`#[naked]` attribute excludes use of `#[repr(align(N))]`

[1millibyte]

I tried this code:

#[repr(align(4))]
#[naked]
extern "C" fn handle_trap_entry() -> !
{ … }

I expected to see this happen:
Code compiling just fine

Instead, this happened:

error[E0736]: attribute incompatible with `#[naked]`
  --> kernel/src/arch/riscv64/trap.rs:30:1
   |
30 | #[repr(align(4))]
   | ^^^^^^^^^^^^^^^^^ the `repr` attribute is incompatible with `#[naked]`
31 | #[naked]
   | -------- function marked with `#[naked]` here

Meta

rustc --version --verbose:

rustc 1.82.0-nightly (a32d4a0e8 2024-08-21)
binary: rustc
commit-hash: a32d4a0e822a29a6682e08b75a8df4c29c7fa9f1
commit-date: 2024-08-21
host: x86_64-unknown-linux-gnu
release: 1.82.0-nightly
LLVM version: 19.1.0

It is a regression, as this was originally supported (last tried on):

rustc 1.81.0-nightly (6be96e386 2024-07-09)
binary: rustc
commit-hash: 6be96e3865c4e59028fd50396f7a46c3498ce91d
commit-date: 2024-07-09
host: x86_64-unknown-linux-gnu
release: 1.81.0-nightly
LLVM version: 18.1.7

Backtrace

N/A

[workingjubilee]

huh, wild. did it actually have the desired codegen properties?

[jdonszelmann]

this combines two unstable features, repr on functions (#82232) and naked functions (#90957), but the fix should be easy by adding it to this allowlist. It seems an artificial limitation, with the allowlist not including repr since that's not a well known attribute on functions. it looks like it should work

[jdonszelmann]

(I just wrote a test for it, if it works I'll make a PR if that's okay)

[folkertdev]

some context: #127853 introduced the allowlist, and this attribute was left out because we were unsure if it would come up. Clearly it has, so this seems fine to allow.