C++ Is a components-based architeture implemented via inheritance considered good practice?

问题

I'm implementing an architecture when i have a container of heterogeneous objects which may have or not some common methods-attribute. I need to cycle through them and apply some functions, update some members, and call some methods over the various interfaces.

I've come up to what i believe a "standard" architecture, based on inheritance:

#include <vector>
#include <memory>
#include <iostream>

using namespace std;

struct Base {
    virtual ~Base() {}
};

struct PositionInterface {
    int x = 0;
    int y = 0;
    virtual ~PositionInterface() {}
};
struct DrawInterface {
    void draw() { cout << "Here i am" << endl; }
    virtual ~DrawInterface() {}
};
struct ChargeInterface {
    int charge = 100;
    virtual ~ChargeInterface() {}
};
struct LifeInterface {
    int life = 100;
    virtual ~LifeInterface() {}
};

struct A: public Base,
          public LifeInterface, public PositionInterface {};
struct B: public Base,
          public DrawInterface, public PositionInterface, public ChargeInterface {};

int main() {
    std::vector<std::shared_ptr<Base>> vec;
    vec.push_back(make_shared<A>());
    vec.push_back(make_shared<B>());

    for (auto & el : vec) {
        auto p = dynamic_cast<PositionInterface *>(el.get());
        if (p) {
            p->x += 10;
            p->y -= 10;
        }
    }
    // ..other stuff
    for (auto & el : vec) {
        auto l = dynamic_cast<LifeInterface *>(el.get());
        if (l) {
            l->life -= 100;
        }
    }
    // ..other stuff
    for (auto & el : vec) {
        auto d = dynamic_cast<DrawInterface *>(el.get());
        if (d) {
            d->draw();
        }
    }
}

Anyway what I'm looking too looks like a component based system. To me seems like that these interfaces could be components added via composition, not inheritance. Something like this:

struct A: public Base {
    LifeInterface l;
    PositionInterface p;
};

But then how could i cycle through the vector of Base objects dynamic_casting to the right interface?

Do you see any drawbacks in this kind of architecture (besides RTTI and public variables :-) ) ?

回答1:

"But then how could i cycle through the vector of Base objects dynamic_casting to the right interface?"

I'd propose to have real abstract interfaces, like this instead:

struct Base {
    virtual ~Base() {}
};

struct PositionInterface {
    virtual int x() const = 0;
    virtual void x(int value) = 0;
    virtual int y() const = 0;
    virtual void y(int value) = 0;
    virtual ~PositionInterface() {}
};

struct DrawInterface {
    virtual void draw() const = 0;
    virtual ~DrawInterface() {}
};

struct ChargeInterface {
    virtual int charge() const = 0;
    virtual ~ChargeInterface() {}
};

struct LifeInterface {
    virtual int life() const {};
    virtual ~LifeInterface() {}
};

---

Base implementation classes that can be used as mixins

class PositionBase : public PositionInterface {
public:
    virtual int x() const { return x_; }
    virtual void x(int value) { x_ = value; }
    virtual int y() const { return y_; }
    virtual void y(int value) { y_ = value; }
    virtual ~PositionBase() {}

protected:
    PositionBase() {}
    int x_;
    int y_;
};

class ChargeBase : public ChargeInterface {
public:
    virtual int charge() const { return charge_; }
    virtual ~ChargeInterface() {}

protected:
    ChargeBase(int charge) : charge_(charge) {}
    const int charge_;
};

class LifeBase : public LifeInterface {
public:
    virtual int life() const { return life_; }
    virtual ~LifeBase() {}

protected:
    LifeBase(int life) : life_(life) {}
    const int life_;
};

And have your implementations like follows

struct A 
: public virtual Base
, public LifeBase
, public PositionBase {
    A() : Base(), LifeBase(100), PositionBase() {}
};

struct B 
: public virtual Base
, public DrawInterface
, public PositionBase
, public ChargeBase {
    B() : Base(), PositionBase(), ChargeBase(100)
    virtual void draw() const {
        // Must implement draw()
    }
};

1. Don't use public member variables, as you can't really control these from inherited classes. Use virtual getter/setter functions as shown above.
2. To check if the right interface is supported, use dynamic_cast<>. To perform operations on these, provide simple templated functions, like e.g.:

---

template<class T> void draw(const T& item) {
    DrawInterface* drawableItem = dyn<mic_cast<DrawInterface*>(&item);
    if(drawableItem) {
        drawableItem->draw();
    }
    // Item isn't drawable, ignore or throw exception
}

回答2:

The main drawback of RTTI is using it, so making it impossible is mostly a good thing.

Use possibly-null pointers, interface-pointer-returning functions, or a boost::optional-like type.

Example:

class Base
{
public:
    virtual LifeInterface* life() { return 0; }
    virtual PositionInterface* position() { return 0; }
};

class A: public Base {
public:
    LifeInterface* life() { return &l; }
private:
    LifeInterface l;
};

// ...

for (auto & el : vec) {
    auto l = el.life();
    if (l) {
        l->life -= 100;
    }
}

来源：https://stackoverflow.com/questions/25084962/c-is-a-components-based-architeture-implemented-via-inheritance-considered-goo