On Mon, 7 Mar 2011 09:08:33 -0500 "Stewart, Robert" <Robert.Stewart@sig.com> wrote: [...]
- pow2, factorial, nan: Same deal. I don't see a problem with making these free functions and specifying the desired return value as an explicit template parameter. It will look the same but make the interface more uniform: typedef integer_t< ... > integer_type; integer_type x = integer_type::pow2(n); vs typedef integer_t< ... > integer_type; integer_type x = pow2< integer_type >(n);
I could easily argue either side of that one. When I originally wrote that code, the current method seemed ever-so-slightly more appealing; now I think the way that you're advocating might be.
I'd rather algorithms be separate. Even if you find that an efficiency problem, it should merely be seen as a catalyst to expand the integer_t interface to regain the lost efficiency.
That's three votes in favor of changing it, counting my own altered opinion. I'll put it on the list.
Chad already knows my feelings on COW, but to be explicit: I'm not necessarily against use of COW in general, but I have yet to be given a use case where COW would outperform move emulation in theory.
I've located two: abs and negate. [...] Both involve changes to the sign bit only, and one only sometimes needs to alter it. [...]
You can do the same, I should think, without COW. All you need is to make your algorithms less naive. The algorithms can manage the sign in a separate variable and refer to an existing magnitude or make a copy, as needed, unless I'm much mistaken. That is a little less convenient, but should obviate COW.
Certainly it could be done without CoW. In an earlier iteration, I had a special subclass for that specific purpose, which kept a reference to an existing integer and a separate negate-the-sign Boolean. But if you work out the implementation, you'll see that it's essentially a poor-man's version of CoW anyway. Arguably less efficient, and definitely less aesthetically appealing, so I replaced it with a true CoW implementation. -- Chad Nelson Oak Circle Software, Inc. * * *