How to implement ICloneable without inviting future object-slicing

孤街浪徒 提交于 2020-08-08 05:31:06

问题


My question is about how to implement the classic ICloneable interface in such a way that it won't lead to inadvertent object-slicing when a future programmer isn't paying close attention. Here's an example of the kind of programming error I'd like to detect (preferably at compile-time):

#include <stdio.h>

class ICloneable
{
public:
   virtual ICloneable * clone() const = 0;
};

class A : public ICloneable
{
public:
   A() {}
   A(const A & rhs) {}

   virtual ICloneable * clone() const {return new A(*this);}
};

class B : public A
{
public:
   B() {}
   B(const B & rhs) {}

   // Problem, B's programmer forget to add a clone() method here!
};

int main(int, char**)
{
   B b;
   ICloneable * clone = b.clone();  // d'oh!  (clone) points to an A, not a B!
   return 0;
}

Is there any way in C++ to convince the compiler to emit an error if B (or any further non-abstract subclasses of B) doesn't define its own clone() method? Short of that, is there any automatic way to detect this error at run-time?


回答1:


It's a while ago that I faced the very same issue in the very same situation without finding a satisfying solution.

Thinking about this again, I found something which might be a solution (at best):

#include <iostream>
#include <typeinfo>
#include <typeindex>

class Base { // abstract
  protected:
    Base() = default;
    Base(const Base&) = default;
    Base& operator=(const Base&) = default;
  public:
    virtual ~Base() = default;
    
    Base* clone() const
    {
      Base *pClone = this->onClone();
      const std::type_info &tiClone = typeid(*pClone);
      const std::type_info &tiThis = typeid(*this);
#if 0 // in production
      assert(std::type_index(tiClone) == type_index(tiThis),
        "Missing overload of onClone()!");
#else // for demo
      if (std::type_index(tiClone) != std::type_index(tiThis)) {
        std::cout << "ERROR: Missing overload of onClone()!\n"
          << "  in " << tiThis.name() << '\n';
      }
#endif // 0
      return pClone;
    }
    
  protected:
    virtual Base* onClone() const = 0;
};

class Instanceable: public Base {
  public:
    Instanceable() = default;
    Instanceable(const Instanceable&) = default;
    Instanceable& operator=(const Instanceable&) = default;
    virtual ~Instanceable() = default;
    
  protected:
    virtual Base* onClone() const { return new Instanceable(*this); }
};

class Derived: public Instanceable {
  public:
    Derived() = default;
    Derived(const Derived&) = default;
    Derived& operator=(const Derived&) = default;
    virtual ~Derived() = default;
    
  protected:
    virtual Base* onClone() const override { return new Derived(*this); }
};

class WrongDerived: public Derived {
  public:
    WrongDerived() = default;
    WrongDerived(const WrongDerived&) = default;
    WrongDerived& operator=(const WrongDerived&) = default;
    virtual ~WrongDerived() = default;

  // override missing
};

class BadDerived: public Derived {
  public:
    BadDerived() = default;
    BadDerived(const BadDerived&) = default;
    BadDerived& operator=(const BadDerived&) = default;
    virtual ~BadDerived() = default;

  // copy/paste error
  protected:
    virtual Base* onClone() const override { return new Derived(*this); }
};

#define DEBUG(...) std::cout << #__VA_ARGS__ << ";\n"; __VA_ARGS__ 

int main()
{
  DEBUG(Instanceable obj1);
  DEBUG(Base *pObj1Clone = obj1.clone());
  DEBUG(std::cout << "-> " << typeid(*pObj1Clone).name() << "\n\n");
  DEBUG(Derived obj2);
  DEBUG(Base *pObj2Clone = obj2.clone());
  DEBUG(std::cout << "-> " << typeid(*pObj2Clone).name() << "\n\n");
  DEBUG(WrongDerived obj3);
  DEBUG(Base *pObj3Clone = obj3.clone());
  DEBUG(std::cout << "-> " << typeid(*pObj3Clone).name() << "\n\n");
  DEBUG(BadDerived obj4);
  DEBUG(Base *pObj4Clone = obj4.clone());
  DEBUG(std::cout << "-> " << typeid(*pObj4Clone).name() << "\n\n");
}

Output:

Instanceable obj1;
Base *pObj1Clone = obj1.clone();
std::cout << "-> " << typeid(*pObj1Clone).name() << '\n';
-> 12Instanceable

Derived obj2;
Base *pObj2Clone = obj2.clone();
std::cout << "-> " << typeid(*pObj2Clone).name() << '\n';
-> 7Derived

WrongDerived obj3;
Base *pObj3Clone = obj3.clone();
ERROR: Missing overload of onClone()!
  in 12WrongDerived
std::cout << "-> " << typeid(*pObj3Clone).name() << '\n';
-> 7Derived

BadDerived obj4;
Base *pObj4Clone = obj4.clone();
ERROR: Missing overload of onClone()!
  in 10BadDerived
std::cout << "-> " << typeid(*pObj4Clone).name() << '\n';
-> 7Derived

Live Demo on coliru

The trick is actually quite easy:

Instead of overriding clone() itself, it is used as trampoline into a virtual onClone() method. Hence, clone() can check the result for correctness before returning it.

This is not a compile-time check but a run-time check (what I consider as second best option). Assuming that every class of a class library in development should hopefully be checked / debugged at least during development I find this quite reliable.


The accepted answer to SO: Reusable member function in C++ showed me a way to make this even more immune against copy/paste errors:

Instead of typing out the class name in every overridden clone(), the class name is obtained via decltype():

class Instanceable: public Base {
  public:
    Instanceable() = default;
    Instanceable(const Instanceable&) = default;
    Instanceable& operator=(const Instanceable&) = default;
    virtual ~Instanceable() = default;
    
  protected:
    virtual Base* onClone() const
    {
      return new std::remove_const_t<std::remove_pointer_t<decltype(this)>>(*this);
    }
};

class Derived: public Instanceable {
  public:
    Derived() = default;
    Derived(const Derived&) = default;
    Derived& operator=(const Derived&) = default;
    virtual ~Derived() = default;
    
  protected:
    virtual Base* onClone() const override
    {
      return new std::remove_const_t<std::remove_pointer_t<decltype(this)>>(*this);
    }
};

Live Demo on coliru




回答2:


Don't make A and B inherit from IClonable. Use a wrapper (BluePrint) instead :

struct IClonable {
    virtual ~IClonable() = default;
    virtual IClonable * clone() const = 0;
};

template<typename T>
class BluePrint final : IClonable {
public:
    explicit BluePrint(T * element) : element(element) {
    }
    IClonable * clone() const override {
        T * copy = element->clone();
        return new BluePrint(copy);
    }
    T * get() const {
        return element;
    }

private:
    T * const element;
};

struct A {
    A * clone() const;
};

struct B : A {
    B * clone() const;
};

You will have to tinker on the code a bit though, since this returns a clone of the wrapper and not immediately a clone of the element to be cloned. Then again, I don't know how you are planning to use the IClonable interface, so I can't complete this example for you.



来源:https://stackoverflow.com/questions/57405375/how-to-implement-icloneable-without-inviting-future-object-slicing

易学教程内所有资源均来自网络或用户发布的内容,如有违反法律规定的内容欢迎反馈
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!