partial_ord_correct and ord_correct sometimes fail

[workingjubilee]

Hello, it's quite alright if you aren't maintaining this crate anymore, but I figured you might be interested in knowing this much:

This is not a regression. Locally the tests partial_ord_correct and ord_correct fail with some chance both on stable and nightly. I've seen all 4 combinations of success and fail on for both versions of Rust. I think it depends on wether quickcheck finds a code path where the properties don't hold, which I assume depends on randomness.

Digging into why it fails, as example input that fails let's take left = [(66, 1)] and right = [(66, 0), (0, 0)]. The tests compare partial_cmp and cmp results. compare_as_iters answers what the keys compared via the the iterator::cmp method yields. In this case [66].iter().cmp([66, 0].iter()) == Ordering::Less. In contrast the hopscotch::Queue::cmp method uses completely different logic and does offset and tag comparisons before going over to value, not key, iter comparison. In this specific instance tag comparison yields Ordering::Greater which causes the mismatch and test failure.

Originally posted by @Voultapher in rust-lang/rust#128899 (comment)