Repository navigation
test: give each rotated Android display state its own baseline, run the Android 16 rotation test again - #1308
Conversation
…rotation On LambdaTest Android 16 the display state after setOrientation does not always follow the state at the start of the session (the rotated page is 952 or 903 pixels wide). Use the viewport after the rotation in the tag, so each rotated state has its own baseline. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The rotated screenshot now has its own baseline for each display state after the rotation, so the skip rule is not needed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Remove before merge. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
…d 16 426x827 state From 6 runs of the LambdaTest Android job (run 37937247476, attempts 1-6): - 8 rotated states, named with the viewport after the rotation (Android 15 and 16, both directions, 2 states each); 7 were seen 2 to 4 times with the same image except the status bar, the Android 16 landscape-to-portrait 426x823 state once. - A third Android 16 start state (426x827), seen once: its 6 baselines. Remove the 4 old Android 15 rotated baselines, which had the viewport at the start of the session in their names. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Reverts 433e928. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Why
The test "should compare a screen after an orientation change" is skipped on LambdaTest Android 16 (#1274). After
setOrientation, the emulator does not always come to the same display state: the rotated page is 952 or 903 pixels wide (display cutout used or left out), and this changes between sessions. The rotated screenshot was named with the viewport at the start of the session (-vp426x848), so one baseline had to match 2 different states (2.674 % mismatch in run 37824184770).On a local Android 16 emulator the state after the rotation is stable (6 of 6 sessions:
900x295), so this happens only on LambdaTest.What
tests/specs/mobile.web.spec.ts: on Android, the tag of the rotated screenshot is the viewport measured after the rotation (-vp${innerWidth}x${innerHeight}), so each rotated state has its own baseline. The Android 16 skip rule for the rotation test is removed.New baselines in
tests/lambdaTestBaseline/pixel_9_pro/, collected from 6 runs of the LambdaTest Android job (run 37937247476, attempts 1–6, with a temporaryautoSaveBaseline: trueand an upload, both reverted in this PR):vp903x322(4×),vp952x322(2×)vp903x322(3×),vp952x322(3×)vp426x848(3×),vp426x823(3×)vp426x848(5×),vp426x823(1×)Images with the same name differ only in the status bar (clock, icons) and Chrome's page-load line under the address bar.
A third Android 16 start state,
426x827(seen once in run 4): its 6 baselines are added. Without them, a session in that state fails all its checks onmaintoo.The 4 old Android 15 rotated baselines (named with the start viewport) are removed.
The spec comment about the display states names the new state.
Test
pnpm test.local.emus.websetup and compare runs, 7/7 passing; the rotated baseline is namedscreenshot-landscape-vp900x295-….autoSaveBaselineoff, the temporary commit reverted), run 37963619201: all 5 jobs pass; the orientation change test passes on Android 15 and 16 in both orientations (Android 14 stays skipped, as before).🤖 Generated with Claude Code