CS193D Winter 2004

Handout 06 January 26, 2004

Copy Constructor and operator=

We already know that the compiler will supply a default (zero-argument) constructor if the programmer does not specify one. This default constructor will call each data members default constructor in order to initialize the object. If the programmer wishes to override that default constructor, he or she simply provides one. There is another type of constructor the compiler generates it is called a copy constructor, and it's called whenever an object needs to be constructed from an existing one. Suppose we had a class to encapsulate strings, and we call this class mystring1 . It might include such data members as the length of the string as well as the characters that would be in the string.
class mystring { public: mystring(const char* s = ""); ~mystring(); ... private: int length; char *str; };

For this example, well assume that the mystring constructor will allocate space for the characters, and the destructor will free that space. The copy constructor may be called when doing simple initializations of a mystring object:
mystring me("Geraldo"); mystring clone = me; // copy constructor gets called.

More importantly, the copy constructor is called when passing an object by value, or returning an object by value. For instance, we might have a function which opens a file:
static void openFile(mystring filename) { // Convert the object to a character string, open a stream... }

Were calling it mystring so we dont get confuse it with the built-in string class.

2 We might declare a string and call the OpenFile function like this:
mystring name(flights.txt); openFile(name);

When passing the name object, the copy constructor is called to copy the mystring object from the calling function to the local parameter in the openFile function. Because we did not specify a copy constructor, the default copy constructor is called. Default Copy Constructor The default copy constructor does a member-wise copy of an object. For our mystring class, the default copy constructor will copy the length integer and the characters pointer. However, the characters themselves are not copied. This is called a shallow copy, because it only copies the data one level deep in a class hierarchy. In memory, what wed have would look like this: 7
f l i g h t s

name

7
filename

This causes a problem because now the two objects share the same characters data. If we changed any of the characters within the filename object, it would change the characters in the name as well. Whats worse is that after the OpenFile function exits, the mystring destructor is called on the filename object, which frees the characters data member. Now the name object has an invalid characters data member! What We Want: The Deep Copy In order to avoid potential disasters like this, what we want is a deep copy of the mystring object. We want a copy of the length, and we want a copy of the entire str data member not just a copy of the pointer.

3 What we want is a picture in memory which looks like this:

name

7
f l i g h t s

7
filename

Now we can manipulate the characters within the filename object and not affect the original name object. When the character array within the copy is freed on exit the openFile function, the original mystring object is still going strong. Declaring a Copy Constructor In order to provide deep-copy semantics for our mystring object, we need to declare our own copy constructor. A copy constructor takes a constant reference to an object of the class type as a parameter and has no return type. We would declare the copy constructor for the mystring within the class declaration like this:
class mystring { public: mystring(const char* s = ); mystring(const mystring& s); ... };

We might implement the copy constructor like this:

mystring::mystring(const mystring& s) { length = s.length; str = new char[length + 1]; strcpy(str, s.str); }

You should provide a copy constructor for any class for which you want deep copy semantics. If the class only has data members that are integral types (that is, it contains no pointers or open files) and/or direct objects (which have properly implemented copy constructors), you can omit the copy constructor and let the default copy constructor handle it.

4 Limitations The copy constructor is not called when doing an object-to-object assignment. For instance, if we had the following code, we would still only get a shallow copy:
mystring betty(Betty Rubble); mystring bettyClone; bettyClone = betty; // Initializes string to Betty Rubble // Initializes string to empty string

This is because the assignment operator is being called instead of the copy constructor. By default, the assignment operator does a member-wise copy of the object, which in this case gives a shallow copy. However, C++ gives us the ability to override the default assignment operator, and well learn that today too. Incidentally, there may be cases when you want to prevent anyone from copying a class object via the copy constructor. By declaring a copy constructor as a private constructor within the class definition and not implementing it, the compiler will prevent the passing of objects of that class by value. Assignment It is possible to redefine the assignment operator for a class. The assignment operator must be defined as a member function in order to guarantee that the left-hand operand is an object. By default, the assignment operator is defined for every class, and it does a member-wise copy from one object to another. This is a shallow copy, so the assignment operator should generally be redefined for any class that contains pointers. You can declare the assignment operator as a private operator and not provide an implementation in order to prevent object-to-object assignment. The key difference between assignment and copy construction is that assignment possibly involves the destruction of embedded resources. We need to overload the assignment operator whenever we want to guarantee that memory (or perhaps other resources) aren't orphaned improperly. The syntax for the assignment method is a trifle quirky, but it's just that: syntax, and you just treat the prototype as boilerplate notation and otherwise implement the method as you would any other.
const mystring& operator=(const mystring& rhs) { if (this != &rhs) { delete[] this->str; // donate back useless memory this->str = new char[strlen(rhs.str) + 1]; // allocate new memory strcpy(this->str, rhs.str); // copy characters this->length = rhs.length; // copy length } return *this; } // return self-reference so cascaded assignment works

5 A good way to look at the assignment operator: Think of it as destruction followed by immediate reconstruction. The delete[] would appear in the destructor as well, whereas the next three lines are very constructor-ish. Caveats 1: The this != &rhs checks to see whether we're dealing with a selfassignment. An expression of the form name = name is totally legal, but there's no benefit in actually destroying the object only to reconstruct it only to take the same form. In fact, the delete[] line would actually release the very memory that we need on the very next line. The return type is const mystring&, and the return statement is return *this, because cascaded assignments involving mystrings should behave and propagate right to left just as they would with ints or double s.
mystring mystring mystring mystring heroine("ginger"); cluck("babs"); quack("rhodes"); meanie("mrs. tweety");

Caveat 2:

meanie = quack = cluck = heroine; cout << heroine << cluck << quack << meanie << endl; // should print: gingergingergingerginger