I tried this code:

use std::mem::transmute;

#[repr(align(16))]
struct Wrap<T>(T);

fn foo<T: ?Sized>(x: Wrap<*const T>) -> *const T {
    unsafe { transmute(x) }
}

fn main() {
    foo(Wrap(std::ptr::null::<i32>()));
    println!("done");
}

The above code shouldn't compile, since Wrap<*const i32> and *const i32 don't have the same size, and transmute is supposed to check the sizes of the two types to be equal at compile time.

Instead, the code compiles and causes a segmentation fault at run time.

The generated assembly is particularly nonsensical:

example::foo::<i32>:
        mov     byte ptr [rax], 1
        ret

Possibly related to #101084. cc @RalfJung @oli-obk

Meta

Reproducible on the playground with version 1.97.0-nightly (2026-04-15 e8e4541ff19649d95afa)