C++ Object Instantiation

Question

On the contrary, you should always prefer stack allocations, to the extent that as a rule of thumb, you should never have new/delete in your user code.

As you say, when the variable is declared on the stack, its destructor is automatically called when it goes out of scope, which is your main tool for tracking resource lifetime and avoiding leaks.

So in general, every time you need to allocate a resource, whether it's memory (by calling new), file handles, sockets or anything else, wrap it in a class where the constructor acquires the resource, and the destructor releases it. Then you can create an object of that type on the stack, and you're guaranteed that your resource gets freed when it goes out of scope. That way you don't have to track your new/delete pairs everywhere to ensure you avoid memory leaks.

The most common name for this idiom is RAII

Also look into smart pointer classes which are used to wrap the resulting pointers on the rare cases when you do have to allocate something with new outside a dedicated RAII object. You instead pass the pointer to a smart pointer, which then tracks its lifetime, for example by reference counting, and calls the destructor when the last reference goes out of scope. The standard library has std::unique_ptr for simple scope-based management, and std::shared_ptr which does reference counting to implement shared ownership.

Many tutorials demonstrate object instantiation using a snippet such as ...

So what you've discovered is that most tutorials suck. ;) Most tutorials teach you lousy C++ practices, including calling new/delete to create variables when it's not necessary, and giving you a hard time tracking lifetime of your allocations.

Though having things on the stack might be an advantage in terms of allocation and automatic freeing, it has some disadvantages.

1. You might not want to allocate huge objects on the Stack.

2. Dynamic dispatch! Consider this code:

#include <iostream>

class A {
public:
  virtual void f();
  virtual ~A() {}
};

class B : public A {
public:
  virtual void f();
};

void A::f() {cout << "A";}
void B::f() {cout << "B";}

int main(void) {
  A *a = new B();
  a->f();
  delete a;
  return 0;
}

This will print "B". Now lets see what happens when using Stack:

int main(void) {
  A a = B();
  a.f();
  return 0;
}

This will print "A", which might not be intuitive to those who are familiar with Java or other object oriented languages. The reason is that you don't have a pointer to an instance of B any longer. Instead, an instance of B is created and copied to a variable of type A.

Some things might happen unintuitively, especially when you are new to C++. In C you have your pointers and that's it. You know how to use them and they do ALWAYS the same. In C++ this is not the case. Just imagine what happens, when you use a in this example as an argument for a method - things get more complicated and it DOES make a huge difference if a is of type A or A* or even A& (call-by-reference). Many combinations are possible and they all behave differently.

Well, the reason to use the pointer would be exactly the same that the reason to use pointers in C allocated with malloc: if you want your object to live longer than your variable!

It is even highly recommended to NOT use the new operator if you can avoid it. Especially if you use exceptions. In general it is much safer to let the compiler free your objects.

I've seen this anti-pattern from people who don't quite get the & address-of operator. If they need to call a function with a pointer, they'll always allocate on the heap so they get a pointer.

void FeedTheDog(Dog* hungryDog);

Dog* badDog = new Dog;
FeedTheDog(badDog);
delete badDog;

Dog goodDog;
FeedTheDog(&goodDog);

Treat heap as a very important real estate and use it very judiciously. The basic thumb rule is to use stack whenever possible and use heap whenever there is no other way. By allocating the objects on stack you can get many benefits such as:

(1). You need not have to worry about stack unwinding in case of exceptions

(2). You need not worry about memory fragmentation caused by the allocating more space than necessary by your heap manager.

The only reason I'd worry about is that Dog is now allocated on the stack, rather than the heap. So if Dog is megabytes in size, you may have a problem,

If you do need to go the new/delete route, be wary of exceptions. And because of this you should use auto_ptr or one of the boost smart pointer types to manage the object lifetime.