Olivier Langlois's blog

This is an advice that can be found in many books such as:

C++ Coding Standards, Exceptional C++, More Effective C++ or The C++ Standard Library

However, despite the abundance of literature giving this advice and explaining why, a lot of people are not aware of this. So if you are one of these persons and you are someone who is spending more time browsing the web than reading books and you have found this blog, this is for you.

At first look, statements like

++i;
i++;

seem equivalent but you have to remember that if i is an object, the increment operator is implemented with a function. Seeing the form of typical operator functions implementation should explain everything:

// Prefix version
T& T::operator++()
{
  //Perform increment
  return *this; // Return a reference on this
}

// Postfix version
T T::operator++(int)
{
  T res(*this); // Make a copy of the current value
  ++*this;      // Call the prefix version
  return res;   // Return the value by result
}

So, as you can see, most of the time the postfix increment operator is implemented in terms of the prefix one so even if you call the postfix operator, you will end up calling the prefix operator function anyway. The rest is just overhead:

• The copy constructor will be invoked at res creation.
• A function call will be done if operator++() is not inline
• The copy constructor might be called a second time because res is returned by value

You'll notice that at the third point, I used 'might'. If you have a good compiler, it might be able to get rid of the construction of the return value by using a technique called the Return Value Optimization (RVO). Most compilers are able to pull off that optimization with an unnamed object (temporary object) like:

return T;

If a return statement is returning an automatic variable (local variable) declared before the return statement as it is done in the operator++(int) function, a lot fewer compilers are able to apply the RVO. Those who are able are said to have named RVO (NRVO). Visual C++ has it only since version 2005 according to this blog entry. GCC supports it since version 3.1 according to this Boost mailing list archive entry.

To conclude, it should now be obvious that unless you absolutely need the postfix increment semantic, you should use the prefix version. This should represents 99% of the increments in a program since usually, they stand alone in their own statement and the return value is just not used at all. This is usually what happens in code using STL containers iterators. One could argue that it does not make a perceptable difference performance wise in a program. Maybe but why choosing the slowest option when both ways are equally easy to write? Herb Sutter and Andrei Alexandrescu think the same and they give the excellent advice (Don't pessimize prematurely) in their book C++ Coding Standards.