Bug in Rustc Version 1.77

[jpx40]

Hey, i noticed an error in One of my deps, this error just started to get up on the Rust Nightly Version.
First Noticed on Version 1.77 2024-01-17

It was tested on Arm M2 MacBook and Arch Linux AMD x86_64.

The error stayed the same.

Repo: https://github.com/time-rs/time

File: https://github.com/time-rs/time/blob/main/time/src/parsing/parsed.rs

Line of the trait Parsed that fails in Line 116.

The Compiler fails to compile the Code of a function that mutates an byte Sequenz.

I tried this code:

 pub fn parse_item<'a>(
        &mut self,
        input: &'a [u8],
        item: &impl sealed::AnyFormatItem,
    ) -> Result<&'a [u8], error::ParseFromDescription> {
        item.parse_item(self, input)
    }

I expected to see this happen: explanation

Instead, this happened: explanation

Meta

rustc --version --verbose:

rustc 1.77.0-nightly (7ffc697ce 2024-01-24)
binary: rustc
commit-hash: 7ffc697ce10f19447c0ce338428ae4b9bc0c041c
commit-date: 2024-01-24
host: aarch64-apple-darwin
release: 1.77.0-nightly
LLVM version: 17.0.6

Backtrace

failures:

---- meta::alignment stdout ----
thread 'meta::alignment' panicked at time/../tests/meta.rs:74:5:
assertion `left == right` failed: alignment of `Parsed` was 8
  left: 16
 right: 8
stack backtrace:
   0: rust_begin_unwind
             at /rustc/7ffc697ce10f19447c0ce338428ae4b9bc0c041c/library/std/src/panicking.rs:647:5
   1: core::panicking::panic_fmt
             at /rustc/7ffc697ce10f19447c0ce338428ae4b9bc0c041c/library/core/src/panicking.rs:72:14
   2: core::panicking::assert_failed_inner
   3: core::panicking::assert_failed
             at /rustc/7ffc697ce10f19447c0ce338428ae4b9bc0c041c/library/core/src/panicking.rs:297:5
   4: tests::meta::alignment
             at ./../tests/meta.rs:74:5
   5: tests::meta::alignment::{{closure}}
             at ./../tests/meta.rs:27:15
   6: core::ops::function::FnOnce::call_once
             at /rustc/7ffc697ce10f19447c0ce338428ae4b9bc0c041c/library/core/src/ops/function.rs:250:5
   7: core::ops::function::FnOnce::call_once
             at /rustc/7ffc697ce10f19447c0ce338428ae4b9bc0c041c/library/core/src/ops/function.rs:250:5



failures:
    meta::alignment

test result: FAILED. 1399 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.05s

error: test failed, to rerun pass `--test tests`
test run_with_all_features ... FAILED

failures:

---- run_with_all_features stdout ----
Error: Error(ExitStatus(unix_wait_status(25856)))

[ehuss]

Can you say more about why this appears to be a new issue? From what I can tell, this test fails on all versions of Rust.

The test appears to be x86-specific. That is, on ARM, the alignment of a 128-bit value is 16, not 8 as it is on x86_64 architectures.

[jpx40]

Can you say more about why this appears to be a new issue? From what I can tell, this test fails on all versions of Rust.

The test appears to be x86-specific. That is, on ARM, the alignment of a 128-bit value is 16, not 8 as it is on x86_64 architectures.

Hey, sry i am not an expert in the byte spectrum. The error first appeared on my x86_64 Linux Machine. Later i tested the Code again on my Arm Notebook. A friend of did test the Code on x86_64 Intel too to close the circle.

Yeah the bug seems to be plattform independent.

[jpx40]

Can you say more about why this appears to be a new issue? From what I can tell, this test fails on all versions of Rust.

The test appears to be x86-specific. That is, on ARM, the alignment of a 128-bit value is 16, not 8 as it is on x86_64 architectures.

sry, i have to sleep, i will test again to more once. again on x86_64 machine to see if the are different Values. But the Code failed on the same Line.

[jpx40]

Can you say more about why this appears to be a new issue? From what I can tell, this test fails on all versions of Rust.

The test appears to be x86-specific. That is, on ARM, the alignment of a 128-bit value is 16, not 8 as it is on x86_64 architectures.

I have no glue. It was an dependency my Code, just figured out which crate failed and started to test the Code of time-rs against different rust Versions. It seems it still works in 1.75 and 1.76. worked (Sry to say it was tested on x86_64). 1.77 failed compiling

[ehuss]

Ah, I see what you are saying now. This was an intentional change as part of #116672 to update the alignment of 128-bit types on x86.

[asquared31415]

Additionally, the alignments of integer types are explicitly not guaranteed.

Most primitives are generally aligned to their size, although this is platform-specific behavior.

source

[jpx40]

I will send the backtrace of my own code later.

[Noratrieb]

Thank you for the report, but this is an expected change as mentioned above. You need to update the test (or remove it, I'm not sure why it would be useful to have).
Closing.

[asquared31415]

It's occasionally useful to have layout tests to ensure that you don't accidentally change layout of something critical. Rustc even has many of these for some of its most commonly used types. But occasionally something like this happens which is out of your control and you just have to update the test and move on.

[jpx40]

It's occasionally useful to have layout tests to ensure that you don't accidentally change layout of something critical. Rustc even has many of these for some of its most commonly used types. But occasionally something like this happens which is out of your control and you just have to update the test and move on.

Ok, unfortunately I couldn't reconstruct the error in my own Code