c++ - Is not calling delete on a dynamically allocated object always a memory leak?

Question

From the discussion started here, I'd like to know whether the following code has a memory leak:

int main()
{
   new int();
   //or
   int* x = new int();
   return 0;
}

I know the memory is reclaimed by the OS, but is it a leak anyway? I believe it is.

What defines a memory leak? I could only find one reference in the standard, and it wasn't very helpful.

EDIT: I don't want to start a debate - "I think that..." is not the kind of answer I'm looking for. I'm mostly interested in sources - what C++ books or websites or whatever have to say about it.

Answer 1

It depends on how you define "leak". According to the most obvious definition, and the only useful one, it is not a leak, at least at the application level. A bucket doesn't leak because you intentionally allow a finite quantity of water to escape. And practically speaking, an application doesn't fail because you intentionally allow a bound set of objects to persist beyond the end of the program.

With regards to memory leaks, our perception of the word has been colored by "leak checkers"---programs like Purify or Valgrind. Their role is to find leaks (amongst other things), but they have no way of knowing what is intentional, and what isn't, and what is bound, and what isn't. So they invent other definitions: an object which is unreachable has "leaked" (and there's a good probability in real code that that's true), or an object which hasn't been deleted after all of the destructors of static objects have been executed has "leaked". In this latter case, the definition is obviously wrong, and sort of useless. But there are enough cases where such things are leaks that it is reasonable to at least warn about them ("possible leaks"), provided there is a way of filtering out specific cases. (Both Purify and Valgrind recognize that not all of these cases are really leaks, and provide various filtering mechanisms for their detection.) All of which is well and good—I'm very happy that we have such tools—but we shouldn't allow them to pervert the language.

And one final reminder: the standard says that the standard iostream objects (std::cout, etc.) will never be destructed. So any buffers they allocate will (probably) never be freed. Certainly no one in their right mind would consider these "leaks".