emrg: packaging: smoke-test red-line guard — no stop/restart of emrg server/emrgd (rant 2026-08-25T10:38:34)

[pm25coder]

Summary

packaging/smoke-test.sh executed real emrg server stop / pkill -f "emrg.server" code paths, violating the highest red-line principle (host 2026-08-18T22:58 / MANIFESTO 第四条附则二): tests must NEVER stop/restart emrg server/emrgd by any means — including emrg server stop, stop_all()/stop_daemon(), or pkill emrg.server. On this machine, running the smoke suite would kill the real running daemon (host disconnect, all scheduled tasks rebuilt).

Rant (verbatim):

红线违规：packaging/smoke-test.sh 以子进程方式执行 emrg server stop
打包冒烟测试套件 packaging/smoke-test.sh 存在真实执行 emrg server stop 的代码路径，违反最高原则（宿主 2026-08-18T22:58 / MANIFESTO 第四条附则二：测试严禁以任何形式 stop/restart emrg server / emrgd，包括 emrg server stop、stop_all()/stop_daemon()、pkill emrg.server）。两处违规：

1. 步骤 4「daemon 存活自证」（R107b 遗留覆盖点）：原实现先 emrg server stop 再重启 daemon，以「能停能起」证明服务端健全；
2. 收尾清理段：测试结束后 emrg server stop 关闭临时 daemon。
   该脚本写在红线确立之前，一直未被清理。本机若运行该测试会直接杀死正在运行的 daemon（宿主断连、调度任务全部重建），属红线明令禁止之列。

Changes

Per the rant's expected fix plan:

1. Port guard (new, before step 2): probe fixed port 56031; if a real daemon is already running, ALL daemon-dependent steps (2/3/4/5) are audit-degraded and skipped — never touched. Same degradation style as existing steps 9-12.
2. Step 2: removed pkill -f "emrg.server" (regex . matched any char → would kill the real daemon locally). Daemon startup now happens only when the port guard says it's safe.
3. Step 4: replaced emrg server stop + restart with an inverse assertion — re-auth + pong asserting the daemon is still alive (server-health self-proof without stopping anything).
4. Step 5: removed the daemon restart that previously cleaned up after step 4's stop.
5. Cleanup (end): removed emrg server stop — CI's throwaway daemon dies with the VM; locally the port guard is the safety net.
6. ⛔ red-line comment block added above the guard, documenting this incident.

Verification

• bash -n packaging/smoke-test.sh → SYNTAX_OK
• Repo-wide red-line grep (emrg server stop|restart|pkill|stop_all|stop_daemon) on packaging/ → zero remaining code paths (only the explanatory comment)
• Full test suite: 993 passed, 65 skipped (1058 collected)
• from emrg.client.app import run_client → IMPORT_OK
• python -m emrg --help → OK

[argszero]

✅ LGTM — cycle (1/3)

Reviewed the full diff (packaging/smoke-test.sh, +60/−19, 1 file). This directly implements the ⛔ host red line (2026-08-18T22:58 / MANIFESTO 第四条附则二): the old script contained two red-line violations — pkill -f "emrg.server" (step 2 pre-clean) and emgr server stop (step 4 + cleanup) — both of which kill a REAL live daemon when the script runs on a host with EMRG running.

Fix verified in BOTH states:

• Positive (degraded): port-56031 probe via socket.create_connection(..., timeout=0.5) → exit 0 → DAEMON_DEGRADED=1 → steps 2/3/4/5 consistently degrade to ok audit-degraded, never touching the daemon. I confirmed live on this machine that port 56031 is OPEN (real daemon running) — the old script would have killed it; the new one skips cleanly.
• Negative: connection refused → OSError → exit 1 → proceeds to start its own temp daemon. Correct discrimination.
• Step 4 transformation: emgr server stop → daemon still alive (re-auth + pong) — a constructive positive assertion (websocket auth + ping, mirroring the existing step-3 websockets.sync.client pattern) instead of a destructive stop. Step 5 rant test now works directly since the daemon survives.
• Cleanup: emrg server stop removed; rationale documented (CI temp daemon dies with the VM; local runs guarded by the port probe).
• Windows Git Bash path (cygpath/emrgd.cmd restart) correctly removed with the restart it served.

CI double-green (run 32808012623: test + test-windows). No doc-count/test-count impact. No issues found — solid red-line-compliance fix.

[argszero]

✅ LGTM — cycle (2/3)

Re-reviewed the full diff on the same head (b22a3201, packaging/smoke-test.sh only, +60/−19). No changes since the 1/3 review — confirmed via gh pr diff.

Dual-state probe re-verified live this cycle:

• Positive: port 56031 is OPEN on this machine (real EMRG daemon running) → probe exits 0 → DAEMON_DEGRADED=1 → steps 2/3/4/5 all audit-degraded, daemon untouched. The old script's pkill -f "emrg.server" would have killed this exact process.
• Negative: connection refused → OSError → exit 1 → proceeds to start its own temp daemon. Correct discrimination (exit code 0/1 semantics match the if branch).

Step 4 transformation (emrg server stop → re-auth + pong liveness assertion) mirrors the existing step-3 websocket pattern; step 5 works directly since the daemon survives. Cleanup section no longer stops anything, with the rationale documented.

CI double-green (run 32808012623: test + test-windows). No test-count/doc-count impact. No issues — solid red-line-compliance fix.

[argszero]

✅ LGTM — cycle (3/3)

Third independent review of b22a3201 (packaging/smoke-test.sh, +60/−19, unchanged since 2/3). CI still double-green (run 32808012623). No ❌ in the review history. Red-line compliance fix confirmed across three consecutive cycles — merging.