With moderate compiler optimization: all three are roughly the same.
Without optimization: the traditional (temp storage) method is significantly FASTER than the other two.

I started to test the performance because Wikepedia is saying that temp storage method is probably faster on current architecture due to instruction pipelining and the other two method must be executed sequentially thus no parallelism can be exploited. And my testing result just proved that.

And, have you given it enough thoughts when you say “The code fails even for the integers when the the sum of two passed variables exceeds the maximum integer limit on that machine.”?

a = a xor b
b = a xor b
a = a xor b

correct(the contract should specify that the references passed in may not be equal), works for all data types, and saves one precious variable.[/quote]

I suppose if someone is at the level of using templates, the concept of exception safety should be introduced somewhere close.

For someone who’s new, I’d let them start with Ruby, Javascript, Logo, Object Pascal, or even Java. C++ sounds like a poor choice for an introductory language.

@Michael,
If you are creating an application in C++ and you want to swap two variables, you most definitely would use std::swap function. If you read carefully, I was talking about people who are starting to learn programming, and hence need to write their own swap, search, sort etc. functions. The exercise clears up points like the ones discussed in this post.
I didn’t even bother to make the swap function inline; my example focussed only on the right way to do the swapping. But I will post the std::swap functions as implemented by major STL implementors to show how simple they actually are.

The only correct way of swapping is don’t bother writing your own swap. Use std::swap directly if you can, call it if you cannot.

std::swap(a, b).

Done. Never reinvent your own wheel.

Your 1st way is wrong because it doesn’t have a “nothrow” guarantee. It’s especially important when memory is tight.