* [ios][precompile] Add use_frameworks resolve method
To be able to handle cocoapods USE_FRAMEWORKS with both dynamic/static linkage and precompiled we needed a common way to resolve this.
The issue was that when using precompiled and USE_FRAMEWORKS our precompiled framework caused the resulting Pods project to only include header files - hence there where no need to change the header_mappings_dir which a lot of the podspecs did.
A method was added that handles this in a common way.
* [ios][precompile] added resolve_use_frameworks to podspecs
Replaced logic for resolving header mappings and module name using the new method `resolve_use_frameworks` in all podspecs.
Also added `React-oscompat` dependency on `React-jsiinspector_modern` which failed when linkage was "dynamic".
* [ios][precompile] added explicit handling of ReactCoden
When using precompiled and building with frameworks (USE_FRAMEWORKS) we need to explicitly add the correct path to ReactCodegen when calling `create_header_search_path_for_frameworks` to ensure libraries can access their codegen files.
This commit adds an explicit check to make sure we add the correct path when using frameworks and the pod is ReactCodegen.
Added includes in the NativeCXXModuleExample.cpp file to test this.
* Update packages/react-native/scripts/cocoapods/utils.rb
Co-authored-by: Riccardo Cipolleschi <cipolleschi@meta.com>
* codereview: removed test include files
* codereview: fixed issue in ruby.rb
After a github `suggestion` we had a superfluous `end`. Sorry for that.
---------
Co-authored-by: Christian Falch <christian.falch@gmail.com>
Co-authored-by: Christian Falch <875252+chrfalch@users.noreply.github.com>
Summary:
When using prebuilts the USE_FRAMEWORKS setting is not really relevant for the React Native code, since there will not be any source code to build frameworks for - and because we already have a framework for the code in React.XCFramework.
This commit adds a new command to the React Native podspecs like we did with the ReactNativeDependencies framework. The method is called `add_rncore_dependency` and it does nothing when building from source - but when linking with the React.XCFramework it explicitly adds linking with the framework.
In addition there are a few places in the ruby code where we check for the USE_FRAMEWORK value and changes some settings - where needed this commit will add a separate check to ensure we're building from source when making these changes.
bypass-github-export-checks
## Changelog:
[IOS] [ADDED] - Added support for using USE_FRAMEWORKS with prebuilt React Native Core
Pull Request resolved: https://github.com/facebook/react-native/pull/52489
Test Plan:
RNTester:
- Builds without USE_FRAMEWORKS as before with/without prebuilt
- Builds with USE_FRAMEWORKS=dynamic as before with source - and now also with prebuilt code.
- Same goes for the static variant.
Reviewed By: mdvacca
Differential Revision: D78012332
Pulled By: cipolleschi
fbshipit-source-id: ea942738ae52b9dceae48fb78a5026f04b7545b8
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/52252
Instead of declaring two different sets of Pods for prebuilt and build from source, this commit now keeps the pod structure the same for both modes so that consuming libraries can expect to have the same pods and header files available - without this, libraries would have to be updated to take advantage of the prebuilds.
This PR does:
- Added React-Core-prebuilt as a pod in React-Core if prebuilt is enabled
- Simplified react_native_pods to keep pods structure and add React-Core-prebuilt pod if prebuilts are enabled
- Added function for selecting source sets based on prebuilt/build from source
To be able to function both in prebuilt and in regular build from source mode, all podspecs are now using the switch function podspec_sources so that they only include header files if we are in prebuild mode.
Also added React-Core-prebuilt as dependency on React-Core if we are in prebuilt mode so that we install the React.XCFramework.
## Changelog:
[IOS] [FIXED] - Added backwards compatible use of prebuild through cocoapods
Pull Request resolved: https://github.com/facebook/react-native/pull/52223
Test Plan:
Tested in RN-Tester both with and without prebuild.
Rollback Plan:
Reviewed By: cortinico
Differential Revision: D77296047
Pulled By: cipolleschi
fbshipit-source-id: f3eb4d56b2a78bfc8e10ad852746be1ceaf828b2
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/52194
Changelog: [Internal]
Introducing a way to include sources and deps via function that react_native_android_dep.
This will help with Fantom OSS build.
Reviewed By: cortinico
Differential Revision: D77146189
fbshipit-source-id: 06b09d433741506bc1f58fbf1f9e6add9a9cff91
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/50317
`rncore`, `FBReactNativeSpec` and `FBReactNativeComponentSpec` contain the same symbols, which leads to conflicts when we try to merge them into a single shared library. Cleanup the duplication and standardize on `FBReactNativeSpec` everywhere. I've left the Android OSS targets names as is, to avoid breaking deps.
This aligns react-native's package.json with the codegen tooling supported across iOS and Android, which is a single target for all all type-derived codegen.
Changelog: [Internal]
Reviewed By: sammy-SC
Differential Revision: D55037569
fbshipit-source-id: dbf3c0a427c9d0df96e439b04e5b123cd1069c51
Summary:
After testing the latest RC and nighly builds, crash appeared when emitting events from turbo modules on 32bit Android devices. The crash is always reproducible only on 32bit devices on signed production builds. Fore more details and the crash log, check the [related issue](https://github.com/facebook/react-native/issues/51628#issue-3094045077).
From what I found, the variadic functions like CallVoidMethod are unsafe on 32bit due to not type checking the passed arguments at compile time. As far as I understand the 64bit cpus and ABIs are more forgiving with alignment and calling conventions. On 32bit the ABIs are strict as arguments are passed on the stack and if there is type/size/alignment issue it reads the wrong memory, which causes the SIGEGV crashes.
## Changelog:
[ANDROID] [FIXED] - emitting event from turbo module crashes on 32bit android
Pull Request resolved: https://github.com/facebook/react-native/pull/51695
Test Plan:
1. Pull the [reproduction demo](https://github.com/vladimirivanoviliev/rn079eventcrash), install the dependencies (v `0.80` is on PR)
2. Run codegen on android
3. Build signed apk. To create it you will need to create new demo key-store.
4. To install the build apk in 32bit mode you can use `adb -s YOURDEVICE install --abi armeabi-v7a android/app/release/app-release.apk`
5. Run the app, create key, save it. Than update the key and save it again. The app crashes when try to emit event from the turbo module.
6. Patch the related `JavaTurboModule.cpp` file with the changes from this PR and enable build from source.
7. Rebuild and reinstall the apk and test again - the issue is now fixed
## Additional notes:
I have tested the app on android using the `rn-tester` demo app, everything works as expected. I also patched our production app and tested more complex scenarios and they works as expected. I have run the tests and linter and they passed.
One thing that I didn't able to setup and run is the iOS `rn-tester` app, due to Hermes engine error `Command PhaseScriptExecution failed with a nonzero exit code`. I haven't found any information how to fix it. I have followed [this guide](https://github.com/facebook/react-native/blob/main/packages/rn-tester/README.md) and installed node modules using yarn and started the `yarn prepare-ios`. I also haven't found any information with what node version and ruby version the react native package is build on CI so I use the same versions locally. If you provide me with updated instructions for those I can contribute by updating the related guides and including `.npmrc`, `.ruby-version` files.
Reviewed By: cortinico
Differential Revision: D75782377
Pulled By: javache
fbshipit-source-id: b94998be6dd51e90ad4137b1d2e38a6850bc3cb2
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/51622
We currently support an optional 3rd params for `target_compile_reactnative_options` which allows to specify a LOG_TAG macro.
No one is actually reading that Macro. The only usage would be logging from the Android SDK which we don't explicitely use.
Here I'm updating our build to specify a LOG_TAG as `ReactNative` for all the targets without allowing to customize it as it just complicates our build setup.
Changelog:
[Internal] [Changed] -
Reviewed By: mdvacca
Differential Revision: D75445577
fbshipit-source-id: a426ce77ba6d1dfd0800e874d9f7838bfdc5b877
Summary:
RuntimeExecutor.h has sync ui thread utils:
* executeSynchronouslyOnSameThread_CAN_DEADLOCK
The ios platform has js -> ui sync calls. This util, when it executes concurrently with those sync calls, deadlocks react native.
On ios, we're going to resolve these deadlocks, which'll involve customizing this util: D74769326. Therefore, this diff forks an implementation of these sync ui thread utils for ios.
Changelog: [Internal]
Reviewed By: mdvacca
Differential Revision: D74901907
fbshipit-source-id: f502df4216e9ba57f458435c696a2f086becf24f
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/51620
The spirit of "Decouple Hermes from JSI nd simplify dependencies" is good, but the problem is that when we ships prebuilds of hermes-engine, it tries to load `libjsi.dylib` and that library is not available.
{F1978594163}
I'll have to investigate more how to properly decouple Hermes prebuilds and how to pass jsi to Hermes and react-native. We might have to decouple it further and have a separate job that builds jsi and feed it to hermes.
This problem will go away with SwiftPM, though, so probably not worth solving this.
## Changelog:
[Internal] - Fix nightlies with a backout
Reviewed By: cortinico
Differential Revision: D75442056
fbshipit-source-id: f1085be8247c0974854254c6bd631ca258488beb
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/51591
This change decouples Hermes from JSI and simplify the dependency graph.
Prior to this change, Hermes was building and providing JSI to React Native.
Hermes and React Native shares the same version of JSI, so that's was not a problem, but this choice added an unnecessary dependency on hermes-engine for some pods. For example, `React-utils` was depending on hermes, although it only needs JSI.
## Changelog:
[Internal] - Decouple hermes-engine from jsi
Reviewed By: cortinico
Differential Revision: D75374285
fbshipit-source-id: 3af6846032e81b6461420dd4f4a9c34b741f31ae
Summary:
Fixes https://github.com/facebook/react-native/issues/51083. Turbo stripped out the dictionary when the value is null. The old architecture transforms null to NSNull. The null seems useful in cases like #51803 for removing the storage of the key. cipolleschi can you please help to review?
## Changelog:
[IOS] [FIXED] - Turbo module: Fixes dictionary stripped out when value is null
Pull Request resolved: https://github.com/facebook/react-native/pull/51103
Test Plan: Repro please see https://github.com/facebook/react-native/issues/51083.
Reviewed By: rshest
Differential Revision: D74208525
Pulled By: javache
fbshipit-source-id: 53d630c265fba15d25309a3e1deb19dca24f298c
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/51366
We inherited this from the legacy native module infra, where we didn't have access to the module instance. Instead we can use the simpler `methodSignatureForSelector` which works correctly with OCMock (needed in D74815079).
Changelog: [Internal]
Reviewed By: sammy-SC
Differential Revision: D74817191
fbshipit-source-id: 6f3f741e9a78dea967a7654e6410ddacfad6d8a3
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/50592
For unclear reasons this is crashing on some devices when dereferencing the module ref. Instead we can just access the existing global_ref to the module instance and avoid any deallocation timing issues.
Also added some additional safeguards and validation against incorrect JNI invocations.
Changelog: [Android][Fixed] Fix crash when TurboModule event emitters are used on arm32
Reviewed By: cortinico
Differential Revision: D72716972
fbshipit-source-id: 3803329048d8388b3d30c9cb55a857e399569eea
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/50479
In D72273163 the only reference to RN feature flag `throwExceptionInsteadOfDeadlockOnTurboModuleSetupDuringSyncRenderIOS` was removed, so this diff removes the feature flag. Also it removes remnants of D69922771 which are no longer needed.
Changelog: [Internal]
Reviewed By: shwanton
Differential Revision: D72427164
fbshipit-source-id: 398583ac14910a396839140c06011efa084fe04d
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/50249
This change introduces a flag to turn off the legacy architecture warning if they become too annoying.
The flag can be set in the Info.plist of the React Native architecture and it is controlled by the key: `RCTLegacyWarningsEnabled`.
* If the key is missing or with a value of `YES`, logs are enabled
* If the key has a value of `NO`, react native will not output any log.
We decided to use the Info.plist file to configure the logs because in that way it will work also with React Native prebuilds.
## Changelog:
[iOS][Added] - Add flag to enable or disable legacy warning.
Reviewed By: cortinico
Differential Revision: D71814001
fbshipit-source-id: b6ae6b032ff7add6bae3d73dba490adeaceffa1f
Summary:
Make RCTScreenSize and RCTScreenScale initialize during React Native init.
After this diff, there shouldn't be any sync dispatches to the main queue. So, we can just introduce an error into RCTUnsafeExecuteOnMainQueueSync.
## Note
If people manually dispatch to the main queue like so, react native can still deadlock:
```
dispatch_sync(dispatch_get_main_queue(), ^{});
```
Changelog: [Internal]
Reviewed By: lyahdav
Differential Revision: D72273163
fbshipit-source-id: 6211851b380b4f5cd556f48f4d717a6f53d65d32
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/50193
This fix makes sure that we convert to JSException only NSException thrwn by sync methods.
Currently, nothing in the stack will be capable of understanding that js error if it is triggered by an exception raised by an asyc method.
See https://github.com/reactwg/react-native-new-architecture/discussions/276 for further details
We need to cherry pick this in 0.78 and 0.79
## Changelog:
[iOS][Fixed] - Make sure the TM infra does not crash on NSException when triggered by async method
Reviewed By: fabriziocucci
Differential Revision: D71619229
fbshipit-source-id: b87aef5dd2720a2641c8da0904da651866370dc6
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/50175
This change logs warning in the RN Dev Tools and in the Xcode console when a legacy module is used through the interop layer.
The `moduleName.methodName` warning is logged only once per usage not to flood the users with Warnings.
## Changelog:
[iOS][Added] - Add warnings when a legacy module is used in the Interop Layer.
Reviewed By: cortinico
Differential Revision: D71561348
fbshipit-source-id: f3ec830ddb07c4d0ab34534ad2baf95e75b1a3b3
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/50170
This is needed in D71470038 and later, where submodules of `jsinspector-modern` need to operate with CDP message payloads. We functionally split out these files as a library to avaoid a dependency cycle.
Changelog: [Internal]
Reviewed By: hoxyq
Differential Revision: D71551561
fbshipit-source-id: 527479399d7563883c1b6599f884b7857e79bd77
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/50106
## Changelog:
[General] [Added] - Create TurboModuleWithJSIBindings interface
So c++ TurboModules can initialize some private members with reference to `jsi::Runtime`
Reviewed By: lenaic
Differential Revision: D71396842
fbshipit-source-id: 59d32e4cbf2c5081912a4c828acc66ceb8702855
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49968
Updates all the core podspec to use the new `ReactNativeDependencies` podspec
## Changelog:
[INTERNAL] - Updates Modules in core to use the new dependencies.
Test Plan:
✅ Run Rn-Tester and verify that it works as expected both building deps from source and using prebuilt tarballs
✅ Add third-party library (react-native-reanimated) and perform the same tests to verify that it works with the changed podspec and utilities
Reviewed By: cortinico
Differential Revision: D71032639
Pulled By: cipolleschi
fbshipit-source-id: e53179a8c1b47ee2b945f5cb5dfaf3f6f2755af2
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49873
In the old architecture, when we were passing a `null` value as a parameter in a function that accepted nullable parameter, the null value was mapped to `nil` on iOS.
After my changes in [d4236791e2](https://github.com/facebook/react-native/commit/d4236791e238a614d2fadf5c5659874d983ab029), in the New Architecture, through the interop layer, legacy modules were receiving an `NSNull` object instead of nil.
This was breaking those modules which started crashing or observing undesired behavior.
This change fixes the issue by making sure that, in those cases, a `nil` value is passed.
Note that nested objects in the old architecture were correctly receiving NSNull, so nested objects were behaving correctly already.
## Changelog:
[iOS][Fixed] - Properly pass `nil` for nullable parameters instead of `NSNull` for legacy modules
Reviewed By: javache
Differential Revision: D70723460
fbshipit-source-id: 384f48b6dbb3f54c369b31b6d2ee06069fa3591c
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49835
The implementation of moduleProviderForName is slightly off.
This method was supposed to replace the previous ternary expression and to enhance with the module provider call.
The ternary expression used to be
```
!RCTTurboModuleInteropEnabled() || [self _isTurboModule:moduleName] ? [self _provideObjCModule:moduleName] : nil
```
However, as you can see from the current implementation, instead of calling `RCTTurboModuleInteropEnabled()` we are calling `RCTTurboModuleEnabled()` which is clearly a mistake.
On top of that, I'm also updating the guard around the `getModuleProvider` selector as it was bypassing the other checks, and that's wrong.
## Changelog:
[Internal] - Fix moduleProviderForName method
Reviewed By: RSNara
Differential Revision: D70569552
fbshipit-source-id: ed4055da9ea385ed10323ed8d7a8772010b3a105
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49786
Most of these call-sites are only used for exceptional scenarios, so we can just rely on std::to_string and do string concatenation. For a few others that may be more perf-sensitive, I switched over to `snprintf`.
Changelog: [Internal]
Reviewed By: sammy-SC
Differential Revision: D70402439
fbshipit-source-id: 3b90ebb13a7bf1c6cf30722ef636e5e8498a5b26
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49803
This refactors the whole ReactCommon to use the `target_compile_reactnative_options` macro we just introduced.
Changelog:
[Internal] [Changed] -
Reviewed By: javache
Differential Revision: D70386740
fbshipit-source-id: 1f15ea7e182d79bbb0bedce653284808e3309e4f
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49738
**Context**
- D70012142 added TM module provider support
- This was causing RN MacOS to silently fail to load any platform modules since it didn't implement the delegate method
Changelog:
[iOS][Fixed] - Add guard for custom module provider lookup in TMManager
Reviewed By: sbuggay
Differential Revision: D70357542
fbshipit-source-id: 5b338616655ecb84cdb3c60e243fdb99444af657
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49611
This Change connects the pod created previously to the `RCTTuboModuleManager` so that it can ask the delegate about the CxxTurboModuleProviders.
We had to introduce a `RCTFullTurboModuleManagerDelegate` that implements both the old `TurboModuleManagerDelegate` and the new Swift compatible one.
## Problem
As of today, it is not possible to create a pure C++ TM and to register it through a Swift AppDelegate
## Solution
We can create a pod that can be imported in a Swift AppDelegate and that offer some pure Objective-C classes.
These classes contains a provider that can be instantiated in Swift.
The TurboModule manager delegate will ask the AppDelegate about the presence of some provider that can instantiate a pure C++ turbomodule with a given name.
The provider has an empty interface, but the implementation contains a function that can actually instantiate the TM. The function is implemented in an Objective-C++ class that imports the pure C++ turbomodule and creates it.
The TMManager extends the provider through a category to attaach the signature of the function that is implemented by the provider.
The last diff in this stack contains an exaple on how to implement this.
## Changelog:
[iOS][Added] - Added the React-SwiftCompatibleNativeModules pod
Reviewed By: javache
Differential Revision: D70012142
fbshipit-source-id: db96c4cd3cdd1062b12f11131b7c6c51ecd74bc7
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49632
changelog: [internal]
Add gating to experiment with removal of `_turboModuleManagerDelegateMutex`.
# Why remove _turboModuleManagerDelegateMutex?
- It is not consistently used in the class. Here is a case where access to _delegate is not protected by the mutex: https://fburl.com/code/9tay04fn
- From inspecting individual implementations of RCTTurboModuleManagerDelegates, it is not needed to synchronise access.
- It may lead to deadlocks under following conditions:
- **A)** JavaScript thread gets hold of the `_turboModuleManagerDelegateMutex` and then tries to dispatch a block on the main thread synchronously (calling `RCTUnsafeExecuteOnMainQueueSync`). For example, when a new native module is created lazily because JavaScript used it and the native module requires main thread setup.-
- **B)** The main thread tries to get hold of `_turboModuleManagerDelegateMutex` when it is creating a native module ([[RCTModuleRegistry moduleForName:lazilyLoadIfNecessary:]](https://fburl.com/code/lioi5zei)) from Obj-C.
Reviewed By: javache, cipolleschi
Differential Revision: D70087779
fbshipit-source-id: 1780b2f48537dd578260a96f8f498dbe11529689
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49587
This is a reland diff for D69805065.
The problems with the original diff were:
- wrong header search path in the `React-NativeModuleApple.podspec`
- adding an implementation `RuntimeExecutor.cpp` file to implement a function that is used by an inline function in the `RuntimeExecutor.h`.
I believe that the last point specifically was creating issues to the linker which was not able to resolve the symbols properly in transitive dependencies.
The fix was to move the implementation of `getMainThreadMutex` to be `inline`.
## Changelog:
[Internal]
Reviewed By: cipolleschi
Differential Revision: D69922771
fbshipit-source-id: 285ce1f232e8c94b99d80aafe1ffbc473921d52f
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49554
For some reason, this diff completely breaks the linker on iOS.
I tried to look for a fix forward, but unsuccessfully.
I'm reverting this diff to get CI green again, but this requires more investigation.
## Changelog:
[Internal] - Revert D69805065
Reviewed By: sammy-SC, huntie
Differential Revision: D69920338
fbshipit-source-id: 8e1d34b5314d8ead51c127208ae2d2250f7d3724
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49509
Currently if we hit a deadlock in sync rendering due to a TurboModule initialization that requires main queue setup we don't get any information about which TurboModule caused the issue.
To help us know which TurboModules we need to fix, this instead will crash with the name of the TurboModule.
Changelog: [Internal]
Reviewed By: sammy-SC
Differential Revision: D69805065
fbshipit-source-id: f75df44f9a603a5f53a008382d32b2b5285c1162
Summary:
> [!NOTE]
> This PR is part of JavaScriptCore Extraction to this repository: https://github.com/react-native-community/javascriptcore
This PR centralizes the setup of js engine dependencies which need to be defined when building with dynamic frameworks. This will allow us to change linked framework if using a third party one in the future
## Changelog:
[INTERNAL] [CHANGED] - centralize JS engine dependency configuration
Pull Request resolved: https://github.com/facebook/react-native/pull/49297
Test Plan: CI Green (Build needs to go properly)
Reviewed By: javache
Differential Revision: D69396641
Pulled By: cipolleschi
fbshipit-source-id: deedd12084f563f73f12f8617fdca0a6d680bf5a
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49250
The TurboModule System decided to ignore the Null values when they are coming to JS. However, in iOS, null value can be mapped to `[NSNull null];` and this value is a valid value that can be used on the native side.
In the old architecture, when the user were sending a null value from JS to a native module, the Native side was receiving the value.
In the New Architecture, the value was stripped away.
This change allow us to handle the `null` value properly in the interop layer, to restore the usage of legacy modules in the New Arch.
I also tried with a more radical approach, but several tests were crashing because some modules do not know how to handle `NSNull`.
See discussion happening here: https://github.com/invertase/react-native-firebase/issues/8144#issuecomment-2548067344
## Changelog:
[iOS][Changed] - Properly handle `null` values coming from NativeModules.
Reviewed By: sammy-SC
Differential Revision: D69301396
fbshipit-source-id: be275185e2643092f6c3dc2481fe9381bbcf69e9
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49072
We have instance of apps crashing when enabling the New Architecture because of the TurboModule interop layer.
What's happening is that when the module is loaded, the TM Interop Layer tries to parse the method definition to expose them in JS. However, for some libraries in the Legacy Architecture, it is possible to define a method in Objective-C and to define a different signature in Swift.
For example, the [`RNBluetoothClassic` library](https://github.com/kenjdavidson/react-native-bluetooth-classic) defines a selector in objective-c which [has the signature](https://github.com/kenjdavidson/react-native-bluetooth-classic/blob/main/ios/RNBluetoothClassic.m#L134-L136)
```
RCT_EXTERN_METHOD(available: (NSString *)deviceId
resolver: (RCTPromiseResolveBlock)resolve
rejecter: (RCTPromiseRejectBlock)reject)
```
And the method is inmplemented in Swift with [the signature](https://github.com/kenjdavidson/react-native-bluetooth-classic/blob/main/ios/RNBluetoothClassic.swift#L502-L505):
```
func availableFromDevice(
_ deviceId: String,
resolver resolve: RCTPromiseResolveBlock,
rejecter reject: RCTPromiseRejectBlock
)
```
When the TurboModule interop layer tries to parse the method, it receives the `accept:resolver:rejecter:` signature, but that signature is not actually defined in as a method in the module instance, and it crashes.
This crash was not happening in the Old Architecture, which was handling this case gracefully. Notice that the specific method from the example is not working in the Old Architecture either. However, the app is not crashing in the old architecture.
This change adds the same graceful behaviors plus it adds a warning in development to notify the developer about which methods couldn't be found in the interface.
Fixes:
- https://github.com/facebook/react-native/issues/47587
- https://github.com/facebook/react-native/issues/48065
## Changelog:
[iOS][Fixed] - Avoid crashing the app when the InteropLayer can't find some methods in the native implementation.
Reviewed By: javache
Differential Revision: D68901734
fbshipit-source-id: 844d1bf29423d5c601b583540e86d57dfffd1428
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/48383
## Changelog:
[Internal] -
This changes the name of `SystraceSection` class to `TraceSection`, the purpose being to make it Systrace/FBSystrace agnostic (and that it can be mapped to e.g. Perfetto instead).
It changes all the internal callsites to the RN code code, and also adds a shim include, `<cxxreact/SystraceSection.h`, for backward compatibility with the external callers for now (which will be addressed separately).
Reviewed By: javache
Differential Revision: D67621914
fbshipit-source-id: 337c63c45a7b075c6e00cfca67ecc06c298c94c0
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/47891
Various C++ utilities for handling callbacks and promises rely on the static LongLivedObjectCollection. Even if a host platform injects it's TurboModuleBinding with a custom LongLivedObjectCollection, there's no guarantee that the platform is not also using C++ TurboModules that use the static LongLivedObjectCollection. Clearing both collections solves issues for TurboModuleBindings with custom LongLivedObjectCollections that also have cross platform C++ TurboModules that use the static collection.
## Changelog
[Internal]
Reviewed By: javache
Differential Revision: D66324044
fbshipit-source-id: a5a973087c35fc00240e40ba757d49125473daf4
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/47598
## Changes
Now:
- in bridgeless, NativeModules.foo will also return turbo modules. And, global.__turboModuleProxy no longer exists.
- in bridge, nothing changes.
| **JS API** | **Bridge** | ***[Before]* Bridgeless** | ***[Before]* Bridgeless w/ Interop**| ***[After]* Bridgeless**
| global.__turboModuleProxy | turbo modules | turbo modules | turbo modules |**deleted**
| global.nativeModuleProxy | legacy modules | error | legacy modules | turbo + legacy modules
## Justification
This reduces the cost for adopting the new architecture:
- Prior, you had to migrate the module itself, **and** all its callsites: NativeModules.foo -> NativeFoo
- Now, you have to migrate the module itself **only**.
This simplifies the interop layer logic in bridgeless: all modules come from the same thing.
Changelog: [General][Breaking] Bridgeless: Make NativeModules.foo load turbomodules (unset turboModuleProxy in bridgeless).
Reviewed By: javache
Differential Revision: D65896934
fbshipit-source-id: 10883c292b78759fceac5bd984e0cdf8a679fc67
Summary:
While the new `installJSIBindings(..)`/`BindingsInstaller` functionality allows you to synchronously set up stuff in the JS runtime before JS runs, there is no access to the JS CallInvoker, meaning you cannot really set up anything that uses callbacks or asynchronous code.
[Nitro](https://github.com/mrousavy/nitro) needs this.
## Changelog:
<!-- Help reviewers and the release process by writing your own changelog entry.
Pick one each for the category and type tags:
[ANDROID|GENERAL|IOS|INTERNAL] [BREAKING|ADDED|CHANGED|DEPRECATED|REMOVED|FIXED|SECURITY] - Message
For more details, see:
https://reactnative.dev/contributing/changelogs-in-pull-requests
-->
[IOS] [ADDED] - Add `CallInvoker` to `installJSIBindings(..)`
[ANDROID] [ADDED] - Add `CallInvoker` to `BindingsInstaller`
Pull Request resolved: https://github.com/facebook/react-native/pull/46851
Test Plan: In `SampleTurboModuleJSIBindings`, we can now use the `CallInvoker`.
Reviewed By: rshest
Differential Revision: D63975546
Pulled By: javache
fbshipit-source-id: 43cd469de2c35581f5362d2500fc27e41d6eab72
Christoph Purrer
generatedunixname89002005287564
Andrew Datsenko
Pieter De Baets
Vladimir Iliev
Nicola Corti
Ramanpreet Nara
zhongwuzw
Samuel Susla
Zeya Peng
Liron Yahdav
Alex Hunt
Shawn Dempsey
Oskar Kwaśniewski
Ruslan Shestopalyuk
Eric Rozell
Alan Lee
Phillip Pan
Marc Rousavy