Regression in #[derive(Eq)] on type with non-Eq members.

[rphmeier]

Playground: https://play.rust-lang.org/?gist=ca6a481e640668153d66259c0ef1dde6&version=stable&backtrace=0

#[derive(Eq)]
pub struct Signature([u8; 65]);

impl PartialEq for Signature {
    fn eq(&self, other: &Self) -> bool {
        &self.0[..] == &other.0[..]
    }
}

fn main() { }

Succeeds on rustc 1.11.0 (9b21dcd6a 2016-08-15), but fails on rustc 1.13.0-beta.1 (cbbeba430 2016-09-28)

Seems to be related to std::cmp::AssertParamIsEq.

[bluss]

Caused by #36384

[bluss]

Explanation: previous ".assert_param_is_eq" used method resolution, finding slice methods on arrays. New code only looks at the type itself (the array).

cc @petrochenkov

[petrochenkov]

Theoretically, this is a bugfix, because the previous implementation didn't enforced the bounds properly.
However, I'm not sure if these restrictions on deriving Eq are necessary at all.

[bluss]

Without this road block, there would be many more Eq implementations on structs with floats in them.

[durka]

Should maybe crater #36384 to see the "bugfix" impact.

[alexcrichton]

Discussed during libs triage, conclusion was that we're likely to let this slide (as it's a bug fix) but we'll wait for the next crater build to ensure that no massive impact is seen.

[alexcrichton]

Ok, ran crater, there are two regressions related to this

• xor_name-0.1.0 (before) (after)
• ethcore-bigint-0.1.0 (before) (after)

I believe the latter is yours, right @rphmeier? The former presumably hasn't been reported yet.

[alexcrichton]

adding back the nominated tag to discuss again.