[Medformatik]

What changed and why

The shell already had processors, durable sessions, effect ports, permissions and recovery, but nothing composed them, so the installed app reported every navigation mode as unavailable and turn-by-turn guidance never started. createForegroundNavigation now builds that graph in the foreground app, and App.tsx sends every page command to its NavigationCoordinator. The one-shot location, settings and sign-in requests still reach the shell services. The handshake reports what the shell can actually do: ground and transit navigation, background location, local notifications once that permission is granted, and speech when the native audio module is present.

• One coordinator. The foreground coordinator is registered, so the background location task feeds the same instance instead of opening a second, headless one beside it.
• Start flow. session.start runs the permission controller with native screens (PermissionPrompts.tsx), because the page may be suspended during OS prompts. On Android 11 and later, the background step uses the system's background-location request, which opens the app's own page with "Allow all the time". The first start also asks once for notification permission, for transit alight alerts.
• Reroute. A reroute effect only starts the request. The answer comes back to the coordinator as a queued applyReroute, which applies it only while the session and its pending request id still match. A newer request aborts the one in flight.

Running it on a phone turned up four more failures:

• Crash on the first location update. expo-task-manager always schedules a persisted JobScheduler job, and Android refuses that without RECEIVE_BOOT_COMPLETED, so the process crashed. The permission is now declared and reviewed in the compliance list. It registers no boot receiver, and the app starts nothing at boot.
• "database is locked" on every location batch. expo-sqlite runs each exclusive transaction on its own connection, so a diagnostic write collided with the batch commit. Writes are now queued on the one handle, with a busy timeout.
• No speech on a cold start. The first cue arrived before Android's TTS engine had initialised and was reported as failed. speak now waits for initialisation, up to 3 s.
• Native screens under the status bar. Their titles sat under the status bar.

Transit refresh and replan effects are still no-ops: a transit session keeps running on its captured itinerary.

Related work

Part of #398. This covers Android only: iOS has not run on a device, and transit refresh and replan are not wired. Third of four stacked PRs. It is based on #757 (shell requests), which is based on #756.

Verification

• Pixel (Android 17) over USB against openmapx.com, with a short driving route:
  • The disclosure, the "Allow all the time" page and the notification prompt appear in order.
  • Native guidance renders from native snapshots. Ending navigation sends session.stopped and takes the location service out of the foreground.
  • A killed process resumes the session on relaunch.
  • Speech reported spoken on three consecutive cold launches. One earlier launch in the same Metro session reported failed, and I could not tell whether it ran the previously installed build, so cold-start speech needs more device runs.
• apps/mobile: jest (new tests for reroute answers, remote work, permission prompts and App composition), tsc, biome, test:native (Swift and Kotlin), assert-permissions, assert-generated, assert-boundary, inventory and pnpm mobile:bundle:check.
• Not checked: a reroute triggered on the road, transit navigation on a device, and iOS.

Release and upgrade impact

Product impact: ordinary