I thought I’d jot down some notes on what I do when designing new classes in C++ to be copyable. I’d be curious to get some feedback as I’m not sure what the C++ community at large does…
Anyway, when creating a new class where objects of that class need to be copyable, the first thing I ask myself is:
1. Is the new class a Plain Old Data type?
My definition of a Plain Old Data type (“POD”) is one that simply includes a fixed amount of publicly accessible data. This rules out anything that would include a member function, anything that has any dynamic memory allocation aspects and anything that would have protected/private data members.
If so, designing the class can be quite simple. I don’t need to worry about:
- copy constructors
- assignment operators
The compiler will supply hidden default (no-argument) constructor and a copy constructor for the above class which will work fine (since all data members are fixed length, no dynamic memory allocation takes place). There is also no need for an explicit destructor since all data members are automatically destroyed.
2. If not a POD, will I really need to copy objects of this class?
Sometimes it’s obvious when you will need to do copies, other times it’s not so obvious but it becomes convenient. For instance, if I want to avoid having to allocate and (remember to) de-allocate memory at runtime for an object I sometimes want to make it part of a class. i.e.:
In the above, I want C2 to contain an object of type C1 when C2 is instantiated. I don’t want to have C2 contain a pointer to C1 and then have to allocate a C1 inside C2’s constructor then remember to de-allocate a C1 when C2 is destroyed (lest I cause a memory leak). Any time I can avoid having to keep track of a pointer, I usually try to (though smart pointers can ease this task and static allocation really has become just a preference of mine).
Another scenario is when you have an array of objects:
Now an object of type C3 contains an array of 2085 C1 objects which are allocated when the object of type C3 is instantiated.
One problem with doing the above is that the C1 objects are statically created when the enclosing class is created using a default (no-argument) constructor. Often-times objects of the contained class (C1) cannot be constructed meaningfully at the time of the enclosing class’ (C2) instantiation, especially if it’s during program startup as opposed to at run-time.
Ultimately, this means you cannot specify any details about the individual C1 objects and often need to “initialize” those objects at some later state (a messy approach) or “reconstruct” those objects and copy them back to data members of the enclosing class.
For scenarios like this, it means I will supply the following:
- A copy constructor
- An assignment operator
I usually throw in a default, no-argument constructor and a destructor at this point too. Here’s my standard approach for this, which I’ve been using for a couple years now:
We supply a member function called “deepcopy” that ensures all data members are properly copied from “rhs” to “this” in a proper fashion.
The reason I return bool from this function is so that with a hierarchy of classes (all of which must be copyable), I can reuse the base class’ deepcopy as a starting point. For example:
Now in a derived class, deep copying can be extended like so:
Of course the drawback of this technique is that every class in your hierarchy has to have a “deepcopy” method. However, it does save time when writing the assignment operator and copy constructors, since they rely on the deepcopy method and their function bodies become completely trivial.
3. Should copying an object cause any side effects to the original object?
This is a little trickier topic, since while the answer seems very obvious at first I’ve come across a scenario where it wasn’t the case. My example was when an object had sole responsibility or control over an external resource and I needed to transfer ownership to the new copy and remove ownership from the old copy. This means that copying an object has a side effect to the copied object on the right-hand-side of the assignment operator.
In this case, you will have to remove the keyword “const” from the deepcopy, copy constructor and assignment operator methods. You should also explicitly document what is going on when a copy is made as a client/user will assume (and rightfully so) that performing a copy will leave the right-hand-side copy unscathed.
Ideally you should avoid these scenarios altogether by redesigning your code in such a way that the right-hand-side object is not affected by the copy (Note that smart pointers may have helped me avoid this scenario).
§72 · March 23, 2005 · C++, Software, Technology, Tips · · [Print]
A copy assignment operator of class is a non-template non-static member function with the name operator= that takes exactly one parameter of type T, T&, const T&, volatile T&, or constvolatile T&. For a type to be , it must have a public copy assignment operator.
|class_nameclass_name ( class_name )||(1)|
|class_nameclass_name ( const class_name )||(2)|
|class_nameclass_name ( const class_name ) = default;||(3)||(since C++11)|
|class_nameclass_name ( const class_name ) = delete;||(4)||(since C++11)|
- Typical declaration of a copy assignment operator when copy-and-swap idiom can be used.
- Typical declaration of a copy assignment operator when copy-and-swap idiom cannot be used (non-swappable type or degraded performance).
- Forcing a copy assignment operator to be generated by the compiler.
- Avoiding implicit copy assignment.
The copy assignment operator is called whenever selected by overload resolution, e.g. when an object appears on the left side of an assignment expression.
Implicitly-declared copy assignment operator
If no user-defined copy assignment operators are provided for a class type (struct, class, or union), the compiler will always declare one as an inline public member of the class. This implicitly-declared copy assignment operator has the form T& T::operator=(const T&) if all of the following is true:
- each direct base of has a copy assignment operator whose parameters are B or const B& or constvolatile B&;
- each non-static data member of of class type or array of class type has a copy assignment operator whose parameters are M or const M& or constvolatile M&.
Otherwise the implicitly-declared copy assignment operator is declared as T& T::operator=(T&). (Note that due to these rules, the implicitly-declared copy assignment operator cannot bind to a volatile lvalue argument.)
A class can have multiple copy assignment operators, e.g. both T& T::operator=(const T&) and T& T::operator=(T). If some user-defined copy assignment operators are present, the user may still force the generation of the implicitly declared copy assignment operator with the keyword .(since C++11)
The implicitly-declared (or defaulted on its first declaration) copy assignment operator has an exception specification as described in dynamic exception specification(until C++17)exception specification(since C++17)
Because the copy assignment operator is always declared for any class, the base class assignment operator is always hidden. If a using-declaration is used to bring in the assignment operator from the base class, and its argument type could be the same as the argument type of the implicit assignment operator of the derived class, the using-declaration is also hidden by the implicit declaration.
Deleted implicitly-declared copy assignment operator
A implicitly-declared copy assignment operator for class is defined as deleted if any of the following is true:
- has a user-declared move constructor;
- has a user-declared move assignment operator.
Otherwise, it is defined as defaulted.
A defaulted copy assignment operator for class is defined as deleted if any of the following is true:
- has a non-static data member of non-class type (or array thereof) that is const;
- has a non-static data member of a reference type;
- has a non-static data member or a direct or virtual base class that cannot be copy-assigned (overload resolution for the copy assignment fails, or selects a deleted or inaccessible function);
- is a union-like class, and has a variant member whose corresponding assignment operator is non-trivial.
Trivial copy assignment operator
The copy assignment operator for class is trivial if all of the following is true:
- it is not user-provided (meaning, it is implicitly-defined or defaulted) , , and if it is defaulted, its signature is the same as implicitly-defined(until C++14);
- has no virtual member functions;
- has no virtual base classes;
- the copy assignment operator selected for every direct base of is trivial;
- the copy assignment operator selected for every non-static class type (or array of class type) member of is trivial;
A trivial copy assignment operator makes a copy of the object representation as if by std::memmove. All data types compatible with the C language (POD types) are trivially copy-assignable.
Implicitly-defined copy assignment operator
If the implicitly-declared copy assignment operator is neither deleted nor trivial, it is defined (that is, a function body is generated and compiled) by the compiler if odr-used. For union types, the implicitly-defined copy assignment copies the object representation (as by std::memmove). For non-union class types (class and struct), the operator performs member-wise copy assignment of the object's bases and non-static members, in their initialization order, using built-in assignment for the scalars and copy assignment operator for class types.
The generation of the implicitly-defined copy assignment operator is deprecated(since C++11) if has a user-declared destructor or user-declared copy constructor.
If both copy and move assignment operators are provided, overload resolution selects the move assignment if the argument is an rvalue (either a prvalue such as a nameless temporary or an xvalue such as the result of std::move), and selects the copy assignment if the argument is an lvalue (named object or a function/operator returning lvalue reference). If only the copy assignment is provided, all argument categories select it (as long as it takes its argument by value or as reference to const, since rvalues can bind to const references), which makes copy assignment the fallback for move assignment, when move is unavailable.
It is unspecified whether virtual base class subobjects that are accessible through more than one path in the inheritance lattice, are assigned more than once by the implicitly-defined copy assignment operator (same applies to move assignment).
See assignment operator overloading for additional detail on the expected behavior of a user-defined copy-assignment operator.
Run this code
The following behavior-changing defect reports were applied retroactively to previously published C++ standards.
|DR||Applied to||Behavior as published||Correct behavior|
|CWG 2171||C++14||operator=(X&)=default was non-trivial||made trivial|