Tracking issue for `#[repr(align(x))]` on enums

[niklasf]

RFC 1358 introduced #[repr(align(x))] on structs (and unions). It mentions:

Custom alignment can only be specified on struct declarations for now. Specifying a different alignment on perhaps enum or type definitions should be a backwards-compatible extension.

It would be a nice convenience to allow this for enums, especially when they are part of a public API.

#[repr(align(x))]
enum Foo {
    // ...
}

would be equivalent to using AlignX<Foo> everywhere:

#[repr(align(x))]
struct AlignX<T>(T);

enum Foo {
    // ...
}

[cramertj]

This compiles today:

#![feature(repr_align_enum)]

#[repr(align(1))]
enum X { X(u64) }

fn main() {
    match &X::X(5) {
        X::X(x) => println!("{:?}", x),
    }
}

I would expect this to be an error-- references to u64 require more than align(1).

[hanna-kruppe]

repr(align(N)) raises the alignment to N, but has no effect if the alignment is already >= N without the annotation (in particular it doesn't lower the alignment or error). Or in other words, it ensures a certain minimum alignment. This is not specific to enums, it works the same for structs and unions. I also think this is intentional, though I can't find mention of it in the RFC at a glance (but in any case I think it's the most sensible semantics -- though perhaps there should be a lint).

[cramertj]

Got it-- it's surprising to me that we wouldn't at least warn on a #[repr(align(X))] attribute that doesn't have any affect (doesn't raise the alignment).

[Centril]

@cramertj Seems like something we could tuck into the unused_attributes lint? cc @oli-obk

[oli-obk]

The effect of the attribute is platform dependent. I don't see how we could make this lint free of false positives. We could add something like it to clippy, but as a rustc lint that is always on, false positives are essentially never accepted.

Suggesting #[cfg_attr(not(current_platform), repr(align(X)))] is possible, technically correct on all platforms, but still seems like it's just noise to me.