> I think you will be surprised when you benchmark this. ;-) @Araq, you are saying that the cost of the branch and object field read aren't that much. I guess I can see this with branch prediction and the compiler optimizing the field read into a single instruction. I'll have to try it.
Also, regarding the allocator proposal: > The only downside is that most existing allocators do not adhere to this > design. :-) Does that meat to use this proposal you need to write your own allocator that makes direct OS requests for memory without going through C/C++ provided API? Is that what current allocators do when the flag to use "malloc" isn't set? Or if your current allocators aren't written in the new way, is that what you plan to do?
