`PinCoerceUnsized` has strange `impl`s

[theemathas]

The PinCoerceUnsized type allows unsize-coercing through Pin.

I found that PinCoerceUnsized has the following impls in the standard library, which are on types that don't even implement Deref, and therefore are useless within a Pin.

unsafe impl<T: ?Sized> PinCoerceUnsized for *const T {}
unsafe impl<T: ?Sized> PinCoerceUnsized for *mut T {}
unsafe impl<T: PointeeSized> PinCoerceUnsized for NonNull<T> {}

unsafe impl<T: ?Sized> PinCoerceUnsized for UnsafeCell<T> {}
unsafe impl<T: ?Sized> PinCoerceUnsized for SyncUnsafeCell<T> {}
unsafe impl<T: ?Sized> PinCoerceUnsized for Cell<T> {}
unsafe impl<T: ?Sized> PinCoerceUnsized for RefCell<T> {}

unsafe impl<T: ?Sized, A: Allocator> PinCoerceUnsized for alloc::rc::Weak<T, A> {}
unsafe impl<T: ?Sized, A: Allocator> PinCoerceUnsized for alloc::sync::Weak<T, A> {}

I don't see any point on why these would be allowed. These impls allows some really strange code to compile in stable rust. For example, the following code compiles:

use std::cell::UnsafeCell;
use std::pin::Pin;

fn foo(x: Pin<UnsafeCell<&i32>>) -> Pin<UnsafeCell<&dyn Sync>> {
    x
}

fn bar(x: Pin<&i32>) -> Pin<*const dyn Send> {
    x
}

Notably, despite bar compiling, it is not allowed to coerce a Pin<&i32> to a Pin<*const i32>.

I'm not sure if these strange behaviors have any soundness implications. (Although the impl on the cell types worryingly lack a T: PinCoerceUnsized bound). But since they serve no purpose, can they be deleted?

@rustbot labels +A-pin +A-coercions

Meta

The code snippets are tested on the playground with version 1.91.0-nightly (2025-08-07 2fd855fbfc8239285aa2)

[theemathas]

Actually, even within the types that do implement Deref, there's this weirdness:

use std::pin::Pin;

// This compiles
fn works(x: Pin<&mut i32>) -> Pin<&dyn Send> {
    x
}

// This does not compile
fn fails(x: Pin<&mut i32>) -> Pin<&i32> {
    x
}

[Darksonn]

Hi I just saw this issue. I added these impls because otherwise the introduction of PinCoerceUnsized would have been a breaking change. I'm in favor of removing them.

[Darksonn]

With regards to the conversion Pin<&mut i32> to Pin<&i32>, please keep in mind that this isn't an unsizing coercion because i32 is a sized type. Unsizing coercions must turn a thin pointer into a wide pointer.