Summary:
#changelog: [internal]
When I built ThreadStorage I didn't know about existence of `thread_local` keyword. Because it achieves the same goal, using built in c++ features is preferred over building our own.
Reviewed By: JoshuaGross, shergin
Differential Revision: D24380680
fbshipit-source-id: e961fc34c6d3f085fc9b918b20bb4827de0d5624
Summary:
changelog: [internal]
Prevents 2 type converions:
1. int <-> size_t
2. int <-> int32_t
# Why is using size_t better when working with indexes.
## 1. Type conversion isn't for free.
Take this example
```
size_t calculate(int number) {
return number + 1;
}
```
It generates following assembly (generated with armv8-a clang 10.0.0):
```
calculate(int): // calculate(int)
sub sp, sp, #16 // =16
str w0, [sp, #12]
ldr w8, [sp, #12]
add w9, w8, #1 // =1
mov w8, w9
sxtw x0, w8
add sp, sp, #16 // =16
ret
```
That's 9 instructions.
If we get rid of type conversion:
```
size_t calculate(size_t number) {
return number + 1;
}
```
Assembly (generated with armv8-a clang 10.0.0):
```
calculate(unsigned long): // calculate(unsigned long)
sub sp, sp, #16 // =16
str x0, [sp, #8]
ldr x8, [sp, #8]
add x0, x8, #1 // =1
add sp, sp, #16 // =16
ret
```
Compiler now produces only 7 instructions.
## Semantics
When using int for indexing, the type doesn't say much. By using `size_t`, just by looking at the type, it gives the reader more information about where it is coming from.
Reviewed By: JoshuaGross
Differential Revision: D24332248
fbshipit-source-id: 87ef982829ec14906ed9e002ea2e875fda4a0cd8
Summary:
We will need it soon.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: sammy-SC
Differential Revision: D24290775
fbshipit-source-id: a312e537a3c3954e709a10c8792b3462b574054a
Summary:
Changelog: [internal]
Fabric uses view commands instead of setNativeProps. This diff removes what's left of setNativeProps from the core.
Reviewed By: JoshuaGross
Differential Revision: D24309999
fbshipit-source-id: 70e54f0a984f8c36f77ba2cd59f59fc6923bc832
Summary:
This diff ended up being a bit more complicated than I anticipated, since the source files in `ReactInternal` were depending on `RCTEventDispatcher`. I made the following changes:
1. Make `RCTEventDispatcher` a `protocol`, keep it in `ReactInternal`.
2. Rename the `RCTEventDispatcher` NativeModule to `RCTEventDispatcherModule`, make it conform to the `RCTEventEmitter` `protocol`, and move it to `CoreModules`.
3. Where necessary, replace categories of `RCTEventDispatcher` with functions.
Changelog:
[iOS][Added] - Make RCTEventDispatcher TurboModule-comaptible
Reviewed By: fkgozali
Differential Revision: D18439488
fbshipit-source-id: b3da15c29459fddf884519f33b0c3b8c036b5539
Summary:
There are a few places where we have JNI table ref leaks, and more places where we can aggressively clean up smart pointers immediately instead of waiting for them to be cleaned up at some later point.
In theory these smart pointers should be cleaned up immediately, but in cases where many components are being measured at once, the JNI table could grow until all measure calls are done. In extreme cases this
could cause a crash, which I want to avoid. At the very least, freeing memory more aggressively in this case can't hurt.
Changelog: [Internal]
Reviewed By: shergin
Differential Revision: D24293775
fbshipit-source-id: 159741ba955e5a6fe02caf6e65d1e4d6d4afadee
Summary:
This adds support for `<Picker style={}/>` prop for text styling. It reuses most of conversion logic from BaseText. This means that it actually supports more styles than Paper Picker supported. (Paper picker only supported ~4 styles, this supports everything that Text supports, so 10+ styles).
The only tricky thing is that Picker supports multiple ways of setting text color. Both
<Picker
itemStyle={{color: '#008BD0'}} >
<Picker.Item label="Java" value="java" />
<Picker.Item label="JavaScript" value="js" />
</Picker>
and
<Picker>
<Picker.Item label="Java" value="java" color={'#008BD0'} />
<Picker.Item label="JavaScript" value="js" />
</Picker>
technically work in Paper. I've decided to maintain this behaviour (since there's lots of product code callsites to both options).
Changelog: [iOS][Fabric] Fabric Picker support
Reviewed By: sammy-SC
Differential Revision: D23980319
fbshipit-source-id: e469a837e28af0ad97cf0e171df26ee19adff3ab
Summary:
This builds on previous diff to properly parse `<Picker.Item/>` into a cpp struct.
This fixes the dummy text and text color TODOs.
Changelog: [iOS][Fabric] Fabric Picker support
Reviewed By: sammy-SC
Differential Revision: D23964557
fbshipit-source-id: f42c6c9cf410bfc5e66ff078645b6378548481de
Summary:
This is a starting point for the handwritten Fabric Picker component. It is incomplete, and needs to be landed with the rest of the stack above it.
In general, this creates a new `ComponentView`, `ComponentDescriptor`, `ShadowNode`, `Props` and a few other boilerplate classes for Picker. A bunch of the logic in `ComponentView` was copied over from the Paper `RCTPicker` and `RCTPickerManager`.
What works in this diff:
- A `<Picker>` with items can be created in JS, and a corresponding `UIPicker` is created in native with placeholder text, default styling and the correct amount of items
What doesn't work yet (implemented in later diffs):
- Parsing items to use correct text and styling in native
- Events/commands
Reviewed By: sammy-SC
Differential Revision: D23941821
fbshipit-source-id: e049ca6004757fbd1361985644d5dbb8f53e1ce6
Summary:
## Previously
- When TurboModules system was on, we'd only log events from the TurboModules system.
- When TurboModules system was off, we'd only log events from the NativeModule system.
This ultimately gives us less data to analyze both systems in production.
## Changes in This Diff
When perf. logging is on, we'll log events from both systems. Each QPL event now include an annotation of which system the event is coming from. Concretely, this will allow us to see how much of the NativeModule system is being exercised in the TurboModule test group.
Changelog: [Internal]
Reviewed By: hramos
Differential Revision: D24232594
fbshipit-source-id: 7dff57bd74fc7ef744d3e06ff174304f25790456
Summary:
Index adjustment is tricky. Seems more reliable to adjust each immediate mutation, and then immediately adjust delayed mutations based on it, rinse and repeat.
Previously it was possible to construct examples where the UI would get into a weird state because index adjustment caused items to be inserted in the wrong location.
Changelog: [Internal]
Reviewed By: kacieb
Differential Revision: D24232926
fbshipit-source-id: f8c445213528c2d2aedacf3e0c73c5bbeb62bc3d
Summary:
The captured `instance_` will result in a call to JVM code to release the reference when the lambda is destroyed. However, Hermes may destroy the lambda in a non-JVM background thread.
Since `instance_` is a member of the class, and `this` is already captured, there is no need to capture it.
Changelog: [Internal]
Reviewed By: RSNara
Differential Revision: D24141179
fbshipit-source-id: e3b268485ed229ee933281a80e5dbe2feba55719
Summary:
Changelog: [internal]
Components can update state multiple times before the state update queue is flushed. This causes unnecessary layout/diff and mount passes. To solve this, drop stale state updates inside `stateUpdateQueue_ ` for specific `ShadowNodeFamily`.
Delivering stale status updates is redundant. Let's take SafeAreaView as an example. It schedules 5-6 state updates before `stateUpdateQueue_` is flushed. That's unnecessary work blocking JS thread. We only care about the latest state update. Same for TextInput and other components using state updates.
Reviewed By: JoshuaGross
Differential Revision: D23987707
fbshipit-source-id: 2e3f92cc93af61d78ac564aa40aef165af64b8c1
Summary:
This compiles SampleTurboModule into RNTester Android. It also adds the NativeModule playground to show case TurboModule system to RNTester examples, just like in iOS.
{F337854369}
Changelog: [Android][TurboModule] Added TurboModule example to RNTester when `USE_CODEGEN` is set
Reviewed By: hramos
Differential Revision: D24004711
fbshipit-source-id: b682dd51fa998ee2e60f8d6ffd8c39220d13a7fe
Summary:
This is the Java/JNI impl of the NativeSampleTurboModule.js, just like on iOS. The files here are supposed to be generated by the react-native-codegen, but they are checked in to the repo for easier build integration with RNTester.
Changelog: [Internal]
Reviewed By: hramos
Differential Revision: D23985746
fbshipit-source-id: 46340d778f3d964efe5b538d15ebe0f2cab04862
Summary:
The sorting function currently forms a partial ordering, not a total ordering. This can cause problems with certain sequences of immediate or conflicting mutations, leading to UI corruption or crashes.
Changelog: [Internal]
Reviewed By: shergin
Differential Revision: D24002668
fbshipit-source-id: edc9b4c1e3104897cb0c5fd6da563ec43d800494
Summary:
Remove the older implementation of image instrumentation in Fabric by removing, RCTImageInstrumentationProxy, ImageInstrumentation from ImageRequest, and trackURLImageContentDidSetForRequest from RCTImageLoaderWithAttributionProtocol.
Changelog: [RN][Fabric][Image] Remove unused Fabric image instrumentation
Reviewed By: fkgozali
Differential Revision: D23990606
fbshipit-source-id: 004d04025d031af11377a73e5bfb64b1e0449962
Summary:
Without this thing some stuff does not compile.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: fkgozali
Differential Revision: D23948622
fbshipit-source-id: f066ada183c0fd6a7b5eec542395d44bbbfe80a3
Summary:
This fixes a recently introduced crash in `colorComponentsFromColor()` (iOS implementation) caused by dereferencing a null pointer.
The fix is just a copy of a code fragment from a previous implementation.
Reviewed By: fkgozali
Differential Revision: D23944812
fbshipit-source-id: 977135dd75c4375affddfd75183e4890618ae819
Summary:
When an animation is completed or a conflicting animation is detected, force props, state, layout, etc to update.
Currently, when a final animation mutation is queued, some attributes can be updated but not others, causing unexpected visual glitches at least on Android.
Some of these are arguably component bugs, but it's easier to just flush all attributes by tricking the platforms into updating all attributes. This will also prevent us from having to track down more of these bugs, potentially.
Changelog: [Internal]
Reviewed By: sammy-SC
Differential Revision: D23886519
fbshipit-source-id: 8e5081bbe3b7843c16c0f283fa07fdec0e211aa8
Summary:
Changelog: [internal]
`LayoutableShadowNode::findNodeAtPoint` was iterating children in incorrect order and wasn't taking zIndex into accout.
Reviewed By: JoshuaGross
Differential Revision: D23814866
fbshipit-source-id: 38eee297147a5c5912304d139bb10f8b16ae2ee1
Summary:
In this diff I simplify index adjustment and add comments to rigorously describe what we're doing at each step of index adjustment.
I've also made unflattening detection more correct, robust, and slightly more efficient.
Bugs that existed before:
1) The reparenting detection that existed in the animations layer had some subtle bugs. I don't have proof that it results in any correctness issues or crashes, but I suspect it.
2) Correctness of index adjustment: there were cases where the Android mounting layer would crash because LayoutAnimations would try to remove a View at an index, but the index was wrong. This is why I sat down and diagrammed the relationships between all the bits of data we have for index readjustment - I believe this to be correct now.
3) Correctness of INSERT index adjustment: I had the insight that when we have batches of INSERTs to consecutive indices, we essentially want them to be treated as a single INSERT for adjustment purposes, so that they're all placed consecutively in the view layer. I added `ConsecutiveAdjustmentMetadata` to deal with this, and there are more comments in the code.
Changelog: [Internal]
Reviewed By: yungsters
Differential Revision: D23806163
fbshipit-source-id: cd9e94945034db8b840f2a806c6377034a91af61
Summary:
In the new Flattening differ, I experimentally verified that these two code paths are not hit (or redundant) and deleted them.
One of the branches did nothing and the other produced duplicate DELETE mutations for the same tag, that is handled elsewhere.
Changelog: [Internal]
Reviewed By: fkgozali
Differential Revision: D23806161
fbshipit-source-id: 9ad2929e2d719a7b9b34640ed35f7a696103604b
Summary:
This is to prepare for enabling TurboModule on Android. This commit compiles in all the core files (C++) into the ReactAndroid NDK build step. This doesn't yet enable TurboModule by default, just compiling in the infra, just like for iOS.
New shared libs:
* libreact_nativemodule_core.so: The TurboModule Android core
* libreact_nativemodule_manager.so: The TurboModule manager/delegate
To be compatible with `<ReactCommon/` .h include prefix, the files had to move to local `ReactCommon` subdirs.
Changelog: [Internal]
Reviewed By: sammy-SC
Differential Revision: D23805717
fbshipit-source-id: b41c392a592dd095ae003f7b2a689f4add2c37a9
Summary:
Changelog: [internal]
In D23648430 (https://github.com/facebook/react-native/commit/a315e4cd30e4b8da841f587650146a62c868f67d) I made a mistake. I prevented calling `onTextLayout` unless there are attachments in the component. It fixed the problem because I unintentionally prevented `onTextLayout` to be called. Therefore, changes from D23648430 (https://github.com/facebook/react-native/commit/a315e4cd30e4b8da841f587650146a62c868f67d) need to be reverted.
To prevent infinite loop in `onTextLayout`, ParagraphEventEmitter checks if `linesMeasurements` have changed before dispatching it to JS.
Reviewed By: shergin
Differential Revision: D23782717
fbshipit-source-id: 0e84ae4f46d79ce0cf4c7340cd32be6f562ae179
Summary:
This diff moves the code of TurboModule Core from ReactCommon/turbomodule to ReactCommon/react/nativemodule
For iOS: Pod spec name stays as "ReactCommon/turbomodule/..." for now, only the source/header location is affected. The target will be renamed/restructured closer to TurboModule rollout.
changelog: [internal] Internal
Reviewed By: RSNara
Differential Revision: D23362253
fbshipit-source-id: c2c8207578e50821c7573255d4319b9051b58a37
Summary:
Exceptions in C++ work quite differently from exceptions in other languages. To make exceptions actually work **correctly** all the code needs to be written with "exceptions in mind" (e.g., see https://www.stroustrup.com/except.pdf). In short, if the code is not "exceptions ready", throwing an exception causes memory leaks, dangling pointers, and invariant violations all over the place, which will probably cause another crashes down the road (which will be especially hard to investigate and attribute to the original issue).
Fabric Core (Layout, Props parsing, ShadowNodes management, and so on) does not use exceptions because in most (all?) the cases the exception is now recoverable. So, if a program detects some internal state invariant violation or missing some resource, *logically* it's fatal. We also don't want to pay code-size and performance tax for exception support, so that's why we don't use them. It's just not the right fit for Fabric Core.
This does not mean that exceptions don't happen though. C++ standard library can throw them... sometimes. And if our library is compiled with exceptions enabled (still the case, unfortunately), an exception can bubble to JavaScript code and losing all context down the road. And it's hard to investigate such crashes. To isolate those occasional exceptions inside C++ core we are marking all C++/JS boundaries with `noexcept` that stops the bubbling.
I hope that will give us much more informative crash reports.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: sammy-SC
Differential Revision: D23787492
fbshipit-source-id: 0822dbf36fc680c15b02b5cd0f2d87328296b642
Summary:
See additional assertion. Tests still pass, so no other change was necessary.
Changelog: [Internal]
Differential Revision: D23775553
fbshipit-source-id: 57d3191f25dd55ab4da189207f6d201a31b175e0
Summary:
This enables a new state auto repeating mechanism built-in mechanism for all state updates which we already use for CK interop. This experiment is supposed to help with T74769670 and co.
This change is gated with MC.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: JoshuaGross
Differential Revision: D23762508
fbshipit-source-id: f535513c724ace9ede570177281324eb507329c5
Summary:
Under Fabric only, we can enter an infinite layout loop where the emitted layout event oscillates between two height values that are off by a very small amount.
The cause is, in part, components that use layoutEvents to determine their dimensions: for example, using onLayout event "height" parameters to determine the height of a child. If the onLayout height changes rapidly, the child's height will change, causing another layout, ad infinitum.
This might seem like an extreme case but there are product use-cases where this is done in non-Fabric and layout stabilizes quickly. In Fabric, currently it may never stabilize.
Part of this is due to a longstanding issue that exists in Fabric and non-Fabric, that we cannot immediately fix: If in a single frame, C++ emits 100 layout events to ReactJS, ReactJS may only process 50 before committing the root. That will trigger more layout events, even though product code has only partially processed the layout events. At the next frame, the next 50 events will be processed which may again change the layout, emitting more events... etc. In most cases the tree will converge and layout values will stabilize, but in extreme cases in Fabric, it might not.
Part of this is because Fabric does not drop *stale* layout events. If 10 layout events are dispatched to the same node, it will process all 10 events in older. Non-Fabric does not have this behavior, so we're changing Fabric to drop stale events when they queue up.
Changelog: [Internal]
Reviewed By: sammy-SC
Differential Revision: D23719494
fbshipit-source-id: e44a3b3e40585b59680299db3a4efdc63cdf0de8
Summary:
LayoutAnimations: at the end of every animation, issue an update mutation - this is so that the props data on the Mounting layer/StubViewTree layer is pointer-identical to the props data on the ShadowTree.
This unblocks iOS debug mode crashes.
Changelog: [Internal]
Reviewed By: sammy-SC
Differential Revision: D23753606
fbshipit-source-id: 407e0c2746a65e6d3ee29c1cce891cd7c1013593
Summary:
iOS needs this function to be marked as static.
Changelog: [internal]
Reviewed By: shergin
Differential Revision: D23749613
fbshipit-source-id: a8c160646853450fc7d849448bdbb45e02beb964
Summary:
Currently, MountingCoordinator's RN_SHADOW_TREE_INTROSPECTION code will crash often because it assumes there is always a "new" tree to compare the old tree to. In the LayoutAnimations context this is not always the case - in fact, the majority of the time, LayoutAnimations is producing mutations for animation without a "new" tree.
Just check that the tree exists before trying to print it.
Changelog: [Internal]
Reviewed By: shergin
Differential Revision: D23747289
fbshipit-source-id: a1ba22aeae32ed8915a53bc33cdc199e8ce5128a
Summary:
In MountingCoordinator override mode (used in LayoutAnimations) we must set the start and end `diff` time when no real diff happens, otherwise we will hit an assert in telemetry later.
I also ensure that the TransactionNumber is incremented in that case.
Changelog: [Internal]
Reviewed By: shergin
Differential Revision: D23746684
fbshipit-source-id: b1fe3864e453fdba89d43cc827bd37434abf7a4d
Summary:
`adjustDelayedMutationIndicesForMutation` asserts that the mutation is either Remove or Insert. At one callsite, we weren't checking the mutation type before calling `adjustDelayedMutationIndicesForMutation`.
Changelog: [Internal]
Reviewed By: shergin
Differential Revision: D23746617
fbshipit-source-id: 6046fec2eb4821094937b1b16f40347bbc55c20e
Summary:
This change maps the three most used colors (black, white, clear) to corresponding predefined values in UIColor. This should meaningfully reduce the overall amount of allocated UIColor/CGColor objects. In my non-scientific measures, it reduces the number of CGColor objects from ~1500 to ~1000. And... it no much at least in terms of kilobytes. However, I still think it's a good idea to implement this because I hope that can remove some work from memory allocation infra and maybe enable some optimizations that UIKit hopefully does for black and white colors. (I tend to believe that this optimization exists because UIKit even has a classes called UIDeviceWhiteColor and UICachedDeviceWhiteColor.)
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: JoshuaGross
Differential Revision: D23753506
fbshipit-source-id: 46e58dc7e6b0dcab3c83d29c7257c90ffbd95246
Summary:
Finally, this diff changes the internal implementation of SharedColor to be `optional<int>`.
Initially, when we started working on the new renderer, it seemed like a good idea to allocated CGColor objects ahead of time and store them with Props. Now, this idea does not look so good, especially because:
* Having `SharedColor` as a `shared_ptr` is a quite big perf overhead for copying this thing. And the size of the object is not small.
* Having `SharedColor` as a `shared_ptr` creates huge interconnectedness among pieces of the core and rendering. E.g. improper releasing a pointer in some component view can cause a crash somewhere in the core (because both places might use the same shared `blackColor`.
On Android, we already use simple `int` as a representation of a color, and this works great. And this diff implements something very similar to Android, but a bit differently: here we use `optional<int>` instead of custom class with a single `int` field and some magic value that represents "empty value".
This approach should fix T75836417 because now we don't have allocation and deallocation when we simply assign color values.
If this solution works fine on iOS, I will do unify all implementations among all platforms.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: JoshuaGross
Differential Revision: D23753507
fbshipit-source-id: 42fd6cee6bf7b39c92c88536da06ba9e8cf4d4db
Summary:
This diff changes the implementation of `RCTCreateCGColorRefFromSharedColor` and `RCTUIColorFromSharedColor` in such a way that they don't rely on the fact that SharedColor is actually a `shared_ptr<CGColorRef>`. Instead, the methods just extract color components from SharedColor and create UIColor and CGColorRef objects on demand.
This allows us to change the implementation of SharedColor without worrying much about the rest of the system, which will do in the next diff.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: JoshuaGross
Differential Revision: D23753510
fbshipit-source-id: 340127527888776ebd5d241ed60c7e5220564013
Summary:
We don't need a shared_ptr here and without it the code will be faster and simpler.
This change is aligned with any clone-line callbacks we have in the Core which accepts a `const &` and return `shared_ptr<>`.
Changelog: [Internal] Fabric-specific internal change.
Reviewed By: JoshuaGross, sammy-SC
Differential Revision: D23725687
fbshipit-source-id: 1cd959f4273913175d342302e2f12752f0114768
Samuel Susla
Valentin Shergin
Paige Sun
Lulu Wu
Joshua Gross
Peter Argany
Ramanpreet Nara
Neil Dhar
Kevin Gozali
generatedunixname89002005287564