polymorphism and inheriting from a base template class

Started by
6 comments, last by phil_t 10 years, 11 months ago

I am trying to design a component-based entity system and I came up with a solution that looks similar to this:


class base_component
{

};

template <class T>
class base_system
{
public:
protected:
	std::map<int, T> m_components;
};

class spacial_component : public base_component
{
public:
};

class spacial_system : public base_system<spacial_component>
{

};

This way I was hoping the base_system class would create an std::map with the components for every component system.

What I am wondering is if it is valid to design it like this and then use polymorphism like the following:

spacial_system ss;

base_system<base_component>* bs = (base_system<base_component>*) ss;

My hope would be that this is possible since my entity_system class should store a vector of all available systems for dispatching messages to the systems using virtual functions. I have it implemented in VS 2012 and it works there. Then I tried to use the same code on VS 2010 and got random crashes. I suspect the cast to base_system<base_component>* is not actually valid C++. So my question is, is the cast valid? Should I use another solution?

Advertisement
This is equivalent to:

class base_component { int foo; };
class spatial_component : public base_component { int bar, baz; };
std::map<int, spatial_component> spatial_system;
std::map<int, base_component>* base_system = reinterpret_cast<std::map<int, base_component>*>( &spatial_system );
And no, this code is not valid. There's nothing in the specification of std::map that says that you should be able to cast a map of type T1 to a map of type T2 and still have it function.

Consider one reason why this might not work -- imagine inside a container class (like std::map) if you were trying to allocate a storage area for 1000 components. If T=base_component or T=spatial_component, you'd allocate different amounts of memory:

void* storage1 = malloc( sizeof(base_component)*1000 );
void* storage2 = malloc( sizeof(spatial_component)*1000 );
When you cast the container to a different type of T, it no longer has the appropriately sized storage area.
Also, it no longer computes the correct offsets to each of the items inside this storage area:

base_component*    item42_1 = ((char*)storage1) + sizeof(base_component)*42;
spatial_component* item42_2 = ((char*)storage2) + sizeof(spatial_component)*42;
assert( item42_1 != item42_2 );//!!!

I'm wondering what you want to achieve with this hierarchy. Isn't the goal of component based design to get away from inheritance as much as possible?

I'm wondering what you want to achieve with this hierarchy. Isn't the goal of component based design to get away from inheritance as much as possible?

The goal is for every component system contain the components without having to define the std::map<int, component> and accessory methods in every system. That is 100's of lines of cut and paste for every component system (create_component(), component_exists(), etc).

Thank you Hodgman for your help, I will rethink my approach.


template <class T>
class base_system
{
public:
protected:
	std::map<int, T> m_components;
};

Since you probably want to use your components in more then one system, the system itself might not be the best place to store your different components.

I'm wondering what you want to achieve with this hierarchy. Isn't the goal of component based design to get away from inheritance as much as possible?

It is a good way to get rid of a lot of inheritance-chains between your different object types, which may lead to code which wont even be used in some objects, but it doesn't mean you should avoid inheritance everywhere, having a basic interface for your systems and components is, depending on your implementation, probably a good idea.

The goal is for every component system contain the components without having to define the std::map and accessory methods in every system. That is 100's of lines of cut and paste for every component system (create_component(), component_exists(), etc).

,>

Ah okay, seems I haven't had enough coffee yet. biggrin.png

But why would you need the upcast to the base system? Could you derive base system from a type agnostic interface that has create_component etc. return the base component type and then downcast that if you need to?


// Pseudocode

interface BaseComponentSystem {
    BaseComponent* createComponent();
    bool hasComponent(id componentId);
}

class SpecialComponentSystem<T> : public BaseComponentSystem {
    map<int, T> components;
    BaseComponent* createComponent() {
        return components.add(generateID(), new T());
    }
    // ...
}

Your entity_system class could store a Map of systems. The Map would use the component type as the key.

Pseudo code


public class entity_system
{
    private Map<Class<?>, base_system<?> systemMap;

    public void registerSystem(Class<T> componentType, base_system<T> system)
    {
        systemMap.put(componentType, system);
    }

    public <T> base_system<T> getSystem(class<T> componentType)
    {
        return systemMap.get(componentType);
    }
}

I agree with the poster who said that systems probably aren't the best place to store your components. Chances are you'll have systems that need to access multiple component types. Similarly, it doesn't really make sense for a system to have methods like createComponent, componentExists, etc...

Instead these methods could be part of the class that holds onto the various systems, or some parent "entity manager" class that holds onto the entities and the various component containers (and the necessary component containers can be passed to each system in the system's constructor, say).

This topic is closed to new replies.

Advertisement