Re: [boost] Statistics code example

From: Neal D. Becker (nbecker_at_[hidden])
Date: 2003-04-14 14:51:48

On Saturday 12 April 2003 09:15 pm, David Abrahams wrote:
> "Neal D. Becker" <nbecker_at_[hidden]> writes:
> > The most promising IMHO is cycle_iterator, which is an adaptor submitted
> > by Gennadiy Rozental.  I believe this is almost ready for a real boost
> > submission.  Last we left it, there were some questions of the semantics
> > of the "distance" between iterators.  In any case, the "distance" between
> > iterators in a cyclic buffer is fuzzy.
>
> A cyclic buffer can never be full, because full is the same as empty
> (begin == end).  Therefore, it can always be re-mapped into an
> equivalent vector by moving all the initial non-empty elements to the
> back, and distance is perfectly well-defined.
>
> Where's the fuzziness?

I thought the distance between 2 iterators is normally interpreted as a signed 
quantity.  In the case of the cyclic iterator, we have an ambiguity.  In 
general, for a container of size N, the distance from "a" to "b" can be:

d = b - a + mN

where m is an arbitrary integer.

I believe it is most useful to always use the interpretation:

d = (b >= a) ? (b - a) : (b - a + N)

That is, the distance from a to b is the same as the number of times I need to 
increment from a to b (assuming of course that increment has the usual 
meaning).

Note that this also means that b - a is always positive.