[scheduled-gate:test.yml] CI is failing on main

[github-actions]

CI has a completed red post-merge run.

Current failure

• Run: 37140685651
• Head: 4e6e68ee0aa2529f4ee1f95e26ea0edad81b6a7f
• Trigger: push

Failing rows

• cargo-test / Run cargo test
• compiler-output-regression / Gate native-ABI proof compiler output
• compiler-output-regression / Gate native-region proof compiler output
• lint / File size limit
• lint / GC header constant consistency
• lint / Object-header shape-descriptor census
• lint / Raw-handle debt ratchet
• lint / Runtime GC-pointer holder custody audit
• lint / SSO string-unboxing inventory
• lint / String payload-access inventory
• lint / Thread-exit custody of address-keyed globals (#11471)
• lint / Unrooted-local shape ratchet
• main-gate / Require the plan and every planned job to have passed
• warnings / rustc warnings (host-compatible, all targets)
• windows-build / GC structural audits (Windows)

Delta from the previous failed run

Previous failure: c17090892e2a

New failing rows:

• None.

Rows that recovered:

• gc-call-effects (macos-aarch64, macos-14) / Build the archives and compare the committed table
• gc-call-effects (windows-x86_64, ubuntu-latest) / Build the archives and compare the committed table
• gc-call-effects-linux / Committed table vs. the archives

Last green main result

No green post-merge run was found in the last 100 results.

Maintained automatically by gate-failure-watch.yml for test.yml. Repeated failures update this issue; the next green run closes it.

[proggeramlug]

Diagnosis for the cargo-test / Run cargo test row, which this tracker names but does not localise.

The job is red on exactly one test:

test standalone_regression ... FAILED
thread 'standalone_regression' (63617) panicked at crates/perry/tests/child_output_late_iterator.rs:15:5:
Error: node-failed-spawn: 1 (signal none)
test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 261.62s
error: test failed, to rerun pass `-p perry --test child_output_late_iterator`

Everything else in the job passes; this single failure is what turns cargo-test red, and cargo-test is a pr-gate input, so every open PR inherits it.

The test is a thin wrapper that shells out:

Command::new("node")
    .arg(root.join("scripts/test-child-output-late-iterator.mjs"))
    .env("PERRY_BIN", env!("CARGO_BIN_EXE_perry"))
    .env("PERRY_TEST_BUILD_RUNTIME", "1")

node-failed-spawn is the driver's own error text, so the failure is inside scripts/test-child-output-late-iterator.mjs rather than in the Rust test — the assertion at line 15 only surfaces the driver's non-zero exit.

Not a recent regression. cargo-test is red on every completed main run I checked back through 33690c5635 (2026-09-16 13:45) — that is 33690c5635, 99363beb1a (×2), 872ce07e24 (×2), f9c3e51e93, fdc437f966 (×2), ff19bd536a, fe0cc15b0f. The test's own last two commits (9143beb74e, 8f51d2b3f7) date to 2026-09-11, before that window, so this is not caused by the merge trains that landed in it.

Separately, and for the record on this tracker: the gap-suite rows are now green on main. All three sweep shards failed on 872ce07e24 and pass from fdc437f966 onward, which is #10387 landing via merge train 206.

[proggeramlug]

Correction to my earlier comment: I overstated the blast radius. Open PRs do not generally inherit this failure.

I wrote that cargo-test feeds pr-gate "so every open PR inherits it". That is wrong. scripts/ci_plan.py:334 sets

"cargo_test_scope": "pr" if tier == "pr" and event == "pull_request" else "full",

so the PR tier runs cargo-test scoped to the diff, and only the sweep/full tiers run it whole. ci_plan.py even self-tests this (check("core PR: cargo-test scoped", core["cargo_test_scope"] == "pr")). A PR whose diff does not name crates/perry/tests/child_output_late_iterator.rs never runs that test.

Confirmed empirically: #10349 is green on cargo-test right now, against the same main where the sweep-tier job is red. #10285 is green on it too, despite touching crates/perry/tests/.

So the accurate statement is narrower: the post-merge sweep on main is red, and main-gate fans that in — which is what this tracker is for — but individual PRs are unaffected unless their diff pulls the test into scope. My apologies for the noise; the diagnosis of which test and why stands, only the impact claim was inflated.

The rest of the earlier comment is unchanged: the failure is standalone_regression in child_output_late_iterator.rs with Error: node-failed-spawn: 1 (signal none), the Rust test is a thin wrapper around node scripts/test-child-output-late-iterator.mjs, and it is red on every completed main run back through 33690c5635, which rules out the recent merge trains.