specialize Iterator::nth for Cycles over ExactSizeIterators to

[llogiq]

reduce clones at the cost of one remainder. This should be faster in most cases and may in some situations reduce memory churn via clones.

[rust-highfive]

r? @withoutboats

(rust_highfive has picked a reviewer for you, use r? to override)

[varkor]

This is subject to the same discussion as in #47533, concerning observability of clones.

[withoutboats]

@bors r? @bluss

[bluss]

This should be just I::Item: Copy style wise. Iterator + ExactSizeIterator is also redundant, just ExactSizeIterator is the same.

[bluss]

Omitting Clone calls for Copy types is fine, that's something we've decided previously (I just don't have a link for that).

Current implementation omits the clones of the iterator itself, so a motivation is needed for that too.

[llogiq]

Thank you for the review! We require clone impls for Copy types to be a memcpy, so eliding them should be OK.

Regarding the iterator, I can change it to require Copy, too, to be on the safe side. On the other hand it is debatable whether cloning the iterator should be an observable side effect of the implementation.

[durka]

Iterators aren't Copy as policy so this impl isn't likely to apply to much of anything.

[llogiq]

Too true, I better revert that one. So since we cannot require I: Copy, we can either declare the number of clones (and thus drops) on the iterator implementation-specified and try to reduce them, or leave stuff as is and have people deal with it in other ways (i.e. by restricting the n parameter manually).

[shepmaster]

Ping from triage, @bluss — new commits up for you to review!

[bluss]

I think the same applies as in the other PRs, see this comment #47533 (comment)

[kennytm]

Ping from triage @bluss@llogiq!

Given the linked comment #47533 (comment), I guess this PR should be closed since clones of the iterator itself are still elided, and an RFC detailing whether clone elision is acceptable should be accepted before this PR can be reconsidered?

[llogiq]

Until we all reach decision
On iterator clone elision
I'm going to close
this PR so those
that remain are awaiting revision