impl TrustedLen for Cycle

[KamilaBorowska]

That was something I noticed in #66531 (comment), so sure, why not.

size_hint method for Cycle<I> can return either (0, Some(0)) or (usize::MAX, None) for TrustedLen implementations ((0, None) is impossible as (0, x) if x != Some(0) returned by inner iterator is impossible by TrustedLen invariant).

[rust-highfive]

r? @dtolnay

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

[dtolnay]

There is no such thing as an unstable trait impl so this would need to be changed to #[stable].

[KamilaBorowska]

TrustedLen isn't a stable trait. Unstable is correct here, I think.

[dtolnay]

This PR is adding:

unsafe impl<I> TrustedLen for Cycle<I> where I: Clone + TrustedLen

where Cycle is the iterator returned by Iterator::cycle.

I don't think this impl is correct because I: TrustedLen does not imply any knowledge about I's Clone impl. Here is a Cycle whose length is neither 0 nor ∞. Returning the wrong length for this Cycle could result in undefined behavior.

use std::iter::TrustedLen;

struct S(usize);

impl Clone for S {
    fn clone(&self) -> Self {
        S(0)
    }
}

impl Iterator for S {
    type Item = ();

    fn next(&mut self) -> Option<Self::Item> {
        if self.0 > 0 {
            self.0 -= 1;
            Some(())
        } else {
            None
        }
    }

    fn size_hint(&self) -> (usize, Option<usize>) {
        (self.0, Some(self.0))
    }
}

unsafe impl TrustedLen for S {}

fn main() {
    for x in S(4).cycle() {
        println!("{:?}", x);
    }
}

FYI @rust-lang/libs

[KamilaBorowska]

Yeah, that's a fair point. While it's strictly speaking fine with built-in implementations, there is nothing to guarantee Clone is sane, which may not be the case with external implantations.