Store derived class objects in base class variables

I would like to store instances of several classes in a vector. Since all classes inherit from the same base class this should be possible.

Imagine this program:

#include <iostream>
#include <vector>
using namespace std;

class Base
{
    public:
    virtual void identify ()
    {
        cout << "BASE" << endl;
    }
};

class Derived: public Base
{
    public:
    virtual void identify ()
    {
        cout << "DERIVED" << endl;
    }
};

int main ()
{
    Derived derived;

    vector<Base> vect;
    vect.push_back(derived);

    vect[0].identify();
    return 0;
}

I expected it to print "DERIVED", because the "identify" method is virtual. Instead 'vect[0]' seems to be a 'Base' instance and it prints

BASE

I guess I could write my own container (probably derived from vector) somehow that is capable of doing this (maybe holding only pointers...). I just wanted to ask if there is a more C++ish method for doing this. AND I would like to be completely vector-compatible (just for convenience if other users should ever use my code).

Answers


What you are seeing is Object Slicing. You are storing object of Derived class in an vector which is supposed to store objects of Base class, this leads to Object slicing and the derived class specific members of the object being stored get sliced off, thus the object stored in the vector just acts as object of Base class.

Solution:

You should store pointer to object of Base class in the vector:

vector<Base*> 

By storing a pointer to Base class there would be no slicing and you can achieve the desired polymorphic behavior as well. Since you ask for a C++ish way of doing this, the right approach is to use a suitable Smart pointer instead of storing a raw pointer in the vector. That will ensure you do not have to manually manage the memory, RAII will do that for you automatically.


You're experiencing slicing. The vector copies the derived object, a new one of type Base is inserted.


TL;DR: You should not inherit from a publicly copyable/movable class.


It is actually possible to prevent object slicing, at compilation time: the base object should not be copyable in this context.

Case 1: an abstract base

If the base is abstract, then it cannot be instantiated and thus you cannot experience slicing.

Case 2: a concrete base

If the base is not abstract, then it can be copied (by default). You have two choices:

  • prevent copy altogether
  • allow copy only for children

Note: in C++11, the move operations cause the same issue.

// C++ 03, prevent copy
class Base {
public:

private:
    Base(Base const&);
    void operator=(Base const&);
};

// C++ 03, allow copy only for children
class Base {
public:

protected:
    Base(Base const& other) { ... }
    Base& operator=(Base const& other) { ...; return *this; }
};

// C++ 11, prevent copy & move
class Base {
public:
    Base(Base&&) = delete;
    Base(Base const&) = delete;
    Base& operator=(Base) = delete;
};

// C++ 11, allow copy & move only for children
class Base {
public:

protected:
    Base(Base&&) = default;
    Base(Base const&) = default;
    Base& operator=(Base) = default;
};

I'd use vector<Base*> to store them. If you say vector<Base>, slicing will occur.

This does mean that you'd have to delete the actual objects yourself after you've removed the pointers from your vector, but otherwise you should be fine.


Need Your Help

Tuples readability : [0,0] vs (0,0)

python tuples readability

I'm using Python since some times and I am discovering the "pythonic" way to code.

Java 8 odd timing/memory issue

java memory jvm java-8 timing

I've run across a rather strange problem that I can create when running Java 8. The problem presents itself as if some sort of timing error is happening within the JVM itself. It is intermittent in