`improper_ctypes` lint triggers on valid `#[repr(transparent)]` types

[madsmtm]

Code

#[repr(C)]
pub struct Foo {
    inner: [u8; 0]
}

extern "C" {
    // Lint does not trigger, as expected
    pub fn foo(_: &Foo);
}

#[repr(transparent)]
pub struct Bar(Foo);

extern "C" {
    // Lint triggers unexpectedly
    pub fn bar(_: &Bar);
}

Current output

warning: `extern` block uses type `Bar`, which is not FFI-safe
  --> <source>:16:19
   |
16 |     pub fn bar(_: &Bar);
   |                   ^^^^ not FFI-safe
   |
   = note: this struct contains only zero-sized fields
note: the type is defined here
  --> <source>:12:1
   |
12 | pub struct Bar(Foo);
   | ^^^^^^^^^^^^^^
   = note: `#[warn(improper_ctypes)]` on by default

warning: 1 warning emitted

Desired output

The lint should not fire

Rationale and extra context

Reproducible from Rust 1.57.0 (where #[repr(transparent)] on a zero-sized struct was made valid) to the current nightly-2023-08-31.

[asquared31415]

I find it odd that this doesn't lint on Foo, it contains only zero sized fields too.

[madsmtm]

If it's an easier pill to swallow, it also lints on *const Bar but not on *const Foo.

Note that using zero-sized types to represent opaque/extern types like this is a documented pattern, so the lint should definitely not fire there.

[surechen]

@rustbot claim

[madsmtm]

Actually, I just saw this Pre-RFC about #[repr(transparent)] to ZSTs, so maybe this pattern was not actually fully specified?

[surechen]

Actually, I just saw this Pre-RFC about #[repr(transparent)] to ZSTs, so maybe this pattern was not actually fully specified?
Hi, Thanks very much for finding the informations. So the current errors reported here is consistent with the reference ?

[surechen]

I find that ignoring the warning conditions for the example seems easy to implement, but there doesn't seem to be a definite conclusion here. I think we should continue the discussion in this issue rust-lang/rust#100954