cmp::max changed behavior in nightly

[emilio]

This program (playground) changed behavior between 97e58c0...1dd1884.

And broke bindgen as a result in funny ways.

use std::cmp;

#[derive(Copy, Clone, Debug, PartialEq, Eq, Ord)]
pub enum SizednessResult {
    NonZeroSized,
    DependsOnTypeParam,
    ZeroSized,
}

impl cmp::PartialOrd for SizednessResult {
    fn partial_cmp(&self, rhs: &Self) -> Option<cmp::Ordering> {
        use self::SizednessResult::*;

        match (*self, *rhs) {
            (x, y) if x == y => Some(cmp::Ordering::Equal),
            (NonZeroSized, _) => Some(cmp::Ordering::Greater),
            (_, NonZeroSized) => Some(cmp::Ordering::Less),
            (DependsOnTypeParam, _) => Some(cmp::Ordering::Greater),
            (_, DependsOnTypeParam) => Some(cmp::Ordering::Less),
            _ => unreachable!(),
        }
    }
}

fn main() {
    println!("{:?}", cmp::max(SizednessResult::ZeroSized, SizednessResult::NonZeroSized))
}

I suspect #64047.

[emilio]

cc @timvermeulen

[emilio]

Sorry, changelog is 97e58c0...ed8b708, which discards the llvm upgrade.

[emilio]

Looking at this further, I think what's going on is that Ord and PartialOrd don't agree. This may just be a bug in bindgen, who expected (looks like, I didn't write that code) PartialOrd::partial_cmp to be called instead of the Ord trait...

[Centril]

@emilio which means that this issue should be closed since it's a contract violation on bindgen's end?

[SimonSapin]

Yes, I think this std change only exposed an existing bug in bindgen. While it is a behavior change, it likely only occurs with contradictory impls:

https://doc.rust-lang.org/std/cmp/trait.Ord.html#how-can-i-implement-ord

Implementations of PartialEq, PartialOrd, and Ord must agree with each other.

Closing, thanks for investigating @emilio!