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
+9 -9
View File
@@ -3736,7 +3736,7 @@ mod tests {
use tempfile::tempdir;
use super::*;
use crate::config::helpers::ENV_MUTEX;
use crate::config::helpers::lock_env;
#[test]
fn test_wizard_creation() {
@@ -3760,7 +3760,7 @@ mod tests {
#[test]
fn test_wizard_owner_id_uses_resolved_env_scope() {
let _guard = ENV_MUTEX.lock().unwrap_or_else(|e| e.into_inner());
let _guard = lock_env();
let _owner = EnvGuard::set("IRONCLAW_OWNER_ID", " wizard-owner ");
let wizard = SetupWizard::new();
@@ -3769,7 +3769,7 @@ mod tests {
#[test]
fn test_wizard_owner_id_uses_toml_scope() {
let _guard = ENV_MUTEX.lock().unwrap_or_else(|e| e.into_inner());
let _guard = lock_env();
let _owner = EnvGuard::clear("IRONCLAW_OWNER_ID");
let dir = tempdir().unwrap(); // safety: test-only tempdir setup
let path = dir.path().join("config.toml");
@@ -3785,7 +3785,7 @@ mod tests {
fn test_try_with_config_and_toml_propagates_invalid_owner_env() {
use std::os::unix::ffi::OsStringExt;
let _guard = ENV_MUTEX.lock().unwrap_or_else(|e| e.into_inner());
let _guard = lock_env();
let original = std::env::var_os("IRONCLAW_OWNER_ID");
unsafe {
std::env::set_var("IRONCLAW_OWNER_ID", OsString::from_vec(vec![0x66, 0x80]));
@@ -4245,7 +4245,7 @@ mod tests {
fn test_build_nearai_model_fetch_config_picks_up_api_key_env() {
use secrecy::ExposeSecret;
let _lock = ENV_MUTEX.lock().unwrap();
let _lock = lock_env();
let _guard = EnvGuard::set("NEARAI_API_KEY", "test-cloud-api-key-12345");
let _guard2 = EnvGuard::clear("NEARAI_BASE_URL");
@@ -4269,7 +4269,7 @@ mod tests {
/// the config should have `api_key: None` (session token path).
#[test]
fn test_build_nearai_model_fetch_config_none_when_no_api_key() {
let _lock = ENV_MUTEX.lock().unwrap();
let _lock = lock_env();
let _guard = EnvGuard::clear("NEARAI_API_KEY");
let _guard2 = EnvGuard::clear("NEARAI_BASE_URL");
@@ -4288,7 +4288,7 @@ mod tests {
/// Regression test for #799: empty NEARAI_API_KEY should be treated as absent.
#[test]
fn test_build_nearai_model_fetch_config_none_when_empty_api_key() {
let _lock = ENV_MUTEX.lock().unwrap();
let _lock = lock_env();
let _guard = EnvGuard::set("NEARAI_API_KEY", "");
let config = build_nearai_model_fetch_config();
@@ -4306,7 +4306,7 @@ mod tests {
fn test_model_discovery_picks_up_injected_var() {
use secrecy::ExposeSecret;
let _lock = ENV_MUTEX.lock().unwrap();
let _lock = lock_env();
let _guard = EnvGuard::clear("NEARAI_API_KEY");
let _guard2 = EnvGuard::clear("NEARAI_BASE_URL");
@@ -4337,7 +4337,7 @@ mod tests {
/// the NEAR AI authentication menu.
#[test]
fn test_build_nearai_model_fetch_config_picks_up_runtime_env() {
let _lock = ENV_MUTEX.lock().unwrap();
let _lock = lock_env();
// Ensure the real env var is unset so the only source is the overlay.
let _guard = EnvGuard::clear("NEARAI_API_KEY");