Bartek's coding blog: Smart pointers gotchas

There are at least several questions about using smart pointers in modern C++11

While learning how to use new C++ standard I came across several issues with smart pointers. In general, you can mess a lot less using those helper objects and thus you should use them in your code instead of raw pointers. Unfortunately, there are some topics you have to understand to take full advantage of them. As in most cases, you get a new tool to solve your problems, but on the other hand this tool introduces another issues as well.

Updated on 27th August 2014: Added section about pointer casting. Plus a clean up.

Some predefines

Let us take a simple Test class with one member field to present further concepts:

class Test
{
public:
   Test():m_value(0) { std::cout << "Test::Test" << std::endl; }
   ~Test() { std::cout << "Test::~Test destructor" << std::endl; }

   int m_value;
};

typedef std::auto_ptr<Test>   TestAutoPtr;
typedef std::unique_ptr<Test> TestUniquePtr;
typedef std::shared_ptr<Test> TestSharedPtr;

Why is `auto_ptr` deprecated?

auto_ptr was one of the first type of smart pointers introduced in C++ (in C++98 to be more precise). It was designed to serve as a simple unique pointer (only one owner, without any reference counter), but people tried to use this also in a form of shared pointer. None of those functionalities were satisfied by auto_ptr's implementation! Quick example below:

void doSomething(TestAutoPtr myPtr) {
    myPtr->m_value = 11;
}

void AutoPtrTest() {
    TestAutoPtr myTest(new Test());
    doSomething(myTest);
    myTest->m_value = 10;
}

Try to compile and run this... what happens? It crashes just after we leave doSomething procedure! We would assume than in doSomething some reference counter for our pointer is incremented, but auto_ptr has no such thing.

The object is destroyed because when we leave doSomething procedure our pointer gets out of scope and it is deleted. To make it work we need to pass a reference to this auto pointer.
Another thing is that we have limited way of deleting more complicated objects, there is no control over it at all, only standard delete can be used here.

Why `unique_ptr` finally works good?

Fortunately with the new standard we got brand new set of smart pointers! When we change auto_ptr to std::unique_ptr<Test> in our previous example, we will get compile (not runtime) error saying that we cannot pass pointer to other function. And this is the proper behaviour.

unique_ptr is correctly implemented because of move semantics basically. We can move (but not copy) ownership from pointer to another. We also need to be aware when and where we pass the ownership.
In our example we can use:

doSomethig(std::move(myTest));

to move the pointer's ownership. That way after the function returns out pointer is also not valid, but we did it on purpose after all. Another nice advantage of this type of pointer is that we can use custom deleters. It is useful when we have some complicated resources (files, textures, etc, etc).

How to use arrays with `unique_ptr?`

First thing to know:

std::unique_ptr<int> p(new int[10]);  // will not work!

The above code will compile, but when resources are about to be deleted only single delete will be called. So how do we ensure that delete[] is called? Fortunately unique pointers have a proper partial specialization for arrays and we can write:

std::unique_ptr<int[]> p(new int[10]);  
p[0] = 10;

For our particular example:

std::unique_ptr<Test[]> tests(new Test[3]);

And we will get the desired output:

Test::Test
Test::Test
Test::Test
Test::~Test destructor
Test::~Test destructor
Test::~Test destructor

As expected :)

Note that if you want to pass address of the first element, you have tu use &(pointerToArray[0]). Writing pointerToArray will not work.

Why create `shared_ptr` with `make_shared?`

Unique pointers provides their features only via wise usage of C++ syntax (using private copy constructor, assignment, etc), they do not need any additional memory. But with shared_ptr we need to associate some reference counter with our object. How to do that efficiently?

When we do:

std::shared_ptr<Test> sp(new Test());
std::shared_ptr<Test> sp2 = std::make_shared<Test>();

We will get the output as expected:

Test::Test
Test::Test
Test::~Test destructor
Test::~Test destructor

So what is the difference? Why not use syntax similar to creation of unique_ptr? The answer lies in the allocation process. With the first construct we need to allocate a space for the object and then for the reference counter. With the second construct there is only one allocation (using placement new) and ref counter shares the same memory block as the pointed object.

VS 2012 local's view

Above you can see a picture with local's view in the VS 2012. Compare the addresses of object data and reference counter block. For the sp2 we can see that they are very close to each other.

To be sure I got proper results I've even asked question on stackoverflow: http://stackoverflow.com/questions/14665935/make-shared-evidence-vs-default-construct

BTW: in C++14 there is a nice improvement: make_unique function ! That way creating smart pointers is a bit more 'unified'. We have make_shared and make_unique.

How to use arrays with `shared_ptr?`

Arrays with shared_ptr are a bit trickier that when using unique_ptr, but we can use our own deleter and have full control over them as well:

std::shared_ptr<Test> sp(new Test[2], [](Test *p) { delete []p;});

We need to use custom deleter (here as a lambda expression). Additionally we cannot use make_shared construction.

Unfortunately, using shared pointers for arrays is not so nice. I suggest taking boost instead. For instance: http://www.boost.org/doc/libs/1520/libs/smartptr/sharedarray.htm or boost: make_shared and allocate_shared for arrays

How to pass smart pointers to functions?

We should use smart pointers as a first class objects in C++, so in general we should pass them by value to functions. That way reference counter will increase/decrease correctly. But we can use some other constructions which seems to be a bit misleading. Here is some code:

void testSharedFunc(std::shared_ptr<Test> sp) {
    sp->m_value = 10;
}

void testSharedFuncRef(const std::shared_ptr<Test> &sp) {
    sp->m_value = 10;
}

void SharedPtrParamTest() {
    std::shared_ptr<Test> sp = std::make_shared<Test>();

    testSharedFunc(sp);
    testSharedFuncRef(sp);
}

The above code will work as assumed, but in testSharedFuncRef we get no benefit of using shared pointers at all! Only testSharedFunc will increase reference counter.
For some performance critical code we, additionally, need to notice that passing by value will need to copy whole pointer block, so maybe it is better to use even raw pointer there.

But, perhaps the second option (with reference) is better? The answer: it depends. The main question is if you want to have full ownership of the object or you want to just 'observe' the object. If observe(for instance you have some generic function that calls methods of the object) then we do not need ownership... simple passing by reference is good and fast method.

It is not only me who get confused. Even Herb Sutter paid some attention to this problem and here is his post on that matter: http://herbsutter.com/2012/06/05/gotw-105-smart-pointers-part-3-difficulty-710/

How to cast smart pointers?

Let's take a common example with a simple inheritance:

class BaseA
{
protected:
    int a{ 0 };
public:
    virtual ~BaseA() { }

    void A(int p) { a = p; }
};

class ChildB : public BaseA
{
private:
    int b{ 0 };
public:
    void B(int p) { b = p; }
};

Without a problem, you can create a smart pointer to BaseA and initialize it with ChildB:

std::shared_ptr<BaseA> ptrBase = std::make_shared<ChildB>();
ptrBase->A(10);

But how to get a pointer to a ChildB class from ptrBase? Although, it is not a good practice, sometimes we know it is needed.

You can try this:

ChildB *ptrMan = dynamic_cast<ChildB *>(ptrBase.get());
ptrMan->B(10);

It should work. But, that way you get a 'normal' pointer only! The use_count for the original ptrBase is not incremented. You can now observe the object, but you are not the owner.

It is better to use casting functions designed for smart pointers:

std::shared_ptr<ChildB> ptrChild = std::dynamic_pointer_cast<ChildB>(ptrBase);
if (ptrChild)
{
    ptrChild->B(20);
    std::cout << "use count A: " << ptrBase.use_count() << std::endl;
    std::cout << "use count B: " << ptrChild.use_count() << std::endl;
}

by using std::dynamic_pointer_cast you get a shared pointer. Now you are also the owner. Use count for ptrBase and ptrChild is '2' in this case.

Take a look at std::static_pointer_cast and std::const_pointer_cast for more information.

What about unique_ptr?

In the previous example you got a copy of the original pointer. But unique_ptr cannot have copies... so it is no sense to provide a casting functions. If you need a casted pointer for observation then you need to do it the old way.

Some additional comments

Smart pointers are very useful, but we, as users, also need to be smart :) I am not as experienced with smart pointers as I would like to be. For instance sometimes I am tempted to use raw pointers: I know what will happen, and at a time I can guarantee that it will not mess with the memory. Unfortunately this can be a potential problem in the future. When code changes my assumptions can be not valid any more and new bugs may occur. With smart pointers it is not so easy to break things.

All this topic is a bit complicated, but as usually in C++, we get something at a price. We need to know what we are doing to fully utilize the particular feature.

Code for the article: https://github.com/fenbf/review/blob/master/smart_ptr.cpp

More...

This article is also hosted on: www.codeproject.com

Thanks for comments on reddit/r/cpp