Taking a shared reference of an array suppresses the unconditional_panic lint

[JanBeh]

I tried this code:

fn main() {
    let xs: [i32; 5] = [1, 2, 3, 4, 5];
    let _ = &xs; // this line suppresses the `unconditional_panic` lint
    let _ = xs[7];
}

I expected to see this happen:

error: this operation will panic at runtime
 --> src/main.rs:4:13
  |
4 |     let _ = xs[7];
  |             ^^^^^ index out of bounds: the length is 5 but the index is 7
  |
  = note: `#[deny(unconditional_panic)]` on by default

Instead, this happened:

thread 'main' panicked at 'index out of bounds: the len is 5 but the index is 7', src/main.rs:4:13

Apparently, taking a shared reference of xs will suppress the unconditional_panic lint. This seems like a false negative. While false negatives can't be avoided entirely, this case seems to be avoidable as taking a shared reference shouldn't change the type of xs (which is [i32; 5]) or modify any data in this example.

Note that

fn main() {
    let xs: [i32; 5] = [1, 2, 3, 4, 5];
    //let _ = &xs;
    let _ = xs[7];
}

will trigger the lint properly.

Meta

rustc --version --verbose:

rustc 1.63.0-nightly (43347397f 2022-06-23)
binary: rustc
commit-hash: 43347397f7c5ca9a670a3bb3890c7187e24a52ab
commit-date: 2022-06-23
host: x86_64-unknown-freebsd
release: 1.63.0-nightly
LLVM version: 14.0.5

[JanBeh]

See also this thread on URLO.

[TheWastl]

@rustbot claim

[poliorcetics]

Note this is a problem because the book has an example which is supposed to fail to compile but this bug makes it fail at runtime instead

[kmdreko]

In case no one has dug into it yet, the error is suppressed in 1.61 but not in 1.60 and before:

$ cargo +1.61 build
   Compiling mycrate v0.1.0 (/rust-tests)
    Finished dev [unoptimized + debuginfo] target(s) in 0.74s

$ cargo +1.60 build
   Compiling mycrate v0.1.0 (/rust-tests)
error: this operation will panic at runtime
   --> src/main.rs:129:13
    |
129 |     let _ = xs[7];
    |             ^^^^^ index out of bounds: the length is 5 but the index is 7
    |
    = note: `#[deny(unconditional_panic)]` on by default

error: could not compile `mycrate` due to previous error
warning: build failed, waiting for other jobs to finish...
error: build failed

[eggyal]

@rustbot label +regression-from-stable-to-stable

[apiraino]

WG-prioritization assigning priority (Zulip discussion).

@rustbot label -I-prioritize +P-medium

[TheWastl]

This is caused by #94934. It only ever worked because the lint ran after optimizations.

I made a fix here. It specializes the length of arrays as constant, even if the array isn't. I'm going to open a PR when I have time, probably tomorrow.

In the long term, I'd like to implement some more extensive tracking of "what parts of a value are constant". For example, the compiler currently assumes that anything that had its reference taken can change, but actually, this can only happen inside UnsafeCell (as I understand it).