Joseph Gauterin wrote:
the important thing is that a call to boost::swap swaps its arguments, not the exact mechanism by which it swaps the arguments.
I would like to add some extra checks to the existing test programs in trunk/libs/utility/swap/test, checking if boost::swap indeed swaps its arguments: swap_test_class object1 = initial_value1; swap_test_class object2 = initial_value2; boost::swap(object1,object2); BOOST_CHECK(object1 == initial_value2); BOOST_CHECK(object2 == initial_value1); Therefore, your swap_test_class would need to become EqualComparible, and it would need to have some data. Hope you think it's okay if I apply the attached patch... :-)
Also, there's a clear precedent with the standard library - the algorithm are all named after what they do, not how they do it (swap, sort, remove etc).
The Standard actually describes std::swap in terms of exchanging values: template<class T> void swap(T& a, T& b); ... Effects: Exchanges values stored in two locations. So we could consider renaming "swap_impl" to "exchange_values". (And have boost::swap calling exchange_values.) What do you think?
// - swap_impl is put outside the boost namespace, to avoid infinite // recursion (stack overflow) when swapping objects of a primitive type.
Does the infinite recursion only happen using primitive types - and if so do you know why?
I just tested moving ::boost_swap_impl::swap_impl to ::boost::swap_impl on my local copy, and running utility/swap/test for MSVC 7.1, MSVC 9.0, and Borland 5.82. On MSVC (both 7.1 and 9.0), moving ::boost_swap_impl::swap_impl to ::boost::swap_impl introduces test failures
If it does, then it might be possible to move the boost_swap_impl namespace inside namespace boost using some sort of SFINAE technique to explicitly call std::swap for primitive types.