Search code examples
c++privateaccess-specifier

Why can I access private variables in the copy constructor?


I have learned that I can never access a private variable, only with a get-function in the class. But then why can I access it in the copy constructor?

Example:

Field::Field(const Field& f)
{
  pFirst = new T[f.capacity()];

  pLast = pFirst + (f.pLast - f.pFirst);
  pEnd  = pFirst + (f.pEnd - f.pFirst);
  std::copy(f.pFirst, f.pLast, pFirst);
}

My declaration:

private:
  T *pFirst,*pLast,*pEnd;

Solution

  • IMHO, existing answers do a poor job explaining the "Why" of this - focusing too much on reiterating what behaviour's valid. "access modifiers work on class level, and not on object level." - yes, but why?

    The overarching concept here is that it's the programmer(s) designing, writing and maintaining a class who is(are) expected to understand the OO encapsulation desired and empowered to coordinate its implementation. So, if you're writing class X, you're encoding not just how an individual X x object can be used by code with access to it, but also how:

    • derived classes are able to interact with it (through optionally-pure virtual functions and/or protected access), and
    • distinct X objects cooperate to provide intended behaviours while honouring the post-conditions and invariants from your design.

    It's not just the copy constructor either - a great many operations can involve two or more instances of your class: if you're comparing, adding/multiplying/dividing, copy-constructing, cloning, assigning etc. then it's often the case that you either simply must have access to private and/or protected data in the other object, or want it to allow a simpler, faster or generally better function implementation.

    Specifically, these operations may want to take advantage of priviledged access to do things like:

    • (copy constructors) use a private member of the "rhs" (right hand side) object in an initialiser list, so that a member variable is itself copy-constructed instead of default-constructed (if even legal) then assigned too (again, if legal)
    • share resources - file handles, shared memory segments, shared_ptrs to reference data etc.
    • take ownership of things, e.g. auto_ptr<> "moves" ownership to the object under construction
    • copy private "cache", calibration, or state members needed to construct the new object in an optimally usable state without having to regenerate them from scratch
    • copy/access diagnostic/trace information kept in the object being copied that's not otherwise accessible through public APIs but might be used by some later exception object or logging (e.g. something about the time/circumstances when the "original" non-copy-constructed instance was constructed)
    • perform a more efficient copy of some data: e.g. objects may have e.g. an unordered_map member but publicly only expose begin() and end() iterators - with direct access to size() you could reserve capacity for faster copying; worse still if they only expose at() and insert() and otherwise throw....
    • copy references back to parent/coordination/management objects that might be unknown or write-only for the client code