Summary:
The change is gated by two QEs (iOS and Android).
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: mdvacca
Differential Revision: D21317797
fbshipit-source-id: 34bfd9546600d9cb3467bddfc61363a16683f81f
Summary:
I spent the last several days thinking about state reconciliation issues, some crashes (T65586949) that suspiciously happen somewhere inside, and a bunch of issues that might be connected to that (possibly, some of T65516263 sub-task).
I cannot see some obvious problems in the current state reconciliation algorithm that might cause the crash (because of some use-after-free or other pure C++ issues), but I suspect some of the problems we experience might be caused by some details of how we reconcile states.
In the current approach, we rank all states based on the "hierarchical" history of their creation (state version is being calculated based on the version of the base tree). That's usually fine but in some cases when trees are being constructed concurrently, a logical version of a based tree does not correspond to the local version of a committed tree. In other words, the linear history of commits does not always correspond to the "hierarchical" history of trees generation that was done by different parties (e.g. React vs native state update pipeline).
In this diff, I tried to change the approach to change the algorithm to follow this logic: If some state is `obsolete` (already been committed and then replaced with newer one), we replace that with the most recent one. This change does not introduce the `obsolete` flag; is already used by State infra to avoid cloning nodes with an outdated state.
Interestingly, it fixes the issue with an empty BottomSheet on Android (T66177144). See the attached video. The hope is that it's also will
This change theoretically might affect all things that use State, so it hard to predict what can break and how. So, if we don't see obvious problems here, I would set up a GK/QE and run the experiment in prod.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: JoshuaGross
Differential Revision: D21295137
fbshipit-source-id: e5613218d3e11a56623cab9bbf2540495b2b24e8
Summary:
`JSCallInvoker` doesn't exist anymore, because it was renamed to `CallInvoker`. So, this `.clang-tidy` dotfile isn't necessary.
Changelog: [Internal]
Reviewed By: PeteTheHeat
Differential Revision: D21318061
fbshipit-source-id: 8098eef670cae35bf3c027c91f18f45eb7bcd33a
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/28779
Creating a JNI wrapper class for RuntimeExecutor so we can pass it to Fabric and TurboModules in Java.
Changelog: [Internal]
Reviewed By: RSNara
Differential Revision: D21049385
fbshipit-source-id: f833004225d9837acf6ffafd3988f89748cf12aa
Summary: Changelog: [Internal] [Yoga] Use double instead of float during rounding process to prevent loss of precision.
Reviewed By: mdvacca
Differential Revision: D21227565
fbshipit-source-id: 380b57535a356624cda8dc2017871a4ef3c882d1
Summary:
We neede that to make the test less flaky.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: mdvacca
Differential Revision: D21261487
fbshipit-source-id: afec45e03be10ba071a5ab808ce13f95b5f475e3
Summary:
Changelog: [Internal]
In `onKeyPress` event, we were not returning `key` property. This diff adds `key` property to `onKeyPress` event and removes other, redundant properties from `onKeyPress` event.
The implementation has been translated from Paper.
Reviewed By: shergin
Differential Revision: D21250411
fbshipit-source-id: f1e31381667acb9dec02d0b33883df8f8f5b2a4b
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/28730
Moving RuntimeExecutor out of react/utils and into its own subdir of ReactCommon. I'm doing this because I'm going to be pulling this into CatalystInstance on Android, and I don't want to pull in all the files we don't need there; also, this should hopefully make the OSS NDK stuff easier (this uses the react/utils prefix to export, and I'm not sure if you can include a '/' in a gradle module name?)
Changelog: [Internal]
Reviewed By: shergin, mdvacca
Differential Revision: D21098528
fbshipit-source-id: 9fbd72901ece522b1caec3ec34fafb8f9b327275
Summary:
TextInput must have `LeafYogaNode` trait to be able contain non-yoga nodes.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: sammy-SC
Differential Revision: D21252322
fbshipit-source-id: 820f3ae4811cb475550419af501739c57c7164e1
Summary:
Because of StateReconciliation (?) it is possible, even likely, that eventually the ShadowNode that ReactJS holds onto will not be the latest ShadowNode, and will not contain the latest LayoutMetrics.
This fixes the problem by always resolving the latest committed version of the ShadowNode before measuring it.
Changelog: [Internal] Fabric measure bugfix
Reviewed By: sammy-SC, mdvacca
Differential Revision: D21248472
fbshipit-source-id: a4f034c36d89a91a804f3b4cdf35e7ae748a1b91
Summary:
It's not immediately obvious from the UI/UX when/if this mechanism breaks, so it's good to have a test.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: JoshuaGross
Differential Revision: D21184718
fbshipit-source-id: 25432a1398cff3ce61f62cf433e3cb73d7a7a93f
Summary:
It's not allowed to return nullptr from the callback. The assert ensures it which is helpful during development.
Probably, we should consider using `gsl::not_null<>` here.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: sammy-SC
Differential Revision: D21149891
fbshipit-source-id: a5f77b35029f22b499491721036405682f812a38
Summary:
Previously, a derived class, WithTuple, was used. This ran
into bugs in MSVC (see https://github.com/microsoft/STL/issues/121).
Instead, use specialization to get the same result using std::tuple
directly. This avoids the bug, and is a cleaner API.
Changelog: [Internal]
Reviewed By: dulinriley
Differential Revision: D21233677
fbshipit-source-id: 1d75991847164e525b4ba70f65a90627e5f8cd56
Summary:
The integration with Yoga was pretty complex from day one. The first attempt to make it simpler was in D19963353 when we removed a bunch of layers of indirection. This is the second iteration that aimed to simplify the structure of methods and their responsibilities.
The only conceptual change (that I am aware of) in this diff is that now we don't support (imaginary) case where a non-leaf YogaLayoutableShadowNode can have a non-YogaLayoutableShadowNode child. In the previous version, it was a no-op, now it's not supported and an assert will fire.
Alongside with refactoring, this diff implements several helper functions that verify the invariants important for the Concurrent Layout in debug mode.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: JoshuaGross
Differential Revision: D21198222
fbshipit-source-id: cc085904948056f861562af5bd2571de45a743b9
Summary:
## Problems:
In my investigation of T65656635, I realized that the TurboModule system has a number of problems:
- In TurboModules, we use 1 lock to create n TurboModules. We should change this setup to n locks for n TurboModules. This way, two threads creating two different NativeModules don't compete for the same lock. Also, this is how it's done in Android (TurboModules & NativeModules), and iOS (NativeModules).
- In TurboModules, we don't calculate "requires main queue setup" faithfully. In the legacy system, if a NativeModule has a custom `init` method or a custom `constantsToExport` method, it "requires main queue setup" with a warning.
- In TurboModules, we don't create the NativeModule on the main queue, if "requires main queue setup" is true. Instead, the NativeModule is always created on the thread that requires it.
- In TurboModules, we don't perform any concurrency control around `id<RCTTurboModule>` setup. We should.
## What this diff does
In this diff, I fixed all the aforementioned issues by re-implementing `provideRCTTurboModule:`.
**Algorithm Notes:**
- **Gist:** When `n` threads race to create NativeModule `x`, only the first thread creates and sets up `x`. All others are told to wait. Once the creator thread finishes its job, it notifies all other waiting threads, which then wake up and return the newly created NativeModule. This algorithm was initially implemented in NativeModules for Android inside (ModuleHolder.java). I modified and implemented it for TurboModules for Android, and now this diff implements it for TurboModules for iOS.
- The TurboModule cache is replace with a TurboModuleHolder map. A TurboModuleHolder manages the creation lifecycle of a TurboModule, and holds a condition variable and mutex for doing concurrency control around it. When the bridge invalidates, in TurboModuleManager, we set the `invalidating` flag to true, which prevents the insertion of new entries into the TurboModuleHolder map.
- I added a `std::mutex` to serialize calls into the TurboModuleManagerDelegate, so we don't get races if the delegate isn't thread-safe.
Changelog:
[iOS][Fixed] - Re-implement RCTTurboModuleManager provideRCTTurboModule:
Reviewed By: shergin
Differential Revision: D21170099
fbshipit-source-id: 8792812c2237d3bfc80c9834c818e011de85b0ea
Summary:
`ShadowNodeFamilyFragment::Value` is a value couter-part type for `ShadowNodeFamilyFragment`.
We need that to be able safely copy data stored inside a `ShadowNodeFamilyFragment` object.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: kacieb
Differential Revision: D21169580
fbshipit-source-id: 1a485e1b2ae47bc7da9476a60466934ac9d61366
Summary:
We need it to be able pass an `EventEmitter` object to constructed concrete State objects.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: JoshuaGross
Differential Revision: D21169581
fbshipit-source-id: 3eef0310de7e2f061108aa85c1a39678a43fe85e
Summary:
Adds `RootTag` as a valid type for arguments and return types in TurboModules (on both Android and iOS).
This will enable us to change `RootTag` into an opaque type. There are two compelling reasons to do this:
- JavaScript will no longer be able to safely depend on `RootTag` being a number (which means we can change this in the future).
- Call sites using `unstable_RootTagContext` will can get a `RootTag`, but call sites using the legacy `context.rootTag` will not. This means the opaque type will give us a strategy for migrating away from legacy context and eventually making `unstable_RootTagContext` the only way to access `RootTag`.
Changelog:
[Internal]
(Note: this ignores all push blocking failures!)
Reviewed By: RSNara
Differential Revision: D21127170
fbshipit-source-id: baec9d7ad17b2f8c4527f1a84f604fc0d28b97eb
Summary:
Easy diff to document in code how to enable logging of Shadow Tree instrospection
changeLog: [Internal] Internal change used on Fabric
Reviewed By: JoshuaGross
Differential Revision: D21150196
fbshipit-source-id: 8eb23ec3ea1d574b79b09333428ab52c851065dd
Summary:
Update differ test so it passes again. Previously to D21111423 (I think) nodes were being incorrectly detected as updated even if they weren't different, so now there are fewer unnecessary Update mutations generated.
Changelog: [Internal]
Reviewed By: mdvacca
Differential Revision: D21148647
fbshipit-source-id: cab6e3ecd0a457e1ac3155b3468bcc56663dab0b
Summary:
Yoga uses a dirty flag to re-layout nodes. In normal, single-threaded approach the policy for dirtying is simple: if a node was changed, we need to dirty it. In the Concurrent Yoga approach, those rules are not so simple, and it seems we haven't formalized those rules yet.
Investigating some layout issues that we have in Fabric, I tend to believe that we don't dirty as much we should. Hense this change adds mode dirtying.
Reviewed By: JoshuaGross
Differential Revision: D21092815
fbshipit-source-id: 4603c97ccb79efcdf5e6a4cc450ebe61b63effb3
Summary:
This is quite a fateful mistake. `getDirtied()` returns the pointer to a function which is obviously a mistake here; we should use `isDirty()` instead.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: mdvacca
Differential Revision: D21028569
fbshipit-source-id: 95212b31f4e32d51c594d5209f295397af3f1252
Summary:
This Hermes-specific mode is similar to Debugger.setPauseOnExceptions
and lets the VM know that it should enter a Pause state whenever a new
script is loaded/executed.
The debugger can then take its time to parse the source map and update
any breakpoints, before automatically continuing.
Changelog: [Internal] Implement a Hermes.setPauseOnLoad CDP call
Reviewed By: bestander
Differential Revision: D20754604
fbshipit-source-id: 7f9d0638706c99e9dcb534699b633f658e364909
Summary:
Until now we've generated scaffolding entirely based on the official devtools
protocol spec. This diff adds support for defining custom domains in `custom.json`
which will be merged with the upstream protocol JSON definition.
ChangeLog: [Internal] Add support for Hermes-specific CDP messages
Reviewed By: bestander
Differential Revision: D20754605
fbshipit-source-id: a8075f81816a40114d1a3332192c7aa076b17848
Summary:
Assigning self as an owner makes a cycle which is obviously a bug.
Changelog: [Internal] Small change in Yoga (should not affect RN).
Reviewed By: SidharthGuglani
Differential Revision: D21111423
fbshipit-source-id: 1835561c055ac827f5ce98a044f25aed0d1845a5
Summary:
Changelog: [Internal]
We were assigned `undefined` value to incorrect edge, instead of `YGEdgeLeft` it should have been `YGEdgeRight`.
If node has `YGEdgeRight` value, it needs to be reassigned to `YGEdgeEnd` and its original value set to undefined.
Reviewed By: mdvacca
Differential Revision: D21095234
fbshipit-source-id: fbecd9b7e6670742ad4a4bb097760aa10eec8685
Summary:
The diff changes how the `empty raw props` optimization works in `ComponentDescriptor::cloneProps()`. Now it only fires only when the base `props` object is null, which is practically all production cases we have (and care about). (I tried, in a normal run there were no cases where the empty raw props were passed with non-null props.) From the other side, the old behavior that may return the same props objects previously several times created bugs and practically unexpected results and practically disallowed to clone props objects easily.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: JoshuaGross
Differential Revision: D21110608
fbshipit-source-id: 884807cd8e9c5c3e6cc1c9e4c1f0227259cc21fb
Summary:
Apparently, `std::this_thread::sleep_for` uses a different clock to measure time which causes ofter misalignment with the clock which Telemery uses which makes the test flaky. Using the same clock should fix it.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: JoshuaGross
Differential Revision: D21116058
fbshipit-source-id: 52dde2e325776d365431a2a957dcc12dfe53f890
Summary:
This pull request adds a Unicode `u8` prefix to the string literal returned in `AttributedString.cpp`'s `Fragment::AttachmentCharacter()`.
This fixes the following error when building on MSVC:
```
react\attributedstring\AttributedString.cpp(21): error C4566: character represented by universal-character-name '\uFFFC' cannot be represented in the current code page (1252)
```
## Changelog
[Internal] [Fixed] - Fabric: Add Unicode prefix to AttachmentCharacter
Pull Request resolved: https://github.com/facebook/react-native/pull/28617
Test Plan: The Fabric test suite has been ran on a Clang-based build of Fabric on macOS, and no regressions in it have been noted.
Differential Revision: D21118078
Pulled By: shergin
fbshipit-source-id: c105de5e4edb67fed97ce44153a75d9d380bf588
Summary:
This pull request changes the include of Glog from `<Glog/logging.h>` to `<glog/logging.h>` in `MountingTest.cpp`. This fixes building on a case-sensitive filesystem.
## Changelog
[Internal] [Fixed] - Fabric: Fix case of Glog include in MountingTest.cpp
Pull Request resolved: https://github.com/facebook/react-native/pull/28616
Test Plan: The `include` of Glog no longer causes issues with building `MountingTest.cpp` on a case-sensitive filesystem.
Differential Revision: D21118085
Pulled By: shergin
fbshipit-source-id: c958c54bf88333fd5001127779c855ce8c2666c3
Summary:
We need to break up the `uimanager` module in order to solve circular dependencies problem (which future diff would have otherwise).
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: JoshuaGross
Differential Revision: D20885646
fbshipit-source-id: b8e3199c0eacc57a5be1481595cf97c84f972293
Summary:
We need to break up the `uimanager` module in order to solve circular dependencies problem (which future diff would have otherwise).
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: JoshuaGross
Differential Revision: D20885645
fbshipit-source-id: 8148bd934879802b076261ed86fa78acf0a07ed3
Summary:
We need to break up the `uimanager` module in order to solve circular dependencies problem (which future diff would have otherwise).
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: JoshuaGross
Differential Revision: D20885163
fbshipit-source-id: 08eb1ba1d408fc0948e8d0da62380786a40973af
Summary:
Yoga uses `std::array` a lot (and `std::array` is not a `std::vector`), so it's useful for printing Yoga values.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: mdvacca
Differential Revision: D21028570
fbshipit-source-id: c6bf114d5362f085ea201ecdc5b7d59646b33ebd
Summary:
This is a debug-only feature that simply should be an assert. When it triggers in debugger and bubbles to some random exception catch block which makes it impossible to understand was exactly it happens. Making it an assert will stop debugger exactly where it happens.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: mdvacca
Differential Revision: D21028571
fbshipit-source-id: 3df4ec0da922026bb9df61081cb71113577e06e9
Summary:
* <Image> must be a leaf node; having a proper trait will fail earlier in case of misuse (mounting something inside).
* <View> must have a `View` trait because it's for what that trait is.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: sammy-SC
Differential Revision: D21028573
fbshipit-source-id: 457716d4661333eb2357f34316f3e495ab4fda24
Summary:
We don't use it as vitrual anymore (setLayoutMetrics is a non-virtual method already), so it does not need to be marker virtual.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: mdvacca
Differential Revision: D21028572
fbshipit-source-id: 99f86fdd4cf2f5972034d9058d7b82bdc8680187
Summary:
Changelog:
[Internal][Yoga] YGDirection variable was initialized incorrectly by casting -1 to YGDirection. Changing it to default value of direction
Same for YGMeasureMode.
Reviewed By: pasqualeanatriello
Differential Revision: D20869042
fbshipit-source-id: 7bfe490193321baae875ef6fb49a938851950c9f
Summary:
## Summary
Please check out D21035208.
## Changes
- Update `RCTTurboModuleManagerDelegate getTurboModule:instance:jsInvoker:nativeInvoker:perfLogger` to use `RCTTurboModuleManagerDelegate getTurboModule:(const ObjCTurboModule::InitParams)`
- Update all implementations of `RCTTurboModuleManagerDelegate` in accordance with this API change
Changelog:
[iOS][Changed] - Make RCTTurboModuleManagerDelegate create TurboModules via ObjCTurboModuleManager::InitParams
Reviewed By: PeteTheHeat
Differential Revision: D21036272
fbshipit-source-id: c16002c47db26e2ba143fc1080afe9e2fe1e7816
Summary:
## Summary
Please check out D21035208.
## Changes
- `ObjCTurboModule::ObjCTurboModule` changed to accept a bag of arguments `const ObjCTurboModule::InitParams` instead of an argument list.
- TurboModule iOS codegen scripts updated to generated `ObjCTurboModule` subclasses that accept a `const ObjCTurboModule::InitParams` object in their constructor, and forward it to `ObjCTurboModule::ObjCTurboModule`.
- All manually checked in code-generated ObjC++ classes (i.e: RCTNativeSampleTurboModule, RCTTestModule, FBReactNativeSpec) are updated.
## Rationale
This way, the code-gen can remain constant while we add, remove, or modify the arguments passed to ObjCTurboModule.
## Commands run
```
function update-codegen() {
pushd ~/fbsource && js1 build oss-native-modules-specs -p ios && js1 build oss-native-modules-specs -p android && popd;
}
> update-codegen
```
Changelog:
[iOS][Changed] Update ObjCTurboModule to use ObjCTurboModule::InitParams
Reviewed By: PeteTheHeat
Differential Revision: D21036266
fbshipit-source-id: 6584b0838dca082a69e8c14c7ca50c3568b95086
Summary:
I'm trying to use JSI for a React Native custom module. I saw these existing examples where the JSI API is used in the context of a CMakeLists.txt:
https://github.com/terrysahaidak/host-object-test/blob/master/libs/android-jsi/test-jsi/src/main/cpp/CMakeLists.txthttps://github.com/ericlewis/react-native-hostobject-demo/pull/4/files#diff-834320be1b4e4016bac27c05dcd17fb9
In both cases, they manually grab the include directories and jsi.cpp from node_modules/react-native, but I also saw that node_modules/react-native/ReactCommon/jsi/jsi already has a CMakeLists.txt that appears to be intended to provide a jsi static lib, so I tried to pull this into my own CMakeLists.txt like this:
```
add_subdirectory(${RN_DIR}/ReactCommon/jsi/jsi ${CMAKE_CURRENT_BINARY_DIR}/jsi)
...
target_link_libraries(MyLib jsi)
```
Unfortunately when doing this, the consuming project still doesn't see the correct include directories. The change I'm proposing here is to use `target_include_directories` and declare that `..` is a public (to consumers) include directory for the library named `jsi`. With this change, you can do what I showed above to consume the jsi lib by just pulling in the CMakeLists.txt file into your own CMakeLists.txt file.
Changelog: [General][Fixed] Fix jsi cmake include dirs
Pull Request resolved: https://github.com/facebook/hermes/pull/207
Reviewed By: willholen
Differential Revision: D21074270
Pulled By: tmikov
fbshipit-source-id: 7d9ec3255f57a16c0b2be489dffa4540727738a1
Summary:
## Problem
Every time we want to add, remove, or change the data passed to JavaTurboModule's constructor, we have to modify the C++ TurboModule codegen. (The same is true of `ObjCTurboModule`).
**Why was this necessary?**
- `JavaTurboModule` is effectively an abstract class whose constructor is always invoked by code-generated C++ classes. These C++ code-generated class constructors accept an argument list, and manually foward each and every item in that list to `JavaTurboModule::JavaTurboModule`.
## The fix
In this diff, I introduce a struct `JavaTurboModule::InitParams`, to represent a bag of arguments:
```
class JSI_EXPORT JavaTurboModule : public TurboModule {
public:
struct InitParams {
std::string moduleName;
jni::alias_ref<JTurboModule> instance;
std::shared_ptr<CallInvoker> jsInvoker;
std::shared_ptr<CallInvoker> nativeInvoker;
};
```
All `JavaTurboModules` will be created with an instance of this `InitParams` struct, instead of a list of arguments. Our code-generated C++ `jsi::HostObject` sublcasses will simply accept `InitParams` in their constructor, and forward it to `JavaTurboModule`'s constructor. This way, the codegen remains oblivious to what arguments JavaTurboModule requires.
## Okay, but why do we need this change now?
In the future, I plan to modify the constructor for `JavaTurboModule` to accept a performance logger, and a `RuntimeExecutor`. Similar modifications are planned for ObjC. For this reason, to avoid these four codemods, and any potential other codemods that occur because we're making modifications to `JavaTurboModule` or `ObjCTurboModule`, I'm launching this codemod, and the codemods in this stack.
## Misc Fix
- Previously, we were generating the TurboModule name from the Spec filename. This is incorrect because that name represents the spec name. Now, the name will be forwarded from TurboModuleManager in the `JavaTurboModule::InitParams` struct.
## Alternative implementations
I initially considered using `ContextContainer`, but decided against it because:
1. There are no type-safety guarantees.
2. I think it's a bit overkill for this scenario. We just need an opaque bag of data, and for our purposes a simple struct does the job fine.
## Commands run
Reviewed By: fkgozali
Differential Revision: D21035208
fbshipit-source-id: 9542cafea192081bc34d337ab3a7a783083eb06c
