emrg: conftest hermeticity guard — un-isolated stop_daemon calls fail loudly (⛔ red line backstop)

[argszero]

Summary

The 2026-08-20 16:18/16:25 incidents killed the live emrgd twice because tests/test_stop_all.py:677 lacked stop_daemon isolation (host red line 2026-08-18T22:58, rant 2026-08-20T16:32:30). PR #889 fixed the specific leak; this PR adds the definitive anti-regression layer.

Changes

1. tests/conftest.py — new autouse fixture _guard_stop_daemon_hermeticity: replaces emrg._stop_all.stop_daemon with a raiser, so any future test that triggers a real stop path fails with AssertionError instead of killing the live daemon. Tests that properly isolate it (monkeypatch.setattr(_stop_all, "stop_daemon", lambda: None)) patch after the fixture and override it as usual.
2. tests/test_stop_all.py — canary test TestStopDaemonHermeticityGuard verifies the guard stays installed (checks function identity only, never calls stop paths).
3. Agent.md — pytest count synced 981 → 982.

Verification

• Full suite: 981 passed + 1 skipped (982 collected), import + CLI OK
• Live daemon (~/.emrg/install/bin/python -m emrg) stayed alive through the entire run — guard confirmed effective
• All existing stop_all tests (656/677/964, all isolated) still green

[argszero]

✅ LGTM — cycle 1 (3 files +37/−1: conftest autouse guard replaces _stop_all.stop_daemon with a raiser — the ⛔ red-line backstop that makes any future un-isolated stop path fail loudly instead of killing the live daemon; canary test checks guard identity without touching stop paths; Agent.md count synced; pytest 981 passed + 1 skipped with daemon alive; test + test-windows CI PASS)

[argszero]

✅ LGTM — cycle 2 (independent re-review: conftest autouse guard correctly replaces _stop_all.stop_daemon with a raiser (⛔ red-line backstop); existing isolated tests patch after the fixture and override as designed; canary checks guard identity only; Agent.md 982 sync correct; test + test-windows CI PASS)

[argszero]

✅ LGTM — cycle 3 (re-review: head 000e4f4 unchanged, 3 files +37/−1; conftest autouse guard is the ⛔ red-line backstop making any un-isolated stop_daemon call fail loudly; canary verifies guard installation; Agent.md 982 synced; test + test-windows CI PASS, mergeable CLEAN)

[argszero]

Summary

The 2026-08-20 16:18/16:25 incidents killed the live emrgd twice because tests/test_stop_all.py:677 lacked stop_daemon isolation (host red line 2026-08-18T22:58, rant 2026-08-20T16:32:30). PR #889 fixed the specific leak; this PR adds the definitive anti-regression layer.

Changes

1. tests/conftest.py — new autouse fixture _guard_stop_daemon_hermeticity: replaces emrg._stop_all.stop_daemon with a raiser, so any future test that triggers a real stop path fails with AssertionError instead of killing the live daemon. Tests that properly isolate it (monkeypatch.setattr(_stop_all, "stop_daemon", lambda: None)) patch after the fixture and override it as usual.
2. tests/test_stop_all.py — canary test TestStopDaemonHermeticityGuard verifies the guard stays installed (checks function identity only, never calls stop paths).
3. Agent.md — pytest count synced 981 → 982.

Verification

• Full suite: 981 passed + 1 skipped (982 collected), import + CLI OK
• Live daemon (~/.emrg/install/bin/python -m emrg) stayed alive through the entire run — guard confirmed effective
• All existing stop_all tests (656/677/964, all isolated) still green

[argszero]

✅ LGTM — cycle 1 (3 files +37/−1: conftest autouse guard replaces _stop_all.stop_daemon with a raiser — the ⛔ red-line backstop that makes any future un-isolated stop path fail loudly instead of killing the live daemon; canary test checks guard identity without touching stop paths; Agent.md count synced; pytest 981 passed + 1 skipped with daemon alive; test + test-windows CI PASS)

[argszero]

✅ LGTM — cycle 2 (independent re-review: conftest autouse guard correctly replaces _stop_all.stop_daemon with a raiser (⛔ red-line backstop); existing isolated tests patch after the fixture and override as designed; canary checks guard identity only; Agent.md 982 sync correct; test + test-windows CI PASS)

[argszero]

✅ LGTM — cycle 3 (re-review: head 000e4f4 unchanged, 3 files +37/−1; conftest autouse guard is the ⛔ red-line backstop making any un-isolated stop_daemon call fail loudly; canary verifies guard installation; Agent.md 982 synced; test + test-windows CI PASS, mergeable CLEAN)

[argszero]

Summary

The 2026-08-20 16:18/16:25 incidents killed the live emrgd twice because tests/test_stop_all.py:677 lacked stop_daemon isolation (host red line 2026-08-18T22:58, rant 2026-08-20T16:32:30). PR #889 fixed the specific leak; this PR adds the definitive anti-regression layer.

Changes

1. tests/conftest.py — new autouse fixture _guard_stop_daemon_hermeticity: replaces emrg._stop_all.stop_daemon with a raiser, so any future test that triggers a real stop path fails with AssertionError instead of killing the live daemon. Tests that properly isolate it (monkeypatch.setattr(_stop_all, "stop_daemon", lambda: None)) patch after the fixture and override it as usual.
2. tests/test_stop_all.py — canary test TestStopDaemonHermeticityGuard verifies the guard stays installed (checks function identity only, never calls stop paths).
3. Agent.md — pytest count synced 981 → 982.

Verification

• Full suite: 981 passed + 1 skipped (982 collected), import + CLI OK
• Live daemon (~/.emrg/install/bin/python -m emrg) stayed alive through the entire run — guard confirmed effective
• All existing stop_all tests (656/677/964, all isolated) still green

[argszero]

✅ LGTM — cycle 1 (3 files +37/−1: conftest autouse guard replaces _stop_all.stop_daemon with a raiser — the ⛔ red-line backstop that makes any future un-isolated stop path fail loudly instead of killing the live daemon; canary test checks guard identity without touching stop paths; Agent.md count synced; pytest 981 passed + 1 skipped with daemon alive; test + test-windows CI PASS)

[argszero]

✅ LGTM — cycle 2 (independent re-review: conftest autouse guard correctly replaces _stop_all.stop_daemon with a raiser (⛔ red-line backstop); existing isolated tests patch after the fixture and override as designed; canary checks guard identity only; Agent.md 982 sync correct; test + test-windows CI PASS)

[argszero]

✅ LGTM — cycle 3 (re-review: head 000e4f4 unchanged, 3 files +37/−1; conftest autouse guard is the ⛔ red-line backstop making any un-isolated stop_daemon call fail loudly; canary verifies guard installation; Agent.md 982 synced; test + test-windows CI PASS, mergeable CLEAN)

[argszero]

Summary

The 2026-08-20 16:18/16:25 incidents killed the live emrgd twice because tests/test_stop_all.py:677 lacked stop_daemon isolation (host red line 2026-08-18T22:58, rant 2026-08-20T16:32:30). PR #889 fixed the specific leak; this PR adds the definitive anti-regression layer.

Changes

1. tests/conftest.py — new autouse fixture _guard_stop_daemon_hermeticity: replaces emrg._stop_all.stop_daemon with a raiser, so any future test that triggers a real stop path fails with AssertionError instead of killing the live daemon. Tests that properly isolate it (monkeypatch.setattr(_stop_all, "stop_daemon", lambda: None)) patch after the fixture and override it as usual.
2. tests/test_stop_all.py — canary test TestStopDaemonHermeticityGuard verifies the guard stays installed (checks function identity only, never calls stop paths).
3. Agent.md — pytest count synced 981 → 982.

Verification

• Full suite: 981 passed + 1 skipped (982 collected), import + CLI OK
• Live daemon (~/.emrg/install/bin/python -m emrg) stayed alive through the entire run — guard confirmed effective
• All existing stop_all tests (656/677/964, all isolated) still green

[argszero]

✅ LGTM — cycle 1 (3 files +37/−1: conftest autouse guard replaces _stop_all.stop_daemon with a raiser — the ⛔ red-line backstop that makes any future un-isolated stop path fail loudly instead of killing the live daemon; canary test checks guard identity without touching stop paths; Agent.md count synced; pytest 981 passed + 1 skipped with daemon alive; test + test-windows CI PASS)

[argszero]

✅ LGTM — cycle 2 (independent re-review: conftest autouse guard correctly replaces _stop_all.stop_daemon with a raiser (⛔ red-line backstop); existing isolated tests patch after the fixture and override as designed; canary checks guard identity only; Agent.md 982 sync correct; test + test-windows CI PASS)

[argszero]

✅ LGTM — cycle 3 (re-review: head 000e4f4 unchanged, 3 files +37/−1; conftest autouse guard is the ⛔ red-line backstop making any un-isolated stop_daemon call fail loudly; canary verifies guard installation; Agent.md 982 synced; test + test-windows CI PASS, mergeable CLEAN)

[argszero]

Summary

The 2026-08-20 16:18/16:25 incidents killed the live emrgd twice because tests/test_stop_all.py:677 lacked stop_daemon isolation (host red line 2026-08-18T22:58, rant 2026-08-20T16:32:30). PR #889 fixed the specific leak; this PR adds the definitive anti-regression layer.

Changes

1. tests/conftest.py — new autouse fixture _guard_stop_daemon_hermeticity: replaces emrg._stop_all.stop_daemon with a raiser, so any future test that triggers a real stop path fails with AssertionError instead of killing the live daemon. Tests that properly isolate it (monkeypatch.setattr(_stop_all, "stop_daemon", lambda: None)) patch after the fixture and override it as usual.
2. tests/test_stop_all.py — canary test TestStopDaemonHermeticityGuard verifies the guard stays installed (checks function identity only, never calls stop paths).
3. Agent.md — pytest count synced 981 → 982.

Verification

• Full suite: 981 passed + 1 skipped (982 collected), import + CLI OK
• Live daemon (~/.emrg/install/bin/python -m emrg) stayed alive through the entire run — guard confirmed effective
• All existing stop_all tests (656/677/964, all isolated) still green

[argszero]

✅ LGTM — cycle 1 (3 files +37/−1: conftest autouse guard replaces _stop_all.stop_daemon with a raiser — the ⛔ red-line backstop that makes any future un-isolated stop path fail loudly instead of killing the live daemon; canary test checks guard identity without touching stop paths; Agent.md count synced; pytest 981 passed + 1 skipped with daemon alive; test + test-windows CI PASS)

[argszero]

✅ LGTM — cycle 2 (independent re-review: conftest autouse guard correctly replaces _stop_all.stop_daemon with a raiser (⛔ red-line backstop); existing isolated tests patch after the fixture and override as designed; canary checks guard identity only; Agent.md 982 sync correct; test + test-windows CI PASS)

[argszero]

✅ LGTM — cycle 3 (re-review: head 000e4f4 unchanged, 3 files +37/−1; conftest autouse guard is the ⛔ red-line backstop making any un-isolated stop_daemon call fail loudly; canary verifies guard installation; Agent.md 982 synced; test + test-windows CI PASS, mergeable CLEAN)

[argszero]

Summary

The 2026-08-20 16:18/16:25 incidents killed the live emrgd twice because tests/test_stop_all.py:677 lacked stop_daemon isolation (host red line 2026-08-18T22:58, rant 2026-08-20T16:32:30). PR #889 fixed the specific leak; this PR adds the definitive anti-regression layer.

Changes

1. tests/conftest.py — new autouse fixture _guard_stop_daemon_hermeticity: replaces emrg._stop_all.stop_daemon with a raiser, so any future test that triggers a real stop path fails with AssertionError instead of killing the live daemon. Tests that properly isolate it (monkeypatch.setattr(_stop_all, "stop_daemon", lambda: None)) patch after the fixture and override it as usual.
2. tests/test_stop_all.py — canary test TestStopDaemonHermeticityGuard verifies the guard stays installed (checks function identity only, never calls stop paths).
3. Agent.md — pytest count synced 981 → 982.

Verification

• Full suite: 981 passed + 1 skipped (982 collected), import + CLI OK
• Live daemon (~/.emrg/install/bin/python -m emrg) stayed alive through the entire run — guard confirmed effective
• All existing stop_all tests (656/677/964, all isolated) still green

[argszero]

✅ LGTM — cycle 1 (3 files +37/−1: conftest autouse guard replaces _stop_all.stop_daemon with a raiser — the ⛔ red-line backstop that makes any future un-isolated stop path fail loudly instead of killing the live daemon; canary test checks guard identity without touching stop paths; Agent.md count synced; pytest 981 passed + 1 skipped with daemon alive; test + test-windows CI PASS)

[argszero]

✅ LGTM — cycle 2 (independent re-review: conftest autouse guard correctly replaces _stop_all.stop_daemon with a raiser (⛔ red-line backstop); existing isolated tests patch after the fixture and override as designed; canary checks guard identity only; Agent.md 982 sync correct; test + test-windows CI PASS)

[argszero]

✅ LGTM — cycle 3 (re-review: head 000e4f4 unchanged, 3 files +37/−1; conftest autouse guard is the ⛔ red-line backstop making any un-isolated stop_daemon call fail loudly; canary verifies guard installation; Agent.md 982 synced; test + test-windows CI PASS, mergeable CLEAN)

[argszero]

Summary

The 2026-08-20 16:18/16:25 incidents killed the live emrgd twice because tests/test_stop_all.py:677 lacked stop_daemon isolation (host red line 2026-08-18T22:58, rant 2026-08-20T16:32:30). PR #889 fixed the specific leak; this PR adds the definitive anti-regression layer.

Changes

1. tests/conftest.py — new autouse fixture _guard_stop_daemon_hermeticity: replaces emrg._stop_all.stop_daemon with a raiser, so any future test that triggers a real stop path fails with AssertionError instead of killing the live daemon. Tests that properly isolate it (monkeypatch.setattr(_stop_all, "stop_daemon", lambda: None)) patch after the fixture and override it as usual.
2. tests/test_stop_all.py — canary test TestStopDaemonHermeticityGuard verifies the guard stays installed (checks function identity only, never calls stop paths).
3. Agent.md — pytest count synced 981 → 982.

Verification

• Full suite: 981 passed + 1 skipped (982 collected), import + CLI OK
• Live daemon (~/.emrg/install/bin/python -m emrg) stayed alive through the entire run — guard confirmed effective
• All existing stop_all tests (656/677/964, all isolated) still green

[argszero]

✅ LGTM — cycle 1 (3 files +37/−1: conftest autouse guard replaces _stop_all.stop_daemon with a raiser — the ⛔ red-line backstop that makes any future un-isolated stop path fail loudly instead of killing the live daemon; canary test checks guard identity without touching stop paths; Agent.md count synced; pytest 981 passed + 1 skipped with daemon alive; test + test-windows CI PASS)

[argszero]

✅ LGTM — cycle 2 (independent re-review: conftest autouse guard correctly replaces _stop_all.stop_daemon with a raiser (⛔ red-line backstop); existing isolated tests patch after the fixture and override as designed; canary checks guard identity only; Agent.md 982 sync correct; test + test-windows CI PASS)

[argszero]

✅ LGTM — cycle 3 (re-review: head 000e4f4 unchanged, 3 files +37/−1; conftest autouse guard is the ⛔ red-line backstop making any un-isolated stop_daemon call fail loudly; canary verifies guard installation; Agent.md 982 synced; test + test-windows CI PASS, mergeable CLEAN)

[argszero]

Summary

The 2026-08-20 16:18/16:25 incidents killed the live emrgd twice because tests/test_stop_all.py:677 lacked stop_daemon isolation (host red line 2026-08-18T22:58, rant 2026-08-20T16:32:30). PR #889 fixed the specific leak; this PR adds the definitive anti-regression layer.

Changes

1. tests/conftest.py — new autouse fixture _guard_stop_daemon_hermeticity: replaces emrg._stop_all.stop_daemon with a raiser, so any future test that triggers a real stop path fails with AssertionError instead of killing the live daemon. Tests that properly isolate it (monkeypatch.setattr(_stop_all, "stop_daemon", lambda: None)) patch after the fixture and override it as usual.
2. tests/test_stop_all.py — canary test TestStopDaemonHermeticityGuard verifies the guard stays installed (checks function identity only, never calls stop paths).
3. Agent.md — pytest count synced 981 → 982.

Verification

• Full suite: 981 passed + 1 skipped (982 collected), import + CLI OK
• Live daemon (~/.emrg/install/bin/python -m emrg) stayed alive through the entire run — guard confirmed effective
• All existing stop_all tests (656/677/964, all isolated) still green

[argszero]

✅ LGTM — cycle 1 (3 files +37/−1: conftest autouse guard replaces _stop_all.stop_daemon with a raiser — the ⛔ red-line backstop that makes any future un-isolated stop path fail loudly instead of killing the live daemon; canary test checks guard identity without touching stop paths; Agent.md count synced; pytest 981 passed + 1 skipped with daemon alive; test + test-windows CI PASS)

[argszero]

✅ LGTM — cycle 2 (independent re-review: conftest autouse guard correctly replaces _stop_all.stop_daemon with a raiser (⛔ red-line backstop); existing isolated tests patch after the fixture and override as designed; canary checks guard identity only; Agent.md 982 sync correct; test + test-windows CI PASS)

[argszero]

✅ LGTM — cycle 3 (re-review: head 000e4f4 unchanged, 3 files +37/−1; conftest autouse guard is the ⛔ red-line backstop making any un-isolated stop_daemon call fail loudly; canary verifies guard installation; Agent.md 982 synced; test + test-windows CI PASS, mergeable CLEAN)