Can put non-Sync data in extern statics

[japaric]

This is not allowed:

static FOO: Cell<i32> = Cell::new(0);

error[E0277]: the trait bound `std::cell::Cell<i32>: std::marker::Sync` is not satisfied
 --> src/lib.rs:5:1
  |
5 | static FOO: Cell<i32> = Cell::new(0);
  | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ the trait `std::marker::Sync` is not implemented for `std::cell::Cell<i32>`
  |
  = note: `std::cell::Cell<i32>` cannot be shared between threads safely
  = note: shared static variables must have a type that implements `Sync`

But this is allowed:

use std::cell::Cell;

extern "C" {
    static FOO: Cell<i32>;
}

warning: found struct without foreign-function-safe representation annotation in foreign module, consider adding a #[repr(C)] attribute to the type
 --> src/lib.rs:4:17
  |
4 |     static FOO: Cell<i32>;
  |                 ^^^^^^^^^
  |
  = note: #[warn(improper_ctypes)] on by default

Using extern "C" doesn't suddenly make FOO safe to share accross threads so this should become an error, IMO. Just like #35112.

cc @nikomatsakis

[retep998]

As a result of #35112, extern statics currently require unsafe to access so this isn't a safety hole. Also, extern statics are often used to reference raw pointers which are definitely !Sync, so changing this to be an error would be a rather huge breaking change.

In my opinion this is working as intended.

[japaric]

As a result of #35112, extern statics currently require unsafe to access so this isn't a safety hole

I agree.

Also, extern statics are often used to reference raw pointers which are definitely !Sync

There's a pattern for such thing:

pub struct NotSyncAtAll<T>(*mut T);

unsafe impl<T> Sync for NotSyncAtAll<T> {}

extern "C" {
    static FOO: NotSyncAtAll<u8>;
}

But if FOO is still unsafe to access after that then maybe that doesn't buy you much. And I don't think extern statics can't ever be safe in the general case as there's no type safety; you can't bind any type to FOO.

so changing this to be an error would be a rather huge breaking change.

Well, that's one reason to leave things as they are.

[nikomatsakis]

I'm also inclined to think this is working as intended.

[japaric]

Agree that the change doesn't seem like it would buy anything and it would just result in breakage so closing.