libc-test panic: unsafe precondition(s) violated: hint::assert_unchecked must never be called when the condition is false

[JohnTitor]

libc-test now panics since 1.78.0-nightly (d44e3b95c 2024-02-09).

Code

repo: https://github.com/rust-lang/libc
example GHA run: https://github.com/rust-lang/libc/actions/runs/7852817358/job/21431727248

Not minimized yet.

gh repo clone rust-lang/libc
cd libc/libc-test
cargo test

Meta

rustc --version --verbose:

rust version 1.78.0-nightly (d44e3b95c 2024-02-09)

Error output

  thread 'main' panicked at library/core/src/panicking.rs:155:5:
  unsafe precondition(s) violated: hint::assert_unchecked must never be called when the condition is false
  note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
  thread caused non-unwinding panic. aborting.

Backtrace

<backtrace>

[asquared31415]

Probably because the condition is now being checked in debug mode even without build-std: #120594

I am inclined to say this is a bug in the calling code.

[asquared31415]

this code is incorrect, it's calling get_unchecked on a slice with length 0 (because the vec had its length set to 0 and is coercing to a slice). Consider raw pointer manipulation with vec.as_mut_ptr() instead.

[JohnTitor]

Thanks for clarifying! Going to close as wontfix then (the linked code is quite old rustc_syntax and yes, it's time that we have to update our test method 😓).

[saethlin]

This is an old pattern that I rooted out of the compiler when I first worked on these checks about a year ago. We've also since stabilized Vec::spare_capacity_mut which is a more ergonomic way to fiddle with the region between len and capacity.

[scottmcm]

Also, while blame says that line is 9 years old (👀), today this is documented

It’s UB to call .get_unchecked(len), even if you immediately convert to a pointer.

https://doc.rust-lang.org/nightly/std/primitive.slice.html#method.get_unchecked