py and testResetitution matches the exact testRestitution test case

> python -m pymunk.tests -f testRestitution

To see all options to the tests command use -h

> python -m pymunk.tests -h

Since the tests cover even the optional parts, you either have to make sure all the optional dependencies are installed, or filter out those tests.

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.

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 to check if its included in the cdef definition in pymunk_extension_build.py. If its not just add it.

cpBool cpBodyIsSleeping(const cpBody *body);

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

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

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

Note that any cdef methods are not part of the public API, and as such might change between minor or patch verisons of Pymunk.

Weak References and free Methods

Internally Pymunk allocates structs from Chipmunk (the c library). For example a Body struct is created from inside the constructor method when a pymunk.Body is created. Because of this its important that the corresponding c side memory is deallocated properly when not needed anymore, usually when the Python side object is garbage collected. Most Pymunk objects use ffi.gc with a custom free function to do this. Note that the order of freeing is very important to avoid errors.