adds better error message for temporary value does not live long enough

[Hiryxx]

When &const { expr } is borrowed for 'static but expr has interior mutability (is not Freeze), emit E0492 instead
of E0716

Closes #154382

[rustbot]

r? @JohnTitor

rustbot has assigned @JohnTitor.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: compiler
• compiler expanded to 69 candidates
• Random selection from 9 candidates

[JohnTitor]

View changes since the review

I don't think it's a good idea to duplicate the message with

rust/compiler/rustc_const_eval/src/errors.rs

Line 296 in 1948ee1

#[diag("interior mutable shared borrows of temporaries that have their lifetime extended until the end of the program are not allowed", code = E0492)]

[Hiryxx]

Yes, I totally agree. Though the borrow checker does not depend on rust_const_eval, what do you do in these situations? Should I move that under rustc_middle so that both can see that?

[Hiryxx]

@JohnTitor I was trying with that but rustc_middle has some errors definitions that are only used inside that module, so I am not sure it would be a good fit. I could also make the borrowchecker depend on const_evals but it seems an overkill for a single diagnostic structure, what do you think? Do you have a place in mind where you would put that struct without needing an "invasive" change.

[JohnTitor]

Could you put FIXME on both sides so that we don't forget to sync and can clean up later?

[Hiryxx]

Sure, thanks

[JohnTitor]

View changes since the review

Please put the issue number in a file title and the link in the comment for reference.

[JohnTitor]

View changes since the review

I guess we could expand the fix for a similar case where promotion failed because the value is not Freeze, like:

use std::cell::Cell;

struct Mutable(Cell<u32>);
impl Mutable {
  const fn new(a: u32) -> Self { Self(Cell::new(a)) }
}

fn foo() -> &'static Mutable {
  &const { Mutable::new(0) }
}

fn main() {}

The current only addresses the case issue mentioned, I guess it's kinda ad hoc. These cases should have the same diagnostics for consistency.

[JohnTitor]

@Hiryxx I don't think this was addressed properly, consider a case like:

fn main() {
    let _: &'static mut _ = &mut const { Cell::new(0) };
}

[JohnTitor]

Because of this return, it could drop a suggestion for a case like:

fn main() {
    let r;
    r = &const { Cell::new(0) };
    r.set(1);
}

I'd keep it.

View changes since the review

[rustbot]

Some changes occurred to the CTFE machinery

cc @RalfJung, @oli-obk, @lcnr

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[Hiryxx]

I had to pull the detection into a span variable up top and reuse it as a flag, this makes it a little verbose for the error output but I couldn't find a better since the alternatives where worse. Let me know what you think @JohnTitor.

[rust-bors]

☔ The latest upstream changes (presumably #159170) made this pull request unmergeable. Please resolve the merge conflicts by rebasing.