Multiple dispatch for collision detection in C++?

Question

I have circles, boxes and lines. And now I want to implement collision detection between them. This means that I have to have a function for each combination of two kinds of shapes. Of course I can use the same for line vs circle and circle vs line, but I think my point still stands. What's the most elegant way to implement this in C++?

double dispatch is done with the visitor pattern as shown in this (java) example: http://en.wikipedia.org/wiki/Visitor_pattern#Java_example — stefaanv, Jan 08 '14 at 16:23
I woul make an oveloaded function like `bool isCOlliding(Circle object1, Line object2);` then `bool isCOlliding(Box object1, Circle object2);` etc, that way you can call the same function name on every combination — GMasucci, Jan 08 '14 at 16:23

LavaScornedOven · Answer 1 · 2014-01-08T17:11:56.000

1

In the Modern C++ Design book, Multimethods chapter explains how to implement them and documents how to use [implementation provided by Loki library][2]. There is also Boost.Multimethod proposal, but it's not there yet. The book demonstrates the power of multimethods exactly on the topic of object collision.

edited Jan 08 '14 at 17:11

answered Jan 08 '14 at 16:30

LavaScornedOven

737
1
11
23

score 0 · Answer 2 · answered Jan 08 '14 at 16:30

Polymorphism

first you have Shape and all the other sub classes:

class Shape;

class Circle : Shape;
class Square : Shape;
...etc

and you want them all to implement intersect with each other, so have one general function:

class Shape{
    bool isIntersect(const Shape&)
    {
        return isIntersecting(this, shape);
    }
};

Now overload isIntersecting with all the pair options, since each case is unique you don't have much of a choice here but to implement each case:

bool isIntersecting(Circle, Square);
bool isIntersecting(Circle, Line);
...

You can also try visitor pattern, but I think it's useless here.

I agree with the comment about visitor pattern, but the important difference between function overloading and multiple dispatch is that function overloading works only if types are known at compile-time. That's something to consider, e.g. in case when objects are kept within container of top-most type (`Shape` in this case). — LavaScornedOven, Jan 08 '14 at 16:41

score 0 · Answer 3 · answered Jan 08 '14 at 16:43

You may do collision detection in a scene and not handle each incident of two shapes. Also, you will face some problems. Among them is type elision and registering geometric shapes:

#include <iostream>
#include <map>

class Shape
{
    public:
    struct Type {
        public:
        typedef unsigned Identifier;

        const char* name;
        Identifier id;

        Type(const char* name)
        :   name(name), id(register_name(name))
        {}

        private:
        static Identifier register_name(const char*);
    };


    virtual const Type& get_type() const = 0;

    virtual ~Shape() {}
};

Shape::Type::Identifier Shape::Type::register_name(const char* name) {
    typedef std::map<const char*, Identifier> Types;
    static std::map<const char*, Identifier> types;
    auto& id = types[name];
    if( ! id) id = types.size();
    return id;
}


class Scene
{
    public:
    typedef bool (*CollisionDetector)(const Shape&, const Shape&);
    void register_collision_detector(CollisionDetector, const Shape::Type&, const Shape::Type&);
    bool collision(const Shape& a, const Shape&, bool strict = false) const;

    private:
    typedef std::pair<Shape::Type::Identifier, Shape::Type::Identifier> Identifier;
    typedef std::map<Identifier, CollisionDetector> CollisionDetectors;
    CollisionDetectors m_detectors;
};

void Scene::register_collision_detector(
    CollisionDetector detector,
    const Shape::Type& t0,
    const Shape::Type& t1)
{
    m_detectors[Identifier(t0.id, t1.id)] = detector;
}

bool Scene::collision(const Shape& s0, const Shape& s1, bool strict) const {
    const Shape::Type::Identifier i0 = s0.get_type().id;
    const Shape::Type::Identifier i1 = s1.get_type().id;
    auto pos = m_detectors.find(Identifier(i0, i1));
    if(pos == m_detectors.end() && ! strict && i0 != i1)
        pos = m_detectors.find(Identifier(i1, i0));
    if(pos != m_detectors.end())
        return pos->second(s0, s1);
    else return false;
}


class Circle : public Shape
{
    public:
    static const Type type;
    virtual const Type& get_type() const  { return type; }
};
const Shape::Type Circle::type("Circle");

class Rectangle : public Shape
{
    public:
    static const Type type;
    virtual const Type& get_type() const { return type; }
};
const Shape::Type Rectangle::type("Rectangle");

bool circle_rectangle_collision(const Shape& circle, const Shape& rectangle) {
    std::cout << "circle_rectangle_collision" << std::endl;
    return false;
}


int main()
{
    Scene scene;
    scene.register_collision_detector(circle_rectangle_collision, Circle::type, Rectangle::type);
    Circle circle;
    Rectangle rectangle;
    scene.collision(circle, rectangle);
    scene.collision(rectangle, circle);
}

Multiple dispatch for collision detection in C++?

3 Answers3