screentinker/docs/109-android-pip-visibility.md
ScreenTinker 10abb3b35f docs(#109): record web PiP (HTML+JS) verification on emulator
Web PiP type loads its WebView and executes JS (a page stamping JS OK · <time>
rendered over live YouTube). No code change — web PiPs don't use the image path
that had the token bug. Completes the image/web/box content-type verification.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-19 14:38:30 -05:00

178 lines
8.9 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# #109 — Android PiP overlay not painting over YouTube
## Symptom
`POST /api/pip` returns `sent: 1`, the overlay's title text appears in
`uiautomator dump` (so the view IS attached, laid out, and on-screen in the
accessibility tree), but **nothing paints on the panel**. The repro is while
**YouTube** content is playing (`R.id.youtubeWebView`).
The PiP title and its media are siblings in one box, so "title in the dump but
nothing on screen" means the **whole box is attached-but-not-painting**, not a
media-only failure.
## The three candidate causes
| # | Cause | What the magenta-box instrumentation shows |
|---|-------|--------------------------------------------|
| 1 | **Surface occlusion** — the YouTube WebView's hardware video plane composites *above* the in-tree overlay | magenta box visible over a static image but **not** over YouTube |
| 2 | **Orientation transform**`rootView`'s rotation/translation pushes the box off-screen | box `getGlobalVisibleRect()` empty / outside the 1920×1080 panel |
| 3 | **Measure / visibility**`pipLayout` not laid out, 0-size, or `GONE` on attach | box not shown / 0-size / `pipLayout` `childCount==0` |
## Phase 1 — Instrumentation (shipped, default OFF)
A `pipDebug` flag (`PipOverlay.pipDebug`, default `false`) is toggled over the
**existing** `device:command` transport — no new transport invented:
```
device:command { "type": "pip_debug", "payload": { "enabled": true } }
```
When on, `PipOverlay.show()`:
- paints the box **solid magenta** (`#CCFF00FF`) with an 8px magenta border and
renders the media on top, so the **box paints even if the media never loads**;
- posts a one-shot Runnable that logs, over `device:log` tag **`pip`**:
- `box` width/height, `getGlobalVisibleRect()`, `isShown`
- `pipLayout` width/height/visibility/childCount + its index in its parent
- `rootView` rotation / translationX / translationY / scaleX / `isHardwareAccelerated`
- `youtubeWebView` visibility + `getGlobalVisibleRect()`
- `loadImageInto` also logs on **success** (bitmap w/h), not just failure.
`pipDebug` is left present and default-false.
## Phase 2 — Reproduce
Enable remote debug logging from the dashboard (so `device:log` is forwarded),
then enable the PiP debug flag, then fire a PiP under each content type and read
the `pip`-tagged lines.
**(a) PiP over a static image**
1. Assign a single still image to the device; let it display.
2. `device:command {type:"pip_debug",payload:{enabled:true}}`.
3. `POST /api/pip {device_id, type:"image", uri:"https://…/x.png", position:"top-right", duration:30}`.
4. Capture the `pip dbg …` lines + a screenshot.
**(b) PiP over YouTube** (the failing repro)
1. Assign a YouTube item; let it play in `R.id.youtubeWebView`.
2. (debug already enabled.)
3. Same `POST /api/pip` as above.
4. Capture the `pip dbg …` lines + a screenshot.
Decision table (compare a-vs-b):
- magenta box visible over the image but **NOT** over YouTube → **(1) surface occlusion**
- box `globalRect` empty / off the 1920×1080 panel → **(2) orientation**
- box not shown / 0-size / not laid out → **(3) measure/visibility**
> Note: on-device capture must be done on the real signage hardware (Fire TV /
> Android TV). The WebView hardware-video-overlay behaviour that drives cause (1)
> is device- and WebView-version-specific and does **not** reproduce on a stock
> emulator, so it cannot be captured from a CI/dev box with no device attached.
## Which cause — and the fix
By elimination the symptom points at **(1) surface occlusion**:
- It is **YouTube-specific** (a WebView playing HTML5 video). An orientation (2)
or measure (3) fault would fail over images too, but the overlay is only
reported broken over YouTube.
- The repro is **landscape** (the default orientation → `rotation = 0`, no
translation), so the box cannot be transformed off-screen → not (2).
- The title shows with **real on-screen bounds** in `uiautomator dump`, so the
box is laid out at non-zero size and `pipLayout` is `VISIBLE` → not (3).
That leaves the video surface compositing above the in-tree overlay.
### Fix (cause 1) — file/line
`pipLayout` previously lived as the **last child of `rootLayout`**, i.e. in the
**same compositing band** as `R.id.youtubeWebView`; the WebView's playing video
surface drew over it. The fix moves the PiP layer to a **top-level view above the
WebView** (the task's option 1a):
- **`MainActivity.onCreate`** (`android/app/src/main/java/com/remotedisplay/player/MainActivity.kt`)
reparents `R.id.pipLayout` out of `rootLayout` up to the window content
(`android.R.id.content`), as a sibling drawn **after** `rootLayout` → it
composites above the WebView.
- **`MainActivity.mirrorTransformToPip()`** copies `rootView`'s current size +
rotation/translation/scale onto `pipLayout` after every transform change
(`applyOrientation` / `applyWallTransform`), so corner positions still track the
rotated content — mirroring how the web/Tizen players apply the same transform
to `#pip` as to `#stage`.
- **`PipOverlay.show()`** (`…/player/PipOverlay.kt`) raises the layer and forces a
layout/redraw on attach (`bringToFront()` + `requestLayout()` + `invalidate()`),
which also covers the cause-(3) measure/visibility path.
- The remote-view screenshot source moved from `rootView` to `captureRoot`
(the window content) so the reparented PiP is still captured.
### Server dispatch logging
`POST /api/pip` and the clear handler (`server/routes/pip.js`) now log one
concise `[pip] …` line each (target kind + id + sent/offline counts) so
`journalctl` shows PiP activity.
## Emulator validation (landscape + portrait)
The fix was exercised end-to-end on an Android emulator (pixel10, API 34) paired
to an isolated local server, with a YouTube item playing in
`R.id.youtubeWebView`:
- **No crash** — provisioning → `MainActivity` → playback ran clean; the reparent
+ `mirrorTransformToPip()` executed (`Applied orientation: landscape … / portrait
(rotation=90.0, swap=true)`).
- **PiP composites above the playing YouTube video** — a `POST /api/pip` box
(magenta via `background_color`) rendered on top of the live video frame
(center and top-right placements both correct, 4% inset honoured).
- **Clear** removed the overlay cleanly; the video kept playing.
- **Portrait** — the overlay rotated *with* the rotated stage and stayed inside
the frame (not off-screen), confirming the transform mirror.
- Server `[pip] show … 1 sent` / `[pip] clear (all) …` dispatch lines appeared.
Caveat (unchanged): the emulator's WebView composites video **inline**, so it
confirms the reparent renders correctly and doesn't regress, but it does **not**
reproduce the Fire TV / Android TV hardware-overlay punch-through that is the
strongest form of cause (1). That still needs the real signage device — use the
`pipDebug` magenta box there to confirm.
### Follow-up bug found in emulator testing: image PiPs never painted the image
Verifying an **image** PiP (a QR PNG) surfaced a separate, pre-existing defect
(present on `main`, unrelated to the occlusion reparent): the image area was
always blank — only the box background + title showed. Root cause in
`PipOverlay.show()`: `teardown()` clears `current` to null, then `loadImageInto`
captured `token = current` (null) as its drop-if-replaced guard, but `current`
was only set to the new `pip_id` *after* the media was built. The decode finishes
on a background thread and posts back **after** `show()` returns — so the guard
`token != current` (null ≠ pip_id) was always true and **every decoded bitmap was
dropped**. (Web PiPs and the box/title were unaffected, which masked it.)
Fix: set `current = pip_id` **before** building the media (so `loadImageInto`'s
token matches). Confirmed on the emulator — the QR now renders in the PiP box
over both a static image and live YouTube.
### Content types verified on the emulator (over live YouTube)
- **image** PiP (a QR PNG) — renders after the token-ordering fix above.
- **web** PiP (an HTML page) — loads in the PiP WebView and **executes JS** (a
page that stamps `JS OK · <time>` into the DOM rendered correctly over the
playing video); composites above the main YouTube WebView. No code change
needed — web PiPs never went through the broken image path.
- title + `background_color` box — paints above the video (the original cause-1
fix).
## If the magenta box is STILL hidden over YouTube on the test device
Then it is the stronger form of cause (1): the WebView places its video on a
**hardware overlay / `SurfaceView` plane** that no in-window view can beat.
Escalate (task options 1b/1c), keeping the first that works on the device:
- host the PiP box in a `SurfaceView` with `setZOrderMediaOverlay(true)` /
`setZOrderOnTop(true)`, or a small `WindowManager` panel sub-window; or
- when YouTube is active, render the PiP via the in-tree image path so no
competing WebView video surface is involved.
The `pipDebug` instrumentation stays in place to make that determination.