Jump to content

  • Log In with Google      Sign In   
  • Create Account





Fun with Modern C++ and Smart Pointers

Posted by Squared'D, 13 February 2014 · 5,247 views

modern c++ smart pointers unique_ptr shared_ptr modern C++
This has been re-posted to http://articles.squaredprogramming.com/2014/02/smartpointers.html

_________________

This is a journal entry not an article so I feel like I can ramble a bit before I get to the main point. I have been programming for some time now and got my CS degree way back in 2003, but in 2007, I decided to take a break from programming, moved to South Korea, and have been an English teach ever since. I started to program again and with the help of an artist friend of mine, made this for last year's IGF. We obviously didn't win and haven't decided whether or not to continue with the project.

Anyway, because of my long hiatus from programming, quite a few things have passed me by, namely modern C++ and all of this smart pointer stuff. I've decided to start writing code not just for myself, but code that others could use so I want to update my knowledge of C++ and explore smart pointers. In this journal entry I want to share what I've learned. I don't claim to be an expert, so if you have any useful insights or corrections, I'd appreciate a "friendly" comment.


Raw Pointer Primer
There are two basic ways to create a variable in C++: stack variables and heap allocations. Stack variables are defined inside a certain scope and as long as you're in that scope, the variable will exist. Heap allocations (dynamically allocated memory), variables typically declared with new or malloc, aren't not defined within a scope and will exist until the memory is freed.

Here's an example:
void foo()
{
    // Object A of class CSomeClass has been declared inside the scope of foo
    CSomeClass A;

    // do some stuff
    ....

    // you can even call other functions and use A as a parameter
    Func1(A); // This could be pass by value or pass by reference depending on Func1's declaration

    Func2(&A); // Passes a pointer to A

    // at the end of this  function, the scope will end and A will automatically be destroyed
}
Now with this function, every time another function calls foo, A will be created and then destroyed when the function exits. Not bad right. What about this?
void foo()
{
    // Object A of class CSomeClass has been declared inside the scope of foo
    CSomeClass *A = new CSomeClass; 

    // do some stuff
    ....

    // you can even call other functions and use A as a parameter
    Func1(*A); // This could be pass by value or pass by reference depending on Func1's declaration

    Func2(A); // Passes pointer A of CSomeClass (Edited)

    // MEMORY LEAK
    // at the end of this  function, the scope will end, but A was created on the heap
    // delete should be called here
}
So with dynamic memory allocations, you must free the memory. So why you might ask do we even need dynamic memory allocations? Well one, to declare variables on the stack, you need to know exactly what you'll need at compile time. If you want to be able to create arrays of various sizes depending on user input, or if you're making a game and want to load variable amount of resources, you'll need to use dynamic memory allocations. Take this example:
int num_students;

// First get the number of students in the class
std::cout << "How many students are in the class?"
std::cin >> num_students;

// Create a dynamic students array
CStudent *student_array = new CStudent[num_students];

// Do some stuff with the data
....

// call the array version of delete to free memory
delete [] student_array;
In the previous situation, you must use dynamic memory because the size of the array is determined by the user. How can smart pointers help?

Smart Pointers
(chief reference: Smart Pointers (Modern C++) on MSDN)

Smart pointers allow you to create dynamic memory allocations but defined them inside a scope or an owner. This way, when the owner goes out of focus, the data will be automatically deleted. Smart pointers have been implemented using templates and to use them, you must include the header <memory>. Smart pointers are in the std namespace. I will only discuss unique_ptr's here. Later, I may talk about the other types of smart pointers in between updates to "A Complete Graphicsless Game"

So how do you create a unique_ptr?
std::unique_ptr<Base> apples(new Base(L"apples")); // Where Base is the class type
You create a unique_ptr like a typical template and then pass the raw pointer to initialize the variable. After that, you can use the unique_ptr, just as you would any other pointer.
class Base
{
    public:
        Base(const std::wstring &string)
        :m_string(string)
        {
        }

        virtual void Display() const
        {
            std::wcout << L"Base:" << m_string << std::endl;
        }

    private:
        std::wstring	m_string;
};

int main()
{
    // declare some unique_ptrs. This pointers can have only one owner and cannot be copied. Only moved.
    std::unique_ptr<Base> apples(new Base(L"apples"));

    apples->Display();
}
unique_ptr's can also be passed to other functions, but you must pass by reference. Passing by value will result in a compiler error. With unique_ptr's, only one can own the pointer, but if you pass by value you will make a copy of the unique_ptr which in essence makes two unique_ptr's that own the same block of memory.

You can also use unique_ptr's with derived class and virtual functions and get the typical C++ behavior.
class Base
{
    public:
        Base(const std::wstring &string)
        :m_string(string)
        {
        }

        virtual void Display() const
        {
            std::wcout << L"Base:" << m_string << std::endl;
        }

    private:
        std::wstring	m_string;
};

class Derived : public Base
{
    public:
        Derived(const std::wstring &string)
        :Base(string)
        {
        }

        virtual void Display() const
        {
            std::wcout << L"Derived:::";
            __super::Display(); // __super is MS specific. Others should use Base::Display();
        }
};

int main()
{
    // declare some unique_ptrs. This pointers can have only one owner and cannot be copied. Only moved.
    std::unique_ptr<Base> apples(new Base(L"apples"));
    std::unique_ptr<Base> oranges(new Derived(L"oranges"));

    apples->Display();
    oranges->Display();
}
This is very useful when dealing with vectors as the next example show.
std::vector <std::unique_ptr<Base>> test_vector;

test_vector.push_back(std::unique_ptr<Base>(new Base(L"apples")));
test_vector.push_back(std::unique_ptr<Base>(new Derived(L"oranges")));
In the above example, you can use the vector of unique_ptr's in the same way you would if it was a vector of raw Base pointers, but there is one nice benefit. If this vector had raw pointers, you'd have to make sure you manual free the pointers before you clear the vector or erase an element, but with unique_ptr's, all of that is handled for you.

Conclusion
I hope this information was helpful. I am in no way an expert so if you anyone sees something glaring that I missed, feel free to leave a "kind" comment. If you would like to know more about smart pointers, I suggest checking out the like on msdn.




Sorry. I tried editing this on my cell and it messed up all my code blocks. I'll fix the rest when I get home and on a proper computer

 

Edit: Fixed

It should be mentioned that __super is MS specific.

in your second foo(), &A returns a pointer to a pointer to CSomeClass, not a pointer to CSomeClass. should be Func2(A).

in your second foo(), &A returns a pointer to a pointer to CSomeClass, not a pointer to CSomeClass. should be Func2(A).

 

Thanks for mentioning that. I think I had just copied and pasted from the first function and forgot to change it. I'll edit the code and I'll add a comment regarding __super.

Thanks for the write up on pointers. I haven't used smart pointers before and have never had the motivation to read up on them. This was perfect for a crash course on the subject, thanks!

Are Smart Pointers in widespread use in gaming studios? Or do professionals prefer normal pointers?

I have the same question as Tispe. Also I viewed the vid and, to me at least, It seems to have a good setup for a story. I say continue the project.

I don't know if the "official" Smart Pointers from the Standard Library are used often, but it is common practice to let the release of a memory block pointed by a raw pointer be handled by some RAII-based wrapper.

Are Smart Pointers in widespread use in gaming studios? Or do professionals prefer normal pointers?

 

Well smart pointers are not much different from normal pointers, they are basically a scoped wrapper around pointers which helps you use scope-based memory management with pointers. They are used pretty much the same as ordinary pointers and I found them very handy when you need temporary heap memory (too large for stack). For objects with a long lifetime I don't think they are as useful but you should have those objects in some kind of data structure, right?

 

Well, actually, smart pointer refers to a large class of pointer wrappers which basically enforce a destruction policy and hence work more like objects ("pointer in a box") than actual raw pointers (scoped pointers are one instance of smart pointers) and behave mostly the same, except they make your life simpler in most cases. I would use them.

Are Smart Pointers in widespread use in gaming studios? Or do professionals prefer normal pointers?

 

If I were you, I think this question could best be answered on the normal forums. You'd probably get more responses. I'm not currently working at a studio so I can't answer this.

It's also worth pointing out that one of the primary motivators for scoped/smart ptrs is to handle correct resource disposal in the presence of exceptions. The new/delete idiom tends to break very easily when exceptions are in play, and writing exception-safe code with raw pointers is extremely challenging.

Thanks for this article!

I'm confused. How is this entry older than the comments posted on it? :P

I think there was a bug in the system. I just updatedsome information and it got added as new. I don't know why and didn't expect it.

One thing of note is that you likely want to be using std::make_unique() when creating unique pointers. They were the first inclusion in C++14 and supported by most compilers now, but even if you are using an environment where it is not available you can easily add the function. For reference see: http://en.cppreference.com/w/cpp/memory/unique_ptr/make_unique

 

Also note in your vector example it would be much better to use test_vector.emplace_back() instead of push_back().

Recent Comments

October 2014 »

S M T W T F S
   1234
567891011
12131415161718
192021222324 25
262728293031 
PARTNERS