fix(phase-1/m2.4b): encrypted imports were a silent no-op on-device

Manual Waydroid pass (phone exports encrypted -> re-imports the same
file through the real UI) caught it: the confirm dialog closed but no
import ran and no snackbar appeared. PassphraseDialog.onConfirm's
shared preamble nulls all URI slots before routing on mode, and the
IMPORT branch never re-armed pendingImportUri -- so the preview
confirm dialog's Import button read null and skipped viewModel.import
entirely. Plain-JSON imports bypass the passphrase dialog, which is
why every scripted desktop leg (M2.1-2.5) stayed green.

Fix: re-arm pendingImportUri in the IMPORT branch before preview;
wrong-pw/dismiss paths still null it and zero the retained chars.

Verified on-device after rebuild + `waydroid app install`: export
snackbar, magic-sniffed "Decrypt backup" sheet, 4-note preview, and
the dedupe snackbar "Imported 0 note(s), skipped 4 already imported."

Also lands the Waydroid test rig used for the pass (tools/WAYDROID_
TESTING.md + tools/inject2.c: uinput tap/move/key/text/swipe injector)
and the full M2.4b write-up in DESKTOP_FEASIBILITY.md. M2 exit
criteria met: bidirectional lossless round-trip plain+encrypted.
This commit is contained in:
Incido Developer committed 2026-10-06 05:40:44 +13:00
1 parent 29288b0370
commit 561d16eb0c
4 files changed
+412

No files matched your search

+184
View File
@@ -0,0 +1,184 @@
# Waydroid: APK Install & Functional Testing Guide
Reference notes from the draft/trash feature testing session (23 Sep 2026),
on this machine: KDE Plasma (Wayland), Waydroid full session, Incido app
(package `org.incido.app`).
---
## 1. Prerequisites / environment check
```bash
waydroid status
# Want: Session: RUNNING, Container: RUNNING
```
If the session is not running (session-only command, no root needed):
```bash
waydroid session start # starts the UI session (runs as your user)
```
Starting the **container** (`waydroid container start`) requires root; it is
normally managed by a systemd unit and already running.
The app's data lives on the host at
`~/.local/share/waydroid/data/data/org.incido.app/` but is owned by root —
reading the SQLite DB from the host needs `sudo` (or `sudo waydroid shell`).
If `sudo` demands a password (as on this machine) you **cannot** use root
paths in automated runs — plan to verify through the UI instead.
## 2. Build & install the APK
```bash
cd ~/Desktop/Apps/Incido
./gradlew :app:assembleDebug -q # exit 0 = success
# APK lands at: app/build/outputs/apk/debug/app-debug.apk
waydroid app install app/build/outputs/apk/debug/app-debug.apk
```
Notes learned the hard way:
- `waydroid app install` needs **no root** — always prefer it over any
`adb install` / `cp` into the container.
- There is no attached adb device by default (`adb devices` → empty), so
plain adb workflows do not work unless you enable
`waydroid root`/adb integration first.
- If the app was already running while reinstalling, relaunch it — the
running process may keep the old code:
```bash
waydroid app launch org.incido.app # also no root needed
waydroid app list | grep -i incido # confirm installed package
```
## 3. Screenshots
`scrot`, `import` (ImageMagick is installed but has no usable X display),
and `gnome-screenshot` are **not available**. Use Spectacle in CLI mode:
```bash
spectacle -b -n -o /tmp/shot.png # -b no border, -n output file
```
Read the PNG back with the file-read tool to visually assert UI state.
## 4. Input injection (clicks / typing)
**xdotool does not work** for injecting into Waydroid under Wayland
(`XDG_SESSION_TYPE=wayland`). KWin exposes an XWayland socket, but window
search/injection via xdotool returns nothing useful.
The working solution is a userspace **uinput** injector:
[`tools/inject2.c`](./inject2.c) kept in this repo. Build once:
```bash
mkdir -p /tmp/incido_test && cd /tmp/incido_test
gcc ~/Desktop/Apps/Incido/tools/inject2.c -o inject2
# needs write access to /dev/uinput (group "input" or udev rule)
```
Usage — `click` etc. are SUBCOMMANDS (bare coordinates fail with "bad args"):
```bash
./inject2 move X Y # warp virtual pointer
./inject2 click X Y # warp + left click
./inject2 press # click at current position
./inject2 drag X1 Y1 X2 Y2
./inject2 wheel N # scroll
./inject2 type "text" # virtual keyboard device
./inject2 key NAME # esc, enter, backspace, up, down, left, right, ...
INJECT_HOLD=500000 ./inject2 click X Y # keep uinput devices alive 0.5s
```
### Coordinate mapping (important!)
Screenshots come back at **2560×1440**, but the injector emits absolute
events calibrated for a **1920×1080** screen. Therefore:
```
inject_x = screenshot_x × 0.75
inject_y = screenshot_y × 0.75
```
Measure the target on the screenshot, multiply by 0.75, then click that.
Known-good targets (already in inject2 space):
hamburger `(30, 35)`, drawer Home `(85, 104)`, drawer Drafts `(85, 206)`,
drawer Trash `(101, 403)`, card overflow menu ≈ `(1868, 111)`,
menu "Delete permanently" row ≈ `(1792, 230)`.
### Workflow that worked well
- One `click` + `sleep 1` + `spectacle` per step; read the PNG to assert.
- After deleting the top list item the list **reflows** into the same
position, so the same coordinates can be reused for repeated deletes.
## 5. Known failure mode: injection silently stops
Mid-session (after ~30 min of successful clicks), injected pointer AND
keyboard events stopped reaching the Waydroid window entirely — screenshots
showed no reaction to any input, while `spectacle` still worked and the app
process was alive (`ps -ef | grep incido`).
Things that were tried and did **not** fix it (don't loop on these):
- Re-activating the window via KWin Scripting DBus
(`qdbus6 org.kde.KWin /Scripting ...loadScript/start` with a JS that sets
`workspace.activeWindow`) — the script runs and logs activation, but
input routing is unaffected.
- `waydroid app launch` relaunch of the app.
- Escape/Back key injection.
- Repeated clicks at freshly computed coordinates.
Likely cause: KWin-side routing of the transient uinput device
(created/destroyed per invocation) gets detached — e.g. after the session
loses/gains focus, or the Python Waydroid UI stops hot-plugging newly
created libinput devices.
**Recovery options, cheapest first:**
1. Ask the user to physically click once inside the Waydroid window — this
re-establishes input focus/routing. Then retry a single click.
2. Restart the Waydroid session UI: `waydroid session stop` then
`waydroid session start` (host UI only; the container survives). App
state resets, so capture anything you still need before doing this.
3. Keep ONE long-lived injector process alive (large `INJECT_HOLD`) instead
of create→click→destroy per command, so the device never disappears
between interactions. Prefer this if the failure recurs often.
**Budget rule for agents:** if 3 consecutive injected inputs produce no
visible change in a screenshot, **stop injecting** — verify state another
way (earlier screenshots, or one manual user click) instead of looping.
## 6. Verifying app state without UI (when root IS available)
```bash
sudo waydroid shell
ls /data/data/org.incido.app/databases/
sqlite3 /data/data/org.incido.app/databases/<db>.db \
"SELECT COUNT(*), SUM(is_draft), SUM(deleted) FROM notes;"
```
(`waydroid shell` refuses without root: `ERROR: Action "shell" needs root
access`. On this machine sudo is password-gated, so this path only works
interactively with the user.)
## 7. Quick recipe (end-to-end)
```bash
cd ~/Desktop/Apps/Incido
./gradlew :app:assembleDebug -q && \
waydroid app install app/build/outputs/apk/debug/app-debug.apk && \
waydroid app launch org.incido.app
mkdir -p /tmp/incido_test && cd /tmp/incido_test
gcc ~/Desktop/Apps/Incido/tools/inject2.c -o inject2
./inject2 click 30 35 # open drawer
sleep 1; spectacle -b -n -o step1.png
```
…then screenshot after every action, and remember the ×0.75 scale when
converting screenshot targets into inject2 targets.