Repository navigation
Quad sometimes stays black: container never laid out after the app is brought to front #56
Description
Seen in device testing (#53, AG-216 Android TV box, Android 10). Deferred: the time-boxed investigation did not find the root cause. #48 adds a mitigation.
Symptom: after Quad is turned on from the menu, the screen stays black. No cell starts a session (no Quad-N: surface ready log). Home → back fixes it.
State while it happens:
dumpsys activity top: the QuadLinearLayoutisVISIBLEbut its bounds are0,0-0,0, and the rows andTextureViews are also 0×0 with the invalidated flag set. The cells'TextureViews therefore never get aSurfaceTexture.ViewRoot: mTraversalScheduled=false: no layout pass is pending. The activity ismResumed=true mStopped=false. No ANR; the main thread is idle.
When: 3 of ~8 attempts, every time after am start … SEND was delivered to the already running activity (the task is brought to the front). On this box the launcher is also known to take focus about 1 s after such a start. It was never seen when Quad was entered without such a start. It may be specific to the test harness, but the app should recover anyway.
Mitigation in #48 (c332a2c): 500 ms after Quad is shown, if the container still has no size, the app forces a layout pass (forceLayout() up the chain, then requestLayout() + invalidate() on the decor view) and logs Quad layout did not happen, forcing a layout pass.
To do later: confirm whether the mitigation fires and is enough; find out why the layout request is lost.
Activity
compot-dev commentedon Oct 12, 2026
Re-test on the final v2 build
55f07ff(2026-10-12)AG-216, Android 10, three cameras in Quad, slot 4 empty.
Quad layout did not happen)am start … SENDdelivered to the running app → menu → Qam start(task brought to front)Not reproducible any more. The fallback layout pass never had to run, so the container was laid out by the normal pass every time.
I checked one hypothesis and it is ruled out:
singleTop(cfbb492) is not what fixed it. The original failures were on4660bba, which already had it. The root cause is still unknown; it is probably a timing race with the launcher taking focus right after the start, as noted above.The mitigation (
c332a2c) stays in v2 and logsQuad layout did not happen, forcing a layout passif it ever runs, so a recurrence will show up in a Device report logcat.compot-dev commentedon Oct 12, 2026
Closing as not reproducible (20 runs on
55f07ff, see above). The fallback layout pass stays in place and logsQuad layout did not happen, forcing a layout pass; if that line shows up in a Device report, reopen this issue.