C++Now 2018: Marcel Ebmer “for_each_adjacent＜N＞”

—
Lightning Talk
—
—
---

*--*

---

[AxelStrem]

wouldn't it make more sense to have an "adjacent_iterator<N>" and simply use it with all the regular algorithms?

[Dth]

Hmm.. thinking about the new ranges, it looks like you're after some transform that takes the first range and gives you a bunch of sub-ranges of n consecutive items.
It's a bit like a sliding-window range adaptor, could be really useful for convolutions!

[ashvardanian]

Why even bother with tuple<T>, if you can use array<T, N>?

[denisyaroshevskiy]

As it was mentioned, your version is not terribly nice for forward_iterators.

However, the std::next(first, Is) trick is really cool and it will be faster for random access (just replace the container type and give it more arguments to work on).
Really nit trick.

Maybe passing adjacent elements as a range would be better though.

[Voy]

Why not deduce the arity of the function and use it instead of explicit N.

[karolyhorvath]

Meh. Weird approach. Create an iterator that consumes in N-ary chunks, then you can use all the algorithms...