ptr-ptr cast does not work well with lifetimes (lifetime widening)

[arielb1]

STR

fn foo<'a>(v: *const &'a ()) {
    let v: *const &'static () = v as *const &'static ();
}

fn main() {}

Expected result

cast will be successful

Actual Result

<anon>:2:33: 2:34 error: mismatched types:
 expected `*const &'static ()`,
    found `*const &'a ()`
(lifetime mismatch) [E0308]
<anon>:2     let v: *const &'static () = v as *const &'static ();
                                         ^
<anon>:2:33: 2:34 help: see the detailed explanation for E0308
<anon>:1:30: 3:2 note: the lifetime 'a as defined on the block at 1:29...
<anon>:1 fn foo<'a>(v: *const &'a ()) {
<anon>:2     let v: *const &'static () = v as *const &'static ();
<anon>:3 }
note: ...does not necessarily outlive the static lifetime
error: aborting due to previous error
playpen: application terminated with error code 101

[steveklabnik]

Why would this cast work? 'a can be shorter than 'static, as the error says.

[bluss]

Raw pointer casts are normally not restricted. *const A as *const B is usally ok, without any checking of any compatibility between A and B.

[steveklabnik]

Ah, I see, thanks @bluss . Seems reasonable.

[ranma42]

Does Rust guarantee that pointers are correctly aligned? Would it be legal in a safe context to create a misaligned pointer?
Neither https://doc.rust-lang.org/book/raw-pointers.html nor https://doc.rust-lang.org/book/unsafe.html mention any constraint on alignment.

[bluss]

Currently, loads and writes through raw pointer dereferences assume the pointer is aligned (this is guarded by unsafe).

Box::new(t) creates a 0x1 pointer if t is zero size, even if it needs more than 1-alignment. For example if it is a [u64; 0] (size 0, align of 8). From the box you can thus get a misaligned reference, however it is one that can't be read/written through (can't read/write something of 0 size).

For guarantees, have to ask https://internals.rust-lang.org/c/language-design/unsafe-code-guidelines

[Rufflewind]

As @talchas on #rust pointed out to me, as seems to forbid lifetime-only changes, but if anything else changes it's totally on board:

fn foo<'a>(v: *const &'a ()) {
    let v: *const &'static () = v as *const () as *const &'static ();
}

[steveklabnik]

Triage: still reproduces.

[fmease]

Triage: No longer reproduces (fail→pass). We could just close this as completed but I'm actually not sure which PR(s) was responsible. Might try to bisect later.

[fmease]

Bisects to 1.75 / nightly-2023-10-27 / PR #113262.
See also issue #113257 which duplicates this one.

Closing as fixed by #113262. No additional tests needed.