fix(android): return from orientation once the display reports the rotation - #2356
Conversation
Size Report
Startup median (7 runs, lower is better):
|
|
The new |
88516f8 to
3104aa0
Compare
|
Addressed, rebased onto main (one commit):
Local: platform-android 645 tests, provider Android scenarios 13, root typecheck, oxlint, oxfmt, fallow (against main) clean. |
|
The bounded probe and provider scenario fixes are in place at 3104aa0. One failure path still returns success: dumpsys display runs with allowFailure, but its exit code is ignored. A failed ADB call with no orientation in stdout therefore looks like a successful probe that lacks the field. Reject failed probes through the existing ADB error handling and add a non-zero-exit regression; preserve the no-field case only for a successful probe. Android smoke also remains red at the later semantic-canary step. |
3104aa0 to
eb426b1
Compare
The first CI failure with the evidence hook (PR #2356, run 34029660070) lost its logcat section: dumping the emulator's whole 2MB buffer took longer than the 5s per-probe bound on the loaded host. The probe now reads the last 4000 lines, which holds the rotation decisions of the last minutes and returns well inside the bound.
|
The failed-display-probe handling is fixed at eb426b1: nonzero exits now fail the operation, with a dedicated regression. The live Android run passed both rotation assertions before failing at automation-press, so that failure is likely unrelated to this change. Ready for human review. |
thymikee
added
the
ready-for-human
…tation `orientation` wrote accelerometer_rotation and user_rotation and returned at once, while the display rotated some time later. On the loaded CI emulator that takes seconds, and accessibility reads hang meanwhile: the Android smoke's `wait text landscape` right after `orientation landscape-left` got a helper request timeout and then no readable capture for its whole 10s budget, with the failed-step snapshot taken afterwards already in landscape (PR #2344, run 34025424834). The command now polls `dumpsys display` for mCurrentOrientation to match the requested rotation before returning, each probe bounded by what is left of the 15s settle budget so a stuck probe ends the settle as a failure. A display that never gets there fails the command with the observed rotation instead of reporting success; a display that reports no rotation at all is left to the setting as before. The provider scenario scripts the display read against the last user_rotation write.
eb426b1 to
cfb529b
Compare
thymikee
removed
the
ready-for-human
|
Addressed at cfb529b (rebased onto main, which now includes #2369):
|
|
Reviewed cfb529b. No remaining code findings. Failed orientation probes now preserve the ADB error, and the single-probe regression remains covered. Android emulator smoke and the remaining checks pass on this head. Good for human review. |
thymikee
added
the
ready-for-human
|
The first CI failure with the evidence hook (PR #2356, run 34029660070) lost its logcat section: dumping the emulator's whole 2MB buffer took longer than the 5s per-probe bound on the loaded host. The probe now reads the last 4000 lines, which holds the rotation decisions of the last minutes and returns well inside the bound.
1 participant
Summary
Android
orientationwroteaccelerometer_rotationanduser_rotationand returned at once, while the display rotated some time later. On the loaded CI emulator that takes seconds, and accessibility reads hang meanwhile.Evidence: the Android Smoke on #2344, https://github.com/callstack/agent-device/actions/runs/34025424834.
wait text landscaperight afterorientation landscape-leftreportedwait_capture_stalledwith zero readable captures: the helper session request timed out at 5s, the one-shot fallback ran into the wait's deadline, and the failed-step snapshot taken right after was already in landscape (nodes 2400px wide). The rotation had landed; the wait's whole budget went to the transition.Change
setAndroidOrientationnow pollsdumpsys displayformCurrentOrientationto match the requested rotation, every 500ms up to 15s, before returning.mCurrentAppOrientation=SCREEN_ORIENTATION_NOSENSOR),orientation landscape-leftpreviously "succeeded" without rotating and now fails after 15s withobservedRotation: 0.mCurrentOrientationat all is left to the setting, as before.Validation
requestedRotation: 1, observedRotation: 0when it never does; leaves a display that reports nothing to the setting. Package suite 641 tests green; typecheck, oxlint, oxfmt, fallow clean.orientation landscape-left1s,mCurrentOrientation=1;orientation portrait1s,mCurrentOrientation=0.What this does not explain
The other Android Smoke failure family (the post-alert canary missing while the device is in landscape although
orientation portraithad been confirmed) is still open; #2350 adds the rotation evidence for it. This change removes the transition cost from the command that starts a rotation, which is also where a later flip would be visible if it is a delayed re-application.