Add .get(range)

[Philippe-Cholet]

Closes #447.

So much changes happened since this work was initiated that rebase on top of master seems nothing less than a nightmare to me (I know, I tried). I therefore looked at the git differences, committed those (after formatting) and credited @wod3 and @TrolledWoods for their respective commits. Then I fixed my own review and a bit more.

---

I thought I would additionally implement IteratorIndex for Either<range, range> (and Box<dyn range> but I could not) but doing so would later result in a nasty breaking change if this was removed after an eventual inclusion to Iterator.

Details about either

    impl<L: Sealed, R: Sealed> Sealed for either::Either<L, R> {}


impl<I, L, R> IteratorIndex<I> for Either<L, R>
where
    I: Iterator,
    L: IteratorIndex<I>,
    R: IteratorIndex<I>,
{
    type Output = Either<<L as IteratorIndex<I>>::Output, <R as IteratorIndex<I>>::Output>;

    fn index(self, iter: I) -> Self::Output {
        // Could be more succinct with `map_either_with` but it would require to bump `either` version.
        match self {
            Either::Left(left) => Either::Left(left.index(iter)),
            Either::Right(right) => Either::Right(right.index(iter)),
        }
    }
}


    /// // It works nicely with either.
    /// use itertools::Either;
    ///
    /// let mut either_range = Either::Left(0..1);
    /// range = vec.iter().get(either_range).copied().collect();
    /// assert_eq!(&range, &[3]);
    ///
    /// either_range = Either::Right(2..);
    /// range = vec.iter().get(either_range).copied().collect();
    /// assert_eq!(&range, &[4, 1, 5]);

[codecov]

Codecov Report

Attention: Patch coverage is 96.36364% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 94.55%. Comparing base (6814180) to head (de08253).
Report is 64 commits behind head on master.

Files	Patch %	Lines
src/iter_index.rs	93.33%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #891      +/-   ##
==========================================
+ Coverage   94.38%   94.55%   +0.16%     
==========================================
  Files          48       49       +1     
  Lines        6665     6945     +280     
==========================================
+ Hits         6291     6567     +276     
- Misses        374      378       +4     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[scottmcm]

Most of these are just things to ponder that might not need changing, but I think at least the overflow thing should be tweaked.

[scottmcm]

This is a very generic name to have as a top-level method. Does it really pull its weight?

[jswrenn]

I think the name is appropriate — it 'gets' elements from the iterator. We use get for consistency with [T]::get, and match its API as closely as possible. There's little risk of conflict, since slices do not directly implement Iterator.

[scottmcm]

In the Itertools trait, I understand.

But it's also pulled in as a function (https://github.com/rust-itertools/itertools/pull/891/files#diff-b1a35a68f14e696205874893c07fd24fdb88882b47c23cc0e0c80a30c7d53759R157). I guess it's unlikely to matter unless people are doing use itertools::*;, which is already discouraged, but as a thing not scoped to an impl it seems surprising.

(But you can feel free to "won't fix" this -- I don't feel that strongly about it.)

[Philippe-Cholet]

Ah fair enough, we don't really need to make the free function get public.

[Philippe-Cholet]

I rebased on master and added 4 commits:

1. Is there a reason to prefer Take<Skip> over Skip<Take>? As Skip<Take> is simpler defined!
2. I don't see how we should really handle a range that includes usize::MAX other than with panic.
3. Clarified when the resulting iterator is ExactSizeIterator and/or DoubleEndedIterator.
4. fix 2nd commit: .get(1..=usize::MAX) should be fine

And removed get as a free function.