Test failures in tests/ui/panics on aarch64-apple-darwin

[Zalathar]

Using an unmodified profile = "compiler" build on aarch64 macOS (14.7.1), I see the following tests fail from ./x test tests/ui/panics:

failures:
    [ui] tests/ui/panics/issue-47429-short-backtraces.rs
    [ui] tests/ui/panics/runtime-switch.rs
    [ui] tests/ui/panics/short-ice-remove-middle-frames-2.rs
    [ui] tests/ui/panics/short-ice-remove-middle-frames.rs

All of the failure diffs look something like this:

diff of run.stderr:

1	thread 'main' panicked at $DIR/issue-47429-short-backtraces.rs:24:5:
2	explicit panic
3	stack backtrace:
-	   0: std::panicking::begin_panic
+	   0: begin_panic<&str>
5	   1: issue_47429_short_backtraces::main
6	note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.
7	


The actual run.stderr differed from the expected run.stderr.

I see the same failures with --stage=1 and --stage=2.

---

Not sure if this is a test output normalization problem, or an actual bug.

Discussion elsewhere indicates that this might be related to #123244.

[Zalathar]

I'm seeing the same failures with all of:

rust.new-symbol-mangling = true
rust.new-symbol-mangling = false
# rust.new-symbol-mangling not set

[jieyouxu]

I cannot repro on x86_64 linux, x86_64 msvc, aarch64 linux, tagging aarch64 + apple unless someone else can repro this on a different environment.

[Zalathar]

I can also confirm that for me the tests pass before #123244, and fail after it.

[jieyouxu]

#123244 added some additional normalizations, FTR

[Mark-Simulacrum]

We probably(?) need some extra normalization in the tests. The symbol in question (begin_panic) is probably found a bit differently by backtrace on this platform?

That said, we run these tests in CI for this platform, so it's not clear whether it's some local delta as well. Perhaps you have e.g. lld enabled or some other optimization setting that's not set in CI?

[madsmtm]

I can reproduce it on my machine too (also using macOS 14.7.1), on both Aarch64 and x86_64 (using Rosetta), so that's at least weak evidence that it's not something environment specific.

Not sure what the underlying issue is yet though.

[Zalathar]

I can make the failures go away by setting rust.debuginfo-level = 1, and then make them appear again by setting rust.debuginfo-level = "line-tables-only", which is the compiler profile default.

That helps to explain why it doesn't show up on CI.

[Zalathar]

Fixed by adding more normalizations in #134781.