New lint: type which is not FFI-safe

[ratijas]

Recent versions of Rust (1.46.0 per the moment of writing) started showing the warning when compiling the project: warning: extern fn uses type std::cell::RefCell<dyn QObject>, which is not FFI-safe. It concerns exactly two functions: RustObject_metaObject and RustObject_destruct.

Full text of warnings:

warning: `extern` fn uses type `std::cell::RefCell<dyn QObject>`, which is not FFI-safe
   --> qmetaobject/src/lib.rs:569:51
    |
569 | pub unsafe extern "C" fn RustObject_metaObject(p: *mut RefCell<dyn QObject>) -> *const QMetaObject {
    |                                                   ^^^^^^^^^^^^^^^^^^^^^^^^^ not FFI-safe
    |
    = note: `#[warn(improper_ctypes_definitions)]` on by default
    = help: consider adding a `#[repr(C)]` or `#[repr(transparent)]` attribute to this struct
    = note: this struct has unspecified layout

warning: `extern` fn uses type `std::cell::RefCell<dyn QObject>`, which is not FFI-safe
   --> qmetaobject/src/lib.rs:575:49
    |
575 | pub unsafe extern "C" fn RustObject_destruct(p: *mut RefCell<dyn QObject>) {
    |                                                 ^^^^^^^^^^^^^^^^^^^^^^^^^ not FFI-safe
    |
    = help: consider adding a `#[repr(C)]` or `#[repr(transparent)]` attribute to this struct
    = note: this struct has unspecified layout

What do we have here is *mut RefCell<dyn QObject>. "RefCell" part is unsized (?Sized) which means 1) it holds any QObject data, 2) it can be stored and accessed only by some kind of "fat-pointer" reference — in this case it's *mut which takes 2 words (16 bytes on x86_64), clearly a trait object with vtable info attached.

Let's see why the warning showed up in the first place. As far as compiler concerned, *mut RefCell<dyn QObject> essentially is:

• a fat pointer (not stable API & ABI, but in practice quite reliable for the time being)...
• to a RefCell (no #[repr], two fields)...
• of a dyn QObject (unsized trait object, type unknown at compile time, unknown ABI).

Seems like a reasonable warning from rustc after all, as most regular Rust programmers should not mess with trait objects representation.

In my opinion, it would be reasonable to silence the warning in these specific functions. Also, it is possible that the issue would get resolved by the fix to #110.