ARM Big Endian partialEq bug

[NielsOuvrard]

At opt-level=0, derived PartialEq on enums produces incorrect results on armebv7r-none-eabi (big-endian) when built with -Z build-std=core. The comparison returns true for non-equal values. opt-level≥1 or manual match-based PartialEq works correctly.
#[repr(u8)] does not mitigate it.

#[derive(Debug, Clone, Copy, PartialEq, Eq)]
#[repr(u8)]
pub enum MyEnum { A, B, C, D }

let a = MyEnum::A;

// BUG: returns true at opt-level=0 on armebv7r-none-eabi
if a == MyEnum::B {
    // reached
}

// OK: match always works
match a {
    MyEnum::A => { /* reached correctly */ }
    MyEnum::B => { /* not reached correctly */ }
    _ => {}
}

Minimal error reproduction's repo:

https://github.com/NielsOuvrard/rust-arm-be-partialeq-bug

Workaround

Set opt-level = 1 in Cargo.toml

[theemathas]

Is the repro readme and code written by an LLM? If so, how much of that is verified by a human? LLMs are known to say false things confidently

[apiraino]

@NielsOuvrard also: is this a regression? Do you have a "good" version where it doesn't reproduce?

[NielsOuvrard]

I indeed use an llm to help me to write this issue, and the readme
But I initially noticed the error while using the library spacepackets. I dig into and find this error.
I reproduced it by my own.

I don't know if this is a regression, I just found this problem, dig it with the help of claude, and found that using opt-level=1 fixed it in my example