ci: hosted 子进程启动耗时证据 (#81/#144/#40)

.github/workflows/conpty-compat.yml

@@ -243,3 +243,57 @@ jobs:
           retention-days: 14
           if-no-files-found: warn
           path: build/test-artifacts/
+
+  windows-2025-startup-latency:
+    # Bounded measurement for #81 / #144 / #40: how long does a plain
+    # PowerShell 5.1 or Node start-up take on this hosted image, and how often
+    # does it cross the fixed fixture deadlines. No deadlines are changed.
+    runs-on: windows-2025
+    timeout-minutes: 30
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+      - uses: actions/setup-python@6d4fd22b20465ccc0f7354476588e8df7c9ca0e0 # v7.1.0
+        with:
+          python-version: '3.12'
+      - name: Measure subprocess start-up latency
+        shell: pwsh
+        run: |
+          @'
+          import json, platform, subprocess, sys, time
+          def samples(command, count, timeout):
+              rows = []
+              for _ in range(count):
+                  started = time.monotonic()
+                  try:
+                      result = subprocess.run(command, capture_output=True, timeout=timeout)
+                      rows.append({"elapsed": round(time.monotonic() - started, 4), "returncode": result.returncode, "timeout": False})
+                  except subprocess.TimeoutExpired:
+                      rows.append({"elapsed": round(time.monotonic() - started, 4), "returncode": None, "timeout": True})
+              return rows
+          def summary(rows):
+              times = sorted(row["elapsed"] for row in rows)
+              def at(fraction):
+                  return times[min(len(times) - 1, int(len(times) * fraction))]
+              return {"count": len(rows), "min": times[0], "p50": at(0.5), "p90": at(0.9), "p99": at(0.99),
+                      "max": times[-1], "over_1s": sum(1 for t in times if t >= 1.0),
+                      "over_5s": sum(1 for t in times if t >= 5.0),
+                      "over_10s": sum(1 for t in times if t >= 10.0),
+                      "timeouts": sum(1 for row in rows if row["timeout"])}
+          ps = samples(["powershell.exe", "-NoLogo", "-NoProfile", "-NonInteractive", "-Command", "exit 0"], 100, 15)
+          node = samples(["node", "--version"], 50, 15)
+          report = {"schema": "termforge-startup-latency-v1", "platform": platform.platform(),
+                    "powershell": summary(ps), "node": summary(node),
+                    "powershell_samples": ps, "node_samples": node}
+          print(json.dumps({k: v for k, v in report.items() if not k.endswith("_samples")}))
+          with open("startup-latency.json", "w", encoding="utf-8") as stream:
+              json.dump(report, stream, indent=2)
+          '@ | Out-File -Encoding utf8 startup-latency.py
+          python startup-latency.py
+      - name: Preserve latency evidence
+        if: always()
+        uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7.0.2
+        with:
+          name: startup-latency-r2
+          retention-days: 14
+          if-no-files-found: warn
+          path: startup-latency.json

research/ui/results/2026-10-30-hosted-startup-latency/README.md

@@ -0,0 +1,31 @@
+# hosted windows-2025 的子进程启动耗时（#81/#144/#40 佐证）
+
+本片回答一个可测子问题：在托管镜像上，**纯进程启动**本身要多久，与 10–12 秒的固定期限相差多少。
+
+## 方法
+
+dispatch-only 作业 `windows-2025-startup-latency`（[工作流](../../../../.github/workflows/conpty-compat.yml)）：
+
+- `powershell.exe -NoLogo -NoProfile -NonInteractive -Command "exit 0"` × **100** 次；
+- `node --version` × **50** 次；
+- 每次新进程，15 秒超时，记录耗时、返回码、是否超时；统计 min/p50/p90/p99/max 与 ≥1s/≥5s/≥10s 计数。
+
+## 结果（run 40935653206，windows-2025）
+
+| 项 | 次数 | min | p50 | p90 | p99 | max | ≥1s | ≥5s | ≥10s | 超时 |
+|---|---|---|---|---|---|---|---|---|---|---|
+| PowerShell 5.1 | 100 | 0.109 s | 0.110 s | 0.125 s | 1.828 s | 1.828 s | 1 | 0 | 0 | 0 |
+| Node | 50 | 0.000 s | 0.016 s | 0.016 s | 0.735 s | 0.735 s | 0 | 0 | 0 | 0 |
+
+## 解释
+
+- 空闲镜像上，PowerShell 5.1 启动中位数 **110 ms**、最慢 **1.828 s**；Node 更快（中位 16 ms、最慢 735 ms）；
+- 因此 `ps51` 在 CI 里出现 **10219 ms** 的 `timed_out`（#81）不能归因于启动成本；Node 夹具的 10 秒超时（#144）同理；
+- 三处（ps51 10.2 s、纯 PowerShell 12 s、Node 10 s）都属于"纯启动/纯脚本、几乎无输出、刚好越过期限"，与本次测得的正常分布相差 1–2 个数量级，指向**宿主级停顿**（安全扫描、VM 调度、存储阻塞）。
+
+## 边界
+
+- 只测空闲镜像；未在完整 CTest 同时段、安全软件扫描中或并行负载下测量；
+- 未修改任何期限、夹具或准备流程；
+- 不能据此断定停顿的具体来源，只排除"启动成本"这一解释；
+- 与 #144 的 Node 夹具计时（warm 63 ms、冷启动 1.47 s）互相印证。

research/ui/results/2026-10-30-hosted-startup-latency/validation.json

@@ -0,0 +1,60 @@
+{
+  "schema": "termforge-hosted-startup-latency-v1",
+  "issues": [
+    81,
+    144,
+    40
+  ],
+  "run": 40935653206,
+  "job": "windows-2025-startup-latency",
+  "platform": "windows-2025 hosted",
+  "commands": {
+    "powershell": [
+      "powershell.exe",
+      "-NoLogo",
+      "-NoProfile",
+      "-NonInteractive",
+      "-Command",
+      "exit 0"
+    ],
+    "node": [
+      "node",
+      "--version"
+    ]
+  },
+  "counts": {
+    "powershell": 100,
+    "node": 50
+  },
+  "results": {
+    "powershell": {
+      "min": 0.109,
+      "p50": 0.11,
+      "p90": 0.125,
+      "p99": 1.828,
+      "max": 1.828,
+      "over_1s": 1,
+      "over_5s": 0,
+      "over_10s": 0,
+      "timeouts": 0
+    },
+    "node": {
+      "min": 0,
+      "p50": 0.016,
+      "p90": 0.016,
+      "p99": 0.735,
+      "max": 0.735,
+      "over_1s": 0,
+      "over_5s": 0,
+      "over_10s": 0,
+      "timeouts": 0
+    }
+  },
+  "interpretation": "Pure start-up cost is 1-2 orders of magnitude below the fixed fixture deadlines, so the observed 10-12 s timeouts cannot be explained by start-up cost.",
+  "limitations": [
+    "idle image only; not under full CTest concurrency or AV scanning",
+    "no deadline or fixture change",
+    "does not identify the stall source, only excludes start-up cost"
+  ],
+  "performance_claim": false
+}