Summary:
Android's onAnimationEnd callback is lossy and ocasionally just does not fire. However the LayoutAnimationController maintains a sparse array of animations (with Strong View refs) that is only cleaned when the onAnimationEnd callback is invoked. This results in a leak of Android View objects over time.
To avoid this, the Strong View refs are migrated to WeakReference's and the associated sparse array is cleaned of any invalid layout animations in response to the reset() call.
This closes two leaks:
1. Unbound growth in LayoutAnimationController::mLayoutHandlers
2. Pinning View objects into memory as the sole remaining GC root
## Changelog:
1. Made OpacityAnimation and PositionAndSizeAnimation classes hold weak refs to views only
2. Added a method to LayoutHandlingAnimation to surface if their view ref is gone
3. Added cleanup for Animation with bad view refs
Pick one each for the category and type tags:
[ANDROID] [Fixed]- Fixes memory leak
Pull Request resolved: https://github.com/facebook/react-native/pull/49959
Test Plan: * Primarily code inspection and regression given the intermittent nature of Android's failure to execute the callback.
Reviewed By: alanleedev
Differential Revision: D71037262
Pulled By: javache
fbshipit-source-id: 1fa4eaa2ca839f347a55cb37e2648db972748586
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/50001
Updates `react-native/.flowconfig` with the following option:
```
react.runtime=automatic
```
This adjusts Flow to model the current behavior of React and JSX, relaxing the requirement that JSX elements have `React` within scope.
Changelog:
[General][Changed] - Changed Flow for the React Native monorepo, so that `React` no longer has to be in scope when using JSX.
Reviewed By: javache
Differential Revision: D71096283
fbshipit-source-id: 65762ac39206c682b7be5d3f512b2b83f7eb1b49
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/50000
This adds some tests that verify that controlled textinput, via text prop, or children creating attributedstring, does not jumble text, so long as JS side does not mutate underlying text content.
The rewriting/max length example does still run into problems here.
Changelog: [Internal]
Reviewed By: joevilches
Differential Revision: D71076703
fbshipit-source-id: 85995e8cc087533b1115f158ba5dad67ed008289
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49999
Changelog: [Internal]
Early initialize call to RCTScreenSize which is used by Modal and will read screen size off the UI Thread to avoid main thread deadlocks
Reviewed By: sammy-SC
Differential Revision: D70999156
fbshipit-source-id: 47fb8666c3330ee67cbcb430060f0cec37ec3c7c
Summary:
We were missing some null checks that could cause a crash.
There seems to be some cases where either the parent or the currently focused View are present on Android's hierarchy but not present on the Shadow Tree, in this cases we can just return a nullptr to fall back to default focusing behavior on Android.
Changelog: [Internal]
Reviewed By: joevilches
Differential Revision: D71050870
fbshipit-source-id: a4b4e533c4e9a59c53190feb4b23db4b436b8ae9
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49987
I went through the whole changelog for 0.79 and applied categories to all the entries + did some significant cleanup (remove some entries that were actually internals, grouped and sorted various items)..
Changelog:
[Internal] [Changed] -
Reviewed By: cipolleschi
Differential Revision: D71003373
fbshipit-source-id: 5a95e619f1cf2ffc84263b92104efee16fd74e53
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49941
This diff introduces annotations into the ios codegen config!
## Before
In the new architecture, people can declare module/component codegen config in [package.json](https://reactnative.dev/docs/the-new-architecture/using-codegen#configuring-codegen)
This config contains the following maps:
- modulesConformingToProtocol
- modulesProvider
- componentProvider
```
"codegenConfig": {
"name": "HelloWorldSampleModule",
"type": "all",
"jsSrcsDir": "specs",
"android": {
"javaPackageName": "com.helloworld"
},
"ios": {
"modulesConformingToProtocol": {
"RCTImageURLLoader": [
"RCTHelloWorldImageURLLoader"
],
"RCTURLRequestHandler": [
"RCTHelloWorldURLRequestHandler"
],
"RCTImageDataDecoder": [
"RCTHelloWorldImageDataDecoder"
]
},
"modulesProvider": {
"HelloWorldImageURLLoader": "RCTHelloWorldImageURLLoader",
"HelloWorldURLRequestHandler": "RCTHelloWorldURLRequestHandler",
"HelloWorldImageDataDecoder": "RCTHelloWorldImageDataDecoder"
},
"componentProvider": {
"FooComponent": "RCTFooComponentClass"
}
}
```
## After
This information is a little bit easier to understand if we group it by module/component (into **annotations**):
(that's what this diff does!)
```
"codegenConfig": {
"name": "HelloWorldSampleModule",
"type": "all",
"jsSrcsDir": "specs",
"android": {
"javaPackageName": "com.helloworld"
},
"ios": {
"modules": {
"HelloWorldImageURLLoader": {
"conformsToProtocols": ["RCTImageURLLoader"],
"className": "RCTHelloWorldImageURLLoader"
},
"HelloWorldURLRequestHandler": {
"conformsToProtocols": ["RCTURLRequestHandler"],
"className": "RCTHelloWorldURLRequestHandler"
},
"HelloWorldImageDataDecoder": {
"conformsToProtocols": ["RCTImageDataDecoder"],
"className": "RCTHelloWorldImageDataDecoder"
}
},
"components": {
"FooComponent": {
"className": "RCTFooComponent"
}
},
}
```
## Migration
The old way is still supported (for now). We will deprecate it soon, and eventually remove it from react native!
Changelog: [iOS][Added] - Codegen: Introduce module/component annotations inside package.json
Reviewed By: mdvacca
Differential Revision: D70822061
fbshipit-source-id: 87ec5664a551c54b9300ba92987f237a601ccfbf
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49940
This will just make sure that we don't unintentionally break this script.
Changelog: [Internal]
Reviewed By: alanleedev
Differential Revision: D70919549
fbshipit-source-id: 6d9f2acca46486f1c920afed69609c70f53e6b18
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49992
changelog: [internal]
View culling does not play along with view flattening/unflattening. This diff adds a test case for it + fix. Previously, view culling would crash.
There are still untested branches of view flattening/unflattening where I am trying to come up with test cases.
Reviewed By: rubennorte
Differential Revision: D71047032
fbshipit-source-id: 9f52473743755e49e4bfda0783803cb79591ca27
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49908
There are various batching mechanisms across react-native platforms. The NativeAnimated startOperationBatch and finishOperationBatch methods are a useful way to signal that complete animation graphs are established for atomic flushes to the UI thread.
Setting up this feature flag for use across non-Android React Native platforms.
## Changelog
[Internal]
Reviewed By: javache
Differential Revision: D70827938
fbshipit-source-id: 4c4a74e9a11f75090152637d6195092f5af866a0
Summary:
This PR (https://github.com/facebook/react-native/pull/48182) introduced skipping hidden folders during Codegen generation.
However, when using pnpm, all files are stored in the `.pnpm` folder (see explanation here: https://pnpm.io/symlinked-node-modules-structure).
As a result, some libraries that support the new architecture but lack the `ios.codegenConfig.componentProvider` field - like [FlashList](https://github.com/Shopify/flash-list/blob/main/package.json) - will be skipped during Codegen generation.
This PR explicitly includes `.pnpm` to prevent this issue.
## Changelog:
[iOS][Fixed] - Check .pnpm folder when looking for third-party components.
Pull Request resolved: https://github.com/facebook/react-native/pull/49983
Test Plan:
Tested on:
RN 0.78.0
PNPM: 10
Flashlist: 1.7.3
Reviewed By: cipolleschi
Differential Revision: D71047936
Pulled By: cortinico
fbshipit-source-id: fa9caab23dea8c92ef5f23c997812d348eb19e08
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49981
Changelog: [internal]
(because MutationObserver isn't a public API yet)
## Context
`MutationObserver` is a [JavaScript API](https://developer.mozilla.org/en-US/docs/Web/API/MutationObserver) used to report mutations in the DOM tree. The mutations available on Web are: changing children, changing attributes and changing text. In React Native, only changing children is supported.
Mutation detection needs to happen synchronously when mutations happen (in ShadowTree commits) and the notification is dispatched as a microtask in JS, which means we can only report mutations happening in JavaScript.
There's an assumption that only React (in JS) can change the structure of the tree in Fabric, so we implemented `MutationObserver` considering this assummption.
Unfortunately, while the assumption is correct (we can only mutate children from React) the implementation didn't take into account that commit hooks were triggered from multiple threads, even if the structure doesn't change (e.g.: with Fabric state updates). In this case, we do the checks but we never dispatch the notifications. This can cause crashes (see T217617393) if we try to check for mutations from the main thread while we add new observers in the JS thread (because that logic wasn't thread safe).
This fixes that crash by, in MutationObserver, not only preventing notifications from commits not coming from React, but also preventing the determination altogether.
In order to do this, this modifies the signature of the commit hooks to also pass the commit options, and adds a new field in the commit options with the source of the commit (for now, just "React" or "Unknown").
In `MutationObserver`, before accessing the data structures of the observer, we check if the commit is coming from React, and return early otherwise.
Reviewed By: javache
Differential Revision: D71036705
fbshipit-source-id: 985c8f903375cbf876dce5174e04563f74d7621a
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49973
We still depend on the Community CLI directly, supporting local development in `packages/rn-tester`. Bump to latest — contains a number of build improvements, and will align us closer to prod/next.
Changelog: [Internal]
Reviewed By: cortinico, cipolleschi
Differential Revision: D71033085
fbshipit-source-id: ddbf7df5a35951b77c36d111cfd12ad64c3b73f7
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49956
This is no longer being used internally so I think we can just remove this for less complexity
Changelog: [Internal]
Reviewed By: philIip
Differential Revision: D70983663
fbshipit-source-id: 61ca5212cb45dd1c8b96aee62964ad62b410d6f1
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49854
Changelog: [GENERAL][ADDED] - Codegen utility functions and types are now exported from the root package
Up to this point, third-party libraries needed to use deep imports to access codegen utilities and types This diff exports them from the main entry point, as we want to move away from supporting deep imports in OSS as a part of JS Stable API project.
Reviewed By: huntie
Differential Revision: D70628408
fbshipit-source-id: 5a6e1cb870465ede0fbc9eb8912d255aa90aa2a8
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49974
Changelog: [internal]
The feature flag system currently logs an error when trying to access common feature flags (flags accessible from everywhere but defined in native) from JS if the native module isn't available.
This forces a pattern in code to check if the module is available before accessing certain feature flags, to avoid showing that error to users in the legacy architecture.
This removes the need for that pattern by adding the check in the feature flag infra itself. If TM infra isn't available, we return default values and don't log the error.
Reviewed By: rozele
Differential Revision: D70975412
fbshipit-source-id: 5544bd96aaaf046ef90ca69bc549bef4e54eb98b
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49855
Changelog: [IOS][FIXED] - Fixed touch events not being dispatched to ScrollView's children when they overflow the content container
Closes https://github.com/facebook/react-native/issues/47740.
Changes the ScrollView's container view to be `RCTViewComponentView` instead of `UIView` and sets custom layout metrics to it in a way that it will propagate touch events to all children, even if they overflow its bounds.
Reviewed By: sammy-SC
Differential Revision: D70619894
fbshipit-source-id: 348a1a369489d5208d6037c8d76b223c4ab2d5f7
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49582
TSIA, this is a fairly straightforward implementation since iOS likes the shape the data is already in. Note I form a stacking context since I need the native view that has `accessibilityElements` to actually be an ancestor of everything under it
Note I am not exposing anything yet on the JS side, since we need to do Android also
Changelog: [Internal]
Reviewed By: NickGerleman
Differential Revision: D69625269
fbshipit-source-id: 836f1bb631646408a5726f126be61c0ff6afe1d0
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49960
Fixes https://github.com/facebook/react-native/issues/49106
RN legacy arch, and web, will clip inline content which appears after elipsized text. This is the correct behavior, compared to new arch, which will put it in a random place depending on the platform.
`line-clamp`: https://jsfiddle.net/7xgdke1b/
`text-overflow`: https://jsfiddle.net/7xgdke1b/2/
Fabric renderer does not, funnily enough, having an `isClipped` field on `TextMeasurement::Attachment` that is never used.
This change propagates state for whether an attachment is beyond elipsized area to this measurement, then when we see it, we set empty layout results with `DisplayType::None` so that we don't create native views. We don't layout child views either, but this seems to work out okay, even when removing and re-adding `numberOfLines`.
Changelog:
[General][Fixed] - Fix New Arch handling of inline views when text truncated
Reviewed By: mdvacca
Differential Revision: D70922174
fbshipit-source-id: 8c1f4aadbf53ff64ce55b44d6c7953d9b2e40bc5
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49543
When using `ReactScrollView` or `ReactHorizontalScrollView` Views with `removeClippedSubviews` keyboard navigation didn't work.
This is because keyboard navigation relies on Android's View hierarchy to find the next focusable element. With `removeClippedSubviews` the next View might've been removed from the hierarchy.
With this change we delegate the job of figuring out the next focusable element to the Shadow Tree, which will always contain layout information of the next element of the ScrollView.
Changelog: [Android][Fixed] - Fix keyboard navigation on lists with `removeClippedSubviews` enabled
Reviewed By: joevilches
Differential Revision: D69618406
fbshipit-source-id: 1df7f90066dfc685e74a89b29222937777714b87
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49961
As preparation for fixing focus on FlatList we need to prevent the currently focused view from getting clipped. This is because in Android, if the currently focused view gets clipped before transferring focus we crash.
Changelog: [Android][Changed] - Prevent currently focused child from getting clipped when `removeClippedSubviews` is enabled
Reviewed By: NickGerleman
Differential Revision: D70994348
fbshipit-source-id: 8c68eeebd7df8e7d8f788e50a3bbe46b0335ace3
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49951
We previously fixed Differentiator generating an incorrect parentTag (https://github.com/facebook/react-native/pull/48055), but this can lead to crashes in Android UI due to reordering that happens in the Android mounting layer. While we have an experiment to disable this reordering (https://github.com/facebook/react-native/pull/46702) this currently has a negative performance impact which needs to be addressed.
As a mitigation, we can make the lookup of parentTag's ViewManager state nullable. We only require this to support `needsCustomLayoutForChildren`, which is not commonly used, and seems acceptable to drop in this scenario.
Changelog: [Android][Changed] Do not crash when parent view state can't be found
Reviewed By: NickGerleman
Differential Revision: D70966621
fbshipit-source-id: 33d0b6a90860788a4c9a8c6cea36c2c72c1392e1
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49937
Adding an assert for recycled views still attached to their parent, which would lead to an exception when the view would be added to a new parent.
Changelog: [Internal]
Reviewed By: javache
Differential Revision: D70922503
fbshipit-source-id: 7d4daf427306203d603c31999ab138b3aee08e83
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49954
We have a different version of OkHttp internally and in OSS so we need to suppress these for now.
Changelog: [Internal]
Reviewed By: Abbondanzo
Differential Revision: D70975145
fbshipit-source-id: cd2a3ab5cdc52f25c78cb608e917fa3ca378e4f1
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49817
changelog: [internal]
Remove feature flag excludeYogaFromRawProps. It did not yield any performance wins and the code to support the feature is spread across multiple classes. Removal simplifies complexity.
Reviewed By: javache
Differential Revision: D70389071
fbshipit-source-id: 21751f56aec8de501ff1e1efafda035069d8ef48
Summary:
Migrate com.facebook.react.uimanager.layoutanimation.OpacityAnimation to Kotlin.
I moved out `OpacityAnimationListener` as well in this PR to separate concerns and align more with the codebase.
## Changelog:
[INTERNAL] - Migrate com.facebook.react.uimanager.layoutanimation.OpacityAnimation to Kotlin
Pull Request resolved: https://github.com/facebook/react-native/pull/49822
Test Plan:
```bash
yarn test-android
yarn android
```
Reviewed By: javache
Differential Revision: D70588884
Pulled By: alanleedev
fbshipit-source-id: 1b70f5e5facd5f47b828090e7c63a218229000cb
Summary:
When using the `login-password` prompt type, there is a TypeScript type mismatch issue. The `callbackOrButtons` parameter returns an object with `{login: string, password: string}` structure, but this type variation is not properly included in the **AlertType** type definition. This causes TypeScript to show type errors when using callback functions that expect credentials in the format `(credentials: {login: string, password: string}) => void`.
## Changelog:
- [General] [Fixed] Add missing type variation `{login: string, password: string}` to **AlertType** type definition to properly support `login-password` prompt callbacks
Pull Request resolved: https://github.com/facebook/react-native/pull/49757
Test Plan: This change is purely type-related and doesn't affect runtime behavior.
Reviewed By: NickGerleman
Differential Revision: D70797036
Pulled By: alanleedev
fbshipit-source-id: 8e27a39f0c8f49083730c683b41b69173715bd68
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49927
This change automates updating App's Info.plist with the new `RCTNewArchEnabled` boolean entry. The value depends on how the pod install is set up. In this way, we maintain the previous UX to enable/disable the New Arch.
## Context
The RCT_NEW_ARCH_ENABLE flag is a compile time flag we used for almost two years to configure the iOS apps and to determine whether the app should build with the New Arch or not.
However, given that we are looking into prebuilding React Native, we have to get rid of all the compilation flags, because they would require us to prebuild a combinatorial number of artifacts for react native. For example:
- New Arch / Hermes
- Old Arch / Hermes
- New Arch / JSC
- Old Arch / JSC
- ...
## Backward compatibility
We are going to keep adding the RCT_NEW_ARCH_ENABLED flag in all the dependencies, through the cocoapods inrastructure, so libraries, which are not prebuilt, will be build for the right architecture by the app itself.
## Changelog:
[iOS][Added] -
Reviewed By: cortinico
Differential Revision: D70888212
fbshipit-source-id: 541a818e02ccb96c822de29b0714646d1e69a5a9
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49924
This change deprecates the RCT_NEW_ARCH_ENABLEd flag to toggle the New Architecture.
The new approach bring iOS closer to Android: to diasable the New Architecture, user needs to modify the App's Info.plist and add a `RCTNewArchEnabled` boolean entry and set it to `NO`.
The absence of the entry implies that the New Arch is enabled. (Defaults to enabled)
This also deprecates the `RCTSetNewArchEnabled` function because it makes no sense now
## Context
The RCT_NEW_ARCH_ENABLE flag is a compile time flag we used for almost two years to configure the iOS apps and to determine whether the app should build with the New Arch or not.
However, given that we are looking into prebuilding React Native, we have to get rid of all the compilation flags, because they would require us to prebuild a combinatorial number of artifacts for react native. For example:
- New Arch / Hermes
- Old Arch / Hermes
- New Arch / JSC
- Old Arch / JSC
- ...
## Backward compatibility
We are going to keep adding the RCT_NEW_ARCH_ENABLED flag in all the dependencies, through the cocoapods inrastructure, so libraries, which are not prebuilt, will be build for the right architecture by the app itself.
## Changelog:
[iOS][Deprecated] - deprecate the `RCT_NEW_ARCH_ENABLED` and the `RCTSetNewArchEnabled`
Reviewed By: cortinico
Differential Revision: D70885454
fbshipit-source-id: 4b8404ead4900d9787049ebd0b8b3c29b272f913
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49929
X-link: https://github.com/facebook/metro/pull/1459
In the next version of Flow, we will stop bundling many of the builtin libdefs, and they have been moved to flow-typed. This diff checks in them to prepare for the deployment of the next version of Flow.
Changelog: [Internal]
Reviewed By: SamChou19815
Differential Revision: D70896122
fbshipit-source-id: aa6bd0b1e653b983a345b0e202c8dad5058a0001
Summary:
GHA passes the version to the script with a `v` prefix. However, when we receive the version from NPM, the `v` prefix is not here.
We can fix the script by dropping the `v` when it is passed to the function.
bypass-github-export-checks
## Changelog:
[Internal] - Fix verifyPackageOnNPM
Pull Request resolved: https://github.com/facebook/react-native/pull/49944
Test Plan: GHA
Reviewed By: cortinico, fabriziocucci
Differential Revision: D70960414
Pulled By: cipolleschi
fbshipit-source-id: 4234103ebe49cf715aea4a1473a8a60978f07a9f
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/49936
Ensure LegacyArchitectureLogger.assertWhenLegacyArchitectureMinifyingEnabled is called only once per class loading in kotlin
changelog: [internal] internal
Reviewed By: alanleedev
Differential Revision: D70922132
fbshipit-source-id: ce4d68dc2ab0f61266e4bb5cddfc175c364f568a
Andrew Knapp
Vitali Zaidman
Tim Yung
Nick Gerleman
Luna Wei
Jorge Cabiedes Acosta
Nicola Corti
Arushi Kesarwani
Ramanpreet Nara
Samuel Susla
Eric Rozell
Kirill Novikov
Rubén Norte
Alex Hunt
Joe Vilches
Peter Abbondanzo
Jakub Piasecki
Sam Zhou
Mateo Guzmán
Pieter De Baets
Nick Lefever
Dawid Bartczak
zhongwuzw
Vojtech Novak
Riccardo Cipolleschi
David Vacca