…TAKClientConnection

…ntegration

Agent-Logs-Url: https://github.com/meshtastic/Meshtastic-Android/sessions/56e05084-d1a0-43f1-8515-bf1b8ae7ad7c

Co-authored-by: jamesarich <2199651+jamesarich@users.noreply.github.com>

- T059: MeshService.onStartCommand() → MeshServiceOrchestrator (commonMain)
- plan.md: 38 → 40 XML fixture files (actual count)

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

…l visibility

- Create TAKMeshIntegrationTest.kt with 10 tests covering lifecycle,
  inbound mesh forwarding, firmware gating, and GeoChat enrichment
- Add 2 broadcastRawXml tests to TAKServerManagerTest (forward + no-op)
- Make TAKServerManagerImpl internal (consumers use interface via Koin)
- Fix backing property lint in TAKServerManagerTest
- Update tasks.md with T078-T080, update test counts (89+ tests)

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

… skill

Document that runTest(UnconfinedTestDispatcher()) is required when
testing SharedFlow collectors launched in backgroundScope. The default
StandardTestDispatcher silently drops emissions, causing tests to
hang or assertions to fail without error.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

- Make TAK preview functions public (required for cross-module screenshot tests)
- Generate 12 reference images (6 previews × light/dark):
  TakConfigCard, TakServerSection (disabled/enabled),
  TakTestCard (idle/running/results)
- Fix MeshService.kt duplicate return block from bad merge

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Security:
- Sanitize untrusted mesh uid in AtakFileWriter (path traversal)

Concurrency:
- Replace catch(Throwable) with catch(Exception) in TAKMeshIntegration
  (7 locations) — prevents swallowing CancellationException/OOM/LinkageError
- Use AtomicBoolean.compareAndSet for start()/stop() race guard
- Add wake lock timeout (30 min) to prevent indefinite battery drain

Observability:
- Log XML parse failures in TAKClientConnection (was silently dropped)
- Escalate SDK compression fallback from debug to warn level
- Log unparseable CoT dates before falling back to now()

Lint:
- Fix all spotless (ktlint) and detekt violations across core:takserver
- Fix broken TAKDefaultsTest (removed references to deleted constants)
- Fix test function naming (ktlint camelCase)

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

- Extract TakSdkCompressor expect/actual to move TAK SDK references
  out of commonMain (fixes compileKotlinIosSimulatorArm64)
- Fix license header year format in TAKConfigItemList.kt (2025-2026 → 2026)
- Add file-level detekt suppressions:
  - TAKConfigItemList.kt: LongMethod, ModifierMissing, ParameterNaming
  - TAKConfigPreviews.kt: PreviewPublic (public for screenshot tests)
  - MeshService.kt: TooGenericExceptionCaught (wake lock error handling)
- Replace catch(Throwable) with catch(Exception) in TakMeshTestRunner
- Run spotlessApply across affected modules

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

- Upgrade from v0.1.3 (fat-jar with bundled protos) to v0.2.1 which
  strips proto classes from the JVM JAR — fixes R8 'Type defined
  multiple times' errors in minifyFdroidReleaseWithR8.
- Exclude xpp3 from jvmAndroidMain (Android provides XmlPullParser as
  a platform class; R8 fails when both library and program classpaths
  define the same type).
- Re-add xpp3 in jvmMain for Desktop JVM which still needs it.
- Replace java.net.BindException with IllegalStateException in
  TAKServerManagerTest to fix iOS test compilation.

Closes meshtastic/TAKPacket-SDK#5

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

TakPermissionHandler invoked the permission launcher directly inside the
composition body when isTakServerEnabled was true, which crashed with:

    java.lang.IllegalStateException: Launcher has not been initialized
        at androidx.activity.compose.ActivityResultLauncherHolder.launch(...)

The launcher's underlying ActivityResultLauncherHolder isn't bound to the
activity until composition completes, so calling launch() mid-composition
hits the not-yet-initialized holder. The iOS and JVM actuals already use
LaunchedEffect — Android was the outlier.

Also gate the request on isLocalNetworkPermissionGranted() so we don't
re-prompt every time the user opens the TAK Server screen with the
server already enabled (it's still requested up-front from the
Connections screen, so the usual path doesn't trigger here at all).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

The permission ACCESS_LOCAL_NETWORK is an Android 17 / API 37 runtime
permission. Both rememberRequestLocalNetworkPermission and
isLocalNetworkPermissionGranted were guarding on Build.VERSION_CODES.S
(API 31, Android 12), so on API 31-36 — including the user's Android 16
test device — the code tried to request a permission the system doesn't
recognize. The launcher's RequestPermission contract returns immediate
denial for unknown permission strings, which triggered the
TakServerScreen's onPermissionResult(false) path and auto-disabled the
TAK Server toggle ~90 ms after enabling it.

Bumped the cutoff to API 37 using an integer literal because the SDK we
compile against (android-37.0) does not yet expose a named constant
past VANILLA_ICE_CREAM (API 35) — no BAKLAVA, no API-37 codename.

Below API 37 we now report the permission as implicitly granted and
skip the request, matching the actual platform behavior where localhost
binding and LAN access work under the INTERNET permission alone.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

The existing v1/v2 send + receive split in TAKMeshIntegration was already
gated on Capabilities.supportsTakV2 (firmware >= 2.8.0) — sendCoTToMesh
dispatches to sendCoTToMeshV2 / sendCoTToMeshV1, and handleMeshPacket
switches on PortNum.ATAK_PLUGIN vs ATAK_PLUGIN_V2. But handleV1Packet
was commented "Backward compat for legacy V1 devices. Will be removed."
which signaled future deletion at odds with our intent to keep 2.7.x
firmware on the v1 stack indefinitely.

Replaced the comment with a doc block describing the v1 receive path's
purpose and stable position. No behavioral change.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>