fix : protect some usages of client to avoid crashes

[bmarty]

Fixes https://github.com/element-hq/element-x-android-rageshakes/issues/3375 and other rageshakes.

[github-actions[bot]]

:iphone: Scan the QR code below to install the build (arm64 only) for this PR. If you can't scan the QR code you can install the build via this link: https://i.diawi.com/TFQFgM

[codecov[bot]]

Codecov Report

Attention: Patch coverage is 56.52174% with 10 lines in your changes missing coverage. Please review.

Project coverage is 82.97%. Comparing base (e76f7fb) to head (0261739). Report is 5 commits behind head on develop.

Files with missing lines	Patch %	Lines
...ndroid/features/roomlist/impl/RoomListPresenter.kt	25.00%	5 Missing and 1 partial :warning:
...ement/android/appnav/loggedin/LoggedInPresenter.kt	62.50%	1 Missing and 2 partials :warning:
...id/libraries/matrix/api/sync/SlidingSyncVersion.kt	66.66%	1 Missing :warning:

Additional details and impacted files

```diff @@ Coverage Diff @@ ## develop #3886 +/- ## =========================================== - Coverage 82.98% 82.97% -0.02% =========================================== Files 1788 1789 +1 Lines 45208 45222 +14 Branches 5334 5337 +3 =========================================== + Hits 37516 37522 +6 - Misses 5818 5825 +7 - Partials 1874 1875 +1 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

---

🚨 Try these New Features:

• Flaky Tests Detection - Detect and resolve failed and flaky tests
• JS Bundle Analysis - Avoid shipping oversized bundles

[ganfra]

Mmmm, this is weird, this means there is an old client still referenced?

[bmarty]

Mmmm, this is weird, this means there is an old client still referenced?

Nothing is done to stop SendQueues. This is maybe another way to fix the issue?

Edit actually the scope should be destroyed, so this should work

[ganfra]

Mmmm, this is weird, this means there is an old client still referenced?

Nothing is done to stop SendQueues. This is maybe another way to fix the issue?

It's bound to the LoggedInFlowNode lifecycle, so this should be stopped at this point :/

[jmartinesp]

According to the logs, the user was signed out but the app wasn't killed, so when the app was opened again it tried restoring the saved navigation state, which I guess either returned a new instance or retained one of our MatrixClient wrappers and then failed when it tried using the underlying Client:

2024-11-16T14:01:24.646079Z ERROR elementx: [Push/Notification/DefaultNotifiableEventResolver] Unable to resolve event: ----.
java.lang.IllegalStateException: NotificationClient object has already been destroyed
    at org.matrix.rustcomponents.sdk.NotificationClient.getNotification(SourceFile:147)
    at io.element.android.libraries.matrix.impl.notification.RustNotificationService$getNotification$2.invokeSuspend(SourceFile:55)
    at kotlin.coroutines.jvm.internal.BaseContinuationImpl.resumeWith(SourceFile:9)
    at kotlinx.coroutines.DispatchedTask.run(SourceFile:107)
    at com.google.android.gms.tasks.zzc.run(SourceFile:49)
    at kotlinx.coroutines.scheduling.TaskImpl.run(SourceFile:3)
    at kotlinx.coroutines.scheduling.CoroutineScheduler$Worker.run(SourceFile:93)
 | 
2024-11-16T14:01:24.647532Z  WARN elementx: [Push/PushHandler] Unable to get a notification data | 
2024-11-16T14:02:31.132845Z  WARN elementx: [MainActivity] onCreate, with savedInstanceState: true | 
2024-11-16T14:02:31.158780Z  WARN elementx: [MainActivity] onNewIntent | 
2024-11-16T14:02:31.159743Z  WARN elementx: [MainActivity] onResume | 
2024-11-16T14:02:31.223778Z DEBUG elementx: Current app migration version: 7. No migration needed. | 
2024-11-16T14:02:31.240437Z  WARN elementx: [MainActivity] onMainNodeInit | 
2024-11-16T14:02:31.241957Z DEBUG elementx: Restore state | 
2024-11-16T14:02:31.242295Z  WARN elementx: Restore with non-empty map | 
2024-11-16T14:02:31.242762Z DEBUG elementx: [Lifecycle] onCreate RootFlowNode | 
2024-11-16T14:02:31.246803Z DEBUG elementx: [Lifecycle] onCreate LoggedInFlowNode | 
2024-11-16T14:02:31.247212Z DEBUG elementx: [Navigation] Navigating to session ---, Current state: Root | 
2024-11-16T14:02:31.247596Z DEBUG elementx: [Navigation] Navigating to space !mainSpace:local. Current state: Session(owner=---, sessionId=---) | 
2024-11-16T14:02:31.248198Z DEBUG elementx: [Lifecycle] onCreate LoggedInAppScopeFlowNode | 
2024-11-16T14:02:31.248975Z DEBUG elementx: Navigating to Room(sessionId=---, roomId=---, threadId=null) | 
2024-11-16T14:02:31.252992Z DEBUG elementx: [Lifecycle] onCreate RoomListNode | 
2024-11-16T14:02:31.253392Z  INFO elementx: setAllSendQueuesEnabled(true) | 
2024-11-16T14:02:31.253413Z DEBUG elementx: [Lifecycle] onCreate LoggedInNode | 
2024-11-16T14:02:31.255261Z DEBUG elementx: [Lifecycle] onCreate RoomFlowNode | 
2024-11-16T14:02:31.256300Z DEBUG elementx: [Lifecycle] onCreate ComposableNode | 
2024-11-16T14:02:31.257910Z DEBUG elementx: Room membership: JOINED | 
2024-11-16T14:02:31.259225Z DEBUG elementx: [Lifecycle] onCreate JoinedRoomFlowNode | 
2024-11-16T14:02:31.260418Z DEBUG elementx: [Lifecycle] onCreate ComposableNode | 
2024-11-16T14:02:31.261355Z DEBUG elementx: Room factory is destroyed, returning null for --- | 
2024-11-16T14:02:31.262613Z DEBUG elementx: Save matrix session keys = [---] | 
2024-11-16T14:02:32.749333Z TRACE elementx: Uncaught exception: java.lang.IllegalStateException: Client object has already been destroyed | 

[ganfra]

So there is probably a scenario where MatrixClientsHolder is not updated correctly

[ganfra]

I've made more investigations on this, and there are some conditions where, while replacing the LoggedInFlowNode, we can get several calls to a destroyed client, the time the node get removed from the composition.

So we have 2 choices :

• either we protect every call to the rust client with try/catch
• either we remove the call to client.destroy and let the GC release the memory when needed

This crash has also emphasised another issue with the refresh token, we need more logs (https://github.com/matrix-org/matrix-rust-sdk/pull/4301)

Edit : the issue with the refresh token might have been fixed https://github.com/matrix-org/matrix-rust-sdk/pull/4304

[sonarcloud[bot]]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud