on Sat Mar 14 2009, Thorsten Ottosen <thorsten.ottosen-AT-dezide.com> wrote:
David Abrahams skrev:
A class that attempts to provide value semantics but doesn't support x = x is putting a big hole in the system of equational reasoning. Justifying that (to me) would take some pretty heavy proof.
I have not really seen code that exhibits x = x. I've seen lot's of discussion (e.g. Sutter & Meyers). Does anybody write such code?
Yes, it happens. It doesn't happen that way, of course. It happens in x = y when x and y happen to refer to the same object. Regardless, as I've been saying, the burden of proof runs in the other direction. You need to show a /realistic/ benchmark that suffers significantly. And don't bias the test unfairly, either: you only need the self-assignment check in your slow branch, the one that does the allocation. The great computer scientists that came before us have shown us the way. http://en.wikipedia.org/wiki/Optimization_(computer_science)#When_to_optimiz... -- Dave Abrahams BoostPro Computing http://www.boostpro.com