Is it possible to use an rvalue reference as pimpl handle? - c++

I want to create a class with a Pimpl (Private implementation). Normally you would do:
class A
{
private:
class B;
B* _pimpl = nullptr;
}
and then I would define it in the .cpp file. But I have to use dynamic allocation. Would it be possible to use an rvalue reference instead?
class A
{
public:
A(); //Constructor to init rvalue reference
private:
class B;
B&& _pimpl;
}
And then in the .cpp file:
class A::B
{
public:
int C = 3u;
}
//and then the constructor of A:
A::A() : _pimpl(B()) { } //now we should have a fresh b pimpl?
I'm currently on holidays and I only have my C++ book for reference. I read about rvalue references and thought it might work. What do you guys think?

If by "work" you mean "compile", then sure.
But _pimpl(B()) is going to initialize _pimpl as a reference to a temporary. Member references don't extend lifetime, so this construct dangles almost immediately. Therefore no, it won't work.
A unique_ptr<B> is the better type to hold a pimpl (as a default choice). The need for dynamic allocation cannot generally be avoided. But the drawbacks may be mitigated if one chooses a good custom allocator.

Related

Object vs Reference for class members and dependency injection

I'm new to C++, with a C# background. I'm trying to use dependency injection commonly used in C# in C++ and I'm confused about the different ways to declare dependencies and pass them in and why I would use one over the other. Let's assume A depends on B.
Approach #1 - declare as object and take object
class A
{
private:
B _b;
public:
A(B b): _b(b) { }
void Foo()
{
_b.Bar();
}
}
Approach #2 - declare as object and take reference
class A
{
private:
B _b;
public:
A(B &b): _b(b) { }
void Foo()
{
_b.Bar();
}
}
Approach #3 - declare as reference and take object - scratch this - undefined behavior
class A
{
private:
B &_b;
public:
A(B b): _b(b) { }
void Foo()
{
_b.Bar();
}
}
Approach #4 - declare as reference and take reference
class A
{
private:
B &_b;
public:
A(B &b): _b(b) { }
void Foo()
{
_b.Bar();
}
}
I've tried all of the above and they all seem to work. Is there any difference between them? and if so why should I choose one over the other? my preference is #1 because I find it the most familiar and intuitive.
I've found a similar discussion discussion here but that's more about Pointers vs References.
One of the main difference between C# and C++ in this regard is that in C# classes are reference types whereas in C++ they are value types.
Approach 1 and 2 are based on copy construction. This could work if the object passed to the constructor is exactly a B. But it might fails due to object slicing if you'd use a derived class of B. Approach 2 might also fail if the source object is a constant.
Approach 3: no longer considered (scratched in your question)
Approach 4: is ok, provided the object passed by reference continues to exist as long as the object. However if function B::Bar() is not a virtual function, B::Bar() would always be called in Foo(), even if you'd use a class derived from B with its own implementation of Bar().
Approach 5: use a smart pointer and clone the object passed by reference (either with a clone() member, or by using templated constructor):
class A
{
private:
unique_ptr<B> _b;
public:
template<class T>A(const T&b): _b(make_unique<T>(b)) { }
void Foo()
{
_b->Bar();
}
};
Online demo
In your case I'd say that best solution from all mentioned is taking and storing reference. The whole point of dependency injection is using pure virtual interface without even knowing how something is implemented. It means that underneath you want vtable lookup that will execute whatever implementation is for object that is referenced. This way you can expect reference for Animal where callers will provide Elephant or Cat implementation of Animal. Moreover, if you use pure virtual interface (with at least one pure virtual function inside) you can't even pass object by value because you can't create object of pure virtual type.
As far as I remmember, C# similarly to Java distinguishes objects from builtin types and objects are passed by reference by default. In C/C++ you explicitly pass everything the way you want, so in your cases you have:
Copy of temporary copy of B You make more copies than you need. If your object is big, it will cause performance loss.
Copy of B If you simply want to store copy of some object it seems to be better way but you created not needed limitation here. If you make copy of an object, you won't modify original object. By passing it by non-const reference you create this limitation and you can't take const objects because of that. If you simply want to take some object by reference and store copy of it, most of the time you want const reference. Exceptions might be in case of noncopyable objects.
reference to temporary copy of B (invalid)
Reference to B Is completely different story. Here you don't store copy of an object but point to original one, so every change you make on this object will be visible to anyone else that has access to object you received. That's more or less how objects work by default in languages like Java. Moreover, you can use virtualization as I mentioned before. The only drawback is that you have to ensure that this object exists as long as you are using it - you are not the owner of B. You just use that but someone has control of it's existence.
As you can see, in first case Second case creates some not needed limitation. You create copy of non-const

Mutual class references in C++

There's some code at my company that takes the following form:
class ObjectA {
public:
ObjectA(ObjectB &objectB): objectB(objectB) { }
void Initialize() { ... }
private:
ObjectB &objectB;
};
class ObjectB {
public:
ObjectB(ObjectA &objectA): objectA(objectA) { }
void Initialize() { ... }
private:
ObjectA &objectA;
};
Basically, two classes have a mutual dependency on the other.
This in itself isn't what bothers me, (although it's not great design IMO), it's that the mutual dependencies are passed through the constructor by reference in both situations.
So, the only way to actually construct this pair of objects (they are both required in the application) is to pass an incomplete reference as a dependency. To account for this, the separate Initialize methods are added as a 2nd stage constructor which are then called after both objects are created. Consequently, the constructors often times do nothing except assign the constructor parameters to the object internals, and initialize everything in the Intizialize method.
Although the code may be valid, my inclination is that this is a fundamentally flawed design for the following reasons:
The state of a dependent object can't be guaranteed in the
constructor
It requires the developers to make an additional call to Initialize
It precludes the use of compiler warnings that check if member variables have been initialized
There's nothing preventing Initialize from being called multiple times, which can result in strange, hard to track down errors
My coworkers tell me that having a separate initialize method simplifies the object construction since the developer doesn't have to worry about what order objects are declared in the application root. They can be declared and constructed in completely arbitrary order since everything is guaranteed to be valid once Initialize gets called.
I've been advocating that object construction should follow the same order of the dependency graph and that these objects should be redesigned to remove the mutual dependency, but I wanted to see what the SO gurus had to say about it. Am I wrong to think this is bad practice?
Edit:
The way these classes actually get constructed is through a factory class like below:
Factory {
public:
Factory():
objectA(objectB),
objectB(objectA) {
}
private:
ObjectA objectA;
ObjectB objectB;
};
This is bad practice yes. Passing a reference to objectB for objectA to work with while it's not properly initialized yet is a no-no.
It might be standard compliant and safe code now because ObjectA doesn't try to access the passed reference to ObjectB in its constructor but that safety is hanging by a thread. If later on someone decides to just initialize in the constructor or access anything from objectB, or change it any other way where objectB will be used you end up with UB.
This sounds like a use case for pointers that are reseated after both constructors have run.
I too don't like the code as posted - it is unnecessarily complicated and fragile. There is usually some concept of ownership when two classes cooperate like this, so, if objects of type A own objects of type B (which they probably should), then you would do something like:
#include <memory>
class A;
class B
{
public:
B (A& a) : a (a) { }
private:
A& a;
};
class A
{
public:
A () { b = std::make_unique <B> (*this); }
private:
std::unique_ptr <B> b;
};
Live demo

C++: Avoid nullptr initialization of members in constructor

A class like
class A{
public:
const B& instanceOfB ;
A(const B& someInstanceOfB):
instanceOfB(someInstanceOfB)
{
};
// some methods depending on instanceOfB here
};
should be extended to provide also some methods which do not depend on instanceOfB, and these methods must be accessible without any instanceOfB (i.e., no instance of B may exist at all).
Hence, a straight forward extension could look like
class A{
public:
B* const instanceOfB ;
A():
instanceOfB(nullptr)
{
}
// some methods not depending on instanceOfB here
void assignInstanceOfB(B* const someInstanceOfB){
instanceOfB = someInstanceOfB;};
// some methods depending on instanceOfB here
};
However, in my opinion, this has several disadvantages:
The original version of the class was easy to use, while for the extended version it must be ensured that assignInstanceOfB is called prior to any other method depending on instanceOfB
Basically, any method depending on instanceOfB must check if the initialization was perfomed. This produces a huge (i.e., ugly) overhead.
A nullptr initialization is used, which seems to be bad practice in my opinion. I generally try to avoid any pointers.
So i am not really satisfied by this design. What could look a more sophisticated version look like? Thank you!
Edit: Due to the comments, i add following additional requirements:
- Splitting of the class is not possible
- The reference to B must be stored, it is not possible to hand it only to the respective methods.
To summarize, it seems that using pointers (including smart pointers) is the recommended or at least usual way.
To address your first concern about having to call the initialize function before the object can be used, you could implement two constuctors. One would take in a pointer to B and the other defaults to nulling the pointer.
class A{
public:
B* const instanceOfB ;
A(B* const b) : instanceOfB(b) { }
A() : A(nullptr) { }
};
In regards to your third concern, using non-owning pointers is the usual way to store a reference to an object that may or may not exist. If you are hell-bent on not using pointers you could use std::optional in a manner such as this:
class A
{
public:
std::optional<std::reference_wrapper<B>> instanceOfB;
A(B * const b = nullptr) : instanceOfB(std::nullopt)
{
if (b != nullptr)
{
instanceOfB = std::make_optional(std::ref(*b));
}
}
void doSomething()
{
if (!instanceOfB)
{
return;
}
instanceOfB->get().something();
}
};
You need to use std::reference_wrapper since std::optional explicitly does not support storing references directly. You'll still need to check if the optional contains anything before using it, so you still end up needing code that's very similar to checking for a null pointer. There's even worse syntax to access the inner reference, as can be seen in doSomething(). In the end this doesn't seem like it gains you anything, and I wouldn't recommend it.

C++ Constructor member initializer lists, Object Slicing

I have two classes
class A {
public:
virtual void doStuff() = 0;
};
class B : public A {
int x;
public:
virtual void doStuff() override { x = x*2;} //just example function
};
And another class that modify and use data from the previous
class Foo {
A a;
public:
Foo::Foo(A &a_) : a(a_) {}
};
now I create the objects, and passes to the Foo class
B b;
// edit b attributes,
Foo foo(b);
So at the argument list for the class constructor I know there is not the problem of object slicing, because is a reference, but what is the case at the moment of assign the variable a(a_)?
Since I don't know how much time the object b is going to live I need to make a secure copy. I have a lot of different derived classes from A, even derived from the derived.
Will there be a object slicing?,
Is there a solution to this, or I need to pass pointers (don't want this approach)?
This causes slicing. C++ built in polymorphism only works with pointer/reference semantics.
In fact:
class Foo {
A a;
that won't even compile, because A is not a concrete class.
To fix this, first make virtual ~A(){}; and then pass smart pointers to A around. Either unique or shared.
Failing that you can use your own bespoke polymorphism. The easiers way is to stuff a pImpl smart pointer as a private member of a class and implement copy/move semantics in the holding class. The pImpl can have a virtual interface, and the wrapping class just forwards the non-overridable part of the behaviour to it.
This technique can be extended with the small buffer optimization, or even bounded size instances, in order to avoid heap allocation.
All of this is harder than just using the built in C++ object model directly, but it can have payoff.
To see a famous example of this, examine std::function<Sig> which is a value type that behaves polymorphically.
There will be object slicing with what you currently have. You're calling the A copy-constructor in Foo's constructor, and there aren't virtual constructors.
Having a member variable of type A only reserves enough space within an instance of Foo for an instance of A. There is only dynamic binding with pointers and references (which are pointers under the hood), not with member variables.
You would have to use pointers to get around this or you could rethink whether you really need a set-up like this.
Yes, there is slicing.
There has to be slicing, because a B does not fit inside a A, but it is an A that you are storing inside the class Foo. The B part is "sliced off" to fit; hence the name.

Declare default parameter circular reference without pointers?

Is there any way to declare these classes in a header file without indirection?
// Forwards declaration of B
class B;
class A
{
public:
// Default parameter referring to B. May return its parameter
const B& func(const B& b = B());
};
class B
{
public:
// B ctors
B() {}
B(const B&) {}
// B has A as a member
A a;
};
Visual C++ 2008 tells me with this:
error C2514: 'B' : class has no constructors
and points to the forward declaration of B ("class B;") and obviously can't see B's constructors below. A can't follow B because B contains A as a member.
If indirection must be used, what's the best way? Perhaps in C++0x B's A could be a unique_ptr member? Or maybe there's a boost class purely to sidestep this issue?
Instead of a default parameter declare two overloads, one that takes a B by reference and one that takes no parameter. In the one that takes no parameter, call the other with B(), which will work because that method can be defined after B is defined.
...
void func();
void func(const B& b);
};
class B...
void A::func() { func(B()); }
void A::func(const B&) { }
Update:
func() returns a const B&...
That's probably not a good idea. With that definition, something like:
const B& foo = a.func();
foo.bar();
would cause the dreaded "undefined behavior" (i.e., crash) because the B to which you have a reference will be destroyed as soon as the first statement is complete. Returning things other than class members by reference is usually a bad idea.
If you really want to do this, then I think you're stuck with forcing the caller to explicitly pass in B(), that is have no default parameter.
a.func(B()).bar();
(This is the only way to avoid undefined behavior with such a function.)
Of course you could just return a copy instead of a reference, but I presume you have a reason for not doing that.
Depending on what you're doing you may be able to set up better semantics using a smart pointer like shared_ptr instead of references so that you can effectively ignore the lifetimes of the objects. You then have to start being careful of reference cycles instead, however.
I can't tell what you're trying to use this for, but you might want to have a look at some Design Patterns to see if there is an established best-practice for it. You may find that this little problem is a symptom of an unfortunate choice of class structure or containment.