fix(tests): eliminate env mutex poison cascade (#1558)

* fix(tests): eliminate env mutex poison cascade and fix test flakiness

The shared ENV_MUTEX used by ~68 config tests would cascade a single
test panic into failures across every module. Replace all .unwrap() /
.expect() lock acquisitions with a poison-recovering lock_env() helper.
Consolidate rogue module-local ENV_LOCK instances (workspace, orchestrator,
bootstrap) onto the shared global mutex to prevent cross-module races.

Also fixes:
- gateway user_id fallback was hardcoded to "default" instead of owner_id
- test_ironclaw_env_path used LazyLock which is order-dependent

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* test(helpers): add regression test for lock_env poison recovery

Satisfies the regression-test-check CI gate by adding a test that
intentionally poisons ENV_MUTEX and verifies lock_env() recovers.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix(ci): detect test changes inside #[cfg(test)] regions

The regression test check relied on git diff -W to expand context to
function boundaries, but git doesn't recognize Rust `mod tests {}` as a
function boundary. Changes to imports, helpers, or lock calls inside
test modules were invisible to the check.

Add a line-level fallback: for each changed .rs file, find where
#[cfg(test)] starts and check if any diff hunk targets a line at or
after that boundary. This catches edits anywhere inside test modules
regardless of git's language awareness.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: address PR review feedback

- Clear ENV_MUTEX poison after regression test so it doesn't leave
  global state dirty for subsequent tests.
- Fix CI regression-test-check to match #[cfg(test)] only when followed
  by `mod` (the test module pattern), avoiding false positives from
  standalone #[cfg(test)] items like statics or functions.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>
This commit is contained in:
Illia Polosukhin
2026-03-22 14:36:24 -07:00
committed by GitHub
co-authored by Claude Opus 4.6
parent fbce9a5fe3
commit 3aa36c8f55
19 changed files with 192 additions and 143 deletions
+3 -7
View File
@@ -164,19 +164,15 @@ pub async fn setup_orchestrator(
#[cfg(test)]
mod tests {
use std::sync::Mutex;
use super::*;
/// Serialize access to `ORCHESTRATOR_PORT` env var across test threads.
static ENV_LOCK: Mutex<()> = Mutex::new(());
use crate::config::helpers::lock_env;
#[test]
fn resolve_orchestrator_port_from_env() {
let _guard = ENV_LOCK.lock().unwrap();
let _guard = lock_env();
// Safety: env-var mutation requires unsafe in edition 2024;
// ENV_LOCK serializes concurrent access from other test threads.
// lock_env() serializes concurrent access from other test threads.
// Absent env var → default 50051
unsafe { std::env::remove_var("ORCHESTRATOR_PORT") };