I wanted to know if anyone knows of a way to force a class hierarchy to be constructible only by the factory, effectively prohibiting the direct use of std::make_shared outside of that factory.
In the example below I have Node as the base class and SceneNode as one of the many derived classes. Node contains a static member function create() which should be the factory and only way to create new instances of Node-derived classes.
#include <iostream>
#include <memory>
class Node {
public:
template <class T, class... Args>
static std::shared_ptr<T> create(Args&&... args)
{
static_assert(std::is_base_of<Node, T>::value, "T must derive from Node");
std::shared_ptr<T> node = std::make_shared<T>(std::forward<Args>(args)...);
return node;
}
protected:
Node() {}
};
class SceneNode : public Node {
public:
SceneNode() : Node()
{
}
};
int main() {
auto a = Node::create<SceneNode>(); // Should be the only way
auto b = std::make_shared<SceneNode>(); // Should be forbidden
}
The classic way of making your factory the only class able to instanciate a given class is to make your class constructor private, and making your factory friend of your class:
class Foo
{
friend class FooFactory;
private:
Foo() = default;
};
class FooFactory
{
public:
static Foo* CreateFoo() { return new Foo(); }
static void DestroyFoo(Foo* p_toDestroy) { delete p_toDestroy; }
};
int main()
{
// Foo foo; <== Won't compile
Foo* foo = FooFactory::CreateFoo();
FooFactory::DestroyFoo(foo);
return 0;
}
EDIT (With some inheritance):
#include <type_traits>
class Foo
{
friend class FooBaseFactory;
protected:
Foo() = default;
};
class Bar : public Foo
{
friend class FooBaseFactory;
protected:
Bar() = default;
};
class FooBaseFactory
{
public:
template <typename T>
static T* Create()
{
static_assert(std::is_base_of<Foo, T>::value, "T must derive from Foo");
return new T();
}
template <typename T>
static void Destroy(T* p_toDestroy)
{
static_assert(std::is_base_of<Foo, T>::value, "T must derive from Foo");
delete p_toDestroy;
}
};
int main()
{
// Foo foo; <== Won't compile
Foo* foo = FooBaseFactory::Create<Foo>();
FooBaseFactory::Destroy<Foo>(foo);
// Bar bar; <== Won't compile
Bar* bar = FooBaseFactory::Create<Bar>();
FooBaseFactory::Destroy<Bar>(bar);
return 0;
}
One solution to this problem is to create a type that only the factory can instantiate, and have an instance of that class be required to construct the base type. You can establish a convention where the first constructor argument for types that derive from Node is a value of or reference to that type which is fed to Node's constructor. Since it's not possible for anyone else to have a NodeKey users can't instantiate anything that derives from Node without going through the factory.
#include <memory>
#include <utility>
// Class that can only be instantiated by the factory type
class NodeKey {
private:
NodeKey() {};
friend class Factory;
};
class Factory {
public:
template<class T, class ... Args>
auto make(Args&&... args) {
auto ptr = std::make_shared<T>(NodeKey{}, std::forward<Args>(args)...);
// Finish initializing ptr
return ptr;
}
};
class Node {
public:
// Can only be called with an instance of NodeKey
explicit Node(const NodeKey &) {};
};
class Foo : public Node {
public:
// Forwards the instance
explicit Foo(const NodeKey & key) : Node(key) {};
};
int main()
{
Factory factory;
auto f = factory.make<Foo>();
}
Related
class Base
{
public:
virtual void foo() = 0;
};
class A : public Base
{
public:
void foo() override { std::cout << "A\n"; }
};
class B : public Base
{
public:
void foo() override { std::cout << "B\n"; }
};
class Registry
{
public:
static Registry& instance()
{
static Registry s_instance;
return s_instance;
}
void register_foo(Base* foo)
{
m_vec.emplace_back(foo);
}
private:
std::vector<Base*> m_vec;
};
template<typename ... T>
class Foo : public T...
{
public:
Foo()
{
Registry::instance().register_foo(this);
}
void test() { (T::foo(), ...); }
};
int main()
{
auto f1 = std::make_unique<Foo<A, B>>();
auto f2 = std::make_unique<Foo<A>>();
f1->test();
f2->test();
}
As you can see I have a Base class, class A and class B.
A and B inherit from Base.
Class Foo is a template class, which is with a variadic template.
The idea is to be able to pass class A and class B into Foo.
Then this Foo is registered in the Registry class / pushed into a vector.
The problem is the following - as you can see I can have both Foo<A> and Foo<A, B>, or Foo<B, A>.
How can I have such a vector which can accept all possible types of Foo?
How about a simple common base class?
class FooBase {
public:
virtual ~FooBase() {}
virtual void test() = 0;
};
template<typename... T>
class Foo : public FooBase, public T...
{
public:
Foo() { }
void test() override { (T::foo(), ...); }
};
int main()
{
auto f1 = std::make_unique<Foo<A, B>>();
auto f2 = std::make_unique<Foo<A>>();
std::vector<std::unique_ptr<FooBase>> foos;
foos.push_back(std::move(f1));
foos.push_back(std::move(f2));
}
A std::vector holds one type of objects. You cannot put objects of different types into the same vector (and objects created from a template with different template arguments are different types).
One option (I'd not recommend it) is having a vector that holds instances of std::any) - works, but cumbersome and inefficient to work with. Another option is a vector of pointers to a common base class and taking advantage of polymorphism. A third option is simply having sepperate vectors for each type of object.
I have a situation that I am attempting to write some c++ code to implement and I think I need to use templates but I am unsure. I don't think I understand templates well enough and just want to know if I am on the right track. I have never implemented a class that requires templates before.
I have a base class that needs to instantiate a new class and store it on a list for future access. The class that needs to be instantiated is derived from a base class that has one virtual method. After the class is instantiated, this method is called and the new instance is stored on a list. Another method is provided to get the instantiated object.
Say the virtual class is named foo:
class foo
{
public:
virtual void process() = 0;
}
the user would create a class bar:
class bar : public foo
{
public:
void process() {};
}
The class that I think needs to be a template class:
class Filter<Foo* T>
{
public:
// use value junk to derive foo
// and put the new instance on a std::list
add(Junk* junk);
T get();
private:
std::list<T> mFooList;
}
void Filter::add(Junk* junk)
{
T* foo = new T();
foo.p(junk);
mFooList.push_back(foo);
}
T* Filter::get()
{
if(!mFoolist.empty())
{
T* res = mFooList.back();
mFooList.pop_back();
return res;
}
return nullptr;
}
You don't need to use a templatevin this situation. In c++ a derived class can be assigned to a pointer of the base class. You can just use foo pointers.
class foo
{
public:
virtual void process();
};
class bar1: public foo
{
public:
void process();
};
class bar2 : public foo
{
public:
void process();
};
class filter
{
private:
std::list<foo*> _foos;
public:
foo* get();
void add(foo* f);
};
void filter::add(foo* f)
{
_foos.push_back(f);
}
foo* filter::get()
{
if(!_foos.empty())
{
return _foos.pop_back();
}
return nullptr;
}
You can then just add and get foos and bars
filter fil;
fil.add(new foo());
fill.add(new bar1());
fill.add(new bar2());
foo f = fill.get();
while(f != nullptr)
{
f->process();
delete f;
f = fil.get();
}
I have the following classes:
class Box{...};
class MyBox : public Box{...};
And the template:
template <type T>
class ObjectManager{...};
Which I use in some other class:
class Engine{
ObjectManager<Box> * manager = nullptr;
...
};
Then I extend (implement) the Engine interface:
class MyEngine : public Engine{...}
And in that implementation (not earlier!) I know that manager should be like that:
MyEngine::MyEngine(){
manager = new ObjectManager<MyBox>();
}
But this gives me an error because of types conflict (conversion between ObjectManager<Box> and ObjectManager<MyBox>), even when MyBox inherits from Box.
Is there any way around that problem? I don't want to modify the Box, MyBox, ObjectManager and Engine classes.
Templatize Engine and then inherit MyEngine with Engine carrying the template instance of Box that you desire. Something like this: (http://codepad.org/SZMSbCRB)
#include <iostream>
using namespace std;
class Box{
};
class MyBox : public Box{
};
template <typename T>
class ObjectManager{
};
template <typename T>
class Engine{
public:
ObjectManager<T*> * manager;
};
class MyEngine : public Engine<MyBox>{
public:
MyEngine(){
manager = new ObjectManager<MyBox*>();
cout<<"myEngine created"<<endl;
}
};
int main() {
MyEngine eng = MyEngine();
return 0;
}
The benefit here is, if tomorrow you create a new Box namely MyBox2 and want to create a custom engine MyEngine2 for that as well, simply inherit MyEngine : public Engine <MyBox2>. Just make sure to take care of type conversions.
As far as I remember, in Java you can declare Engine as something like Engine<extends T> which lets you instantiate Engine with any subtype of the template T provided. That is a safer and better way to do it, but I do not know if C++ provides something like that.
Hope it helps.
Consider using a wrapper for your ObjectManager.
NOTE marks the lines which will fail to compile if T and BaseT are not compatible.
Assuming:
template<class T>
ObjectManager{
T* objectAtIndex(size_t i); // As an example.
void insertAtIndex(T* object, size_t i); // As an example.
};
Wrapper:
template<class T, class BaseT>
class MyObjectManager
{
public:
ObjectManager<BaseT>* manager; // public for simplicity only!
T* objectAtIndex(size_t i){
return static_cast<T*>(manager->objectAtIndex(i)); // <- NOTE
}
void insertAtIndex(T* object, size_t i){
manager->insertAtIndex(object, i); // <- NOTE
}
};
Usage:
class MyEngine : public Engine
{
MyObjectManager<MyBox, Box> my_manager;
MyEngine(){
// Setup the manager(s).
manager = new ObjectManager<Box>();
my_manager.manager = manager;
// Example usage.
my_manager.insertAtIndex(new MyBox(), 0);
MyBox* p = my_manager.objectAtIndex(0);
}
};
This isn't possible without altering the Design in some way I think.
ObjectManager<Box> just isn't ObjectManager<MyBox> and it is not a base class of it but only the template arguments are base and derived of each other.
If your ObjectManager uses a pointer or smart pointer for the "managed" object...
template <class T>
class ObjectManager
{
T * object;
public:
ObjectManager(T * ptr) : object(ptr) { }
};
... you can construct the ObjectManager using a pointer to a derived object:
class MyEngine : public Engine
{
MyEngine()
{
manager = new ObjectManager<Box>(new MyBox);
}
};
If Engine is an abstract interface you could also have an own manager in MyEngine and use it to implement MyEngine
class MyEngine : public Engine
{
ObjectManager<MyBox> * mymanager;
MyEngine() : Engine(), mymanager(new ObjectManager<MyBox>)
{ }
};
You could provide an implementation to support the conversion. This should be similar to the way a std::unique_ptr can be implicitly converted from a derived pointer type to a base pointer type.
Example Code
#include <iostream>
#include <memory>
#include <vector>
class Box
{
public:
virtual ~Box() {}
virtual void foo() = 0;
};
class MyBox : public Box
{
public:
virtual ~MyBox() {}
virtual void foo() override
{
std::cout << "MyBox::foo()\n";
}
};
template<typename T>
class ObjectManager
{
public:
ObjectManager() {}
void add(T *object)
{
objects.emplace_back(object);
}
template<typename U>
ObjectManager<T> &operator=(ObjectManager<U> &other)
{
return *this;
}
std::size_t size() const
{
return objects.size();
}
T& operator[](std::size_t i)
{
return *objects[i];
}
private:
std::vector<std::unique_ptr<T>> objects;
};
class Engine
{
public:
ObjectManager<Box> manager;
};
class MyEngine : public Engine
{
public:
MyEngine()
{
manager = ObjectManager<MyBox>();
manager.add(new MyBox());
}
};
int main()
{
MyEngine engine;
for (std::size_t i = 0; i < engine.manager.size(); ++i)
{
engine.manager[i].foo();
}
return 0;
}
Example Output
MyBox::foo()
Another Example
This might help illustrate how to implement the conversion.
Example Code
template<typename T>
class MyPointer
{
public:
MyPointer() :
mPointer(nullptr)
{
// Do nothing
}
template<typename U>
MyPointer(MyPointer<U> &other) :
mPointer(other.mPointer)
{
other.mPointer = nullptr;
}
MyPointer(T *pointer) :
mPointer(pointer)
{
// Do nothing
}
template<typename U>
MyPointer<T> &operator=(MyPointer<U> &other)
{
mPointer = other.mPointer;
other.mPointer = nullptr;
return *this;
}
~MyPointer()
{
delete mPointer;
}
T* operator->()
{
return mPointer;
}
private:
template<typename U> friend class MyPointer;
T* mPointer;
};
int main()
{
{
MyPointer<MyBox> myBox(new MyBox());
MyPointer<Box> box;
box = myBox;
box->foo();
}
{
MyPointer<MyBox> myBox(new MyBox());
MyPointer<Box> box(myBox);
box->foo();
}
return 0;
}
Example Output
MyBox::foo()
MyBox::foo()
This question is regarding copying and pointer polymorphism. Consider the code below. We have two classes: Base and Derived, which are just regular objects. Then we've got class Foo, which has a pointer to Base as its only member.
The typical usage of Foo is described in the main function. The input to Foo::SetMemberX may or may not be a temporary object.
The problem is that I want Foo::SetMember to create a proper copy of the passed object, and assign its address as a Base* to Foo::mMember.
I've managed to come up with 4 possible solution, none of which seem very elegant to me. The first three are shown in the code below in Foo::SetMember1, Foo::SetMember2, and Foo::SetMember3. The 4th option is to leave the memory allocation to the user (ex. foo.SetMember(new Derived())), which is not very desirable for obvious memory safety issues. Foo should be responsible for memory management, not the user.
#include <iostream>
template <typename tBase, typename tPointer>
void ClonePointer(tBase*& destination, const tPointer* pointer)
{
destination = static_cast<tBase*>(new tPointer(*pointer));
}
// Base can be a virtual class
class Base
{
public:
virtual void Function()
{
std::cout << "Base::Function()" << std::endl;
}
virtual Base* Clone() const = 0;
};
class Derived : public Base
{
public:
virtual void Function()
{
std::cout << "Derived::Function()" << std::endl;
}
virtual Base* Clone() const
{
return new Derived(*this);
}
};
class Foo
{
public:
Foo() : mMember(NULL) { }
~Foo()
{
if (mMember != NULL)
{
delete mMember;
mMember = NULL;
}
}
template <typename T>
void SetMember1(const T& t)
{
if (mMember != NULL)
{
delete mMember;
mMember = NULL;
}
ClonePointer(mMember, &t);
}
void SetMember2(const Base& b)
{
mMember = b.Clone();
}
template <typename T>
void SetMember3(const T& t)
{
if (mMember != NULL)
{
delete mMember;
mMember = NULL;
}
mMember = new T(t);
}
Base& GetMember()
{
return *mMember;
}
private:
Base* mMember;
};
int main(int argc, char** argv)
{
{
Foo f1;
Foo f2;
Foo f3;
// The input may or may not be a tempoary/RValue reference
f1.SetMember1(Derived());
f2.SetMember2(Derived());
f3.SetMember3(Derived());
f1.GetMember().Function();
f2.GetMember().Function();
f3.GetMember().Function();
}
// Output:
// Derived::Function();
// Derived::Function();
// Derived::Function();
system("pause"); // for quick testing
}
The problem with the first method (Foo::SetMember1) is that I have a random, free, template function, and a template accessore (see the problem with the third method below).
The problem with the second method (Foo::SetMember2) is that every derived class must implement its own Clone function. This is too much boilerplate code for the class user, as there will be a lot of classes deriving from Base. If I could somehow automate this, or create a base Cloneable class (without each Base-derived class having to explicitly calling it) with an implemented template Clone function, this would be the ideal solution.
The problem with the third method (Foo::SetMember3) is that I'd need a template accessor for Foo. This may not always be possible, especially because of how virtual template methods are not allowed in non-template classes (Foo cannot be a template itself), which is a functionality that might be required.
My questions are:
Are these the only options I have?
Is there a better, more elegant solution to this problem that I'm missing?
Is there any way to create a base Cloneable class and derive Base from it, and have cloning automagically happen for DerivedType::Clone()?
Here's a more or less robust Clonable class that can be inherited to any depth. It uses CRTP and Alecsandrescu-style interleaving inheritance pattern.
#include <iostream>
// set up a little named template parameters rig
template <class X> struct Parent{};
template <class X> struct Self{};
template<class A, class B> struct ParentChild;
// can use ...< Parent<X>, Self<Y> >...
template<class A, class B> struct ParentChild< Parent<A>, Self<B> >
{
typedef A parent_type;
typedef B child_type;
};
// or ...< Self<Y>, Parent<X> >
template<class A, class B> struct ParentChild< Self<B>, Parent<A> >
{
typedef A parent_type;
typedef B child_type;
};
// nothing, really
struct Nada
{
// except the virtual dtor! Everything clonable will inherit from here.
virtual ~Nada() {}
};
// The Clonable template. Accepts two parameters:
// the child class (as in CRTP), and the parent class (one to inherit from)
// In any order.
template <class A, class B = Parent<Nada> > class Clonable :
public ParentChild<A,B>::parent_type
{
public:
// a nice name to refer to in the child class, instead of Clonable<A,B>
typedef Clonable Parent;
// this is our child class
typedef typename ParentChild<A,B>::child_type child_type;
// This is the clone() function returning the cloned object
// Non-virtual, because the compiler has trouble with covariant return
// type here. We have to implemens something similar, by having non-virtual
// that returns the covariant type calling virtual that returns the
// base type, and some cast.
child_type* clone()
{
return static_cast<child_type*>(private_clone());
}
// forward some constructor, C++11 style
template<typename... Args> Clonable(Args&&... args):
ParentChild<A,B>::parent_type(args...) {}
private:
// this is the main virtual clone function
// allocates the new child_type object and copies itself
// with the copy constructor
virtual Nada* private_clone()
{
// we *know* we're the child_type object
child_type* me = static_cast<child_type*>(this);
return new child_type(*me);
};
};
// Test drive and usage example
class Foo : public Clonable < Self<Foo> >
{
public:
Foo (int) { std::cout << "Foo::Foo(int)\n"; }
Foo (double, char) { std::cout << "Foo::Foo(double, char)\n"; }
Foo (const Foo&) { std::cout << "Foo::Foo(Foo&)\n"; }
};
class Bar : public Clonable < Self<Bar>, Parent<Foo> >
{
public:
// cannot say Bar (int i) : Foo(i), unfortunately, because Foo is not
// our immediate parent
// have to use the Parent alias
Bar (int i) : Parent(i)
{ std::cout << "Bar::Bar(int)\n"; }
Bar (double a, char b) : Parent(a, b)
{ std::cout << "Bar::Bar(double, char)\n"; }
Bar (const Bar& b) : Parent(b)
{ std::cout << "Bar::Bar(Bar&)\n"; }
~Bar() { std::cout << "Bar::~Bar()\n"; }
};
int main ()
{
Foo* foo1 = new Bar (123);
Foo* foo2 = foo1->clone(); // this is really a Bar
delete foo1;
delete foo2;
}
For 2nd method, you can use CRTP and you won't need to write clone method in every derived class:
struct Base {
virtual ~Base() {}
virtual Base *clone() const = 0;
};
template <typename Derived>
struct CloneableBase : public Base {
virtual Base *clone() const {
return new Derived(static_cast<Derived const&>(*this));
}
};
struct Derived: CloneableBase<Derived> {};
I have simple base and derived class that I want both have shared_from_this().
This simple solution:
class foo : public enable_shared_from_this<foo> {
void foo_do_it()
{
cout<<"foo::do_it\n";
}
public:
virtual function<void()> get_callback()
{
return boost::bind(&foo::foo_do_it,shared_from_this());
}
virtual ~foo() {};
};
class bar1 : public foo , public enable_shared_from_this<bar1> {
using enable_shared_from_this<bar1>::shared_from_this;
void bar1_do_it()
{
cout<<"foo::do_it\n";
}
public:
virtual function<void()> get_callback()
{
return boost::bind(&bar1::bar1_do_it,shared_from_this());
}
};
Causes exception tr1::bad_weak_ptr in following code:
shared_ptr<foo> ptr(shared_ptr<foo>(new bar1));
function<void()> f=ptr->get_callback();
f();
So after "googling" I have found following solution:
class bar2 : public foo {
void bar2_do_it()
{
cout<<"foo::do_it\n";
}
shared_ptr<bar2> shared_from_this()
{
return boost::static_pointer_cast<bar2>(foo::shared_from_this());
}
public:
virtual function<void()> get_callback()
{
return boost::bind(&bar2::bar2_do_it,shared_from_this());
}
};
And now it works.
Is there any better and more convinient and correct way to enable_shared_from_this for both parent and child?
Thanks
The OP solution can be made more convenient by defining the following on the base class.
protected:
template <typename Derived>
std::shared_ptr<Derived> shared_from_base()
{
return std::static_pointer_cast<Derived>(shared_from_this());
}
This can be made more convenient by placing it in a base class (for reuse).
#include <memory>
template <class Base>
class enable_shared_from_base
: public std::enable_shared_from_this<Base>
{
protected:
template <class Derived>
std::shared_ptr<Derived> shared_from_base()
{
return std::static_pointer_cast<Derived>(shared_from_this());
}
};
and then deriving from it as follows.
#include <functional>
#include <iostream>
class foo : public enable_shared_from_base<foo> {
void foo_do_it()
{
std::cout << "foo::do_it\n";
}
public:
virtual std::function<void()> get_callback()
{
return std::bind(&foo::foo_do_it, shared_from_base<foo>());
}
};
class bar1 : public foo {
void bar1_do_it()
{
std::cout << "bar1::do_it\n";
}
public:
virtual std::function<void()> get_callback() override
{
return std::bind(&bar1::bar1_do_it, shared_from_base<bar1>());
}
};
Sorry, but there isn't.
The problem is that shared_ptr<foo> and shared_ptr<bar1> are different types. I don't understand everything that's going on under the hood, but I think that when the constructor returns and is assigned to a shared_ptr<foo>, the internal weak_ptr<bar1> sees that nothing is pointing to it (because only a shared_ptr<bar1> would increment the counter) and resets itself. When you call bar1::shared_from_this in get_callback, you get the exception because the internal weak_ptr isn't pointing to anything.
Essentially, enable_shared_from_this only seems to work transparently from a single class in a hierarchy. If you try implementing it manually, the problem should become obvious.
A similar solution to #evoskuil that reduces boilerplate in derived classes should you want to implement a shared_from_this() function, resulting in the following code at the point of use in the class:
auto shared_from_this() {
return shared_from(this);
}
This uses 'shim' functions outside of the class. By doing it that way it also provides a clean way to do this for classes who's interface can't be modified but derive from enable_shared_from_this - e.g.
auto shared_that = shared_from(that);
Note: Use of auto for return types here will depend upon the age of your compiler.
Shim functions that could be placed in a library header:
template <typename Base>
inline std::shared_ptr<Base>
shared_from_base(std::enable_shared_from_this<Base>* base)
{
return base->shared_from_this();
}
template <typename Base>
inline std::shared_ptr<const Base>
shared_from_base(std::enable_shared_from_this<Base> const* base)
{
return base->shared_from_this();
}
template <typename That>
inline std::shared_ptr<That>
shared_from(That* that)
{
return std::static_pointer_cast<That>(shared_from_base(that));
}
The above code relies on the fact that the type passed to shared_from(...) inherits from std::enable_shared_from_this<Base> at some point in its ancestry.
Calling shared_from_base will figure out what type that ultimately was. Since we know that That inherits from Base, a static downcast can be made.
Probably there are some pathological corner cases with classes having type conversion operators.. but that's unlikely to occur in code not designed to break this.
Example:
struct base : public std::enable_shared_from_this<base> {};
struct derived : public base
{
auto shared_from_this() {
return shared_from(this);
}
// Can also provide a version for const:
auto shared_from_this() const {
return shared_from(this);
}
// Note that it is also possible to use shared_from(...) from
// outside the class, e.g.
// auto sp = shared_from(that);
};
template <typename X>
struct derived_x : public derived
{
auto shared_from_this() {
return shared_from(this);
}
};
Compilation test:
int main()
{
auto pbase = std::make_shared<base>();
auto pderived = std::make_shared<derived>();
auto pderived_x = std::make_shared<derived_x<int> >();
auto const& const_pderived = *pderived;
const_pderived.shared_from_this();
std::shared_ptr<base> test1 = pbase->shared_from_this();
std::shared_ptr<derived> test2 = pderived->shared_from_this();
std::shared_ptr<derived_x<int> > test3 = pderived_x->shared_from_this();
return 0;
}
https://onlinegdb.com/SJWM5CYIG
Prior solution that I posted, kept to make the comments still make sense - this placed the functions in the base class which had some problems - particularly non-uniformity between the required implementation for 'normal' classes and template classes.
Additionally the implementation in the base class would need to be repeated for new class hierarchies which is not all that DRY.
Furthermore the base class function suffered from the possibility of misuse by supplying a base class pointer from a different object. The newer scheme above avoids this entirely and the runtime assert(...) check goes.
Old implementation:
#include <cassert>
#include <memory>
class base : public std::enable_shared_from_this<base>
{
protected:
template <typename T>
std::shared_ptr<T> shared_from(T* derived) {
assert(this == derived);
return std::static_pointer_cast<T>(shared_from_this());
}
};
class derived : public base
{
public:
auto shared_from_this() {
return shared_from(this);
}
};
template <typename X>
class derived_x : public derived
{
public:
auto shared_from_this() {
return this->template shared_from(this);
}
};
int main()
{
auto pbase = std::make_shared<base>();
auto pderived = std::make_shared<derived>();
auto pderived_x = std::make_shared<derived_x<int> >();
std::shared_ptr<base> test1 = pbase->shared_from_this();
std::shared_ptr<derived> test2 = pderived->shared_from_this();
std::shared_ptr<derived_x<int> > test3 = pderived_x->shared_from_this();
return 0;
}
Quite easy; inherit public shared_from_this only in your base class. Implement an accessor in your derived class that casts to the appropriate type;
std::shared_ptr<Derived> shared()
{
return std::dynamic_pointer_cast<Derived>(Base::shared_from_this());
}
With c++23 deducing this, things become much easier. https://godbolt.org/z/j499WK58Y
#include <memory>
#include <iostream>
#include <functional>
using namespace std;
struct new_enable_shared_from_this :
public std::enable_shared_from_this<new_enable_shared_from_this> {
template <typename Self>
auto new_shared_from_this(this Self& self) {
return std::static_pointer_cast<Self>(self.shared_from_this());
}
};
class foo : public new_enable_shared_from_this {
void foo_do_it()
{
cout<<"foo::do_it\n";
}
public:
virtual function<void()> get_callback()
{
return bind(&foo::foo_do_it,new_shared_from_this());
}
virtual ~foo() {};
};
class bar1 : public foo {
void bar1_do_it()
{
cout<<"foo::do_it\n";
}
public:
virtual function<void()> get_callback()
{
return bind(&bar1::bar1_do_it,new_shared_from_this());
}
};
int main() {
auto pf = std::make_shared<foo>();
pf->get_callback()();
auto pb = std::make_shared<bar1>();
pb->get_callback()();
}
This is the previous answer:
Well, I don't like virtual function. Virtual function is just for type erasing, but we don't need type erasing all the time. So, provide a mechanism for type erasing is enough. Here is an example that don't use virtual function:
#include <iostream>
#include <functional>
#include <memory>
using namespace std;
template<typename derived>
class foo_imp {
void foo_do_it()
{
cout<<"foo::do_it\n";
}
public:
function<void()> get_callback()
{
auto&& d = static_cast<derived&>(*this);
return bind(&foo_imp::foo_do_it, d.shared_from_this());
}
};
template<typename derived>
class bar_imp {
void bar_do_it()
{
cout<<"bar::do_it\n";
}
public:
function<void()> get_callback()
{
auto&& d = static_cast<derived&>(*this);
return bind(&bar_imp::bar_do_it, d.shared_from_this());
}
};
struct foo : public foo_imp<foo>, public enable_shared_from_this<foo> {};
struct bar : public bar_imp<bar>, public enable_shared_from_this<bar> {};
struct v_foo {
virtual function<void()> get_callback() = 0;
};
template <typename T>
std::shared_ptr<v_foo> convert(const std::shared_ptr<T>& st) {
struct _ : public v_foo {
_(const std::shared_ptr<T>& st) : _st{st} {}
function<void()> get_callback() override {
return _st->get_callback();
}
std::shared_ptr<T> _st;
};
return std::make_shared<_>(st);
}
int main() {
auto sf = make_shared<bar>();
sf->get_callback()();
auto svf = convert(sf);
svf->get_callback()();
auto sb = make_shared<foo>();
sb->get_callback()();
auto svb = convert(sb);
svb->get_callback()();
}
#include <memory>
template<class T>
class Base : public std::enable_shared_from_this<T> {
};
class Derived : public Base<Derived> {
std::shared_ptr<Derived> getDerived() { return shared_from_this(); }
};