Re: Copy constructor, or assingment operator not being called.

[Guy Harrison]

manu wrote:

> I have a strange problem with a program.

-ansi -pedantic -W -Wall -Werror

Hint: "return *this"

> I've reduced the code to the following working program.
> 
> ----- Code -----
> 
> class a
> {
> public:
>         a () : _i (0)
>         {
>                 ++_s_count;
>                 std::cout << "Creating a." << std::endl;
>         }
>         a (int i)
>                 : _i (i)
>         {
>                 ++_s_count;
>                 std::cout << "Creating a with " << i << " value." <<
> std::endl;
>         }
>         a (const a & p) : _i (p._i)
>         {
>                 ++_s_count;
>                 std::cout << "Copying a, value " << p._i << "." <<
> std::endl;
>         }
>         ~a ()
>         {
>                 --_s_count;
>                 std::cout << "Destroying a." << std::endl;
>         }
>         a & operator= (const a & p)
>         {
>                 std::cout << "Assigning a, value " << p._i << "." <<
> std::endl;
>         }
>         int print () { return _i; }
>         void dump () { std::cout << "_s_count=" << _s_count <<
> std::endl; }
> private:
>         static int _s_count;
>         int _i;
> };
> 
> int a::_s_count = 0;
> 
> a makeA ()
> {
>         a z(1);
>         return z;
> }
> 
> int main (void)
> {
>         a na = makeA();
> 
>         std::cout << "Value: " << na.print() << std::endl;
> 
>         na.dump ();
> 
>         return 0;
> }
> 
> ----- End code -----
> 
> The problem is that in the makeA function, no copy constructor is
> being called. And I don't know why!

'z' has an int ctor so is constructed directly.

//hint: a z(1),x(z);

> It should make an a object, and then a copy to pass it as return
> value, but it doesn't.

Strictly, yes. Compiler optimisation. No point making an exact duplicate of
something that can be passed back directly.

> Output is:
> 
> Creating a with 1 value.
> Value: 1
> _s_count=1
> Destroying a.
> 
> It should be at least 2 objects, may be even 3. (In function 1,
> parameter 2 and assignment in main 3)
> 
> Am I wrong? Am I missing something?

Nope. If an object never got created then it's a safe bet that "not
destroying it" will work. Provided 'na' ends up with contents of 'z' then
job done. For a correctly written object, even if it does invoke all the
theoretical steps it'll still work.

> I'm working with GCC version 3.3.4 (Debian).

Handy development tip...

For any class, plonk its cctor and ator in the private section. That'll
cause the compiler to moan if *any* other code tries to copy construct or
assign off that object (if you want it you can implement it but finding an
unwanted default one is pita).