r? @matthewjasper

(rust-highfive has picked a reviewer for you, use r? to override)

I don't believe we should allow enums with struct or tuple variants (even if they don't have fields) to omit the repr.

(Ideally, I'd also prefer if such enums couldn't be as-cast to their discriminant, but that seems like it'd break backwards compatibility.)

Personally, I'm hoping that #81642 helps improve this substantially, so that using as becomes less necessary.

@fee1-dead Do you think you'd be able to prepare a PR that rejected casting enums with 0-field tuple/struct variants? (Per the discussion at https://rust-lang.zulipchat.com/#narrow/stream/213817-t-lang/topic/arbitrary.20enum.20discriminants/near/251899236) I can maybe give it a try myself if you'd like.

I'm busy with other stuff. You can do it if you'd like.

☔ The latest upstream changes (presumably #89884) made this pull request unmergeable. Please resolve the merge conflicts.

merge conflicts

discussed at T-compiler meeting; we think this is waiting on lang team decision regarding e.g. #88621 and PR #89234

Hello. This was discussed at the lang team meeting today.

• The main open question was if this change affects Stable or not. Currently, having a discriminant on a non-unit enum variant requires the arbitrary_enum_discriminant feature. This change should likely be "no more stable" than that feature, and it's unclear.
• Also, would as casting work for enums declared like this? In the meeting it was assumed "no".

would as casting work for enums declared like this? In the meeting it was assumed "no".

Actually, the answer is "yes". https://play.rust-lang.org/?version=stable&mode=debug&edition=2021&gist=f38e9d525cd1ca67ea8836aa65d6f7b9

#[repr(u8)]
pub enum Foo {
    A,
    B(),
    C{},
}

pub fn a(foo: Foo) -> u8 {
    foo as u8
}