New operator

Don't really understand what the new operator does. My books say it's a memory allocation operator which makes perfect sense, but does it simply free up memory to store a placeholder for a pointer? If so, what is the usefulness of reserving memory instead of just initializing a pointer and thus using the necessary amount of memory? What is the essential difference between:

Code:

int x = 100;
int *p = &x;
//do stuff
delete p;

and

Code:

int x = 100;
int *p = new int; //really not too sure about this syntax, do you need to specify the type of the pointer when using new?
//…
*p = &x;
//do stuff
delete p;

If your program needs pointers to hold memory addresses, why don't you just initialize the pointers when they are needed? Maybe I just have no idea what new does.

The new operator allocates objects on the heap (1) which exists as long as you don't call delete, while a normal object is allocated on the stack (2), which is destructed on scope exit.

Code:

int* foo() //(2)
{
    int x = 100;
    return &x; //Undefined behaviour, as x does not exist after the function exits
}

int* foo()//(1)
{
    int * x = new int(100);
    return x;//Okay, but you *must* call delete after you are done with the return value.
}

Also, stack space is very limited on normal computers, while the heap has all (not really, read up on virtual memory if you want details) the free memory in your RAM.

Originally Posted by manasij7479

The new operator allocates objects on the heap (1) which exists as long as you don't call delete, while a normal object is allocated on the stack (2), which is destructed on scope exit.

Code:

int* foo() //(2)
{
    int x = 100;
    return &x; //Undefined behaviour, as x does not exist after the function exits
}

int* foo()//(1)
{
    int * x = new int(100);
    return x;//Okay, but you *must* call delete after you are done with the return value.
}

Also, stack space is very limited on normal computers, while the heap has all (not really, read up on virtual memory if you want details) the free memory in your RAM.

Thanks for the quick reply. Surprisingly, I think I fully understand now (actually I didn't know the difference between heap and stack memory, but I do now). Right now I'm reading through Effective C++, just read the Item called "Don't try to return a reference when you must return an object" yesterday, so I understand how returning the memory address of a local variable doesn't work.

Does new have any use in the global scope though?

Of course it has, it dynamically allocates memory (as much as you need) !

But also know that new should be used sparingly and only when really necessary. For one, pointer semantics are a lot less convenient and more error prone than value semantics (smart pointers alleviate that pain to some degree). Also, in multi-threaded applications there may be certain performance issues attached to the use of new. The free store from which new gets memory has to be protected from concurrent access, which means some sort of locking may be involved. This can potentially hurt performance.

Originally Posted by manasij7479

Of course it has, it dynamically allocates memory (as much as you need) !

This is where I get confused. Dynamically allocates it for what? If you need to hold the memory address of an object, won't you just initialize a pointer or make a reference to it? What does blank allocated memory exactly do? I realize it is not truly blank, because the 2/4/6/whatever bytes of data are already reserved, but there is no data there, right?

semi-EDIT: I understand that there actually is data there, as I misunderstood the fundamental use of new (I thought it reserved memory instead of writing data to it). But instead of deleting the post I'm gonna leave it out there just in case I don't understand (which is not unlikely).

Originally Posted by antred

But also know that new should be used sparingly and only when really necessary. For one, pointer semantics are a lot less convenient and more error prone than value semantics (smart pointers alleviate that pain to some degree). Also, in multi-threaded applications there may be certain performance issues attached to the use of new. The free store from which new gets memory has to be protected from concurrent access, which means some sort of locking may be involved. This can potentially hurt performance.

Yeah, this is not going to become my new favorite C++ thing or anything, I just wasn't sure about it and wanted to learn. About the second part, why would there be locking involved in multi-threaded applications and not in single-threaded ones (if you know)?

Originally Posted by ex-mortis

Yeah, this is not going to become my new favorite C++ thing or anything, I just wasn't sure about it and wanted to learn. About the second part, why would there be locking involved in multi-threaded applications and not in single-threaded ones (if you know)?

I can't really give you a general answer to that as a lot of it depends on the compiler / operating system used, but apparently on Windows, EVERY access to the free store involves a lock (even in single-threaded applications, if I understand correctly ... though I'm by no means sure about that one). Maybe someone more knowledgeable than myself can clear that up.

In the meantime, here's a link to a StackOverflow question related to this topic. multithreading - In multithreaded C/C++, does malloc/new lock the heap when allocating memory - Stack Overflow

Originally Posted by ex-mortis

This is where I get confused. Dynamically allocates it for what? If you need to hold the memory address of an object, won't you just initialize a pointer or make a reference to it? What does blank allocated memory exactly do? I realize it is not truly blank, because the 2/4/6/whatever bytes of data are already reserved, but there is no data there, right?

semi-EDIT: I understand that there actually is data there, as I misunderstood the fundamental use of new (I thought it reserved memory instead of writing data to it). But instead of deleting the post I'm gonna leave it out there just in case I don't understand (which is not unlikely).

The memory allocated by new contains a full constructed object of the type you have specified (here, int and it'll contain 0);

Yeah, this is not going to become my new favorite C++ thing or anything, I just wasn't sure about it and wanted to learn.

yes, learn to use the standard containers.

About the second part, why would there be locking involved in multi-threaded applications and not in single-threaded ones (if you know)?

When you go to a public restroom, do you lock the door?
You can, of course do the same when you are alone at your home, but that doesn't serve any purpose.

Originally Posted by manasij7479

When you go to a public restroom, do you lock the door?
You can, of course do the same when you are alone at your home, but that doesn't serve any purpose.

LOL, that's a great analogy. I'll use that the next time this question comes up.

Originally Posted by antred

LOL, that's a great analogy. I'll use that the next time this question comes up.

Just wait till grumpy says that a public toilet can't protect itself from concurrent access !

(The analogy actually still holds up!)

Originally Posted by manasij7479

When you go to a public restroom, do you lock the door?
You can, of course do the same when you are alone at your home, but that doesn't serve any purpose.

Get it, thanks.

Originally Posted by manasij7479

yes, learn to use the standard containers.

This line confuses me greatly. Really don't know if it's sarcastic or not. If you mean I should prefer the standard types for holding data to pointers, then I already do because I know they can be dangerous and I only use them if necessary. But you said I should learn to use the standard containers. What do you mean by that?

:facepalm:
By standard containers. I mean std::vector, std::string, std::list. std::deque, std::map, etc.
They greatly simplify the task of high level programming, and frees the programmer of having to deal with memory management (new-delete) directly.

Originally Posted by manasij7479

:facepalm:
By standard containers. I mean std::vector, std::string, std::list. std::deque, std::map, etc.
They greatly simplify the task of high level programming, and frees the programmer of having to deal with memory management (new-delete) directly.

Alright I'll read up on them, thanks.

Just to clarify, he meant the Standard Template Library (STL) part of C++, wherein you dont have to "new-delete" stuff and can have dynamic size even during run-time

Part of this thread was moved to a thread specifically about the STL.