Regression: Fail to evaluate static slice with reference to internally-mutable data

[nvzqz]

Code

I tried this code:

use std::sync::OnceLock;

static LAZY_INIT: OnceLock<u32> = OnceLock::new();

static LAZY_INIT_REF: &[&OnceLock<u32>] = &[&LAZY_INIT];

I expected to see this happen: it compiles with LAZY_INIT being reachable through LAZY_INIT_REF.

Instead, this happened:

error[E0080]: it is undefined behavior to use this value
 --> src/main.rs:5:1
  |
5 | static LAZY_INIT_REF: &[&OnceLock<u32>] = &[&LAZY_INIT];
  | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ constructing invalid value at .<deref>[0]: encountered reference to mutable memory in `const`
  |
  = note: The rules on what exactly is undefined behavior aren't clear, so this check might be overzealous. Please open an issue on the rustc repository if you believe it should not be considered undefined behavior.
  = note: the raw bytes of the constant (size: 8, align: 8) {
              ╾─────alloc4<imm>─────╼                         │ ╾──────╼
          }

note: erroneous constant encountered
 --> src/main.rs:5:43
  |
5 | static LAZY_INIT_REF: &[&OnceLock<u32>] = &[&LAZY_INIT];
  |                                           ^^^^^^^^^^^^^

Version it worked on

It most recently worked on: nightly-2024-02-10

Version with regression

nightly-2024-02-11

rustc --version --verbose:

rustc 1.78.0-nightly (6cc484351 2024-02-10)
binary: rustc
commit-hash: 6cc4843512d613f51ec81aba689180c31b0b28b6
commit-date: 2024-02-10
host: aarch64-apple-darwin
release: 1.78.0-nightly
LLVM version: 17.0.6

@rustbot modify labels: +regression-from-stable-to-nightly +A-const-eval +T-compiler -regression-untriaged

[nvzqz]

Like #120949, this issue is not contrived since I also use this technique in Divan in code generation for benchmarks of generic functions, specifically for caching BenchOptions via OnceLock. The simple example I gave is just a minimal reproduction of the issue. This appears to happen with any internally-mutable types such as AtomicUsize.

[oli-obk]

Yea, this is a duplicate of #120949

[nvzqz]

@oli-obk this seems similar but the error messages suggest to me that it's not a duplicate. This is about referencing mutable data while the other is a cycle caused by self-referencing.

[oli-obk]

Let's keep it open an check