impl_ptr is a smart pointer for classes using the PImpl idiom or similar use cases.

It owns the pointed-to data and supports copy and move operations.

This is a header only library that provides a impl_ptr<T> class template, useful as the pointer inside a PImpl class or similar classes that contain pointers to data they own.

It assumes the data it points to is only a pointer for technical reasons but should behave like a normal member as much as possible. So when the impl_ptr is created it allocates the pointed to object, when it is destroyed it destroys the pointed to object. When it is copied, moved or copy/move assigned it does that action for the pointed-to object.

It does not share data between instances (unlike std::shared_ptr<>) and allows copying (unlike std::unique_ptr<>).

Assignment and destruction are noexcept if the pointed-to type permits this.

impl_ptr can be used with incomplete types, as is often desired in Pimpl classes.

The basic usage looks like this:

#include "impl_ptr.hpp"

struct details {
   std::string somedata;

   explicit details(const std::string &sd) : somedata(sd) {
   }
};

struct example {
   impl_ptr<details> ptr;

   // The impl_ptr passes constructor arguments to the pointed-to class
   example() : ptr("abc") {
   }
};

int main() {
   example e1;
   assert(e1.ptr->somedata == "abc");

   // the pointed to data can be accessed with the usual operators -> and *
   e1.ptr->somedata = "xyz";

   // move/copy default implementations work with impl_ptr<> members
   example e2(e1);
   assert(e1.ptr->somedata == "xyz");
   assert(e2.ptr->somedata == "xyz");
   e2.ptr->somedata = "abc";
   assert(e1.ptr->somedata == "xyz");
   assert(e2.ptr->somedata == "abc");
}

The impl_ptr<> can be declared pointing to an incomplete type, but the complete type is required when the pointed-to data is accessed. This includes construction, destruction and assignment, even in their default implementation. Implementing those methods directly in the class definition will lead to errors with an incomplete type, the need to be defined where the complete type is available.

In the header file there would be a forward declaration of the pointed to data type and declarations of the constructors/... without implementations:

// example.hpp
#include "impl_ptr.hpp"

struct detail;

class example {
   impl_ptr<detail> ptr;

public:
   example();
   example(const example &);
   example(example &&);
   ~example();

   example& operator=(const example &);
   example& operator=(example &&);
};

The associated .cpp file then contains the method definitions. In this case they can all be declared as = default since detail and example don't contain any special members:

// example.cpp
#include "example.hpp"

struct detail {
   std::string somedata;
};

example::example() = default;
example::example(const example &) = default;
example::example(example &&) = default;
example::~example() = default;
example& example::operator=(const example &) = default;
example& example::operator=(example &&) = default;

The methods could of course also be implemented by hand if necessary or declared as = delete in the header, if they are not desired.