加快门禁：删除测试窗口、释放桩线程并并行运行三个阶段

关闭后未删除的测试窗口让每次主题切换重排全部残留控件，ui_check 每次关窗还要等桩线程 5 秒。Linux 上结论不变，test_ui_smoke 75→64 s、ui_check 52→16 s；Windows 门禁结果见 .scratch/gate-speed/。

README.md

@@ -118,8 +118,9 @@ is meant to be unskippable:
 powershell -NoProfile -ExecutionPolicy Bypass -File tools\gate.ps1
 ```
 
-It runs `tools/lint_gate.py` over the changed files, the full studio suite, the full
-shared suite, and `ui_check.py`; writes each one's output to
+It runs `tools/lint_gate.py` over the changed files, then the full studio suite, the full
+shared suite, and `ui_check.py` at the same time (`-Serial` runs those three one after
+another with live output, for reading a failing stage); writes each one's output to
 `.scratch/gate/gate-<run-id>/`; and holds the run to three contracts — an **empty**
 `tools/known_failures.txt`, the skip ceilings and collection floors in
 `tools/skip_budget.txt`, and zero findings anywhere. It exits 0 only when all of

apps/studio/tests/conftest.py

@@ -97,6 +97,8 @@ def make_window(qt_app, tmp_path, monkeypatch):
 
     yield _make
 
+    from PySide6.QtCore import QCoreApplication, QEvent
+
     for window in created:
         try:
             window.close()
@@ -104,6 +106,14 @@ def make_window(qt_app, tmp_path, monkeypatch):
             # Reported rather than swallowed: a window that cannot close is a Qt lifetime
             # problem worth seeing, but it must not turn a green test red.
             logging.getLogger(__name__).debug("could not close a test window", exc_info=True)
+        # ``close()`` only hides a window: something in the window keeps it referenced,
+        # so it stayed alive for the rest of the session. Every leaked window's ~600
+        # widgets are then re-polished by each later ``app.setStyleSheet`` — measured on
+        # Linux, one theme switch takes 0.07 s with one window and 0.85 s after ten closed
+        # ones, and three theme tests took 76 s of a 231 s Windows suite. Delete it.
+        window.deleteLater()
+    created.clear()
+    QCoreApplication.sendPostedEvents(None, QEvent.Type.DeferredDelete)
 
 
 @pytest.fixture

apps/studio/tests/test_ui_smoke.py

@@ -1048,7 +1048,9 @@ def test_the_domain_boundary_is_latched_for_the_session(make_window):
         "the previous session's island survived into a new run"
     )
     window._run_timer.stop()
-    window._stop_run(reason="user_stop")
+    # Release the stand-in worker as well: the second run re-armed its hang, and a
+    # window closed on a hanging worker waits out its 5 s join in teardown.
+    _finish(window, stuck)
 
 
 # ------------------------------------------------------------- experiment record

apps/studio/tools/ui_check.py

@@ -414,6 +414,22 @@ DECORATIVE_PAIRS: tuple[tuple[str, tuple[str, ...]], ...] = (
 )
 
 
+def _close_window(window) -> None:
+    """Close a check window, releasing its stand-in worker first.
+
+    A ``_StuckWorker`` hangs on purpose, and the window's shutdown path stops its
+    worker with ``wait=True``, which joins that thread for up to 5 s. Every scene that
+    closed a window on a still-hanging stub paid those 5 s (measured: 7 joins, 35 s of
+    a 52 s run). What those scenes check happens before the close, so the stub is let
+    go here; the shutdown itself is checked separately in a subprocess.
+    """
+    for name in ("_box_worker", "_finger_worker"):
+        worker = getattr(window, name, None)
+        if isinstance(worker, _StuckWorker):
+            worker.release()
+    window.close()
+
+
 def _srgb_channel(value: int) -> float:
     channel = value / 255.0
     if channel <= 0.03928:
@@ -836,7 +852,7 @@ def main() -> int:
                 len(window.prefs.get("splitter_sizes", [])) == 4,
                 "splitter sizes not persisted",
             )
-            window.close()
+            _close_window(window)
             app.processEvents()
 
     section("small window")
@@ -885,7 +901,7 @@ def main() -> int:
                 f" in {type(parent).__name__} {parent.width()}x{parent.height()}"
             )
     check(not oversize, f"widgets overflow their parent: {oversize}")
-    window.close()
+    _close_window(window)
     app.processEvents()
 
     section("top area")
@@ -1072,7 +1088,7 @@ def main() -> int:
             not window.toolbar.settings_expanded(),
             f"{theme}: the settings area could not be collapsed again",
         )
-        window.close()
+        _close_window(window)
         app.processEvents()
 
     # English labels are the widest case ("Start recording", "Sensor config"),
@@ -1096,7 +1112,7 @@ def main() -> int:
         f"the sensor summary is not stated in full: "
         f"{english_window.toolbar.btn_sensor.text()!r}",
     )
-    english_window.close()
+    _close_window(english_window)
     app.processEvents()
     set_locale("zh_CN")
 
@@ -1153,7 +1169,7 @@ def main() -> int:
             f"  reset label at 1280x720, {language}: live {live_width}px, "
             f"replay {replay_width}px (button width {button.width()}px)"
         )
-        measured.close()
+        _close_window(measured)
         app.processEvents()
     set_locale("zh_CN")
 
@@ -1320,7 +1336,7 @@ def main() -> int:
     window.toolbar.set_record_checked(False)
     window._stop_run(reason="user_stop")
     stuck.release()
-    window.close()
+    _close_window(window)
     app.processEvents()
 
     section("stopping a hung connect")
@@ -1388,7 +1404,7 @@ def main() -> int:
         "Esc did not cancel the hung connect immediately",
     )
     check(not window._running, "Esc left the session running")
-    window.close()
+    _close_window(window)
     app.processEvents()
 
     # Let the deliberately-stuck thread finish before the interpreter tears
@@ -1491,7 +1507,7 @@ def main() -> int:
         "the depth colour scale re-fitted per frame instead of staying fixed",
     )
 
-    window.close()
+    _close_window(window)
     app.processEvents()
     stuck.release()
 
@@ -1797,14 +1813,14 @@ def main() -> int:
         "the 2D image area does not follow the output image's aspect ratio: "
         f"{image.width()}x{image.height()}",
     )
-    finger_window.close()
+    _close_window(finger_window)
     app.processEvents()
     check(
         window.toolbar.record_chip.text() == t("record.state_idle"),
         f"the record strip did not fall back to idle: "
         f"{window.toolbar.record_chip.text()!r}",
     )
-    window.close()
+    _close_window(window)
     app.processEvents()
 
     section("chart header and unit chip")
@@ -1892,7 +1908,7 @@ def main() -> int:
     # The locale is process-wide: leave it where the other sections expect it.
     header_window._set_language("zh_CN")
     app.processEvents()
-    header_window.close()
+    _close_window(header_window)
     app.processEvents()
 
     section("acquisition correctness")
@@ -2156,7 +2172,7 @@ def main() -> int:
     pane_window.replay.close()
     pane_window._stop_run(reason="user_stop")
     stuck.release()
-    pane_window.close()
+    _close_window(pane_window)
     app.processEvents()
     _shutil.rmtree(replay_root, ignore_errors=True)
 
@@ -2323,7 +2339,7 @@ def main() -> int:
     )
     check(meta["stats"]["frames"] == 31, f"unexpected frame count: {meta['stats']['frames']}")
 
-    window.close()
+    _close_window(window)
     app.processEvents()
     stuck.release()
     _shutil.rmtree(scratch, ignore_errors=True)
@@ -2430,7 +2446,7 @@ def main() -> int:
             f"the session phase is {fail_window._session_phase!r} after a failed recording",
         )
 
-    fail_window.close()
+    _close_window(fail_window)
     app.processEvents()
     stuck_fail.release()
     _app_module.recordings_dir = lambda: scratch
@@ -2605,7 +2621,7 @@ def main() -> int:
         f"a matching calibration was not requested: {stuck_identity.calibration_requests}",
     )
 
-    identity_window.close()
+    _close_window(identity_window)
     app.processEvents()
     _shutil.rmtree(record_dir, ignore_errors=True)
 
@@ -2657,7 +2673,7 @@ def main() -> int:
         units_plots.history.sample_count > 0,
         "the rebuilt history lost every sample",
     )
-    units_window.close()
+    _close_window(units_window)
     units_window._box_worker = None
     stuck_units.release()
     app.processEvents()
@@ -2716,7 +2732,7 @@ def main() -> int:
         and not any(_math.isfinite(v) for v in one_state.values[:, 1]),
         "FX carried a value without a calibration curve",
     )
-    one_window.close()
+    _close_window(one_window)
     one_window._box_worker = None
     stuck_one.release()
     app.processEvents()
@@ -2903,7 +2919,7 @@ def main() -> int:
     check(plots._source_ready, "switching units detached the curve source")
     check(plots._unit == "N", f"the unit switch did not reach the charts: {plots._unit}")
 
-    window.close()
+    _close_window(window)
     app.processEvents()
     stuck.release()
 
@@ -2996,7 +3012,7 @@ def main() -> int:
         "the way back to the camera did not take effect",
     )
 
-    window.close()
+    _close_window(window)
     app.processEvents()
     stuck.release()
 
@@ -3141,7 +3157,7 @@ def main() -> int:
         pane.close_view()
         check(True, "closing the 3D pane twice is safe")
     finally:
-        pane_window.close()
+        _close_window(pane_window)
         app.processEvents()
 
     section("replay")
@@ -3325,7 +3341,7 @@ def main() -> int:
     )
     other_window.replay_ctl.stop_activity()
     app.processEvents()
-    other_window.close()
+    _close_window(other_window)
     app.processEvents()
     replay_window.activateWindow()
     app.processEvents()
@@ -3706,7 +3722,7 @@ def main() -> int:
     )
 
     replay_window.replay.close()
-    replay_window.close()
+    _close_window(replay_window)
     app.processEvents()
 
     _shutil.rmtree(replay_root, ignore_errors=True)

docs/test-execution-plan.zh.md

@@ -24,7 +24,7 @@
 
 代理开始验证前说明“改动 → 所选测试 → 扩大理由”。定向测试通过后结束验证；没有新增改动、失败或未解决风险时，不追加全量。全量只在最终候选状态运行一次；之后仅复验新改动涉及的测试，除非新改动再次触及公共基础。
 
-**提交前跑一次 `tools/gate.ps1`。** 定向测试用于开发中，门禁用于“这一版可以交给别人”的那一刻：它依次跑改动文件的 lint、studio 全量、shared 全量、`ui_check.py`，并要求 `tools/known_failures.txt` 为空、跳过数与收集数在 `tools/skip_budget.txt` 之内。可选 `pre-push` 钩子会调用它（安装方式见 [README](../README.md)，不自动安装）。门禁平台是 **Windows**：测试渲染真实 Qt 控件，而 [测试修复与门禁计划](test-repair-plan.zh.md) 第 3 节的 11、13 号只在 Linux 上失败，Linux 的通过不等于同一结论。
+**提交前跑一次 `tools/gate.ps1`。** 定向测试用于开发中，门禁用于“这一版可以交给别人”的那一刻：它先跑改动文件的 lint，再同时跑 studio 全量、shared 全量与 `ui_check.py`（`-Serial` 改为逐个运行并实时显示输出），并要求 `tools/known_failures.txt` 为空、跳过数与收集数在 `tools/skip_budget.txt` 之内。可选 `pre-push` 钩子会调用它（安装方式见 [README](../README.md)，不自动安装）。门禁平台是 **Windows**：测试渲染真实 Qt 控件，而 [测试修复与门禁计划](test-repair-plan.zh.md) 第 3 节的 11、13 号只在 Linux 上失败，Linux 的通过不等于同一结论。
 
 不能通过跳过失败、删除断言或把必需测试标成可选来提速。测试缺依赖、收集为空或意外跳过必须报告。间歇性错误先保留证据，查明影响后只重跑受影响范围；不循环重跑全量碰运气。
 

tools/gate.ps1

@@ -3,13 +3,17 @@
     Merge gate for orisys-app: lint the change, run both test suites, run the UI check.
 
 .DESCRIPTION
-    One command that a change cannot skip. It runs, in order:
+    One command that a change cannot skip. It runs:
 
       1. tools/lint_gate.py      the narrow lint rule set over the changed files
       2. pytest apps/studio/tests full, with the ``timing`` tests deselected
       3. pytest shared/tests      full, with the ``timing`` tests deselected
       4. apps/studio/tools/ui_check.py  the layout / theme / record-channel checks
 
+    Lint runs first; 2-4 then run at the same time (they write to separate places), so
+    the gate takes as long as the slowest of them. -Serial runs 2-4 one after another
+    with live output; the verdict is the same either way.
+
     and then holds the run to three contracts:
 
       * tools/known_failures.txt must be empty. Any content fails the gate: a known
@@ -44,6 +48,11 @@
 .PARAMETER SkipLint
     Run only the suites and the UI check (used when diagnosing a red gate).
 
+.PARAMETER Serial
+    Run the studio suite, the shared suite and ui_check one after another with live
+    output, instead of at the same time. The verdict is the same; use it to read a
+    failing stage as it runs, or when a parallel run is suspected of interfering.
+
 .EXAMPLE
     powershell -NoProfile -ExecutionPolicy Bypass -File tools/gate.ps1
 
@@ -58,7 +67,8 @@ param(
     [string]$Task = 'gate',
     [string]$Python = '',
     [string]$LintBase = 'origin/main',
-    [switch]$SkipLint
+    [switch]$SkipLint,
+    [switch]$Serial
 )
 
 $ErrorActionPreference = 'Stop'
@@ -218,22 +228,21 @@ try {
     }
 
     # -------------------------------------------------------------- test suites
-    function Invoke-Suite([string]$Name, [string]$Path) {
+    function Get-SuiteArguments([string]$Name, [string]$Path) {
         $safe = $Name.Replace(' ', '-')
-        $log = Join-Path $Scratch "$safe.log"
-        $xml = Join-Path $Scratch "$safe.xml"
-        $base = Join-Path $Scratch "pytest-$safe"
-        Write-Step "$Name"
-        $exit = Invoke-Native $Python @(
+        return @(
             '-m', 'pytest', $Path,
             '-m', 'not timing',
             '-p', 'no:cacheprovider',
-            "--basetemp=$base",
-            "--junitxml=$xml"
-        ) $log
+            "--basetemp=$(Join-Path $Scratch "pytest-$safe")",
+            "--junitxml=$(Join-Path $Scratch "$safe.xml")"
+        )
+    }
 
+    function Read-Suite([string]$Name, [int]$Exit) {
+        $xml = Join-Path $Scratch "$($Name.Replace(' ', '-')).xml"
         $counts = [pscustomobject]@{
-            Tests = 0; Failures = 0; Errors = 0; Skipped = 0; Exit = $exit; Xml = (Test-Path -LiteralPath $xml)
+            Tests = 0; Failures = 0; Errors = 0; Skipped = 0; Exit = $Exit; Xml = (Test-Path -LiteralPath $xml)
         }
         if ($counts.Xml) {
             try {
@@ -252,6 +261,51 @@ try {
         return $counts
     }
 
+    # A stage started in the background: stdout and stderr go to two files, joined into
+    # <stage>.log when it ends, so every stage leaves the same one log as a serial run.
+    function Start-Stage([string]$Name, [string[]]$Arguments) {
+        $safe = $Name.Replace(' ', '-')
+        $out = Join-Path $Scratch "$safe.out.tmp"
+        $err = Join-Path $Scratch "$safe.err.tmp"
+        # Start-Process joins -ArgumentList with spaces and quotes nothing, so each
+        # argument is quoted here ('not timing' is one argument).
+        $quoted = @($Arguments | ForEach-Object { '"' + ($_ -replace '"', '\"') + '"' })
+        $options = @{
+            FilePath = $Python
+            ArgumentList = $quoted
+            WorkingDirectory = $Repo
+            RedirectStandardOutput = $out
+            RedirectStandardError = $err
+            NoNewWindow = $true
+            PassThru = $true
+        }
+        $began = Get-Date
+        $process = Start-Process @options
+        # Windows PowerShell 5.1 reports no ExitCode for a -PassThru process unless its
+        # handle was read while the process ran.
+        $null = $process.Handle
+        return [pscustomobject]@{
+            Name = $Name; Process = $process; Out = $out; Err = $err; Began = $began
+            Log = (Join-Path $Scratch "$safe.log")
+        }
+    }
+
+    function Complete-Stage($Stage) {
+        $Stage.Process.WaitForExit()
+        $exit = $Stage.Process.ExitCode
+        # No exit code is not a pass: treat it like a crash that left no answer.
+        if ($null -eq $exit) { $exit = -1 }
+        $text = New-Object System.Collections.ArrayList
+        foreach ($part in @($Stage.Out, $Stage.Err)) {
+            if (Test-Path -LiteralPath $part) {
+                foreach ($line in (Get-Content -LiteralPath $part -Encoding UTF8)) { [void]$text.Add($line) }
+                Remove-Item -LiteralPath $part -Force -ErrorAction SilentlyContinue
+            }
+        }
+        Set-Content -LiteralPath $Stage.Log -Value $text -Encoding UTF8
+        return $exit
+    }
+
     #: An access violation while Qt tears the window down after the last assertion has
     #: been seen (exit 0xC0000005 as a signed int). It is allowed and reported; every
     #: other non-zero exit from a suite with a clean JUnit report is a gate failure.
@@ -311,18 +365,48 @@ try {
         exit 2
     }
 
-    Write-Step 'studio suite'
-    $studio = Invoke-Suite 'studio suite' 'apps/studio/tests'
+    # -------------------------------------------------------------- suites and ui_check
+    # The three stages are independent processes and write to separate places (studio
+    # tests: apps/studio/outputs/_test_record*, shared tests: shared/tests/_scratch,
+    # ui_check: apps/studio/outputs/_*_check, each pytest run its own --basetemp), so by
+    # default they run at the same time and the gate takes as long as the slowest one.
+    # Child output is UTF-8: a finding with Chinese text must not crash a cp1252 console.
+    $env:PYTHONIOENCODING = 'utf-8'
+    $stages = @(
+        [pscustomobject]@{ Name = 'studio suite'; Args = (Get-SuiteArguments 'studio suite' 'apps/studio/tests') },
+        [pscustomobject]@{ Name = 'shared suite'; Args = (Get-SuiteArguments 'shared suite' 'shared/tests') },
+        [pscustomobject]@{ Name = 'ui_check'; Args = @((Join-Path $Repo 'apps\studio\tools\ui_check.py')) }
+    )
+    $exits = @{}
+    $durations = @{}
+    if ($Serial) {
+        foreach ($stage in $stages) {
+            Write-Step $stage.Name
+            $log = Join-Path $Scratch "$($stage.Name.Replace(' ', '-')).log"
+            $began = Get-Date
+            $exits[$stage.Name] = Invoke-Native $Python $stage.Args $log
+            $durations[$stage.Name] = ((Get-Date) - $began).TotalSeconds
+        }
+    } else {
+        Write-Step 'studio suite, shared suite and ui_check, in parallel (-Serial for live output)'
+        $running = @(foreach ($stage in $stages) { Start-Stage $stage.Name $stage.Args })
+        foreach ($stage in $running) {
+            $exits[$stage.Name] = Complete-Stage $stage
+            # StartTime cannot be read once a process has exited; the launch time is ours.
+            $durations[$stage.Name] = ($stage.Process.ExitTime - $stage.Began).TotalSeconds
+            Write-Host ("  {0,-13} exit {1}  {2:n0} s  ({3})" -f $stage.Name, $exits[$stage.Name], $durations[$stage.Name], $stage.Log)
+        }
+    }
+
+    $studio = Read-Suite 'studio suite' $exits['studio suite']
     Test-Suite 'studio suite' $studio $budget 'studio_min_tests' 'studio_max_skips'
 
-    Write-Step 'shared suite'
-    $shared = Invoke-Suite 'shared suite' 'shared/tests'
+    $shared = Read-Suite 'shared suite' $exits['shared suite']
     Test-Suite 'shared suite' $shared $budget 'shared_min_tests' 'shared_max_skips'
 
     # -------------------------------------------------------------- ui_check
-    Write-Step 'ui_check'
     $uiLog = Join-Path $Scratch 'ui_check.log'
-    $uiExit = Invoke-Native $Python @((Join-Path $Repo 'apps\studio\tools\ui_check.py')) $uiLog
+    $uiExit = $exits['ui_check']
     # ui_check prints "scenes: 4   checks: 631   failures: 2"; the last number is the
     # verdict. Its exit code is also the verdict: a script that dies partway through
     # (missing worker signature, an import error) must not pass on a stale summary line.
@@ -358,6 +442,9 @@ try {
     [void]$lines.Add("Platform : Windows")
     [void]$lines.Add("Python   : $Python")
     [void]$lines.Add("LintBase : $LintBase")
+    $mode = if ($Serial) { 'serial' } else { 'parallel' }
+    $timing = ($stages | ForEach-Object { "{0} {1:n0} s" -f $_.Name, $durations[$_.Name] }) -join ', '
+    [void]$lines.Add("Stages   : $mode; $timing")
     [void]$lines.Add("Scratch  : $Scratch")
     [void]$lines.Add('')
     [void]$lines.Add('| Check | Result | Detail |')