Wide raw pointer casts let me assert marker trait bounds but not lifetime bounds

[RalfJung]

The following code compiles just fine:

trait Trait {}

fn assert_send(ptr: *mut dyn Trait) -> *mut (dyn Trait + Send) {
    ptr as _
}

This basically unsafely asserts that our underlying type is actually Send.

However, strangely, the following code is rejected:

fn assert_static<'a>(ptr: *mut (dyn Trait + 'a)) -> *mut (dyn Trait + 'static) {
    ptr as _
}

This does basically the same as the above, it unsafely assertions that our underlying type satisfies the 'static bound. So I would expect both of these to be accepted. But somehow the borrow checker is getting in the way and says something about variance:

error: lifetime may not live long enough
 --> src/lib.rs:8:5
  |
7 | fn assert_static<'a>(ptr: *mut (dyn Trait + 'a)) -> *mut (dyn Trait + 'static) {
  |                  -- lifetime `'a` defined here
8 |     ptr as _
  |     ^^^^^^^^ returning this value requires that `'a` must outlive `'static`
  |
  = note: requirement occurs because of a mutable pointer to `dyn Trait`
  = note: mutable pointers are invariant over their type parameter
  = help: see <https://doc.rust-lang.org/nomicon/subtyping.html> for more information about variance
help: consider changing the trait object's explicit `'static` bound to the lifetime of argument `ptr`
  |
7 | fn assert_static<'a>(ptr: *mut (dyn Trait + 'a)) -> *mut (dyn Trait + 'a) {
  |                                                                       ~~
help: alternatively, add an explicit `'static` bound to this reference
  |
7 | fn assert_static<'a>(ptr: *mut (dyn Trait + 'static)) -> *mut (dyn Trait + 'static) {
  |                           ~~~~~~~~~~~~~~~~~~~~~~~~~~

To work around this, one has to use a transmute, and that's always a bad sign. This even happens in the standard library.

Cc @rust-lang/types

[lcnr]

I expect the following to happen here:

• the cast works in HIR typeck
• during MIR building, we throw out identity casts. This looks at the types returned by HIR typeck which have erased regions
• this means the MIR used by borrowck doesn't have cast

I think this should be fixable 🤔

[lukas-code]

related: #113257

[RalfJung]

Oh interesting, so this fails even when the pointers are not wide -- yes that definitely looks like a bug to me.

Closing as a duplicate of #113257.