Skip to content

testing.md

Latest commit

 

History

History
44 lines (29 loc) · 5.58 KB

testing.md

File metadata and controls

44 lines (29 loc) · 5.58 KB
Raw

Validation

Offline app changes

npm ci
npm run check
npm run check:tools
npx playwright install chromium
npm run check:browser

check runs ES5 runtime lint, validates package identity/version and bundled resources, then tests the production device adapter with a synthetic API/clock. It does not contact a TV. check:browser uses a fresh Playwright Chromium profile and its own loopback preview. It checks List and Dock navigation, single focus, Home controls, grouped Settings, explicit OK application, applied checkmarks, pending-load cancellation, retries, persistence, all fifteen wallpaper bundles, available fonts, 720p scaling, migrations, the Home editor and honest browser-only launch errors. Reports and screenshots are in ignored .local/qa/. A failed assertion exits with a nonzero status.

For public-source validation, set CLEANFRAME_PUBLIC_FONTS=1 when running check:browser; the preview helper then excludes the local proprietary overlay. Run npm run check:public for the bounded source/history checks and npm run export:public for a widget that always excludes private fonts. The GitHub Actions workflow uses this public path. Browser/package checks do not clear third-party asset rights or validate TV behaviour.

check:tools uses synthetic credentials and a fake SDK in isolated temporary folders. It checks local configuration, identity overlays, owner-only password files, both signing modes, failed-build handling, private-file rejection, public signature metadata and real Git index/history blobs. It contacts no TV and reads no real signing credentials. Temporary fixtures are retained for inspection instead of permanently deleted.

check:public checks working files, index blobs and all reachable Git refs (including fetched PR/checkpoint history) for the shared private-file policy and recognisable credential patterns. A working-file edit cannot hide a staged credential. Packaging and public export apply the same checks to widget contents; public signing certificates and signature XML are allowed. These checks are bounded and do not detect every possible secret. For release validation, also use a clean clone of the candidate branch without unrelated review refs. The generated-path policy rejects any bin or obj directory in source, plus widget/DLL/debug-symbol/executable files; it is deliberately broader than the native-helper folder.

Optional local commit/push guards are in .githooks/. To enable them deliberately for this clone, run git config --local core.hooksPath .githooks after checking that this will not replace an existing hooks setup. Run npm run check:public before enabling them. That all-ref audit must pass: unrelated fetched PR/checkpoint history can block every commit/push even when the working tree is clean. Use a clean candidate clone when retaining rejected review refs elsewhere. The hooks run check:public before committing or pushing. They are not enabled automatically and can be bypassed. Pre-push is the last local preventive check; CI runs after upload and can only detect exposure at that point. A later fix commit does not remove sensitive data from earlier commits. Keep CI and review as additional checks.

If the default browser/cache location is unavailable, use a writable local path for both commands:

PLAYWRIGHT_BROWSERS_PATH="$PWD/.local/playwright" npx playwright install chromium
PLAYWRIGHT_BROWSERS_PATH="$PWD/.local/playwright" npm run check:browser

For manual review, use npm run dev and open /home/. Use the arrow keys, Enter and Escape. Check the changed flow in List and Dock, on Photo, Light and Dark backgrounds. Check a single app, nine apps and a longer order when changing app layout. In Settings, browse without applying, apply with OK, return with both Back routes, and cancel a pending resource load. Check editor placement/undo, wallpaper rotation pausing during panels/background time, and restored focus on return when those flows change. Browser preferences are separate from TV preferences.

CLEANFRAME_PUBLIC_FONTS=1 npm run check:performance runs the existing 4× CPU-throttled Chromium regression screen. Compare navigation work, frame stalls, retained DOM/heap and live-filter checks before and after rendering changes. This is not a physical-TV benchmark.

Physical TV checks

Run these only under a task that includes TV testing. Announce any installation or launch first, then follow development.md. SDK success alone is not a visible result. Ask this TV's owner to confirm the normal remote behaviour and record the version, firmware, result and any open checks in nu8000.md and the summary in tv-status.md.

  1. Move in both directions through each layout and Settings; check focus, response and readable labels from the sofa.
  2. Launch Netflix/YouTube, press physical Home and confirm CleanFrame opens on this fork's currently configured NU8000. Check the selected app. If the separate Home assignment has been removed, confirm Samsung Home instead. Treat streaming apps' own Back handling separately.
  3. Check Other apps, TV settings and Settings → Samsung → Samsung Home; verify the actual visible destinations. Check Art Mode only on a TV that has the feature.
  4. Check saved appearance/app order/wallpaper after relaunch. Check timed rotation pauses while a panel is open or the app is hidden.
  5. Test normal power-on after a streaming app only when that work is included. This NU8000 has no native Home reassignment or recovery baseline; never use the upstream baseline for it.

The current open checks are summarized in tv-status.md. Browser CPU throttling is a development comparison, not a TV performance measurement.