Summary:
ShadowView, ShadowViewMutation, and Differentiator were decoupled to separate module.
That enables us to use ShadowView more widely without facing a circular dependency problem.
Reviewed By: mdvacca
Differential Revision: D13205229
fbshipit-source-id: 7373864bf153a7813c2f97edb263a41454ce0b88
Summary:
Previously, we stored a pointer to ShadowNode inside NSAttributedString's attributes to make possible retrieving an EventEmitter associated with some text fragment.
That worked fine besides only one caveat: the internal implementation of NSAttributedString is quite strange and that causes a memory leak. Because of some reason, NSAttributedString does not release stored attributes after own destruction (maybe OS uses some kind of caching).
So, now, instead of storing a strong pointer to ShadowNode inside NSAttributedString, we store a weak pointer to EventEmitter. Storing a weak pointer is okay because a desired lifetime of EventEmitter is guaranteed by LocalData stored inside a View. Storing a weak EventEmitter instead of weak ShadowNode will also help us with migration to ShadowView (we cannot store ShadowView weakly because it's a stack allocated object).
Reviewed By: sahrens
Differential Revision: D13196886
fbshipit-source-id: f8714e4b3709765629d6456edf0c635bf5f7c53b
Summary: Over-retaining a LocalData object inside the View can cause a crash during tearing down JS VM because LocalData can indirectly retain EventEmitter objects which were not properly "disabled".
Reviewed By: sahrens
Differential Revision: D13196887
fbshipit-source-id: 001d9fadf775c89f750c84fe8da9b84d4636631c
Summary: RCTViewComponentView retains an EventEmitter, so we have to clear this up after we recyled the view.
Reviewed By: sahrens
Differential Revision: D13196884
fbshipit-source-id: e9f2e2400be864c5c6177227255012101ed8c4d1
Summary: View recycling can be pretty aggressive and memory intensive, so we can properly react on system memory-pressure notification.
Reviewed By: mdvacca
Differential Revision: D13176278
fbshipit-source-id: 38ea1b27da988aeaaa5db6ac0b94389a0bd2799e
Summary:
Apparently, the previous behavior brings more problems than some *possible-in-the-future* features and flexibility.
The new model allows us to easily implement "nested text" feature.
(We temporary hope the old behavious for Android only for compatibility reasons.)
Reviewed By: mdvacca
Differential Revision: D13176277
fbshipit-source-id: 01f7bfb3c2e70cc89d76ecb78add016ee91cbd63
Summary:
The whole mounting iOS infra now uses `ComponentHandle` instead of `std::string` as a reference to particular `ComponentView` implementation. All changes are pretty straightforward, we use a different thing/type to refer to the particular class; no changes in the logic besides a new `RCTComponentViewFactory` that serves the same role of classes registry as Objective-C runtime served previously.
That has several benefits:
* It should be slightly faster, mostly because we don't need to convert `char *` strings to `std::string` and then to `NSString *`.
* We don't need string-based component-name maps anymore (at least on this layer). We can call classes as we want and it will work because of classes are now explicit about which ShadowNodes they are compatible with.
* Most importantly, it's explicit now! That means that no runtime magic is involved anymore and we can rely on static linting tool now and not be afraid of improper code stripping/overoptimization.
Reviewed By: mdvacca
Differential Revision: D13130760
fbshipit-source-id: aadf70525a1335b96992443abae4da359efdc829
Summary: The new method in the protocol enforces view component classes to expose a component handle of the component that the view component represents. That will allow us to wire up those classes with shadow views in runtime explicitly and in a much more performant way than it is now.
Reviewed By: mdvacca
Differential Revision: D13114663
fbshipit-source-id: 853187d978aab200f85719d9c1d9fea2e3ad4e55
Summary: Future::then taking a value-taking function is deprecated and being deleted. This cleans up a few more callsites.
Reviewed By: yfeldblum, shergin
Differential Revision: D13163277
fbshipit-source-id: 98d1f78c5b34ca34603cc24d79157a4718573576
Summary: That's a temporary change that useful only while we don't have deeper/proper intergration with UIKit's navigation infra.
Reviewed By: mdvacca
Differential Revision: D13104172
fbshipit-source-id: 024e71e04470d56d2c5e9b3f6c3392555ce50b63
Summary:
Apparently, if a gesture recognizer got 'reset', OS does not call `touchesCancelled:` method, so we have to do it manually.
To implement this we have to split `_dispatchTouches:eventType:` into two methods: the first converts Objective-C touches to C++ touches, the second consumes that. We have to do this because during reset we don't have a collection of UIKit touches.
Reviewed By: mdvacca
Differential Revision: D13072807
fbshipit-source-id: 677e2febf9f96dcdfaadfadf5b9ab3893f93e796
Summary: Every C++ engineer (except me several months ago) knows that `operator []` can mutate the collection (Yeah! Don't ask), so this is especially dangerous if your hash function is broken (see the previous diff).
Reviewed By: mdvacca
Differential Revision: D13072805
fbshipit-source-id: 4436a8ff12fb27a57bfb6ee0ff986d7a9a825549
Summary:
Without this fix the PointerHasher hashed not a pointer, but an address where the pointer is stored, which is obviously incorrect.
The incorrect hash function compromised the whole `_activeTouches` caused many issues with logically invalid invariants (e.g. removing element from the collection silently didn't work).
Reviewed By: mdvacca
Differential Revision: D13072804
fbshipit-source-id: d68289e940fe21e2df08a31439619b7e2fe0fa13
Summary: This diff changes the behavior of the Scheduler.schedulerDidRequestPreliminaryViewAllocation to avoid pre-allocating views that are non-layoutables
Reviewed By: shergin
Differential Revision: D12962008
fbshipit-source-id: cb2670beafdcbd2116fbdaf2dc5d1b4726330ec2
Summary: We are moving to more stable APIs removing all mentiones of the effort name from the codebase.
Reviewed By: mdvacca
Differential Revision: D12912894
fbshipit-source-id: 4a0c6b9e7454b8b14e62d419e9e9311dc0c56e7a
Summary: This diff exposes rootTag as part of SchedulerDelegate.schedulerDidRequestPreliminaryViewAllocation(). This will be necessary to be able to pool views per Surface in Android
Reviewed By: shergin
Differential Revision: D12875656
fbshipit-source-id: d2a8c1f9bcc6b14c17b34bf59085da44ae3c3416
Summary: Trivial. We don't use it anymore.
Reviewed By: mdvacca
Differential Revision: D12876743
fbshipit-source-id: dc979aaea1fef443b8caf2e58d44b0c7aad90246
Summary:
We double down on JSI in Fabric. So, practically, JSI is now a hard dependency for Fabric. I hope it's for good.
Now `jsi::Runtime` is coupled with scheduling via `EventExecuter`, so we have to make `jsi::Runtime` a part of `EventBeat` to proxy runtime reference to bindgings.
Reviewed By: sahrens
Differential Revision: D12837225
fbshipit-source-id: 98edc33d6a3358e6c2905f2f03ce0004a9ca0503
Summary: Now we use RuntimeExecutor instead of MessageQueue; that's more clear and remove a dependency from Bridge.
Reviewed By: sahrens
Differential Revision: D12837226
fbshipit-source-id: 0ea3782ce2f49c7f3a91425880863e3b3ea37712
Summary:
It works great on iOS, and mostly works on Android, and is now OTA'able as part of the screen config! Haven't done template view yet. One remaining issue:
Layout is borked on Android. I'm guessing the issue has to do with the timing of setting the constraints in `updateRootLayoutSpecs` and calling `mBinding.startSurface` which actually builds the shadow tree. If I try to call `updateRootLayoutSpecs` earlier, it just crashes immediately. Here's the layout it spits out, which clearly has -440 for the x of 420006, which is the RCTText component, causing it to get cut off on the left of the screen:
```
updateLayoutMountItem for reactTag: 420006 x: -440, y: -13, width: 931, height: 78
updateLayoutMountItem for reactTag: 420010 x: 26, y: 79, width: 0, height: 1651
updateLayoutMountItem for reactTag: 420012 x: 0, y: 26, width: 0, height: 158
updateLayoutMountItem for reactTag: 420016 x: 0, y: 210, width: 454, height: 454
updateLayoutMountItem for reactTag: 420018 x: 454, y: 210, width: 455, height: 454
updateLayoutMountItem for reactTag: 420022 x: 0, y: 690, width: 454, height: 454
updateLayoutMountItem for reactTag: 420024 x: 454, y: 690, width: 455, height: 454
updateLayoutMountItem for reactTag: 420028 x: 0, y: 1171, width: 454, height: 454
updateLayoutMountItem for reactTag: 420030 x: 454, y: 1171, width: 455, height: 454
updateLayoutMountItem for reactTag: 420032 x: 0, y: 1651, width: 0, height: 0
```
Reviewed By: mdvacca
Differential Revision: D12813192
fbshipit-source-id: 450d646af4883ff25184141721351da67b091b7c
Summary:
This diff introduces a new integration concept (called RuntimeExecutor) which consolidates `Runtime` and the dispatching mechanism.
As simple as that:
`using RuntimeExecutor = std::function<void(std::function<void(facebook::jsi::Runtime &runtime)> &&callback)>;`
Reviewed By: fkgozali
Differential Revision: D12816746
fbshipit-source-id: 9e27ef16b98af861d494fe50c7e50bd0536e6aaf
Summary:
change RCTCxxBridge to use JSIExecutorFactory+JSCRuntime
instead of JSCExecutorFactory. Also remove JSC usage from RN in other
files. This allows deleting files, too, which is done further down
the stack.
Reviewed By: RSNara
Differential Revision: D9369111
fbshipit-source-id: 67ef3146e3abe5244b6cf3249a0268598f91b277
Summary: The code fragment `super.accessibilityLabel` always meant "use stored value which came from Props". But after we override the implementation of this getter in the base class, this starts working differently (wrong). This change basically reverts that to original intent.
Reviewed By: sahrens
Differential Revision: D10350597
fbshipit-source-id: 913951eb08c4ede76fc0e9be76b48d86599bcc62
Summary: Empty string in AccessibilityProps basically means same as `nil` in iOS Accessibility API.
Reviewed By: sahrens
Differential Revision: D10350596
fbshipit-source-id: fad9cdc914388c72e1b8261b27f14cbfa9a037db
Summary: We didn't have support for them... and now we have it.
Reviewed By: sahrens
Differential Revision: D10280430
fbshipit-source-id: 7275d4617ed3994366f673a17c24b823293d7092
Summary:
This diff fixes previously broken custom border rendering.
We need a dedicated CALayer for border bitmap in order to fully support all UIView capabilities in case if some subclass uses that. Otherwise, any call of `drawRect:` method can override any content which is stored inside `contents` property of CALayer.
Q&A:
How does it work in current RN? - It does not. All `drawRect:` methods in RCTView subclasses are dysfunctional.
How does text view work in current RN? - RCTTextView does not inherit RCTView, so it does not have this problem.
How does text view support custom borders in current RN then? - It does not.
Reviewed By: PeteTheHeat
Differential Revision: D10228805
fbshipit-source-id: 22bc31f41ab1914a97f3a5981cd1b24ebca725cd
Summary:
Size constraints are essential part of the running Surface, decoupling them from starting process means that we will have to perform additional commit later.
This and previous couple diffs fix a problem with initial zero size of the surface and following visible "jumpy" relayout.
Reviewed By: sahrens
Differential Revision: D10174280
fbshipit-source-id: 0ec48692cb814fd46cc3a1d044c5eb8ab9ecb031
Summary:
New `ShadowTree::synchronize` method allows to perform operations on ShadowTree without a risk of an unsuccessful commit. To make it happen, the `commitMutex_` is now recursive and `synchronize` acquires it before calling the callback.
Using that we finally can implement reliable `constraintLayout`.
Reviewed By: mdvacca
Differential Revision: D10174281
fbshipit-source-id: 9864ebb5343d40e2da205272a834710f0ab730db
Summary:
Setting the right expectations: setting layout constraints might fail. Nothing really changed.
Implementing a reliable `constraintLayout` which locks instead of returning immediately requires some additional work and new/additional API.
Reviewed By: mdvacca
Differential Revision: D10159457
fbshipit-source-id: bb23c7de105629ef086ae0b04667ff32c6ffb81d
Summary: That's why we need the previous three diffs. Synchonous executor deadlocks if the beat is missing.
Reviewed By: sahrens
Differential Revision: D10081501
fbshipit-source-id: 9986d0a1844e642048b6f37a1fcb5f623a267663
Summary:
Don't ask.
Really, all those descriptions from official docs like below are useless:
* `__bridge_transfer` Moves a Core Foundation pointer to Objective-C with transfer of the ownership to ARC.
* `__bridge` Transfers a pointer between Objective-C and Core Foundation with no transfer of ownership
All that is totally confusing and useless. At the end of the day, we only have to think about which additional `CFRetain` and `CFRelease` ARC will add or will not add for our pointers.
So, following official docs recommendation, we would like to add `__bridge_transfer` because of course, we do want to ARC managing the variable after we introduced it to ARC here. But we also want to have shared ownership of this. That's the key. If we use `__bridge_transfer` ARC will assume that this variable already retained once (because it exists) and will call CFRelease at the end of the scope. Right before that when we pass this variable down to call stack ARC will retain and then manage the variable according to the rest of the code. But still, from this point, we will have zero-balanced reference counter; the owning by `shared_ptr` bump is already compensated with `CFRelease` at the end of the scope. As soon as the rest of the code release the object, it will be incorrectly deallocated.
So, instead of using `__bridge_transfer` we have to use `__bridge`. That will indicate that *in this block* ARC does not manage the reference counter of the variable (which is kinda true because having `shared_ptr` inside the block already retains that) and will not add `CFRelease` at the end of the block.
Reviewed By: mdvacca
Differential Revision: D10054241
fbshipit-source-id: 6e82c5270fe5d53f1ed68e167b94f70dc4367a9f
Summary: Apparently, after updating CALayer props we have to request redrowing on top of it manually.
Reviewed By: mdvacca
Differential Revision: D10053340
fbshipit-source-id: f87311399bab809c9e13a3076f526bbe3f7f3fb4
Summary:
Instead of wrapping all public methods into The controller you requested could not be found. blocks/mutexes, RCTSurfacePresenter utilizes a different thread-safety pattern where all instance variable are granularly thread-safe.
The names of all internal methods were prefixed by '_'.
Reviewed By: mdvacca
Differential Revision: D10033407
fbshipit-source-id: 97fd2879c879dd9ef8d9ece24e25af00f749a871
Summary:
There is no need to make JS calls to start or stop React Native apps; Scheduler does it automatically. Yay!
With this change (because we have to change Scheduler API) we are starting slow process migrating away from using term `reactRootTag` when we refer to running a ReactNative app (aka Surface). We will use `surfaceId` instead. We plan to slowly and gracefully retire `reactTag` term entity replacing it with several appropriate entities specific for particular usage, e.g. `viewId` (some id which makes sense for mounting), `surfaceId`, `nodeId` (unique id representing nodes which were cloned from the original one), or `eventTarget`.
Reviewed By: mdvacca
Differential Revision: D9999336
fbshipit-source-id: bbc7303c195b070b8c235c9ca35546d1dc693e98
Summary: Besides that it's more simple and straight-forward now, we need that to always instantiate Scheduler with a context full of fresh valid objects derived from the new instance of the bridge.
Reviewed By: mdvacca
Differential Revision: D9995780
fbshipit-source-id: 534a314152d93562b08dd7857962f174b0d06886
Summary:
The original design of RCTSurface implied that the Surface starts on initialization and stops on deallocation. Recently I realized that this not sufficient API in some cases when the application uses ARC with autorelease pools (that can postpone object deallocations, which is highly undesirable).
And that's simply handy to have those methods sometimes.
Reviewed By: mdvacca
Differential Revision: D9982356
fbshipit-source-id: baa3bd24804d3708606ebd00b8795f2d5c9d4de9
Summary: We call this method in a constructor before the actual object is beeing constructed, so it's incorrect; it should be class method.
Reviewed By: mdvacca
Differential Revision: D9931315
fbshipit-source-id: 304ba8e2354f3f408cfa2bf1729266525a08f951
Summary: All integration with Bridge was removed from RCTFabricSurface, now it's Surface's responsibility to start and stop JS app and register the ShadowTree in the Scheduler.
Reviewed By: mdvacca
Differential Revision: D9931317
fbshipit-source-id: 55a682f0afb1c542a904e1a8570029e4690967cc
Summary: Now we have a way to enumerate surfaces stored in RCTSurfaceRegistry.
Reviewed By: mdvacca
Differential Revision: D9931316
fbshipit-source-id: 6b886c4b52cebddf15fef0239190fb2730d682cc
Summary:
Several reasons:
* We are fulfilling a promise that RCTScheduler is just a very thin interop proxy between C++ and Objective-C;
* We have to pass all parameters down to Scheduler anyway, so instead of creating all of them separately and then passing one-by-one, we consolidate them into Context created where we have all those values.
In the future, we probably will move it to some dedicated place.
Reviewed By: mdvacca
Differential Revision: D9884892
fbshipit-source-id: f1d5744e4044bc4bdfe53ec9a97ee61dcf0c60c2
Summary:
To dispatch events synchonously on the main thread we still have to block the Message Queue.
We actually dispatch events on the Message Queue while the main thread is waiting.
Reviewed By: mdvacca
Differential Revision: D9799853
fbshipit-source-id: 8033be36f27371ad2f1dc7210d564fbca1174910
Summary: This change drops the year from the copyright headers and the LICENSE file.
Reviewed By: yungsters
Differential Revision: D9727774
fbshipit-source-id: df4fc1e4390733fe774b1a160dd41b4a3d83302a
Summary:
* Now all `RCTViewComponentView` subclasses are required to set `_props` instance variable in the constructor with a default value;
* `RCTViewComponentView`'s `_props` instance variable now has `ShadredViewProps` type (that enforced by `static_assert` in `ConcreteViewShadowNode` template);
* New we use `static_pointer_cast` instead of `dynamic_pointer_cast` for casting props.
Reviewed By: mdvacca
Differential Revision: D9734199
fbshipit-source-id: b0a0939c936f8b5b540fa5fa1e4a2f1037346fc5
Summary:
When text nodes are nested, as below, `onPress` handlers need to be correctly invoked:
```
render() {
return (
<Text onPress={() => console.warn('hi')}>
hi
<Text onPress={() => console.warn('ramanpreet')}>ramanpreet</Text>
</Text>
);
}
```
In the above example, clicking on "hi" should warn "hi", and clicking on "ramanpreet" should warn "ramanpreet". This diff implements that behaviour.
Reviewed By: shergin
Differential Revision: D9696905
fbshipit-source-id: 2daf24e76c3b3c37aa36cd1540e54876a367faf7
Summary:
This diff includes two changes:
1. Even though `RCTViewComponentView` conforms to `RCTTouchableComponentViewProtocol`, it never implemented this protocol. This diff makes the correction.
2. Make the necessary changes to `RCTSurfaceTouchHandler` required by the changes in `RCTTouchableComponentViewProtocol`. This means modifying the `ActiveTouch` struct to hold `SharedTouchEventEmitter`s. It also means passing in the touch point to the `componentView`'s `touchEventEmitterAtPoint` method.
Reviewed By: shergin
Differential Revision: D9696909
fbshipit-source-id: 3d4a833f7dbae6d0238a0807eb2220250ccbae3d
