codereport / An-Algorithm-Library

A C++ algorithm library that extends the C++ standard algorithms found in <algorithm> & <numeric>
Boost Software License 1.0
22 stars 4 forks source link

[Question, Suggestion] Would a variadic `zip()` be useful for the library? #2

Open juntuu opened 3 years ago

juntuu commented 3 years ago

A variadic zip() might make implementing other variadic algorithms easy in terms of existing stl algorithms. It might also be useful on its own.

The function would take variadic number of ranges, and return a range that iterates over all the input ranges, advancing all the input range iterators at each step, and yielding a tuple of values when dereferencing. The zipped range would only iterate until the shortest input range is consumed.

For example:

auto a = std::vector{1, 2, 3, 4};
auto b = std::vector{1, 2};
auto zipped = zip(a, b);
auto it = zipped.begin();

*it == std::tuple{*a.begin(), *b.begin()};
++it;
*it == std::tuple{*(a.begin()+1), *(b.begin() + 1)};
++it;
(it != zipped.end()) == false; // b only had 2 elements

I played around a bit with the idea and wrote a little poc implementation https://godbolt.org/z/9oGWs9.

Ps. boost seems to have something similar https://www.boost.org/doc/libs/1_75_0/libs/iterator/doc/zip_iterator.html

Sebanisu commented 3 years ago

So would this be like a range adapter/view? So would this be like a version of what we're getting in c++23? http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2020/p2214r0.html#the-zip-family

juntuu commented 3 years ago

Something like that, yes. I wasn't aware of the plans for it in c++23, thanks for pointing that out

It could also be made with iterator pairs like zip(a.begin(), a.end(), b.begin(), b.end(), c.begin(), c.end(), ...) or less safe zip(a.begin(), a.end(), b.begin(), c.begin(), ...), but using ranges might be nicer.

codereport commented 3 years ago

I think in general our policy should be to add algorithms when we need them. So if this comes up in the future we can add it.

juntuu commented 3 years ago

I totally agree, algorithm with no use is essentially dead code.

The main reason I started this discussion, was because I was thinking whether or not zip() would be beneficial for implementing the other variadiac algorithms.