Michel Fortin wrote: > On 2010-03-05 11:32:27 -0500, Jonathan M Davis <[email protected]> > said: > >>> Why couldn't a container have the same interface as a range, only with >>> the added ability to push elements into it? >> >> Probably because it would become way too easy to accidentally use the >> container as a range, which in many cases would end up emptying it when >> passing it to one of the std algorithms. > > Yes, I understand that. > > My point is that the problem already exists with ranges: by exposing a > stream as a range, you'll empty the stream when iterating over it. If > that's acceptable for streams, why is it not for containers? > > And perhaps you actually want to empty the container as you iterate > over it. If you use your container as a queue, that's what you'll > expect. In fact, a stream is just like a queue, and you want algorithms > to work on streams don't you? >
It would probably be better in such a case to have a range wrap the container. That way, you can have range semantics for the container if you want them, but you don't risk using them on a container where you don't. As for streams, I don't know. I do like the idea of using ranges with streams, but I haven't really looked at what the downsides to that might be. It depends a lot on what the std algorithms do. With some it might be fine while with others it would be a really bad idea. I'd have to look into it more to say, but at least with a stream, it's expected that iterating over it will alter it. With a container, that's not usually the case. And in cases where it is, either that container could be a special case and be a range as well, or it you could have a wrapper range do it. Personally, I'd prefer the wrapper idea. It would probably allow you to create generic wrappers to use on pretty much any container (or at least containers with a particular set of functions) rather than special casing particular containers. Regardless, I think that making containers ranges in the general case is a bad idea. - Jonathan M Davis
