Re-enable assertions on macOS alt builds

[madsmtm]

These were previously disabled, in part for performance reasons, in part due to needing availability symbols __isPlatformVersionAtLeast and __isOSVersionAtLeast that compiler-builtins did not provide, see #62592 (comment) and #134275 (comment) for failed checks.

Since #138944 though, std now provides these symbols, so we should be able to re-enable LLVM assertions, debug assertions and overflow checks.

Fixes #59637.

try-job: *apple*

[madsmtm]

@bors try

[rust-bors]

💔 Test for beeae75 failed: CI. Failed jobs:

• try - dist-apple-various-alt (web logs, extended logs)

[madsmtm]

Hmm, might need a stage0 bump, waiting for that:
@rustbot blocked

EDIT: #146636

[madsmtm]

@rustbot author
@bors try

[rustbot]

Reminder, once the PR becomes ready for a review, use @rustbot ready.

[madsmtm]

@bors try cancel
I need to rebase first probably

[rust-bors]

Try build cancelled. Cancelled workflows:

• https://github.com/rust-lang/rust/actions/runs/18057828350

[madsmtm]

@bors try

[rust-bors]

💔 Test for 56e5725 failed: CI. Failed jobs:

• try - aarch64-apple-alt (web logs, extended logs)

[rust-bors]

☀️ Try build successful (CI)
Build commit: 2ec5766 (2ec5766d21d2e578d3ba452d7d12814b21e8140b, parent: ade84871f718ea20a6460d28e82290353b4bf3d2)

[madsmtm]

So, I think this works!
@rustbot ready

r? @Mark-Simulacrum since you're on t-infra and were assigned to #59637 in the past.

Also CC @jieyouxu, since you've tried to do this in the past.

[madsmtm]

I'm unsure if it has a performance impact, comparing it against the auto build on 959b450 gives:

	Previously	This PR
CI Run	Link	Link
dist-x86_64-apple	1h 40m	1h 56m
dist-apple-various	1h 5m	1h 5m
dist-aarch64-apple	2h 2m	1h 37m
aarch64-apple	1h 49m	2h 3m

Which feels like wayy too high a variance to really be able to say anything conclusive?

[rustbot]

This PR changes a file inside tests/crashes. If a crash was fixed, please move into the corresponding ui subdir and add 'Fixes #' to the PR description to autoclose the issue upon merge.

[nikic]

I'm a bit confused. This appears to enable assertions on a bunch of dist targets, but distributed toolchains are not supposed to enable assertions.

Should this change be limited to only aarch64-apple, which is a test target and as such should enable assertions?

View changes since this review

[madsmtm]

I'm a bit confused. This appears to enable assertions on a bunch of dist targets, but distributed toolchains are not supposed to enable assertions.

Right, the dist- builds set DEPLOY. Only the -alt builds get debug assertions, overflow checks and LLVM assertions in here. So this PR as-is only actually affects aarch64-apple, the rest still have all the assertions disabled.

That said, it this will allow others to enable -alt if they find they need to test with debug assertions. I tested an -alt build in this CI run, it currently passes.

[Mark-Simulacrum]

@bors r+ rollup=never

This seems reasonable to me. Thanks! I agree that the CI times don't seem affected (at least within noise reasonable to achieve from a single PR's measurements).

[bors]

📌 Commit 9878be7 has been approved by Mark-Simulacrum

It is now in the queue for this repository.

[bors]

⌛ Testing commit 9878be7 with merge f957826...

[bors]

☀️ Test successful - checks-actions
Approved by: Mark-Simulacrum
Pushing f957826 to master...

[github-actions]

What is this?

This is an experimental post-merge analysis report that shows differences in test outcomes between the merged PR and its parent PR.

Comparing c8905ea (parent) -> f957826 (this PR)

Test differences

Show 35 test diffs

Stage 2

• [codegen] tests/codegen-llvm/binary-heap-peek-mut-pop-no-panic.rs: pass -> ignore (ignored when std is built with debug assertions) (J0)
• [codegen] tests/codegen-llvm/mem-replace-big-type.rs: pass -> ignore (ignored when std is built with debug assertions) (J0)
• [codegen] tests/codegen-llvm/mem-replace-simple-type.rs: ignore (only executed when the architecture is x86_64 ((to not worry about usize differing))) -> ignore (ignored when std is built with debug assertions) (J0)
• [codegen] tests/codegen-llvm/range-loop.rs: pass -> ignore (ignored when std is built with debug assertions) (J0)
• [codegen] tests/codegen-llvm/slice-reverse.rs: ignore (only executed when the architecture is x86_64) -> ignore (ignored when std is built with debug assertions ((debug assertions prevent generating shufflevector))) (J0)
• [codegen] tests/codegen-llvm/swap-small-types.rs: ignore (only executed when the architecture is x86_64) -> ignore (ignored when std is built with debug assertions ((ptr::swap_nonoverlapping has one which blocks some optimizations))) (J0)
• [codegen] tests/codegen-llvm/vec-in-place.rs: pass -> ignore (ignored when std is built with debug assertions ((FIXME: checks for call detect scoped noalias metadata))) (J0)
• [codegen] tests/codegen-llvm/vec-shrink-panik.rs: pass -> ignore (ignored when std is built with debug assertions ((plain old debug assertions))) (J0)
• [codegen] tests/codegen-llvm/vec-with-capacity.rs: pass -> ignore (ignored when std is built with debug assertions) (J0)
• [codegen] tests/codegen-llvm/vecdeque-drain.rs: pass -> ignore (ignored when std is built with debug assertions ((FIXME: checks for call detect scoped noalias metadata))) (J0)
• [codegen] tests/codegen-llvm/vecdeque_no_panic.rs: pass -> ignore (ignored when std is built with debug assertions ((plain old debug assertions))) (J0)
• [crashes] tests/crashes/108428.rs: ignore (ignored if rustc wasn't built with debug assertions) -> pass (J0)
• [crashes] tests/crashes/118778.rs: ignore (ignored if rustc wasn't built with debug assertions) -> pass (J0)
• [crashes] tests/crashes/118784.rs: ignore (ignored if rustc wasn't built with debug assertions) -> pass (J0)
• [crashes] tests/crashes/120175.rs: ignore (ignored if rustc wasn't built with debug assertions) -> ignore (ignored when the target vendor is Apple ((raw-dylib doesn't work on Apple targets yet))) (J0)
• [crashes] tests/crashes/123862.rs: ignore (ignored if rustc wasn't built with debug assertions) -> pass (J0)
• [crashes] tests/crashes/130395.rs: ignore (ignored if rustc wasn't built with debug assertions) -> pass (J0)
• [crashes] tests/crashes/133965.rs: ignore (ignored if rustc wasn't built with debug assertions) -> pass (J0)
• [crashes] tests/crashes/134061.rs: ignore (ignored if rustc wasn't built with debug assertions) -> pass (J0)
• [crashes] tests/crashes/137514.rs: ignore (ignored if rustc wasn't built with debug assertions) -> pass (J0)
• [crashes] tests/crashes/139381.rs: ignore (ignored if rustc wasn't built with debug assertions) -> pass (J0)
• [crashes] tests/crashes/139387.rs: ignore (ignored if rustc wasn't built with debug assertions) -> pass (J0)
• [mir-opt] tests/mir-opt/pre-codegen/mem_replace.rs: pass -> ignore (ignored when std is built with debug assertions) (J0)
• [mir-opt] tests/mir-opt/pre-codegen/ptr_offset.rs: pass -> ignore (ignored when std is built with debug assertions ((precondition checks are under cfg(debug_assertions)))) (J0)
• [mir-opt] tests/mir-opt/pre-codegen/slice_iter.rs: pass -> ignore (ignored when std is built with debug assertions ((precondition checks on ptr::add are under cfg(debug_assertions)))) (J0)
• [ui] tests/ui/associated-consts/issue-93775.rs#current: pass -> ignore (ignored when rustc is built with debug assertions) (J0)
• [ui] tests/ui/associated-consts/issue-93775.rs#next: pass -> ignore (ignored when rustc is built with debug assertions) (J0)
• [ui] tests/ui/errors/emitter-overflow-bad-whitespace.rs: ignore (ignored if rustc wasn't built with debug assertions) -> pass (J0)
• [ui] tests/ui/mir/static-by-value-dyn.rs: ignore (ignored if rustc wasn't built with debug assertions) -> pass (J0)
• [ui] tests/ui/mir/unreachable-loop-jump-threading.rs: ignore (ignored if rustc wasn't built with debug assertions) -> pass (J0)
• [ui] tests/ui/print_type_sizes/niche-filling.rs: pass -> ignore (ignored when std is built with debug assertions ((debug assertions will print more types))) (J0)
• registry::reach_max_unpack_size: [missing] -> pass (J0)

Additionally, 3 doctest diffs were found. These are ignored, as they are noisy.

Job group index

• J0: aarch64-apple

Test dashboard

Run

cargo run --manifest-path src/ci/citool/Cargo.toml -- \
    test-dashboard f957826bff7a68b267ce75b1ea56352aed0cca0a --output-dir test-dashboard

And then open test-dashboard/index.html in your browser to see an overview of all executed tests.

Job duration changes

1. aarch64-apple: 5057.5s -> 9296.2s (83.8%)
2. dist-armhf-linux: 4979.7s -> 7862.3s (57.9%)
3. aarch64-gnu-llvm-20-2: 2690.6s -> 2166.5s (-19.5%)
4. pr-check-1: 1757.8s -> 1434.7s (-18.4%)
5. x86_64-gnu-llvm-20: 2792.4s -> 2384.4s (-14.6%)
6. x86_64-rust-for-linux: 3103.8s -> 2723.7s (-12.2%)
7. i686-gnu-2: 6382.0s -> 5607.5s (-12.1%)
8. x86_64-gnu-llvm-20-2: 5503.0s -> 6158.6s (11.9%)
9. pr-check-2: 2633.3s -> 2334.5s (-11.3%)
10. i686-gnu-1: 8167.3s -> 7354.5s (-10.0%)

How to interpret the job duration changes?

Job durations can vary a lot, based on the actual runner instance
that executed the job, system noise, invalidated caches, etc. The table above is provided
mostly for t-infra members, for simpler debugging of potential CI slow-downs.

[rust-timer]

Finished benchmarking commit (f957826): comparison URL.

Overall result: ❌ regressions - no action needed

@rustbot label: -perf-regression

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	0.0%	[0.0%, 0.0%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

Results (primary -1.6%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	0.5%	[0.5%, 0.5%]	1
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-3.6%	[-3.6%, -3.6%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-1.6%	[-3.6%, 0.5%]	2

Cycles

Results (primary 2.0%, secondary 2.4%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	2.0%	[2.0%, 2.0%]	1
Regressions ❌ (secondary)	2.4%	[2.4%, 2.4%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	2.0%	[2.0%, 2.0%]	1

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 470.874s -> 470.189s (-0.15%)
Artifact size: 387.66 MiB -> 387.63 MiB (-0.01%)

[madsmtm]

Okay, so this did change the timings fairly significantly:

aarch64-apple: 5057.5s -> 9296.2s (83.8%)

Is this within an acceptable threshold? Or should we revert this? Or maybe split the aarch64-apple CI step into two?

[nikic]

That result is probably not representative, as this change will have triggered an uncached LLVM rebuild.

[madsmtm]

Re perf regression, I think that's spurious, this is only affecting the macOS runners, which aren't perf-tested, and it shouldn't affect the dist- builds (which are the ones that perf would test, and those are still built without assertions).