Desktop E2E (Windows) — Edho Ferdian Mode

Where this sits

e2e-testing-edho-ferdian owns the process: Phase 1 journey mapping, Phase 2 Page Object Model, Phase 3 flake quarantine, Phase 4 artifact capture. All four are driver-agnostic and apply unchanged here — do not restate or fork them. This skill supplies only what changes when the driver is native Windows automation instead of a browser:

Concept	Browser (Playwright)	Here (pywinauto + UIA)
Locator	role / data-testid / CSS	AutomationId → Name → ClassName+index
Wait	auto-waiting locators	spec.wait("visible") / wait_until(...)
Artifact	screenshot / video / trace	window capture / ffmpeg gdigrab / step trace
Isolation	browser context	fresh process + redirected APPDATA/TEMP

Not for: web apps (use Playwright), Electron/CEF/WebView2 (the HTML layer needs browser automation, UIA only sees the shell), mobile, or anything that does not need a running GUI.

How it works

Everything routes through Windows UI Automation, an accessibility API built into Windows. Each UI framework ships a UIA provider of varying quality:

pytest → pywinauto (uia backend) → Windows UI Automation API → app's UIA provider → running .exe

Framework	AutomationId source	Reliability
WPF	x:Name maps directly	Excellent
UWP / WinUI 3	native	Excellent
Qt 6.x	setAccessibleName; accessibility on by default	Excellent
WinForms	AccessibleName	Good
Qt 5.15+	needs QT_ACCESSIBILITY=1	Good
Qt 5.7–5.14	needs QT_ACCESSIBILITY=1; manual objectName	Fair
Win32 / MFC	resource control IDs appear as numeric strings	Fair

Phase 0′ — Environment check (before anything else)

pip install pywinauto pytest pytest-html pytest-timeout Pillow

from pywinauto import Desktop
Desktop(backend="uia").windows()   # must list top-level windows

Install Accessibility Insights for Windows (free, Microsoft) — it is the DevTools equivalent for this stack. Inspect the UIA tree there before writing a locator; guessing at AutomationIds is the main time sink here.

If UIA is unreachable, or the target turns out to be Electron/WebView2, stop and say so — report VERIFIED: NO — no usable native automation driver, the same convention the rest of this ecosystem uses. Never fabricate a run.

Testability first

The single highest-leverage action is giving every interactive control a stable AutomationId before writing tests. Per framework:

• WPF — x:Name="btnLogin" in XAML; nothing else needed.
• WinForms — btnLogin.AccessibleName = "btnLogin";
• Qt — set both, via one helper, and centralize the string constants in a header so tests and app cannot drift:

  void setTestId(QWidget* w, const char* id) {
      w->setObjectName(id);
      w->setAccessibleName(id);   // becomes the UIA Name property
  }
  

• Win32 / MFC — resource IDs surface as numeric AutomationId strings; prefer SetWindowText for a readable Name.

A test suite written against ClassName + index because the app has no AutomationIds will be rewritten within a release. Fixing testability in the app is cheaper than the flakiness it prevents — say so when the app owner resists.

References

• references/pywinauto-harness.md — the Page Object base class, locator priority, wait patterns, artifact capture, the opt-in per-step trace, and the flaky-cause table.
• references/isolation-and-ci.md — the three isolation tiers (tmp_path env redirect / Job Object / Windows Sandbox), hang prevention, and the GitHub Actions windows-latest workflow.
• references/qt-and-fallback.md — Qt-specific quirks (combo popups, dialogs as top-level windows, table cells, self-drawn widgets) and the screenshot-matching fallback with its DPI rules.

Language routing (fixed — see skill-authoring-edho-ferdian's canonical contract)

Communication to the user in Bahasa Indonesia; test code, page-object names, and assertions in English — fixed, never ask. Full contract: skill-authoring-edho-ferdian §7.

Global rules

1. AutomationId first, always. Name second, ClassName+index only when forced, coordinates never as a primary strategy.
2. Never time.sleep() as synchronization. wait_visible / wait_gone / wait_until exist for exactly this.
3. Fresh process per test. A session-scoped app fixture leaks state between tests and turns every failure into an archaeology exercise.
4. Assert content and state, never pixel geometry. lblStatus == "Logged in" is a test; btn.rectangle().left == 120 is a trap.
5. Redact typed text in traces. Never enable trace text capture on a login or payment flow.
6. Quarantine, don't delete — the protocol in e2e-testing-edho-ferdian/references/flake-quarantine.md applies verbatim; the Python spelling is @pytest.mark.skip(reason="...") / @pytest.mark.skipif(os.environ.get("CI") == "true", reason="...").