Hide/Show Outline

Chipmunk Physics 6.1.0

First of all, Chipmunk is a 2D rigid body physics library distributed under the MIT license. It is intended to be blazingly fast, portable, numerically stable, and easy to use. For this reason it’s been used in hundreds of games across just about every system you can name. This includes top quality titles such as Night Sky for the Wii and many #1 sellers on the iPhone App Store! I’ve put thousands of hours of work over many years to make Chipmunk what it is today. If you find Chipmunk has saved you a lot of time, please consider donating. You’ll make an indie game developer very happy!

First of all, I would like to give a Erin Catto a big thank you, as Chipmunk’s impulse solver was directly inspired by his example code way back in 2006. (Now a full fledged physics engine all it’s own: Box2D.org). His contact persistence idea allows for stable stacks of objects with very few iterations of the solver. My previous solver produced mushy piles of objects or required a large amount of CPU to operate stably.

Why a C Library?

A lot of people ask me why I wrote Chipmunk in C instead of pick your favorite language here. I tend to get really excited about different programming languages. Depending on the month, take your pick of Scheme, OCaml, Ruby, Objective-C, OOC, Lua, Io… the list goes on. The one common factor between most any language is that they are usually dead simple to make bindings to C code. I also wanted Chipmunk to be fast, portable, easy to optimize and easy to debug. Writing Chipmunk in C made it simpler to acheive all of those goals.

That said, I’ve never developed a whole game in C and I probably never will. There are much more interesting and fun languages than C with all sorts of nice features like garbage collection, closures and all sorts of unique object oriented runtimes. Check out the Bindings and Ports page to see if you can use Chipmunk from your language of choice. Because Chipmunk is written in a subset of C99 it compiles cleanly as C, C++, Objective-C and Objective-C++ code, making it easy to integrate into projects in those languages.

If you are writing iPhone games using Chipmunk, you should check out the Objective-Chipmunk wrapper that we’ve developed for Chipmunk. It integrates with the Objective-C memory model and provides a number of high level APIs that make writing physics based games for the iPhone even easier. While we do charge for Objective-Chipmunk, it will almost certainly save you more time than the small cost to license it. As a bonus, you’ll be helping to ensure that we can afford to continue to work on Chipmunk improvements.

Limitations of a C API:

Chipmunk does provide overloaded operators for *, +, and - (unary and binary) if you are using C++, but falls back to using functions such as cpvadd() and cpvsub() for C code. This is a little harder to read, but works OK once you get used to it. Most of the interesting vector operations that are possible don’t have a symbol for them anyway (at least not on a keyboard).

Another problem for a C API is access restriction. There are many structs, fields and functions in Chipmunk that are only meant to be used internally. To work around this, I have a separate header full of Chipmunk’s private API, chipmunk_private.h. I also use a macro, CP_PRIVATE() to mangle names in public structures. While you can feel free to include this header or use the macro in your own code to access the private API, be aware that the fields and functions may be renamed or disappear without warning and I have no plans to document or support usage of the private API.

Get it, Compile it:

If you haven’t downloaded it yet, you can always download the newest version of Chipmunk here. Inside you’ll find a command line build script that works with CMake, a XCode project and project files for Visual Studio ’09 and ’10.

Debug or Release?

Debug mode might be slightly slower, but will include a lot of error checking assertions that can help you find bugs quicker such as removing objects twice or doing things that might cause unsolvable collisions. I highly recommend you use a debug build until you are ready to ship the game and only then switch to a release build.

XCode (Mac/iPhone)

The included XCode project has targets for building a static library for the Mac or iOS. Additionally, you might want to just run the macosx/iphonestatic.command or macosx/macstatic.command to build you a directory with the headers and debug/release static libraries that you can just drop right into your projects. Including Chipmunk in your project with all the correct compiler flags applied couldn’t be simpler. The iPhone script generates a “fat” library that can be used with both the iOS simulator and devices. The device version is compiled as release, and the simulator version is compiled as debug.

MSVC

I rarely use MSVC, but others have chipped in to help me maintain Visual Studio project files. The MSVC 10 project should work, as I usually remember to test that before making a stable release. The MSVC 9 project may not as I don’t have access to that version. Let me know if there are any issues.

Command Line

The CMake build script should work on any system (Unix/Win/Mac) as long as you have CMake installed. It can even generate XCode or MSVC projects if you want (see CMake’s documentation for more information).

To compile a Chipmunk debug build on the command line, all you need to do is run:

cmake -D CMAKE_BUILD_TYPE=Debug .
make

If the -D CMAKE_BUILD_TYPE=Debug option is left out, it will make a release build instead.

Hello Chipmunk (World)

Hello world Chipmunk style. Create a simple simulation where a ball falls onto a static line segment, then rolls off. Print out the coordinates of the ball.

Hide/Show Hello Chipmunk Example

Support:

The best way to get support is to visit the Chipmunk Forums. There are plenty of people around using Chipmunk on the just about every platform I’ve ever heard of. If you are working on a commercial project, Howling Moon Software (my company) is available for contracting. We can help with implementing custom Chipmunk behaviors, as well as priority bug fixes and performance tuning.

Contact:

If you find any bugs in Chipmunk, errors or broken links in this document, or have a question or comment about Chipmunk you can contact me at slembcke(at)gmail(dot)com. (email or GTalk)

License:

Chipmunk is licensed under the MIT license.

Copyright (c) 2007-2011 Scott Lembcke

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

This means that you do not have to buy a license or pay to use Chipmunk in commercial projects. (Though we really appreciate donations)

Links:

Chipmunk Basics:

Overview:

There are 4 basic object types you will use in Chipmunk.

There is often confusion between rigid bodies and their collision shapes in Chipmunk and how they relate to sprites. A sprite would be a visual representation of an object, while a collision shape is an invisible property that defines how objects should collide. Both the sprite’s and the collision shape’s position and rotation are controlled by the motion of a rigid body. Generally you want to create a game object type that ties these things all together.

Memory Management the Chipmunk way:

For most of the structures you will use, Chipmunk uses a more or less standard and straightforward set of memory management functions. Take the cpSpace struct for example:

You are responsible for freeing any structs that you allocate. Chipmunk does not do reference counting or garbage collection. If you call a new function, you must call the matching free function or you will leak memory.

Additionally if you need more control over allocation and initialization because you are allocating temporary structs on the stack, writting a language binding, or working in a low memory environment you can also use the following functions. Most people will never have any need to use these functions.

Like calls to the new and free functions. Any memory allocated by an alloc function must be freed by cpfree() or similar. Any call to an init function must be matched with it’s destroy function.

To further ease integration with garbage collectors or other memory management constraints, Chipmunk has a number of compile time defines (cpcalloc(), cprealloc(), and cpfree()) that can be overriden. If you aren’t using Chipmunk from a garbage collected language, I’d highly recommend using libGC. It provides nearly transparent garbage collection for C based languages.

Basic Types:

chipmunk_types.h defines a number of basic types that Chipmunk uses. These can be changed at compile time to better suit your needs:

If you are writting a game engine or language binding on top of Chipmunk, you might want to choose to use object references instead of integers for collision type and group. I often use class pointers for collision types and game object pointers for groups. It’s much simpler than keeping a table of enumerations around.

Note: On the iPhone, cpFloat is defined as float and cpVect is an alias for CGPoint for performance and compatibility reasons.

Math the Chipmunk way:

First of all, Chipmunk uses double precision floating point numbers throughout it’s calculations by default. This is likely to be faster on most modern desktop processors, and means you have to worry less about floating point round off errors. You can change the floating point type used by Chipmunk when compiling the library. Look in chipmunk_types.h.

Chipmunk defines a number of aliases for common math functions so that you can choose to use floats or doubles for Chipmunk’s floating point type. In your own code, there probably isn’t a strong reason to use these aliases unless you expect you might want to change Chipmunk’s floating point type later and a 2% performance hit from using the wrong float/double version of math functions will matter.

There are a few unique functions you will probably find very useful:

Floating point infinity is defined as INFINITY. This is defined by many math libraries, but is not actually part of the C standard library.

Chipmunk Vectors: cpVect

Struct Definition, Constants and Constructors:

typedef struct cpVect{
	cpFloat x, y;
} cpVect

2D vector packed into a struct. No surprises here.

static const cpVect cpvzero = {0.0f,0.0f};

Constant for the zero vector.

cpVect cpv(const cpFloat x, const cpFloat y)

Convenience constructor for creating new cpVect structs.

Operations:

Chipmunk Axis Aligned Bounding Boxes: cpBB

Struct Definition and Constructors:

typedef struct cpBB{
	cpFloat l, b, r ,t;
} cpBB

Simple bounding box struct. Stored as left, bottom, right, top values.

cpBB cpBBNew(const cpFloat l, const cpFloat b, const cpFloat r, const cpFloat t)

Convenience constructor for cpBB structs. Like cpv() this function returns a copy and not a malloced pointer.

cpBB cpBBNewForCircle(const cpVect p, const cpFloat r)

Convenience constructor for making a cpBB fitting a circle at position p with radius r.

Operations:

Chipmunk Rigid Bodies: cpBody

Rogue and Static Bodies:

Normally when you create a rigid body, you add it to a space so the space will start simulating it. This means it will update it’s position and velocity, apply forces to it, be affected by gravity, etc. A body that isn’t added to a space (and not simulated) is called a rogue body. The most important use for rogue bodies are as static bodies, but you can also use them to implement directly controlled objects such as moving platforms.

Static bodies are rogue bodies, but with a special flag set on them to let Chipmunk know that they never move unless you tell it. Static bodies have two purposes. Originally they were added for the sleeping feature. Because static bodies don’t move, Chipmunk knows that it’s safe to let objects that are touching or jointed to them fall asleep. Objects touching or jointed to regular rogue bodies are never allowed to sleep. The second purpose for static bodies is that Chipmunk knows shapes attached to them never need to have their collision detection data updated. Chipmunk also doesn’t need to bother checking for collisions between static objects. Generally all of your level geometry will be attached to a static body except for things like moving platforms or doors.

In previous versions of Chipmunk before 5.3 you would create an infinite mass rogue body to attach static shapes to using cpSpaceAddStaticShape(). You don’t need to do any of that anymore, and shouldn’t if you want to use the sleeping feature. Each space has a dedicated static body that you can use to attach your static shapes to. Chipmunk also automatically adds shapes attached to static bodies as static shapes.

Memory Management Functions:

cpBody *cpBodyAlloc(void)
cpBody *cpBodyInit(cpBody *body, cpFloat m, cpFloat i)
cpBody *cpBodyNew(cpFloat m, cpFloat i)

void cpBodyDestroy(cpBody *body)
void cpBodyFree(cpBody *body)

Standard set of Chipmunk memory management functions. m and i are the mass and moment of inertia for the body. Guessing the mass for a body is usually fine, but guessing a moment of inertia can lead to a very poor simulation. Be careful not to free a body before any shapes or constraints attached to it have been removed from a space.

Creating Additional Static Bodies:

While every cpSpace has a built in static body that you can use, it can be convenient to make your own as well. One potential use is in a level editor. By attaching chunks of your level to static bodies, you can still move and rotate the chunks independently of each other. Then all you have to do is call cpSpaceRehashStatic() to rebuild the static collision detection data when you are done.

For more information on rogue and static bodies, see Chipmunk Spaces.

cpBody *cpBodyAlloc(void);
cpBody *cpBodyInitStatic(cpBody *body)
cpBody *cpBodyNewStatic()

Create additional static bodies with infinite mass and moment of inertia.

Properties:

Chipmunk provides getter/setter functions for a number of properties on rigid bodies. Setting most properties automatically wakes the rigid bodies up if they were sleeping. You can also set the fields directly on the cpBody struct if you wish. They are documented in the headers.

cpFloat cpBodyGetMass(const cpBody *body)
void cpBodySetMass(cpBody *body, cpFloat m)

Mass of the body.

cpFloat cpBodyGetMoment(const cpBody *body)
void cpBodySetMoment(cpBody *body, cpFloat i)

Moment of inertia (MoI or sometimes just moment) of the body. The moment is like the rotational mass of a body. See below for function to help calculate the moment.

cpVect cpBodyGetPos(const cpBody *body)
void cpBodySetPos(cpBody *body, cpVect pos)

Position of the center of gravity of the body. When changing the position you may also want to call cpSpaceReindexShapesForBody() to update the collision detection information for the attached shapes if plan to make any queries against the space.

cpVect cpBodyGetVel(const cpBody *body)
void cpBodySetVel(cpBody *body, const cpVect value)

Linear velocity of the center of gravity of the body.

cpVect cpBodyGetForce(const cpBody *body)
void cpBodySetForce(cpBody *body, const cpVect value)

Force applied to the center of gravity of the body.

cpFloat cpBodyGetAngle(const cpBody *body)
void cpBodySetAngle(cpBody *body, cpFloat a)

Rotation of the body in radians. When changing the rotation you may also want to call cpSpaceReindexShapesForBody() to update the collision detection information for the attached shapes if plan to make any queries against the space.

cpFloat cpBodyGetAngVel(const cpBody *body)
void cpBodySetAngVel(cpBody *body, const cpFloat value)

The angular velocity of the body in radians per second.

cpFloat cpBodyGetTorque(const cpBody *body)
void cpBodySetTorque(cpBody *body, const cpFloat value)

The torque applied to the body.

cpVect cpBodyGetRot(const cpBody *body)

The rotation vector for the body. Can be used with cpvrotate() or cpvunrotate() to perform fast rotations.

cpFloat cpBodyGetVelLimit(const cpBody *body)
void cpBodySetVelLimit(cpBody *body, const cpFloat value)

Velocity limit of the body. Defaults to INFINITY unless you set it specifically. Can be used to limit falling speeds, etc.

cpFloat cpBodyGetAngVelLimit(const cpBody *body)
void cpBodySetAngVelLimit(cpBody *body, const cpFloat value)

Angular velocity limit of the body in radians per second. Defaults to INFINITY unless you set it specifically.

cpSpace* cpBodyGetSpace(const cpBody *body)

Get the cpSpace that body has been added to.

cpDataPointer cpBodyGetUserData(const cpBody *body)
void cpBodySetUserData(cpBody *body, const cpDataPointer value)

User data pointer. Use this pointer to get a reference to the game object that owns this body from callbacks.

Moment of Inertia and Area Helper Functions:

Use the following functions to approximate the moment of inertia for your body, adding the results together if you want to use more than one.

Hide/Show Moments Example

Use the following functions to get the area for common Chipmunk shapes if you want to approximate masses or density or whatnot.

Coordinate Conversion Functions:

Many things are defined in coordinates local to a body meaning that the (0,0) is at the center of gravity of the body and the axis rotate along with the body.

Applying Forces and Torques:

People are sometimes confused by the difference between a force and an impulse. An impulse is basically a very large force applied over a very short period of time, like a ball hitting a wall or cannon firing. Chipmunk treats impulses as if they occur instantaneously by simply adding directly to the velocity of an object. Both impulses and forces are affected the mass of an object. Double the mass of the object and halve the effect.

Note: Both the cpBodyApplyForce() cpBodyApplyImpulse() functions take a force or impulse in absolute coordinates and applies it at a relative offset in absolute coordinates. (The offset is relative to the center of gravity, but is not rotated with the body)

Sleeping Functions:

Chipmunk supports a sleeping feature so that it stops using CPU time simulating groups of objects that stop moving. Read more about it in the cpSpace section.

void cpBodySleepWithGroup(cpBody *body, cpBody *group)

When objects in Chipmunk sleep, they sleep as a group of all objects that are touching or jointed together. When an object is woken up, all of the objects in it’s group are woken up. cpBodySleepWithGroup() allows you group sleeping objects together. It acts identically to cpBodySleep() if you pass NULL as group by starting a new group. If you pass a sleeping body for group, body will be awoken when group is awoken. You can use this to initialize levels and start stacks of objects in a pre-sleeping state.

Hide/Show Sleeping Example

Iterators

typedef void (*cpBodyShapeIteratorFunc)(cpBody *body, cpShape *shape, void *data)
void cpBodyEachShape(cpBody *body, cpBodyShapeIteratorFunc func, void *data)

Call func once for each shape that is attached to body and added to a space. data is passed along as a context value. It is safe to remove shapes using these callbacks.

typedef void (*cpBodyConstraintIteratorFunc)(cpBody *body, cpConstraint *constraint, void *data)
void cpBodyEachConstraint(cpBody *body, cpBodyConstraintIteratorFunc func, void *data)

Call func once for each constraint that is attached to body and added to a space. data is passed along as a context value. It is safe to remove constraints using thes callbacks.

typedef void (*cpBodyArbiterIteratorFunc)(cpBody *body, cpArbiter *arbiter, void *data)
void cpBodyEachArbiter(cpBody *body, cpBodyArbiterIteratorFunc func, void *data)

This one is more interesting. Calls func once for each collision pair that body is involved in. Calling cpArbiterGet[Bodies|Shapes]() or CP_ARBITER_GET_[BODIES|SHAPES]() will return the body or shape for body as the first argument. You can use this to check all sorts of collision information for a body like if it’s touching the ground, another particular object, how much collision force is being applied to an object, etc. Sensor shapes and arbiters that have been rejected by a collision handler callback or cpArbiterIgnore() are not tracked by the contact graph.

Note: If your compiler supports blocks (such as Clang), there are an alternate set of functions you can call. cpBodyEachShapeBlock(), etc. See chipmunk.h for more information.

Integration Callbacks:

This section is a stub. For now you can look at the Planet demo for an example of how to use integration callbacks to implement planetary gravity.

Misc Functions:

Notes:

Chipmunk Collision Shapes: cpShape

There are currently 3 collision shape types:

You can add as many shapes to a body as you wish. That is why the two types are separate. This should give you the flexibility to make any shape you want as well providing different areas of the same object with different friction, elasticity or callback values.

When creating different types of shapes, you will always be given a cpShape* pointer back. This is because Chipmunk shapes are meant to be opaque types. Think of the specific collision types such as cpCircleShape, cpSegmentShape and cpPolyShape as private subclasses of cpShape. You can still read some properties from them using the getter functions, but you are not intended to cast cpShape pointers to their specific types.

Properties:

Chipmunk provides getter/setter functions for a number of properties on collision shapes. Setting most properties automatically wakes the rigid body they are attached to up if it was sleeping. You can also set some of the fields directly on the cpShape struct if you wish. They are documented in the headers.

cpBody * cpShapeGetBody(const cpShape *shape)
void cpShapeSetBody(cpShape *shape, cpBody *body)

The rigid body the shape is attached to. Can only be set when the shape is not added to a space.

cpBB cpShapeGetBB(const cpShape *shape)

The bounding box of the shape. Only guaranteed to be valid after cpShapeCacheBB() or cpSpaceStep() is called. Moving a body that a shape is connected to does not update it’s bounding box. For shapes used for queries that aren’t attached to bodies, you can also use cpShapeUpdate().

cpBool cpShapeGetSensor(const cpShape *shape)
void cpShapeSetSensor(cpShape *shape, cpBool value)

A boolean value if this shape is a sensor or not. Sensors only call collision callbacks, and never generate real collisions.

cpFloat cpShapeGetElasticity(const cpShape *shape)
void cpShapeSetElasticity(cpShape *shape, cpFloat value)

Elasticity of the shape. A value of 0.0 gives no bounce, while a value of 1.0 will give a “perfect” bounce. However due to inaccuracies in the simulation using 1.0 or greater is not recommended however. The elasticity for a collision is found by multiplying the elasticity of the individual shapes together.

cpFloat cpShapeGetFriction(const cpShape *shape)
void cpShapeSetFriction(cpShape *shape, cpFloat value)

Friction coefficient. Chipmunk uses the Coulomb friction model, a value of 0.0 is frictionless. The friction for a collision is found by multiplying the friction of the individual shapes together. Tables of friction coefficients.

cpVect cpShapeGetSurfaceVelocity(const cpShape *shape)
void cpShapeSetSurfaceVelocity(cpShape *shape, cpVect value)

The surface velocity of the object. Useful for creating conveyor belts or players that move around. This value is only used when calculating friction, not resolving the collision.

cpCollisionType cpShapeGetCollisionType(const cpShape *shape)
void cpShapeSetCollisionType(cpShape *shape, cpCollisionType value)

You can assign types to Chipmunk collision shapes that trigger callbacks when objects of certain types touch. See the callbacks section for more information.

cpGroup cpShapeGetGroup(const cpShape *shape)
void cpShapeSetGroup(cpShape *shape, cpGroup value)

Shapes in the same non-zero group do not generate collisions. Useful when creating an object out of many shapes that you don’t want to self collide. Defaults to CP_NO_GROUP.

cpLayers cpShapeGetLayers(const cpShape *shape)
void cpShapeSetLayers(cpShape *shape, cpLayers value)

Shapes only collide if they are in the same bit-planes. i.e. (a->layers & b->layers) != 0 By default, a shape occupies all bit-planes. Wikipedia has a nice article on bitmasks if you are unfamiliar with how to use them. Defaults to CP_ALL_LAYERS.

cpSpace* cpShapeGetSpace(const cpShape *shape)

Get the cpSpace that shape has been added to.

cpDataPointer cpShapeGetUserData(const cpShape *shape)
void cpShapeSetUserData(cpShape *shape, cpDataPointer value)

A user definable data pointer. If you set this to point at the game object the shapes is for, then you can access your game object from Chipmunk callbacks.

Filtering Collisions:

Chipmunk has two primary means of ignoring collisions: groups and layers.

Groups are meant to ignore collisions between parts on a complex object. A ragdoll is a good example. When jointing an arm onto the torso, you’ll want them to allow them to overlap. Groups allow you to do exactly that. Shapes that have the same group don’t generate collisions. So by placing all of the shapes in a ragdoll in the same group, you’ll prevent it from colliding against other parts of itself.

Layers allow you to separate collision shapes into mutually exclusive planes. Shapes can be in more than one layer, and shapes only collide with other shapes that are in at least one of the same layers. As a simple example, say shape A is in layer 1, shape B is in layer 2, and shape C is in layer 1 and 2. Shape A and B won’t collide with each other, but shape C will collide with both A and B.

Layers can also be used to set up rule based collisions. Say you have four types of shapes in your game. The player, the enemies, player bullets and enemy bullets. The are that the player should collide with enemies, and bullets shouldn’t collide with the type (player or enemy) that fired them. Making a chart would look like this:

Player Enemy Player Bullet Enemy Bullet
Player (1) (2)
Enemy (3)
Player Bullet
Enemy Bullet

The ‘-’s are for redundant spots in the chart, and the numbers are spots where types should collide. You can use a layer for each rule that you want to define. Then add the layers to each type: The player should be in layers 1 and 2, the enemy should be in layers 1 and 3, the player bullets should be in layer 3, and the enemy bullets should be in layer 2. Treating layers as rules this way, you can define up to 32 rules. The default cpLayers type is unsigned int which has a resolution of 32 bits on most systems. You can redefine the cpLayers type in chipmunk_types.h if you need more bits to work with.

There is one last way of filtering collisions using collision handlers. See the section on callbacks for more information. While collision handlers can be more flexible, they are also the slowest method. So you try to use groups or layers first.

Memory Management Functions:

void cpShapeDestroy(cpShape *shape)
void cpShapeFree(cpShape *shape)

Destroy and Free functions are shared by all shape types. Allocation and initialization functions are specific to each shape type. See below.

Misc Functions:

Working With Circle Shapes:

cpCircleShape *cpCircleShapeAlloc(void)
cpCircleShape *cpCircleShapeInit(cpCircleShape *circle, cpBody *body, cpFloat radius, cpVect offset)
cpShape *cpCircleShapeNew(cpBody *body, cpFloat radius, cpVect offset)

body is the body to attach the circle to, offset is the offset from the body’s center of gravity in body local coordinates.

cpVect cpCircleShapeGetOffset(cpShape *circleShape)
cpFloat cpCircleShapeGetRadius(cpShape *circleShape)

Getters for circle shape properties. Passing as non-circle shape will throw an assertion.

Working With Segment Shapes:

cpSegmentShape* cpSegmentShapeAlloc(void)
cpSegmentShape* cpSegmentShapeInit(cpSegmentShape *seg, cpBody *body, cpVect a, cpVect b, cpFloat radius)
cpShape* cpSegmentShapeNew(cpBody *body, cpVect a, cpVect b, cpFloat radius)

body is the body to attach the segment to, a and b are the endpoints, and radius is the thickness of the segment.

cpVect cpSegmentShapeGetA(cpShape *shape)
cpVect cpSegmentShapeGetA(cpShape *shape)
cpVect cpSegmentShapeGetNormal(cpShape *shape)
cpFloat cpSegmentShapeGetRadius(cpShape *shape)

Getters for segment shape properties. Passing a non-segment shape will throw an assertion.

Working With Polygon Shapes:

cpPolyShape *cpPolyShapeAlloc(void)
cpPolyShape *cpPolyShapeInit(cpPolyShape *poly, cpBody *body, int numVerts, cpVect *verts, cpVect offset)
cpShape *cpPolyShapeNew(cpBody *body, int numVerts, cpVect *verts, cpVect offset)

body is the body to attach the poly to, verts is an array of cpVect structs defining a convex hull with a clockwise winding, offset is the offset from the body’s center of gravity in body local coordinates. An assertion will be thrown the vertexes are not convex or do not have a clockwise winding.

int cpPolyShapeGetNumVerts(cpShape *shape)
cpVect cpPolyShapeGetVert(cpShape *shape, int index)

Getters for poly shape properties. Passing a non-poly shape or an index that does not exist will throw an assertion.

Boxes:

Because boxes are so common in physics games, Chipmunk provides shortcuts to create box shaped polygons. The boxes will always be centered at the center of gravity of the body you are attaching them to. If you want to create an off-center box, you will need to use cpPolyShapeNew() or cpPolyShapeInit().

cpPolyShape *cpBoxShapeInit(cpPolyShape *poly, cpBody *body, cpFloat width, cpFloat height)
cpShape *cpBoxShapeNew(cpBody *body, cpFloat width, cpFloat height)

Poly Shape Helper Functions:

Convex Hull Helper Functions:

int cpConvexHull(int count, cpVect *verts, cpVect *result, int *first, cpFloat tol)

Calculate the convex hull of a given set of points. Returns the count of points in the hull. result must be a pointer to a cpVect array with at least count elements. If result is NULL, then verts array wil be reduced instead. first is an optional pointer to an integer to store where the first vertex in the hull came from (i.e. verts[first] == result[0]) tol is the allowed amount to shrink the hull when simplifying it. A tolerance of 0.0 creates an exact hull.

#define CP_CONVEX_HULL(inputCount, inputVerts, outputCount_varName, outputVerts_varName)

Convenience macro for using cpConvexHull(). Creates an array on the stack using alloca() and then calls cpConvexHull(). Because the output array is created on the stack it doesn’t need to be freed.

Hide/Show cpConvexHull Example

Modifying cpShapes:

The short answer is that you can’t because the changes would be only picked up as a change to the position of the shape’s surface, but not it’s velocity. The long answer is that you can using the “unsafe” API as long as you realize that doing so will not result in realistic physical behavior. These extra functions are defined in a separate header chipmunk_unsafe.h.

Notes:

Chipmunk Spaces: cpSpace

Spaces in Chipmunk are the basic unit of simulation. You add rigid bodies, shapes and constraints to it and then step them all forward through time together.

What Are Iterations, and Why Should I care?

Chipmunk uses an iterative solver to figure out the forces between objects in the space. What this means is that it builds a big list of all of the collisions, joints, and other constraints between the bodies and makes several passes over the list considering each one individually. The number of passes it makes is the iteration count, and each iteration makes the solution more accurate. If you use too many iterations, the physics should look nice and solid, but may use up too much CPU time. If you use too few iterations, the simulation may seem mushy or bouncy when the objects should be solid. Setting the number of iterations lets you balance between CPU usage and the accuracy of the physics. Chipmunk’s default of 10 iterations is sufficient for most simple games.

Sleeping

New in Chipmunk 5.3 is the ability of spaces to disable entire groups of objects that have stopped moving to save CPU time as well as battery life. In order to use this feature you must do 2 things. The first is that you must attach all your static geometry to static bodies. Objects cannot fall asleep if they are touching a non-static rogue body even if it’s shapes were added as static shapes. The second is that you must enable sleeping explicitly by choosing a time threshold value for cpSpace.sleepTimeThreshold. If you do not set cpSpace.idleSpeedThreshold explicitly, a value will be chosen automatically based on the current amount of gravity.

Properties:

int cpSpaceGetIterations(const cpSpace *space)
void cpSpaceSetIterations(cpSpace *space, int value)

Iterations allow you to control the accuracy of the solver. Defaults to 10. See above for more information.

cpVect cpSpaceGetGravity(const cpSpace *space)
void cpSpaceSetGravity(cpSpace *space, cpVect value)

Global gravity applied to the space. Defaults to cpvzero. Can be overridden on a per body basis by writing custom integration functions.

cpFloat cpSpaceGetDamping(const cpSpace *space)
void cpSpaceSetDamping(cpSpace *space, cpFloat value)

Amount of simple damping to apply to the space. A value of 0.9 means that each body will lose 10% of it’s velocity per second. Defaults to 1. Like gravity can be overridden on a per body basis.

cpFloat cpSpaceGetIdleSpeedThreshold(const cpSpace *space)
void cpSpaceSetIdleSpeedThreshold(cpSpace *space, cpFloat value)

Speed threshold for a body to be considered idle. The default value of 0 means to let the space guess a good threshold based on gravity.

cpFloat cpSpaceGetSleepTimeThreshold(const cpSpace *space)
void cpSpaceSetSleepTimeThreshold(cpSpace *space, cpFloat value)

Time a group of bodies must remain idle in order to fall asleep. The default value of INFINITY disables the sleeping feature.

cpFloat cpSpaceGetCollisionSlop(const cpSpace *space)
void cpSpaceSetCollisionSlop(cpSpace *space, cpFloat value)

Amount of overlap between shapes that is allowed. It’s encouraged to set this as high as you can without noticable overlapping as it improves the stability. It defaults to 0.1.

cpFloat cpSpaceGetCollisionBias(const cpSpace *space)
void cpSpaceSetCollisionBias(cpSpace *space, cpFloat value)

Chipmunk allows fast moving objects to overlap, then fixes the overlap over time. Overlapping objects are unavoidable even if swept collisions are supported, and this is an efficient and stable way to deal with overlapping objects. The bias value controls what percentage of overlap remains unfixed after a second and defaults to ~0.2%. Valid values are in the range from 0 to 1, but using 0 is not recommended for stability reasons. The default value is calculated as cpfpow(1.0f - 0.1f, 60.0f) meaning that Chipmunk attempts to correct 10% of error ever 1/60th of a second. Note: Very very few games will need to change this value.

cpTimestamp cpSpaceGetCollisionPersistence(const cpSpace *space)
void cpSpaceSetCollisionPersistence(cpSpace *space, cpTimestamp value)

The number of frames the space keeps collision solutions around for. Helps prevent jittering contacts from getting worse. This defaults to 3 and very very very few games will need to change this value.

cpFloat cpSpaceGetCurrentTimeStep(const cpSpace *space)

Retrieves the current (if you are in a callback from cpSpaceStep()) or most recent (outside of a cpSpaceStep() call) timestep.

cpFloat cpSpaceIsLocked(const cpSpace *space)

Returns true when in a callback meaning that you cannot add/remove objects from the space. Can be used to choose to create a post-step callback instead.

cpDataPointer cpSpaceGetUserData(const cpSpace *space)
void cpSpaceSetUserData(cpSpace *space, cpDataPointer value)

A user definable data pointer. It’s useful to point this at the gamestate object or scene management object that owns the space.

cpBody * cpSpaceGetStaticBody(const cpSpace *space)

A dedicated static body for the space. You don’t have to use it, but because it’s memory is managed automatically with the space it’s very convenient. You can set its user data pointer to something helpful if you want for callbacks.

Memory Management Functions:

cpSpace* cpSpaceAlloc(void)
cpSpace* cpSpaceInit(cpSpace *space)
cpSpace* cpSpaceNew()

void cpSpaceDestroy(cpSpace *space)
void cpSpaceFree(cpSpace *space)

More standard Chipmunk memory functions.

void cpSpaceFreeChildren(cpSpace *space)

This function will free all of the shapes, bodies and joints that have been added to space. Does not free space. You will still need to call cpSpaceFree() on your own. You will probably never use this in a real game, as your gamestate or game controller should manage removing and freeing objects from the space.

Operations:

cpShape *cpSpaceAddShape(cpSpace *space, cpShape *shape)
cpShape *cpSpaceAddStaticShape(cpSpace *space, cpShape *shape)
cpBody *cpSpaceAddBody(cpSpace *space, cpBody *body)
cpConstraint *cpSpaceAddConstraint(cpSpace *space, cpConstraint *constraint)

void cpSpaceRemoveShape(cpSpace *space, cpShape *shape)
void cpSpaceRemoveBody(cpSpace *space, cpBody *body)
void cpSpaceRemoveConstraint(cpSpace *space, cpConstraint *constraint)

cpBool cpSpaceContainsShape(cpSpace *space, cpShape *shape)
cpBool cpSpaceContainsBody(cpSpace *space, cpBody *body)
cpBool cpSpaceContainsConstraint(cpSpace *space, cpConstraint *constraint)

These functions add and remove shapes, bodies and constraints from space. The add/remove functions cannot be called from within a callback other than a postStep() callback (which is different than a postSolve() callback!). Attempting to add or remove objects from the space while cpSpaceStep() is still executing will throw an assertion. See the callbacks section for more information. The add functions return the thing being added so that you can create and add something in one line. Be careful not to free bodies before removing shapes and constraints attached to them or you will cause crashes.. The contains functions allow you to check if an object has been added to the space or not.

Deprecated:

cpShape *cpSpaceAddStaticShape(cpSpace *space, cpShape *shape)
void cpSpaceRemoveStaticShape(cpSpace *space, cpShape *shape)

Shapes attached to static bodies are automatically treated as static. There isn’t really a good reason to explicitly add static shapes anymore.

Spatial Indexing:

Chipmunk 6 officially supports 2 spatial indexes. The default is an axis-aligned bounding box tree inspired by the one used in the Bullet Physics library, but extended with my own collision pair caching to give the tree very good temporal coherence. The tree requires no tuning, and most games will find that they get much better performance using it. The other available index available is a spatial hash, which can be much faster when you have a very large number (1000s) of objects that are all the same size.

Occasionally, you might want to update the collision detection data for a shape. If you move a static shape or a static body you must do this to let Chipmunk know it needs to have it’s collision detection data updated. You may also want to manually update the collision data for normal shapes if you move them and still want to perform queries.

Iterators:

typedef void (*cpSpaceBodyIteratorFunc)(cpBody *body, void *data)
void cpSpaceEachBody(cpSpace *space, cpSpaceBodyIteratorFunc func, void *data)

Call func for each body in the space also passing along your data pointer. Sleeping bodies are included, but static and rogue bodies are not as they aren’t added to the space.

Hide/Show cpSpaceEachBody Example
typedef void (*cpSpaceShapeIteratorFunc)(cpShape *shape, void *data)
void cpSpaceEachShape(cpSpace *space, cpSpaceShapeIteratorFunc func, void *data)

Call func for each shape in the space also passing along your data pointer. Sleeping and static shapes are included.

typedef void (*cpSpaceConstraintIteratorFunc)(cpConstraint *constraint, void *data)
void cpSpaceEachConstraint(cpSpace *space, cpSpaceConstraintIteratorFunc func, void *data)

Call func for each constraint in the space also passing along your data pointer.

Note: If your compiler supports blocks (such as Clang), there are an alternate set of functions you can call. cpSpaceEachBodyBlock(), etc. See chipmunk.h for more information.

Simulating the Space:

void cpSpaceStep(cpSpace *space, cpFloat dt)

Update the space for the given time step. Using a fixed time step is highly recommended. Doing so can greatly increase the quality of the simulation. The easiest way to do constant timesteps is to simple step forward by 1/60th of a second (or whatever your target framerate is) for each frame regardless of how long it took to render. This works fine for many games, but a better way to do it is to separate your physics timestep and rendering. This is a good article on how to do that.

Enabling and Tuning the Spatial Hash:

If you have thousands of objects that are all approximately the same size, the spatial hash may be for you.

void cpSpaceUseSpatialHash(cpSpace *space, cpFloat dim, int count)

Switch the space to use a spatial hash instead of the bounding box tree.

The spatial hash data is fairly size sensitive. dim is the size of the hash cells. Setting dim to the average collision shape size is likely to give the best performance. Setting dim too small will cause the shape to be inserted into many cells, setting it too low will cause too many objects into the same hash slot.

count is the suggested minimum number of cells in the hash table. If there are too few cells, the spatial hash will return many false positives. Too many cells will be hard on the cache and waste memory. the Setting count to ~10x the number of objects in the space is probably a good starting point. Tune from there if necessary.

Using the spatial has visualization in the demo program you can see what I mean. The grey squares represent cells in the spatial hash. The darker the cell, the more objects have been mapped into that cell. A good dim size is when your objects fit nicely into the grid:

Notice the light grey meaning that each cell doesn’t have too many objects mapped onto it.

When you use too small a size, Chipmunk has to insert each object into a lot of cells. This can get expensive.

Notice that the grey cells are very small compared to the collision shapes.

When you use too big of a size, a lot of shapes will fit into each cell. Each shape has to be checked against every other shape in the cell, so this makes for a lot of unnecessary collision checks.

Notice the dark grey cells meaning that many objects are mapped onto them.

Chipmunk 6 also has an experimental single axis sort and sweep implementation. It can be very efficient on mobile games if your world is very long and flat like a racing game. See the code for cpSpaceUseSpatialHash() if you want to try enabling it.

Notes:

Chipmunk Constraints: cpConstraint

A constraint is something that describes how two bodies interact with each other. (how they constrain each other) Constraints can be simple joints that allow bodies to pivot around each other like the bones in your body, or they can be more abstract like the gear joint or motors.

What constraints are and what they are not:

Constraints in Chipmunk are all velocity based constraints. This means that they act primarily by synchronizing the velocity of two bodies. A pivot joint holds two anchor points on two separate bodies together by defining equations that say that the velocity of the anchor points must be the same and calculating impulses to apply to the bodies to try and keep it that way. A constraint takes a velocity as it’s primary input and produces a velocity change as it’s output. Some constraints, (joints in particular) apply velocity changes to correct differences in positions. More about this in the next section.

A spring connected between two bodies is not a constraint. It’s very constraint-like as it creates forces that affect the velocities of the two bodies, but a spring takes distances as input and produces forces as it’s output. If a spring is not a constraint, then why do I have two varieties of spring constraints you ask? The reason is because they are damped springs. The damping associated with the spring is a true constraint that creates velocity changes based on the relative velocities of the two bodies it links. As it is convenient to put a damper and a spring together most of the time, I figured I might as well just apply the spring force as part of the constraint instead of having a damper constraint and having the user calculate and apply their own spring forces separately.

Properties:

cpBody * cpConstraintGetA(const cpConstraint *constraint)
cpBody * cpConstraintGetB(const cpConstraint *constraint)

Getters for the two bodies the constraint is attached to.

cpFloat cpConstraintGetMaxForce(const cpConstraint *constraint)
void cpConstraintSetMaxForce(cpConstraint *constraint, cpFloat value)

The maximum force that the constraint can use to act on the two bodies. Defaults to INFINITY.

cpFloat cpConstraintGetErrorBias(const cpConstraint *constraint)
void cpConstraintSetErrorBias(cpConstraint *constraint, cpFloat value)

The percentage of joint error that remains unfixed after a second. This works exactly the same as the collision bias property of a space, but applies to fixing error (stretching) of joints instead of overlapping collisions.

cpFloat cpConstraintGetMaxBias(const cpConstraint *constraint)
void cpConstraintSetMaxBias(cpConstraint *constraint, cpFloat value)

The maximum speed at which the constraint can apply error correction. Defaults to INFINITY.

cpSpace* cpConstraintGetSpace(const cpConstraint *constraint)

Get the cpSpace that constraint has been added to.

cpDataPointer cpConstraintGetUserData(const cpConstraint *constraint)
void cpConstraintSetUserData(cpConstraint *constraint, cpDataPointer value)

User data pointer. Use this pointer to get a reference to the game object that owns this constraint from callbacks.

cpFloat cpConstraintGetImpulse(cpConstraint *constraint)

The most recent impulse that constraint applied. To convert this to a force, divide by the timestep passed to cpSpaceStep(). You can use this to implement breakable joints to check if the force they attempted to apply exceeded a certain threshold.

Hide/Show BreakableJoint Example

To access properties of specific joint types, use the getter and setter functions provided (ex: cpPinJointGetAnchr1()). See the lists of properties for more information.

Error correction by Feedback:

Joints in Chipmunk are not perfect. A pin joint can’t maintain the exact correct distance between it’s anchor points, nor can a pivot joint hold it’s anchor points completely together. Instead, they are designed to deal with this by correcting themselves over time. In Chipmunk 5, you have a fair amount of extra control over how joints correct themselves and can even use this ability to create physical effects that allow you to use joints in unique ways:

There are three properties of cpConstraint structs that control the error correction, maxForce, maxBias, and biasCoef. maxForce is pretty self explanatory, a joint or constraint will not be able to use more than this amount of force in order to function. If it needs more force to be able to hold itself together, it will fall apart. maxBias is the maximum speed at which error correction can be applied. If you change a property on a joint so that the joint will have to correct itself, it normally does so very quickly. By setting a maxSpeed you can make the joint work like a servo, correcting itself at a constant rate over a longer period of time. Lastly, biasCoef is the percentage of error corrected every step before clamping to a maximum speed. You can use this to make joints correct themselves smoothly instead of at a constant speed, but is probably the least useful of the three properties by far.

Hide/Show JointRecipies Example

Constraints and Collision Shapes:

Neither constraints or collision shapes have any knowledge of the other. When connecting joints to a body the anchor points don’t need to be inside of any shapes attached to the body and it often makes sense that they shouldn’t. Also, adding a constraint between two bodies doesn’t prevent their collision shapes from colliding. In fact, this is the primary reason that the collision group property exists.

Video Tour of Current Joint Types. (Requires connection to YouTube)

Shared Memory Management Functions:

void cpConstraintDestroy(cpConstraint *constraint)
void cpConstraintFree(cpConstraint *constraint)

Destroy and Free functions are shared by all joint types. Allocation and initialization functions are specific to each joint type.

Constraint Types:

Pin Joints:

cpPinJoint *cpPinJointAlloc(void)
cpPinJoint *cpPinJointInit(cpPinJoint *joint, cpBody *a, cpBody *b, cpVect anchr1, cpVect anchr2)
cpConstraint *cpPinJointNew(cpBody *a, cpBody *b, cpVect anchr1, cpVect anchr2)

a and b are the two bodies to connect, and anchr1 and anchr2 are the anchor points on those bodies. The distance between the two anchor points is measured when the joint is created. If you want to set a specific distance, use the setter function to override it.

Properties

Slide Joints:

cpSlideJoint *cpSlideJointAlloc(void)

cpSlideJoint *cpSlideJointInit(
	cpSlideJoint *joint, cpBody *a, cpBody *b,
	cpVect anchr1, cpVect anchr2, cpFloat min, cpFloat max
)

cpConstraint *cpSlideJointNew(cpBody *a, cpBody *b, cpVect anchr1, cpVect anchr2, cpFloat min, cpFloat max)

a and b are the two bodies to connect, anchr1 and anchr2 are the anchor points on those bodies, and min and max define the allowed distances of the anchor points.

Properties

Pivot Joints:

cpPivotJoint *cpPivotJointAlloc(void)
cpPivotJoint *cpPivotJointInit(cpPivotJoint *joint, cpBody *a, cpBody *b, cpVect pivot)
cpConstraint *cpPivotJointNew(cpBody *a, cpBody *b, cpVect pivot)
cpConstraint *cpPivotJointNew2(cpBody *a, cpBody *b, cpVect anchr1, cpVect anchr2)

a and b are the two bodies to connect, and pivot is the point in world coordinates of the pivot. Because the pivot location is given in world coordinates, you must have the bodies moved into the correct positions already. Alternatively you can specify the joint based on a pair of anchor points, but make sure you have the bodies in the right place as the joint will fix itself as soon as you start simulating the space.

Properties

Groove Joint:

cpGrooveJoint *cpGrooveJointAlloc(void)

cpGrooveJoint *cpGrooveJointInit(
	cpGrooveJoint *joint, cpBody *a, cpBody *b,
	cpVect groove_a, cpVect groove_b, cpVect anchr2
)

cpConstraint *cpGrooveJointNew(cpBody *a, cpBody *b, cpVect groove_a, cpVect groove_b, cpVect anchr2)

The groove goes from groov_a to groove_b on body a, and the pivot is attached to anchr2 on body b. All coordinates are body local.

Properties

Damped Spring:

cpDampedSpring *cpDampedSpringAlloc(void)

cpDampedSpring *cpDampedSpringInit(
	cpDampedSpring *joint, cpBody *a, cpBody *b, cpVect anchr1, cpVect anchr2,
	cpFloat restLength, cpFloat stiffness, cpFloat damping
)

cpConstraint *cpDampedSpringNew(
	cpBody *a, cpBody *b, cpVect anchr1, cpVect anchr2,
	cpFloat restLength, cpFloat stiffness, cpFloat damping
)

Defined much like a slide joint. restLength is the distance the spring wants to be, stiffness is the spring constant (Young’s modulus), and damping is how soft to make the damping of the spring.

Properties

Damped Rotary Spring:

cpDampedRotarySpring *cpDampedRotarySpringAlloc(void)

cpDampedRotarySpring *cpDampedRotarySpringInit(
	cpDampedRotarySpring *joint, cpBody *a, cpBody *b,
	cpFloat restAngle, cpFloat stiffness, cpFloat damping
)

cpConstraint *cpDampedRotarySpringNew(cpBody *a, cpBody *b, cpFloat restAngle, cpFloat stiffness, cpFloat damping)

Like a damped spring, but works in an angular fashion. restAngle is the relative angle in radians that the bodies want to have, stiffness and damping work basically the same as on a damped spring.

Properties

Rotary Limit Joint:

cpRotaryLimitJoint *cpRotaryLimitJointAlloc(void)
cpRotaryLimitJoint *cpRotaryLimitJointInit(cpRotaryLimitJoint *joint, cpBody *a, cpBody *b, cpFloat min, cpFloat max)
cpConstraint *cpRotaryLimitJointNew(cpBody *a, cpBody *b, cpFloat min, cpFloat max)

Constrains the relative rotations of two bodies. min and max are the angular limits in radians. It is implemented so that it’s possible to for the range to be greater than a full revolution.

Properties

Ratchet Joint:

cpRatchetJoint *cpRatchetJointAlloc(void);
cpRatchetJoint *cpRatchetJointInit(cpRatchetJoint *joint, cpBody *a, cpBody *b, cpFloat phase, cpFloat ratchet);
cpConstraint *cpRatchetJointNew(cpBody *a, cpBody *b, cpFloat phase, cpFloat ratchet);

Works like a socket wrench. ratchet is the distance between “clicks”, phase is the initial offset to use when deciding where the ratchet angles are.

Properties

Gear Joint:

cpGearJoint *cpGearJointAlloc(void);
cpGearJoint *cpGearJointInit(cpGearJoint *joint, cpBody *a, cpBody *b, cpFloat phase, cpFloat ratio);
cpConstraint *cpGearJointNew(cpBody *a, cpBody *b, cpFloat phase, cpFloat ratio);

Keeps the angular velocity ratio of a pair of bodies constant. ratio is always measured in absolute terms. It is currently not possible to set the ratio in relation to a third body’s angular velocity. phase is the initial angular offset of the two bodies.

Properties

Simple Motor:

cpSimpleMotor *cpSimpleMotorAlloc(void);
cpSimpleMotor *cpSimpleMotorInit(cpSimpleMotor *joint, cpBody *a, cpBody *b, cpFloat rate);
cpConstraint *cpSimpleMotorNew(cpBody *a, cpBody *b, cpFloat rate);

Keeps the relative angular velocity of a pair of bodies constant. rate is the desired relative angular velocity. You will usually want to set an force (torque) maximum for motors as otherwise they will be able to apply a nearly infinite torque to keep the bodies moving.

Properties

Notes:

Overview of Collision Detection in Chipmunk:

In order to make collision detection in Chipmunk as fast as possible, the process is broken down into several stages. While I’ve tried to keep it conceptually simple, the implementation can be a bit daunting. Fortunately as a user of the library, you don’t need to understand everything about how it works. Though if you are trying to squeeze every ounce of performance out of Chipmunk, understanding this section can be helpful.

Spatial Indexing:

A for loop that checks every object against every other object in the scene would be very slow. So the first stage of the collision detection (or broadphase as it is commonly called), is to use a high level spatial algorithm to figure out which objects should be even be checked for collisions. Currently Chipmunk supports two spatial indexes, an axis-aligned bounding box tree and a spatial hash. These spatial indexes are able to quickly identify which shapes are near each other and should be checked for a collision.

Collision Filtering:

After the spatial index figures out which pairs of shapes are likely to be near each other, it passes them back to the space to perform some additional filtering on the pairs. Before doing anything else, Chipmunk performs a few quick tests to check if shapes should collide.

Primitive Shape to Shape Collision Detection:

The most expensive test is to actually check for overlap based on their geometry. Circle to circle and circle to line collisions are pretty quick. Poly to poly collisions get more expensive as the number of vertexes increases. Simpler shapes make for faster collisions (and more importantly fewer collision points for the solver to run). Chipmunk uses a small dispatch table to figure out which function to use to check if the shapes overlap.

Collision Handler Filtering:

After checking if two shapes overlap Chipmunk will look to see if you have defined a collision handler for the collision types of the shapes. This is vital to process collisions events for the gameplay, but also gives you a very flexible way to filter out collisions. The return value of the begin() and preSolve() callbacks determines whether or not the colliding pair of shapes is discarded or not. Returning true will keep the pair, false will discard it. Rejecting a collision from a begin() callback is permanent, rejecting it from the preSolve() only applies to the step it occured in. If you don’t define a handler for the given collision_types, Chipmunk will call the space’s default handler, which by default is defined to simply accept all collisions.

While using callbacks to filter collisions is the most flexible way, keep in mind that by the time your callback is called all of the most expensive collision detection has already been done. For simulations with a lot of colliding objects each frame, the time spent finding collisions is small compared to the time spent solving the physics for them so it may not be a big deal. Still, use layers or groups first if you can.

Collision Callbacks:

A physics library without any events or feedback would not be very useful for games. How would you know when the player bumped into an enemy so that you could take some health points away? How would you know how hard the car hit something so you don’t play a loud crash noise when a pebble hits it? What if you need to decide if a collision should be ignored based on specific conditions, like implementing one way platforms? Chipmunk has a number of powerful callback systems that you can plug into to accomplish all of that.

Collision Handlers:

A collision handler is a set of 4 function callbacks for the different collision events that Chipmunk recognizes. The event types are:

Collision callbacks are closely associated with cpArbiter structs. You should familiarize yourself with those as well.

Note: Shapes tagged as sensors (cpShape.sensor == true) never generate collisions that get processed so collisions between sensors shapes and other shapes will never call the postSolve() callback. They still generate begin(), and separate() callbacks, and the preSolve() callback is also called every frame even though there is no real collision.

Note #2: preSolve() callbacks are called before the sleeping algorithm runs. If an object falls asleep, it’s postSolve() callback won’t be called until it’s reawoken.

Collision Handler API:

typedef int (*cpCollisionBeginFunc)(cpArbiter *arb, struct cpSpace *space, void *data)
typedef int (*cpCollisionPreSolveFunc)(cpArbiter *arb, cpSpace *space, void *data)
typedef void (*cpCollisionPostSolveFunc)(cpArbiter *arb, cpSpace *space, void *data)
typedef void (*cpCollisionSeparateFunc)(cpArbiter *arb, cpSpace *space, void *data)

Collision handler function types. While all of them take an arbiter, space, and a user data pointer, only the begin() and preSolve() callbacks return a value. See above for more information.

void cpSpaceAddCollisionHandler(
	cpSpace *space,
	cpCollisionType a, cpCollisionType b,
	cpCollisionBeginFunc begin,
	cpCollisionPreSolveFunc preSolve,
	cpCollisionPostSolveFunc postSolve,
	cpCollisionSeparateFunc separate,
	void *data
)

Add a collision handler for given collision type pair. Whenever a shapes with collision type (cpShape.collision_type) a and collision type b collide, these callbacks will be used to process the collision. data is a user definable context pointer that is passed to each of the callbacks. NULL can be provided for callbacks you do not wish to implement, however Chipmunk will call it’s own default versions for these and not the default ones you’ve set up for the space. If you need to fall back on the space’s default callbacks, you’ll have to provide them individually to each handler definition.

void cpSpaceRemoveCollisionHandler(cpSpace *space, cpCollisionType a, cpCollisionType b)

Remove a collision handler for a given collision type pair.

void cpSpaceSetDefaultCollisionHandler(
	cpSpace *space,
	cpCollisionBeginFunc begin,
	cpCollisionPreSolveFunc preSolve,
	cpCollisionPostSolveFunc postSolve,
	cpCollisionSeparateFunc separate,
	void *data
)

Register a default collision handler to be used when no specific collision handler is found. The space is given a default handler when created that returns true for all collisions in begin() and preSolve() and does nothing in the postSolve() and separate() callbacks.

Post-Step Callbacks:

Post-step callbacks are the one place where you can break the rules about adding or removing objects from within a callback. In fact, their primary function is to help you safely remove objects from the space that you wanted to disable or destroy in a collision or query callback.

Post step callbacks are registered as a function and a pointer that is used as a key. You can only register one postStep() callback per key. This prevents you from accidentally removing an object more than once. For instance, say that you get a collision callback between a bullet and object A. You want to destroy both the bullet and object A, so you register a postStep() callback to safely remove them from your game. Then you get a second collision callback between the bullet and object B. You register a postStep() callback to remove object B, and a second postStep() callback to remove the bullet. Because you can only register one callback per key, the postStep() callback for the bullet will only be called once and you can’t accidentally try to remove it twice.

typedef void (*cpPostStepFunc)(cpSpace *space, void *obj, void *data)

Function type used for postStep() callbacks. space is the space the callback was registered on, obj is the pointer value you supplied as the key, and data is a user definable pointer you can use to pass in as a context value.

void cpSpaceAddPostStepCallback(cpSpace *space, cpPostStepFunc func, void *obj, void *data)

Add func to be called before cpSpaceStep() returns. obj and data will be passed to your function. Only the last callback registered for any unique value of obj will be recorded. You can add postStep() callbacks from outside of other callback functions, but there isn’t a good reason to and they won’t be called until the next time cpSpaceStep() is finishing.

Note: Post-step callbacks are not run in any particular order. If you need to sequence a number of events, you’ll need to put them in a single callback.

Examples:

See the callback examples for more information.

Chipmunk Collision Pairs: cpArbiter

Chipmunk’s cpArbiter struct encapsulates a pair of colliding shapes and all of the data about their collision.

Why are they called arbiters? The short answer is that Box2D called them that way back in 2006 when I was looking at the source for it’s solver. An arbiter is like a judge, a person that has authority to settle disputes between two people. It was a fun, fitting name and was shorter to type than CollisionPair which I had been using. It was originally meant to be a private internal structure only, but evolved to be useful from callbacks.

Memory Management:

You will never need to create or free an arbiter. More importantly, because they are entirely managed by the space you should never store a reference to an arbiter as you don’t know when they will be recycled. Use them within the callback where they are given to you and then forget about them or copy out the information you need.

Properties:

cpFloat cpArbiterGetElasticity(const cpArbiter *arb)
void cpArbiterSetElasticity(cpArbiter *arb, cpFloat value)

The calculated elasticity for this collision pair. Setting the value in a preSolve() callback will override the value calculated by the space.

cpFloat cpArbiterGetFriction(const cpArbiter *arb)
void cpArbiterSetFriction(cpArbiter *arb, cpFloat value)

The calculated friction for this collision pair. Setting the value in a preSolve() callback will override the value calculated by the space.

cpVect cpArbiterGetSurfaceVelocity(const cpArbiter *arb)
void cpArbiterSetSurfaceVelocity(cpArbiter *arb, cpVect value)

The calculated surface velocity for this collision pair. Setting the value in a preSolve() callback will override the value calculated by the space.

int cpArbiterGetCount(const cpArbiter *arb)
cpVect cpArbiterGetNormal(const cpArbiter *arb, int i)
cpVect cpArbiterGetPoint(const cpArbiter *arb, int i)
cpFloat cpArbiterGetDepth(const cpArbiter *arb, int i)

Get the number of contacts tracked by this arbiter or the specific collision point, collision normal or penetration depth of a collision point.

cpBool cpArbiterIsFirstContact(const cpArbiter *arb)

Returns true if this is the first step the two shapes started touching. Can be used

void cpArbiterGetShapes(const cpArbiter *arb, cpShape **a, cpShape **b)
void cpArbiterGetBodies(const cpArbiter *arb, cpBody **a, cpBody **b)

Get the shapes or bodies in the order that they were defined in the collision handler associated with this arbiter. If you defined the handler as cpSpaceAddCollisionHandler(space, 1, 2, ...), you you will find that a->collision_type == 1 and b->collision_type == 2.

Hide/Show CollisionCallback Example

Contact Point Sets:

Contact point sets make getting contact information simpler.

cpContactPointSet cpArbiterGetContactPointSet(const cpArbiter *arb)

Get a contact point set struct from an arbiter.

You might do something like the following to get and process a contact point set:

cpContactPointSet set = cpArbiterGetContactPointSet(arbiter);
for(int i=0; i<set.count; i++){
	// get and work with the collision point normal and penetration distance:
	set.points[i].point
	set.points[i].normal
	set.points[i].dist
}

Helper Functions:

void cpArbiterGetShapes(cpArbiter *arb, cpShape **a, cpShape **b)
void cpArbiterGetBodies(const cpArbiter *arb, cpBody **a, cpBody **b)

Get the shapes (or their bodies) in the order that they were defined in the collision handler associated with this arbiter. If you defined the handler as cpSpaceAddCollisionHandler(space, 1, 2, ...), you you will find that a->collision_type == 1 and b->collision_type == 2. The convenience macro defines and initializes the two shape variables for you. The default collision handler doesn’t use collision types so the order is undefined.

#define CP_ARBITER_GET_SHAPES(arb, a, b) cpShape *a, *b; cpArbiterGetShapes(arb, &a, &b)
#define CP_ARBITER_GET_BODIES(arb, a, b) cpBody *a, *b; cpArbiterGetBodies(arb, &a, &b);

Shortcut macros for defining variables for and retrieving the shapes/bodies for an arbiter.

cpVect cpArbiterTotalImpulseWithFriction(cpArbiter *arb);
cpVect cpArbiterTotalImpulse(cpArbiter *arb);

Returns the impulse that was applied this step to resolve the collision. These functions should only be called from a postStep() or cpBodyEachArbiter() callback, otherwise the result is undefined. If in doubt which function to use, use cpArbiterTotalImpulseWithFriction().

cpFloat cpArbiterTotalKE(const cpArbiter *arb);

Calculate the amount of energy lost in a collision including static, but not dynamic friction. This function should only be called from a postSolve(), postStep() or cpBodyEachArbiter() callback.

Queries:

Chipmunk spaces support four kinds of spatial queries, nearest point, segment, shape and fast bounding box queries. Any type can be performed efficiently against an entire space, and point and segment queries can be performed against individual shapes. All types of queries take a collision group and layer that are used to filter matches out using the same rules used for filtering collisions between shapes. See cpShape for more information. If you don’t want to filter out any matches, use CP_ALL_LAYERS for the layers and CP_NO_GROUP as the group.

Nearest Point Queries:

Point queries are useful for things like mouse picking and simple sensors. They allow you to check if there are shapes within a certain distance of a point, find the closest point on a shape to a given point or find the closest shape to a point.

typedef struct cpNearestPointQueryInfo {
	/// The nearest shape, NULL if no shape was within range.
	cpShape *shape;
	/// The closest point on the shape's surface. (in world space coordinates)
	cpVect p;
	/// The distance to the point. The distance is negative if the point is inside the shape.
	cpFloat d;
} cpNearestPointQueryInfo;

Nearest point queries return the point on the surface of the shape as well as the distance from the query point to the surface point.

cpFloat cpShapeNearestPointQuery(cpShape *shape, cpVect p, cpNearestPointQueryInfo *out)

Find the distance from point to shape. If the point is inside of the shape, the distance will be negative and equal to the depth of the point.

typedef void (*cpSpaceNearestPointQueryFunc)(cpShape *shape, cpFloat distance, cpVect point, void *data);

void cpSpaceNearestPointQuery(
	cpSpace *space, cpVect point, cpFloat maxDistance,
	cpLayers layers, cpGroup group,
	cpSpaceNearestPointQueryFunc func, void *data
)

Query space at point for shapes within the given distance range filtering out matches with the given layers and group. func is called for each shape found along with the distance to the closest point on the shape’s surface, the distance to that point and the data argument passed to cpSpaceNearestPointQuery(). Sensor shapes are included. If a maxDistance of 0.0 is used, the point must lie inside a shape. Negative maxDistance is also allowed meaning that the point must be a under a certain depth within a shape to be considered a match.

cpShape *cpSpaceNearestPointQueryNearest(cpSpace *space, cpVect point, cpFloat maxDistance, cpLayers layers, cpGroup group, cpNearestPointQueryInfo *out)

Query space at point and return the closest shape within maxDistance units of distance. out is an optional pointer to a cpNearestPointQueryInfo if you want additional information about the match.

Segment Queries:

Segment queries are like ray casting, but because not all spatial indexes allow processing infinitely long ray queries it is limited to segments. In practice this is still very fast and you don’t need to worry too much about the performance as long as you aren’t using extremely long segments for your queries.

typedef struct cpSegmentQueryInfo {
	/// The shape that was hit, NULL if no collision occured.
	cpShape *shape;
	/// The normalized distance along the query segment in the range [0, 1].
	cpFloat t;
	/// The normal of the surface hit.
	cpVect n;
} cpSegmentQueryInfo;

Segment queries return more information than just a simple yes or no, they also return where a shape was hit and it’s surface normal at the hit point. t is the percentage between the query start and end points. If you need the hit point in world space or the absolute distance from start, see the segment query helper functions farther down.

cpBool cpShapeSegmentQuery(cpShape *shape, cpVect a, cpVect b, cpSegmentQueryInfo *info)

Perform a segment query from a to b against a single shape shape. info must be a valid pointer to a cpSegmentQueryInfo structure which will be initialized with the raycast info.

typedef void (*cpSpaceSegmentQueryFunc)(cpShape *shape, cpFloat t, cpVect n, void *data)

void cpSpaceSegmentQuery(
	cpSpace *space, cpVect start, cpVect end,
	cpLayers layers, cpGroup group,
	cpSpaceSegmentQueryFunc func, void *data
)

Query space along the line segment from start to end filtering out matches with the given layers and group. func is called with the normalized distance along the line and surface normal for each shape found along with the data argument passed to cpSpacePointQuery(). Sensor shapes are included.

cpShape *cpSpaceSegmentQueryFirst(
	cpSpace *space, cpVect start, cpVect end,
	cpLayers layers, cpGroup group,
	cpSegmentQueryInfo *info
)

Query space along the line segment from start to end filtering out matches with the given layers and group. Only the first shape encountered is returned and the search is short circuited. Returns NULL if no shape was found. The info struct pointed to by info will be initialized with the raycast info unless info is NULL. Sensor shapes are ignored.

Segment Query Helper Functions:

cpVect cpSegmentQueryHitPoint(cpVect start, cpVect end, cpSegmentQueryInfo info)

Return the hit point in world coordinates where the segment first intersected with the shape.

cpFloat cpSegmentQueryHitDist(cpVect start, cpVect end, cpSegmentQueryInfo info)

Return the absolute distance where the segment first hit the shape.

AABB Queries:

AABB queries give you a fast way to check roughly which shapes are in an area.

typedef void (*cpSpaceBBQueryFunc)(cpShape *shape, void *data)

void cpSpaceBBQuery(
	cpSpace *space, cpBB bb,
	cpLayers layers, cpGroup group,
	cpSpaceBBQueryFunc func, void *data
)

Query space to find all shapes near bb filtering out matches with the given layers and group. func is called for each shape whose bounding box overlaps bb along with the data argument passed to cpSpaceBBQuery(). Sensor shapes are included.

Shape Queries:

Shape queries allow you to check if shapes in a space are overlapping a specific area. You can use this to check if an object already exists at a location you want to add another shape, or to use as sensor queries for AI.

You can either create a body/shape pair before querying, or you can create a shape passing NULL for the body and position the shape using cpShapeUpdate() to set the position and rotation of the shape.

typedef void (*cpSpaceShapeQueryFunc)(cpShape *shape, cpContactPointSet *points, void *data);

cpBool cpSpaceShapeQuery(cpSpace *space, cpShape *shape, cpSpaceShapeQueryFunc func, void *data);

Query space to find all shapes overlapping shape. Matches are filtered using the layers and group of shape. func is called for each overlapping shape along with a pointer to a temporary cpContactPointSet and the data argument passed to cpSpaceBBQuery(). Sensor shapes are included.

Blocks:

If your compiler supports blocks (such as Clang), there are an alternate set of functions you can call. cpSpaceNearestPointQueryBlock(), etc. See chipmunk.h for more information.

Examples:

See the query examples for more information.