IOS - Crash when reloading app using RNN with new Architecture

[snaerth]

What happened?

When using React Native Navigation (RNN 8.3.0) with RN 0.77.3 arch enabled on IOS, quickly reloading the app (pressing Reload multiple times in rapid succession) can cause a native crash (see log) with a simplest app possible

Screen.Recording.2025-08-18.at.11.37.54.AM.mov

However on an more complex app like this one, this happens every time I reload and also often the app crashes even in production.

The crash occurs in Fabric’s SurfaceHandler during unregisterSurface, typically triggered by Navigation.setRoot being called before the previous root has fully torn down I assume.

Can someone please take a look at this. Thank you for your contribution

What was the expected behaviour?

App should just reload like normally

Was it tested on latest react-native-navigation?

• I have tested this issue on the latest react-native-navigation release and it still reproduces.

Help us reproduce this issue!

Test Repo

In what environment did this happen?

React Native Navigation version: 8.3.0
React Native version: 0.77.3
Has Fabric (React Native's new rendering system) enabled: yes
Node version: v20.19.0
Device model:
iOS version: 18.4

[Hukutus]

We're having the same issue when completing our app's onboarding and when resetting the app, both calling Navigation.setRoot(...) to set a new stack. There's also a previous closed issue #7780, which might be related?

After setting a breakpoint to -[RCTMountingTransactionObserverCoordinator unregisterViewComponentDescriptor:] in Xcode and typing po componentViewDescriptor.view in the debug console the crash seems to point to RCTScrollViewComponentView, which I assume could point to unregistering a ScrollView (?):

<RCTScrollViewComponentView: 0x7fe3b87b9860; frame = (0 0; 393 671.333); clipsToBounds = YES; tag = 1192; backgroundColor = UIExtendedSRGBColorSpace 0.956863 0.956863 0.960784 1; layer = <CALayer: 0x600001681c40>>

[snaerth]

Yes this is very frustrating.
What would really help us is guidance on where in the library we could reliably trigger a teardown of the active Surface before a new instance of the app is bootstrapped. Right now it’s not clear to me where that should live, and having direction from the maintainers would save a lot of trial and error.

[Hukutus]

I managed to fix our app reset crash by removing modalPresentationStyle: OptionsModalPresentationStyle.overCurrentContext from our iOS "Do you want to log out?" modal. Maybe something to do with animations?

We still have the crashing issue when completing our app's onboarding, which I've been investigating but haven't been able to fix. When the app starts without onboarding I can see a bunch of logs like this in the Xcode console:

E0829 09:49:10.923298 108556288 conversions.h:1112] react_native_expect failure: isMap E0829 09:49:10.924577 108556288 conversions.h:1112] react_native_expect failure: isMap
E0829 12:19:58.989418 147337216 LeakChecker.cpp:53] [LeakChecker] Surface with id: 41 has leaked 241 components out of 254

The app still seems to still work normally. After completing onboarding the app crashes due to:

*** Terminating app due to uncaught exception 'NSInternalInconsistencyException', reason: 'RCTComponentViewRegistry: Attempt to query unregistered component.'

Maybe something stays behind when the stack gets reset?

[Hukutus]

Is anyone investigating this?

Since there hasn't been any news we upgraded our Android SDK to 35 separately due to the Google Play Store requirement, but now there's also the Nov 1 requirement for 16 KB memory page size, which basically means we'll need RN 0.77 by then.

[Belobopka]

Same problem. For our project the main reason of this crash is - surfaces unregistering and then registering again via app reload, and here we have race condition - hot reload of surfaces(main tabs) and dealloc of these surfaces.
So the crash occurs when SurfaceHandler::start -> app reload -> SurfaceHandler::setUIManager(nullptr ui manager) -> Scheduler::registerSurface -> SurfaceHandler::setUIManager(WITH ui manager) -> RCTFabricSurface.dealloc (which calls unregister surface)

[Belobopka]

For testing purposes I commented/removed these lines. And it seems the app reload started working fine, but it needs more testing and validating.
React-RuntimeApple -> RCTHost.mm -> _reloadWithShouldRestartSurfaces

if (shouldRestartSurfaces) {
       [_instance callFunctionOnBufferedRuntimeExecutor:[surface](facebook::jsi::Runtime &_) { [surface start]; }];
}

For setRoot crash - I couldn't reproduce it for now.

[Hukutus]

Just to update on the issue the crashes I reported earlier were actually related to Reanimated entering and exiting animations reported in #8096

The fast reload crash still happens, so the issue is different here.

[gosha212]

@Hukutus Thanks for your investigation. This are different issues. There is a problem in iOS timings of multiple surfaces creation. We don't have a proper solution for that now. We are working on it with React Native Team.
You can see that the start surface is happening on jsi thread and stop surfaces happens on the main thread. That what causes rase condition in surface states.

[markdevocht]

@Belobopka I'm trying to reproduce this issue on a newer version of RNN (8.4.3) and can't seem to get it to crash. Can you give it a quick try and see if the latest version fixes the problem?

[Belobopka]

@markdevocht Hello. It seems that this problem has been fixed for us. Thanks.

[Belobopka]

@markdevocht Hello. It seems the crash is still exists. Using the above mentioned app with updated rnn to 8.5.0 it reproducible 100% with spamming "R" button

surface_crash.mov