#[repr(align(…))] should allow arbitrary constant expressions, not just integer literals

[comex]

For example:

const ALIGN_OF_FOO: usize = 32;
#[repr(align(ALIGN_OF_FOO))]
struct Foo;

Currently this produces an error:

error[E0552]: unrecognized representation hint
 --> src/main.rs:2:8
  |
2 | #[repr(align(ALIGN_OF_FOO))]
  |        ^^^^^^^^^^^^^^^^^^^

A potential use case is for opaque struct definitions.

I could submit this as a new RFC, but I can't think of any reason not to allow it or any ambiguity as to how it should work, so I'm hoping an issue report may be sufficient.

[newpavlov]

I also would love for this to be implemented. Currently I have to use a custom macro and static asserts to check that a correct alignment was indeed used, which is far from ideal.

[dancrossnyc]

Please please please I'd love this.

[sunshowers]

Another use case for this is to align types with respect to other types, similar to alignas(T) in C or C++.

#[repr(align(64))]
struct Foo;

#[repr(align(std::mem::align_of::<Foo>()))]
struct Bar;

[chrisnc]

Theelain crate implements a solution using a trait and associated type, for those interested. https://docs.rs/crate/elain/

Supporting this in the language would be great, though. C/C++ allow constant expressions in __attribute__((aligned())) which is also useful for some contexts where the alignment depends in a non-trivial way on the size of an object.