on Mon Oct 31 2011, "Peter Dimov" <pdimov-AT-pdimov.com> wrote:
Dave Abrahams wrote:
I think you're missing the point, which is:
When the specification for an algorithm exposes the actual expressions used in a given implementation of an algorithm, it tends to lock down the implementation in ways that don't necessarily make sense (and can prevent optimizations). Tiny details that one usually doesn't consider, like whether part of the expression is an lvalue or an rvalue, const or non-const, become encoded into the specification of the algorithm.
None of this has anything to do with the fact that you want to require operator== to be an equivalence relation, and it doesn't have to be.
Again, missing my point. I long ago stipulated that this algorithm could be useful even in places where op== is not an equivalence relation, and that it was desirable to support those uses. My instinct to require equivalence was part of a reaction against writing the algorithm's specification in terms of valid expressions. I think such specifications need to be given scrutiny than they have been, historically. -- Dave Abrahams BoostPro Computing http://www.boostpro.com