#[repr(C)] and Drop causes undefined behavior in FFI code

[mahkoh]

extern {
    fn f(x: *const X);
}

#[repr(C)]
struct X {
    i: i32,
}

impl Drop for X {
    fn drop(&mut self) { }
}

fn main() {
    let x = [X { i: 0 }, X { i: 0 }];
    println!("{}", std::mem::size_of_val(&x));
    unsafe { f(x.as_ptr()); }
}

Prints 16.

[huonw]

It seems like the improper_ctypes lint should check for repr(C) on the RHS of a Drop impl. "warning: implementing Drop changes the representation of the type, conflicting with repr(C)".

(It should presumably also check for #[unsafe_no_drop_flag] and not warn if it exists.)

[mahkoh]

This has to be fixed before 1.0.

[pnkfelix]

nominating; we are unlikely to implement non-zeroing dynamic drop before the 1.0 release (and certainly not before the 1.0 beta). So we probably need to address tis in some way, perhaps a lint as @huonw suggests.

Nominating for 1.0 polish.

[brson]

Adding a lint is easy.

[nikomatsakis]

This is to some extent a dup of #5016

[pnkfelix]

We won't object to someone adding a lint (which would be temporary assuming nonzeroing drop lands eventually).

But it also wont block the release.

P-high, not 1.0.