`#[repr(align(8))]` is not accepted for arrays

[manuthambi]

Code

struct Data {
    #[repr(align(8))]
    x: [usize; 20]
}

#[repr(align)] was accepted for arrays in rust v1.50, but is rejected in v1.51.

error[E0517]: attribute should be applied to a struct, enum, or union
 --> src/main.rs:3:12
  |
3 |     #[repr(align(8))]
  |            ^^^^^^^^
4 |     x: [usize; 20]
  |     -------------- not a struct, enum, or union

I am not entirely sure whether the attribute actually did the alignment in 1.50, or the compiler just ignored it. Regardless, is this expected behavior, and if it is how do I align an array?

[ehuss]

Extra validation was added for situations where attributes were ignored (in this case it was #80641). Previous versions issued an unused-attribute warning.

If you want to align an array, I think you can wrap it in a struct with #[repr(align(8))] struct Arr([usize; 20]).

[camelid]

Nominating to see if this was an intended "lint -> hard error" change.

(If there's another way to ask if a change is intended that people on T-compiler prefer, please let me know :)

@rustbot label: +I-nominated

[rustbot]

Error: Parsing nominate command in comment failed: ...'t nominate' | error: no team specified at >| ''...

Please let @rust-lang/release know if you're having trouble with this bot.

[rustbot]

Error: The feature nominate is not enabled in this repository.
To enable it add its section in the triagebot.toml in the root of the repository.

Please let @rust-lang/release know if you're having trouble with this bot.

[manuthambi]

Extra validation was added for situations where attributes were ignored (in this case it was #80641). Previous versions issued an unused-attribute warning.

1.50 didn't warn I think. I haven't specifically configured unusued_attributes in my project. Anyways, it is a good idea to lint/error for unused attributes.

If you want to align an array, I think you can wrap it in a struct with #[repr(align(8))] struct Arr([usize; 20]).

That works, if the compiler guarantees that there will not be any padding at the beginning of Arr struct. Does it?

[scottmcm]

I am not entirely sure whether the attribute actually did the alignment in 1.50, or the compiler just ignored it.

It just ignored it: https://rust.godbolt.org/z/4rso564PP

(Note also the warning for "#[warn(unused_attributes)] on by default".)

It's my understanding that this is known and expected fallout from fixing compiler bugs where it had previously forgotten to run validation in certain positions.