DEV Community

Discussion on: What's unique pointer in C++

Collapse
 
rsa profile image
Ranieri Althoff

The destructor is overly complex :P delete nullptr is a no-op by standard and, after the destruction, the value can't be accessed anymore, so it could just be:

~unique_ptr() {
  delete m_ptr;
}
Enter fullscreen mode Exit fullscreen mode
Collapse
 
gapry profile image
Gapry • Edited

Hi, Ranieri Althoff

I agree with you the destructor is overly complex since the expression m_ptr = nullptr; is redundant.

Assume we design the destructor as you describe, someone call the ~unique_ptr() twice or above. The destructor of m_ptr will be called not only once, hence we need to check whether m_ptr is nullptr.

Finally, the design of the destructor should be as following description.

~unique_ptr() {
  if (m_ptr) {
    delete m_ptr;
  }   
}

Thank you for your rely.

Best regards, Gapry.

Collapse
 
dwd profile image
Dave Cridland

If someone is calling your destructor twice, you have bigger problems - and m_ptr won't be set to nullptr if it's called twice.

Also, you want to delete the implicit copy constructor (otherwise you'll have multiple owners), and similarly with the assignment operator - and you need a move constructor and a move-assignment operator to make the whole thing more useful (and conformant).

Thread Thread
 
gapry profile image
Gapry • Edited

Hi, Dave Cridland

Thank you for your suggestion and insights. I think the similar design may be like this.

class noncopyable
{
public:
  noncopyable() = default;
  ~noncopyable() = default;

  noncopyable(const noncopyable& rhs) = delete;
  noncopyable& operator=(const noncopyable& rhs) = delete;
};

template <class T>
class unique_ptr : public noncopyable  {
  T* m_ptr = nullptr;

public:
  unique_ptr (unique_ptr&& rhs)
    : m_ptr(std::move(rhs.m_ptr)) 
  {} 

  unique_ptr& operator=(unique_ptr&& rhs) {
    std::move(rhs.m_ptr);
    return *this;
  }
};

Thank you for your rely.

Best regards, Gapry.

Thread Thread
 
rsa profile image
Ranieri Althoff

How is it possible to call a destructor twice if that's an unique pointer? Double destructor call wouldn't impliy that you are sharing the pointer?

Thread Thread
 
gapry profile image
Gapry • Edited

Hi, Ranieri Althoff

Assume we're in the multi-threading and the environment is Linux. The m_ptr points to the shared-memory which normally is the heap in Linux.

auto task(void) -> void {
  auto pt = beta::unique_ptr<point<float>>(new point<float>(1.2f, 2.3f, 3.4f, 4.5f));
  // skip the left detail implementation.
}

auto worker1 = std::thread(task);
auto worker2 = std::thread(task);
auto worker3 = std::thread(task);

For now, we have three threads need to be executed and their have the same task. If we don't check the m_ptr, we call the destructor not only once.

I find a picture that you may be interested so I want to shared it with you. The picture sources from here.

Thank you for your rely.

Best regards, Gapry.

Thread Thread
 
dwd profile image
Dave Cridland

No, because the point instances are unique in each thread.

If you put the following into the task lambda, you'll see:

  // ...
  std::cout << &pt << ' ' << &*pt << std::endl;
  // ...

That will show different pointer values for all three unique pointers, and the objects they point to.

You literally cannot (implicitly) call a destructor more than once without some really esoteric and low-level behaviour (such as using the clone() system call on Linux to fork into a thread).

Thread Thread
 
gapry profile image
Gapry • Edited

Hi, Dave Cridland

Agree!! You're right !!!! I'm wrong, sorry.

If I change the code as following, what do you think?

auto task(point* pt_) -> void {
  auto pt = beta::unique_ptr<point<float>>(pt_);
  // skip the left detail implementation.
}
auto pt = new point<float>(1.2f, 2.3f, 3.4f, 4.5f);
auto worker1 = std::thread(task, pt);
auto worker2 = std::thread(task, pt);
auto worker3 = std::thread(task, pt);
// skip the left detail implementation.

Thank you for your rely.

Best regards, Gapry.

Thread Thread
 
dwd profile image
Dave Cridland

Right! If you do that, you will get a double-free, and essentially nothing can protect you (because the m_ptr of each unique_ptr is independent). But also, you don't need to use threads to do it - and it's why best practise with smart pointers is to use std::make_unique:

auto pt = std::make_unique<point<float>>(1.2f, 2.3f, 3.4f, 4.5f);
Thread Thread
 
gapry profile image
Gapry

Hi, Dave Cridland

Thank you for your rely, insights and suggestion!

Best regards, Gapry.