Raise the harness timeout of the cargo_autoharness_fmt_impls test

[tautschnig]

Pass --harness-timeout 5m to script-based-pre/cargo_autoharness_fmt_impls, as the other autoharness tests (bounded, bounds, byte_str, c_str, formatter, wtf8) already do.

Why. On the macOS x86_64 runners (regression (macos-15-intel)), the LowerExp, UpperExp and Pointer harnesses of this test take 40–60 s, right at kani autoharness's default --harness-timeout of 60 s. When one of them times out, CBMC reports CBMC timed out instead of the expected failure, the test's Failed Checks: "lower exp" / "upper exp" / "pointer" line is missing, and the test fails. The same harnesses take about 10 s on Linux.

Evidence. I went through the logs of the regression (macos-15-intel) jobs between 2026-09-28 and 2026-09-30. 33 of them failed. This test failed in all 33, with 1–3 of those three harnesses timing out each time, and in 32 of them it was the only failing test. The one exception, #4903's merge-queue push run, also had cargo_autoharness_wtf8's len harness time out, at 5 m. On main pushes alone, the job failed 7 times out of 12 on 2026-09-30. The failures are independent of the changes under test:

• Today's automated toolchain upgrade, Automatic toolchain upgrade to nightly-2026-09-25 #4917: run 36663872235 (LowExp, UpExp, Ptr timed out; every other regression job passed)
• main pushes: 36654469481, 36656240395, 36657849257, 36659060566, 36670355214, 36674524840
• Merge queue: Re-enable the std_codegen test #4895 36650793954, Re-enable loop_assigns_for_vec with an invariant that establishes the length #4896 36650908721, Autoharness: support &Wtf8 arguments under --bounded-arguments #4809 36654883143, Fetch a candidate's MIR with the provider that serves its body kind #4848 36663305755, Strip the crate prefix from scanner names to match kani list #4903 36667236036, Fix ICE on pattern types over wide pointers under -Z valid-value-checks #4856 36670723600, LLBC: translate wrapping arithmetic as wrapping, and cover what the Charon bump will touch #4881 36596324505, LLBC: name inherent methods instead of aborting, and deduplicate the name builders #4882 36612997613
• Pull requests: Fix assert! and cover! expansions in expression contexts #4875 36427316996, Re-enable the std_codegen test #4895 36443349794, Report the failure exit code under --quiet #4771 36620184925

The runner image version (20260819.586) was the same in passing and failing runs.

Manual testing. Locally the three harnesses take about 10 s each. Running the test with --harness-timeout 3s reproduces the CI failure exactly: the same three harnesses time out, their Failed Checks lines are missing, and the summary still reports 9 failures. With this change, cargo run -p compiletest -- --suite script-based-pre --mode exec cargo_autoharness_fmt_impls passes.

If the macOS Intel runner gets much slower, a longer timeout alone may not be enough; making these three harnesses cheaper would be the next step.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 and MIT licenses.

[Copilot]

Copilot review overview

🟢 Approval recommended

The focused change matches existing autoharness test conventions and correctly addresses the observed timeout failures.

Review effort: Balanced
Findings: None

What changed in this PR

Raises the formatting autoharness test timeout to prevent macOS Intel CI flakes.

Changes:

• Sets a five-minute harness timeout.
• Enables the required unstable option and documents the rationale.

File	Description
tests/​script-based-pre/​cargo_autoharness_fmt_impls/​fmt-impls.sh	Configures and explains the increased timeout.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[feliperodri]

let's merge and see if this decreases the noise. I'll keep an eye on those slow ones.