Question

Which way is good for returning an array of classes?

class BigSizeClass {
    int data[1024 * 1024];
};

1. Very bad and expensive way:

   CArray<BigSizeClass> test() {
     CArray<BigSizeClass> ret;
     for(int i = 0; i < 1024; i++) {
       BigSizeClass c;
       ret->Add(c);
     }
     return ret;
   }
   

2. may be I forget delete returning value:

   CArray<BigSizeClass>* test() {
     CArray<BigSizeClass>* ret = new CArray<BigSizeClass>();
     for(int i = 0; i < 1024; i++) {
       BigSizeClass c;
       ret->Add(c);
     }
     return ret;
   }
   

3. may be I do like Expensive:

   // CArray<BigSizeClass> r = test();  --> Expensive
   // CArray<BigSizeClass>& r = test(); --> Not Expensive (Correct - because of &)
   
   CArray<BigSizeClass>& test() {
       CArray<BigSizeClass>* ret = new CArray<BigSizeClass>();
       for(int i = 0; i < 1024; i++) {
           BigSizeClass c;
           ret->Add(c);
       }
       return *ret;
   }
   

4. may be I forget remove objects:

   CArray<BigSizeClass*> test()
   {
     CArray<BigClassSize> ret;
     for(int i = 0; i < 1024; i++)
     {
       BigSizeClass* c = new BigSizeClass;
       ret->Add(c);
     }
     return ret;
   }
   

Answer 1

Way 1 is just fine. RVO and NRVO will kick in here. Even if it doesn’t you can still “swaptimize”, assuming that CArray does the smart thing and has a swap function.

CArray<BigSizeClass> result;
test().swap(result);

However, I’d have to recommend against the use of CArray- it doesn’t use the copy constructor but instead uses memcpy, which is hideously undefined behaviour for any non-POD classes.