On Saturday, 1 June 2013 at 08:12:00 UTC, bearophile wrote:
David Piepgrass:
In fact, most STL algorithms require exactly two iterators--a
range--and none require only a single iterator<
I think there are some C++ data structures that store many
single iterators. If you instead store ranges you double the
data amount.
Hashmaps would be the most common example. Usually implemented as
a linked list of key-value pairs along with a vector of list
iterators.
Iterators are also useful for constructing sub-ranges, which
proves useful in the implementation of some algorithms. Writing
std::next_permutation in D with ranges is quiet frustrating
compared to C++.
https://github.com/D-Programming-Language/phobos/blob/master/std/algorithm.d#L10901
http://gcc.gnu.org/onlinedocs/gcc-4.6.2/libstdc++/api/a01045_source.html#l03619
Notice that in D you need to maintain a count of the elements
popped off the back and then use takeExactly to retrieve that
portion again. In C++ you just move the iterators around and
create "ranges" from pairs of iterators as you need them.
When I implemented nextPermutation in D, I constantly felt as if
I was fighting with ranges instead of working with them - I knew
exactly what I wanted to do, but ranges only provide a roundabout
way of doing it.
