Leave promoteds untainted by errors when borrowck fails

[tmiasko]

Previously, when borrowck failed it would taint all promoteds within the MIR body. An attempt to evaluated the promoteds would subsequently fail with spurious "note: erroneous constant used". For example:

...
note: erroneous constant used
 --> tests/ui/borrowck/tainted-promoteds.rs:7:9
  |
7 |     a = &0 * &1 * &2 * &3;
  |         ^^

note: erroneous constant used
 --> tests/ui/borrowck/tainted-promoteds.rs:7:14
  |
7 |     a = &0 * &1 * &2 * &3;
  |              ^^

note: erroneous constant used
 --> tests/ui/borrowck/tainted-promoteds.rs:7:19
  |
7 |     a = &0 * &1 * &2 * &3;
  |                   ^^

note: erroneous constant used
 --> tests/ui/borrowck/tainted-promoteds.rs:7:24
  |
7 |     a = &0 * &1 * &2 * &3;
  |                        ^^

Borrowck failure doesn't indicate that there is anything wrong with promoteds. Leave them untainted.

Fixes #110856.

[rustbot]

r? @lcnr

(rustbot has picked a reviewer for you, use r? to override)

[rustbot]

Some changes occurred to MIR optimizations

cc @rust-lang/wg-mir-opt

[compiler-errors]

Yeah, I have no idea why I made the tainted_by_errors propagate into the promoted constant in #93691.

I don't really think there are ways of triggering a (meaningful) borrowck error on a promoted constant anyways, though happy to be disproven by an example...

I checked, it also gets rid of the "erroneous constant" note at the end of:

fn main() {
    println!("{}", {
        let x: &str;
        x
    });
}

error[E0381]: used binding `x` isn't initialized
 --> /home/gh-compiler-errors/test.rs:4:9
  |
3 |         let x: &str;
  |             - binding declared here but left uninitialized
4 |         x
  |         ^ `x` used here but it isn't initialized
  |
help: consider assigning a value
  |
3 |         let x: &str = todo!();
  |                     +++++++++

note: erroneous constant used
 --> /home/gh-compiler-errors/test.rs:2:14
  |
2 |     println!("{}", {
  |              ^^^^

r=me when green

[matthewjasper]

There's the error in #71587, not sure if that counts as "meaningful" or not.

[compiler-errors]

@matthewjasper: Sorry, by "meaningful", I guess I should've been more clear. What I mean is borrowck errors that are likely to cause spurious const-eval errors or ICEs. It doesn't seem like that's the case with the class of issues you fixed in #71587, though, and it doesn't seem like any other existing UI tests were affected by this change, so I consider this PR to be pretty low-risk. The only possible issue is that we might see are spurious const eval errors or ICEs from no longer tainting these promoteds, thus leading to them being evaluated...

@bors r+

[bors]

📌 Commit c678acd has been approved by compiler-errors

It is now in the queue for this repository.