[proggeramlug]

What

A minutes-scale gate for every PR, plus the two pre-existing defects that validating it exposed.

Why

The only workflow here is the Linux shared-runtime proof. It needs LLVM 22, a Perry checkout, provider builds and cgroup delegation, and it is budgeted at two hours. Nothing in this repository gave a contributor feedback before that finished.

perch-host-abi, perch-worker and perch-cli have no Perry dependency — the worker dlopens application images rather than linking them — so they build, lint and test on a stock toolchain in minutes. perch-daemon and perry-stdlib-shared do need Perry and stay with the proof job.

The two defects it found before landing

perch-cli did not compile. Not just its tests — the binary:

error[E0599]: no method named `env` found for struct `Arg`
  --> crates/perch-cli/src/main.rs:19:17
   |  #[arg(long, env = "PERCH_URL", default_value = "http://127.0.0.1:80")]

clap's env feature was not enabled, so the derive expanded to a call that does not exist. One feature flag.

plugin_host.rs tripped items_after_test_module — a #[cfg(test)] mod boundary_tests sat mid-file with five production items after it. Moved to the end of the file.

Both are fixed here so the gate lands green. A gate that arrives red is a gate everyone learns to ignore.

Deliberate details

• --no-fail-fast is mandatory. Without it the run stops at the first failing target and silently skips the remaining worker test binaries — which reads as a pass.
• A step surfaces skip: lines. Several suites self-skip when fixtures are absent, and a skip is an early return that counts as a pass in the summary. A green run should not be able to hide that nothing executed.
• Cancellation is scoped to pull requests. Cancelling main runs means a busy merge day can cancel every run before it reaches a runner, leaving a gate that has never actually executed. Worth considering the same one-line change for linux-shared-runtime.yml, which currently cancels main runs on a 120-minute job.
• Provider suites excluded by name, not pattern — plugin_roundtrip, provider_integrity, binary_http_roundtrip — so adding a fourth fails loudly here instead of being silently skipped.

Separately worth knowing

While validating, plugin_roundtrip SIGBUSed (signal 10) after four tests passed — a crashed target, not a test failure, which is why cargo exited 101 while the summary line still said 0 failed.

It is the suite containing the recently un-ignored 50,000-dispatch churn test, running against a Perry pin that is 58 commits stale and predates the FUNCTION_CLASS_IDS dead-key fix (PerryTS/perry#8168). That is the second independent sighting of that bug class in Perch — the other being a benchmark shape dying under sustained load with TypeError: value is not a function.

Not addressed here; it argues for bumping perry-main.lock once #8168 lands.

Verified locally

fmt, clippy -D warnings, and 14 tests across the four in-scope targets — all green.

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Free

Run ID: f30329c8-8479-4ffe-ac60-19bbe73bc539

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Note

🎁 Summarized by CodeRabbit Free

Your organization is on the Free plan. CodeRabbit will generate a high-level summary and a walkthrough for each pull request. For a comprehensive line-by-line review, please upgrade your subscription to CodeRabbit Pro by visiting https://app.coderabbit.ai/login.

_{Comment @coderabbitai help to get the list of available commands.}