Summary: iOS TurboModules work in OSS. I pulled out JSCallInvoker from `React-turbomodule-core`, as a part of D15055511. In this diff, I'm adding it back in.
Reviewed By: mdvacca
Differential Revision: D15195007
fbshipit-source-id: 64ee294a68c0a63ba400b8a829864c6619d3311a
Summary:
`CatalystInstanceImpl.cpp` now depends on `JSCallInvoker` and `JavaJSCallInvokerHolder`. Therefore, we need to correctly adjust the OSS builds to include these dependencies into the `libreactnativejni.so` file.
I made `ReactCommon/turbomodule/jscallinvoker` a static library `libjscallinvoker.a`. I then made `ReactAndroid/src/main/jni/react/jni`'s `libreactnativejni.so` depend on that static library. Also, because the Android NDK build system doesn't support header namespaces, I had to use the filesystem to simulate them. This is why all the `.cpp` and `.h` files for `JSCallInvoker` were moved to a `jsireact` folder.
Reviewed By: mdvacca
Differential Revision: D15194821
fbshipit-source-id: 0cee682e41db53d0619f56ad017d5882f6d554aa
Summary:
`TurboModuleManagerDelegate` is an abstract base class with the following API:
```
public TurboModule getModule(String name, ReactApplicationContext reactApplicationContext);
public CxxModuleWrapper getLegacyCxxModule(String name, ReactApplicationContext reactApplicationContext);
```
```
std::shared_ptr<TurboModule> getTurboModule(std::string name, jni::global_ref<JTurboModule> turboModule, std::shared_ptr<JSCallInvoker> jsInvoker) override;
std::shared_ptr<TurboModule> getTurboModule(std::string name, std::shared_ptr<JSCallInvoker> jsInvoker) override;
```
On the C++ side, when asked to provide a TurboModule with name `name`:
1. First, is this a CxxModule? If so:
1. Create the CxxModule and return
2. Otherwise:
1. Somehow get a Java instance of the TurboModule
2. If this Java object represents a CxxModule, like `FbReactLibSodiumModule`:
1. Grab the C++ part of this object, and wrap it in a `TurboCxxModule.cpp` and return.
3. Otherwise:
1. Wrap the Java object in a C++ HostObject and return.
This pseudocode demonstrates how we'd use `TurboModuleManagerDelegate` to implement `__turboModuleProxy`.
```
__turboModuleProxy(name, jsCallInvoker):
let cxxModule = TurboModuleManagerDelegate::getTurboModule(name, jsCallInvoker)
if (!cxxModule) {
return cxxModule;
}
// JNI Call that forwards to TurboModuleManagerDelegate.getLegacyCxxModule
let javaCxxModule : CxxModuleWrapper = TurboModuleManager.getLegacyCxxModule(name)
if (!javaCxxModule) {
return std::shared_ptr<react::TurboCxxModule>(javaCxxModule.getModule())
}
// JNI Call that forwards to TurboModuleManagerDelegate.getModule
let javaModule : TurboModule = TurboModuleManager.getModule(name)
if (!javaCxxModule) {
return TurboModuleManagerDelegate::getTurboModule(name, javaModule, jsCallInvoker)
}
return null
```
Reviewed By: mdvacca
Differential Revision: D15111335
fbshipit-source-id: c7b0aeda0e4565e3a2729e7f9604775782b6f893
Summary:
JSCallInvoker requires a `std::weak_ptr<Instance>` to create. In our C++, `CatalystInstance` is responsible for creating this `Instance` object. This `CatalystInstance` C++ initialization is separate from the `TurboModuleManager` C++ initialization. Therefore, in this diff, I made `CatalystInstance` responsible for creating the `JSCallInvoker`. It then exposes the `JSCallInvoker` using a hybrid class called `JSCallInvokerHolder`, which contains a `std::shared_ptr<JSCallInvoker>` member variable. Using `CatalystInstance.getJSCallInvokerHolder()` in TurboModuleManager.java, we get a handle to this hybrid container. Then, we pass it this hybrid object to `TurboModuleManager::initHybrid`, which retrieves the `std::shared_ptr<JSCallInvoker>` from the `JavaJSCallInvokerHandler`.
There were a few cyclic dependencies, so I had to break down the buck targets:
- `CatalystInstanceImpl.java` depends on `JSCallInvokerHolderImpl.java`, and `TurboModuleManager.java` depends on classes that are packaged with `CatalystInstanceImpl.java`. So, I had to put `JSCallInvokerHolderImpl.java` in its own buck target.
- `CatalystInstance.cpp` depends on `JavaJSCallInvokerHolder.cpp`, and `TurboModuleManager.cpp` depends on classes that are build with `CatalystInstance.cpp`. So, I had to put `JavaJSCallInvokerHolder.cpp` in its own buck target. To make things simpler, I also moved `JSCallInvoker.{cpp,h}` files into the same buck target as `JavaJSCallInvokerHolder.{cpp,h}`.
I think these steps should be enough to create the TurboModuleManager without needing a bridge:
1. Make `JSCallInvoker` an abstract base class.
2. On Android, create another derived class of `JSCallInvoker` that doesn't depend on Instance.
3. Create `JavaJSCallInvokerHolder` using an instance of this new class somewhere in C++.
4. Pass this instance of `JavaJSCallInvokerHolder` to Java and use it to create/instatiate `TurboModuleManager`.
Regarding steps 1 and 2, we can also make JSCallInvoker accept a lambda.
Reviewed By: mdvacca
Differential Revision: D15055511
fbshipit-source-id: 0ad72a86599819ec35d421dbee7e140959a26ab6
Summary: Diffing algorithm uses a small map for every layer of shadow tree; that's a lot of maps. Luckily, it does not need a complex feature-full map (which is not free to allocate and use), it needs a tiny map with a dozen values. Why do we need to pay for what we don't use? This diff introduces a trivial map optimized for constraints that we have here. (See more details in the code.)
Reviewed By: mdvacca
Differential Revision: D15200495
fbshipit-source-id: d859b68b9543253840b403e7430f945a0b76d87b
Summary:
This is a small micro-optimization in Diffing algorithm.
Seems we don't need to store full ShadowView objects in `insertedPairs` map, we can store only pointers to them. That can save memory and CPU cycles because we will not need to store full objects and copy shared pointers (which is somewhat expensive).
Reviewed By: mdvacca
Differential Revision: D15200498
fbshipit-source-id: 2a268c3ee80755555bff3317e10e679be1cf9830
Summary: Diffing is already pretty fast, but using move semantic should make it even faster. ShadowViews have shared pointers, so moving them can save us atomic counter bumps.
Reviewed By: mdvacca
Differential Revision: D15200496
fbshipit-source-id: 6fb0eb79e07cd6ae9b3100713497c634f306bc18
Summary: Convert FabricUIManager.measure params to floats. Currently we convert parameters to ints across the JNI boundary, and then back to floats several times in Java. This is unnecessary and actually makes measurements trickier. The new implementation uses floats across the JNI boundary and uses Float.POSITIVE_INFINITY to represent unconstrained values, which is consistent with Fabric C++ as well.
Reviewed By: shergin, mdvacca
Differential Revision: D15176108
fbshipit-source-id: cf849b3773007637f059279460163872f300a4aa
Summary: Looks like FBJNI exports a C Macro that does exactly what `throwIfJNIReportsPendingException` does. Therefore, I'm replacing `throwIfJNIReportsPendingException` with calls to `FACEBOOK_JNI_THROW_PENDING_EXCEPTION()`.
Reviewed By: mdvacca
Differential Revision: D15174820
fbshipit-source-id: 9dfb519352cbd5f37527675323cbabad05e31d4a
Summary:
`jclass` in `JNI` is just a regular local reference. Therefore, it's unsafe to keep a static reference to it. Link: http://journals.ecs.soton.ac.uk/java/tutorial/native1.1/implementing/refs.html.
This bug made it so that when you clicked on `getConstants` twice in the TurboModule playground, the app would crash.
Reviewed By: mdvacca
Differential Revision: D15174821
fbshipit-source-id: 13b2b8726473acc9b07306558044d26bed0db92d
Summary: Previously, we'd override the `TurboModule::get` method inside the `JavaTurboModule` class to return a special `jsi::Function` in the case that the property being accessed was "getConstants". We really don't need to do this because we can simply special-case the invocation of the `getConstants` method inside the `JavaTurboModule::invokeJavaMethod` method.
Reviewed By: mdvacca
Differential Revision: D15174822
fbshipit-source-id: 0ee705be841757d3870c908da911c3872b977a9f
Summary: We conducted an experiment with different measure cache sizes. This has now been deallocatedi (D15183473). Remove the necessary APIs.
Reviewed By: SidharthGuglani
Differential Revision: D15183486
fbshipit-source-id: a38fa5a3ab0321c2521265f7d1cd6b495efd76cf
Summary:
@public
`YGConfig::YGConfig(YGConfig*)` was not initializing the same fields as the default constructors.
Here, we make the default constructor delegate to the more specialized one to remove duplication.
Reviewed By: SidharthGuglani
Differential Revision: D15164599
fbshipit-source-id: 27247709091b7664386057d09ac67d481877871f
Summary:
* invokeMethod() ends up not useful because each platform has its own way of invoking the platform methods
* invalidate() is not necessary because there's already the destructor of each C++ class
Reviewed By: mdvacca
Differential Revision: D15187833
fbshipit-source-id: 9478ed1e6288da30c67179e03a7bc7da6043280b
Summary:
@public
We want to enable tooling, instrumentation, and statistics within Yoga without coupling these functionalities to our core code.
This commit introduces the foundations of a simple, global event system.
For the time being, we will only support a single subscriber. Should we require more than one, we can add support for it later.
Reviewed By: SidharthGuglani
Differential Revision: D15153678
fbshipit-source-id: 7d96f4c8def646a6a1b3908f946e7f81a6dba9c3
Summary:
Previously we computed the list of nodes that need to be notified about layout changes using a list of mutation instructions. That was fine, but that's not really compatible with some other changes that I plan to make, so I decided to change it (make it better).
Besides the better design (debatable; fewer dependencies to unrelated moving pieces), here is why I believe the new way is more performant:
* The new approach has no `dynamic_casts`, whereas the previous has tons of them (two per a mutation). If a `dynamic_cast` takes 10 ns, for 500 nodes it can take up to 5ms only for casts. (Non-scientific assumption.)
* After removing dependency to mutation instruction, we can enable flattening for views which have `onLayout` event.
Reviewed By: mdvacca
Differential Revision: D15110725
fbshipit-source-id: 31a657ccfd02441734ad1d71a833653223163289
Summary: Instrumentation tests are expensive and flaky. Luckly this one does not need to be instrumentation one.
Reviewed By: mdvacca
Differential Revision: D15158985
fbshipit-source-id: 3c88e5a0d82db2cd00f5866c3f9956409cc8fc7f
Summary:
@public
Makes bitfield getters/setters part of the bitfield ref template.
Since we introduced the tracking bit as template parameter in D14933022, every bitfield ref is an individual class anyway, and having function pointers doesn’t potentially lead to less code generation anyway.
Furthermore, this change can (in the absence of tracking bits) avoid less specialized templates dealing with refs, and to dynamic dispatch as a consequence.
Reviewed By: SidharthGuglani
Differential Revision: D15085495
fbshipit-source-id: 0dd70fa05e9d43a29e38a619cddb642c9ca3f7ab
Summary:
@public
In order to optimise property storage, we have to know how style properties are used in our apps.
Here, we add a bitmask that allows us to track which properties are set explicitely, and use that for our analysis.
Reviewed By: SidharthGuglani
Differential Revision: D14933022
fbshipit-source-id: 1ab8af562b14baba1d02057e527aa36d5c9a7823
Summary:
@public
The extra overload of `updateStyle` introduced in D15078961 can also handle `BitfieldRef`.
That means that we can remove the more specific implementation previously introduced for `BitfieldRef`
Reviewed By: SidharthGuglani
Differential Revision: D15081069
fbshipit-source-id: 98f1f3478627974c5273c85d268ca07350f303d7
Summary:
@public
Change style property accessors to return `Ref` instances instead of references to `CompactValue`.
This will allow to track assignments to properties later on, e.g. for instrumentation or dynamic property storage.
Reviewed By: SidharthGuglani
Differential Revision: D15078961
fbshipit-source-id: 259f05f7d30f093c04bf333c5bd4fb3601b8e933
Summary: This diff exposes the Legacy method UIManager.measureInWindow as part of Fabric
Reviewed By: shergin
Differential Revision: D15110795
fbshipit-source-id: 2b4bf47452f7272fd3edc4e580e65ae7ec2f2622
Summary:
Different frameworks use different kinds of floats, optional floats, and floats with assigned unit names. All those approaches use different ways to represent undefined and empty values. To deal with it we need to have some helper functions.
So, this diff changes some ways that we convert some corner values (like NaN and empty value). That change is motivated by recent personal discoveries in this field that shifted my vision on that. E.g. ComponentKit does not use `CGFloatMax` value as `Infinite` value. UIKit is also (surprisingly to me) okay with using `Infitite` instead of `CGFloatMax`. And, in general, seems using really conceptually appropriate values (instead of UIKit-inspired ones) it's the right thing to do.
Reviewed By: mdvacca
Differential Revision: D15155189
fbshipit-source-id: 33e15141f1ca3efb400a7160811224335de34ba1
Summary: `kFloatUndefined` means "no value here", but in this particular case, we have to have `Infinity` value that represents maximum available space.
Reviewed By: mdvacca
Differential Revision: D15155190
fbshipit-source-id: d2de20681ad04da7444331eff44b93d2bd0200e3
Summary:
We don't need to have those constants because this functionality is available in STL via `std::numeric_limits<YourParticularFloatType>::infinity()` (or `::min()` and `::max()`).
At the same time usage of `kFloatMax` was replaced with `Infinity` (which is a different value). Using `max` instead of `Infinity` was an attempt to mimic iOS/UIKit model where `Infinity` and `NaN` values usually are not being used. However, now this does not seem like a good idea. This concept is not used anywhere else (even in CK which is totally incompatible with it) and de-facto in RN we use it only in few places. So, let's use Infinity in places where it's logically appropriate.
Reviewed By: mdvacca
Differential Revision: D15155191
fbshipit-source-id: 4d24350c7540cec074a8b040d7c13f257aa812e7
Summary: This diff exposes the Legacy method UIManager.measureLayout as part of Fabric
Reviewed By: shergin
Differential Revision: D15103117
fbshipit-source-id: 4cf7ab3776f6a541cf0d6a00789420a0bb008fae
Summary: This adds support for specifying the exact selector name for each exposed ObjC method. This allows us to avoid dynamic method lookup every time a method is called from JS.
Reviewed By: RSNara
Differential Revision: D15141611
fbshipit-source-id: ed2820782ab013369e4e1f22dbce31d9838a17bb
Summary: Apparently we can/should not have in RCTConversions because it creates unnecessary dependency to core iOS module.
Reviewed By: mdvacca
Differential Revision: D15055325
fbshipit-source-id: 507f5a40c03b5c261967de4504297d31ecd02783
Summary: Sometimes we don't know for sure if `ContextContainer` has a value or not (and that's perfectly legit use case). In those cases now we can use `findInstance` method that returns an optional intead of throwing an exeption.
Reviewed By: JoshuaGross
Differential Revision: D15039137
fbshipit-source-id: 95ba8cc7b76e37d1bd17e18c0098e56350ff3fef
Summary: It turns out that just only props is not enought to build an initial state value in some cases for some component. Seems we need at least `surfaceId` and `eventEmitter` in some cases (which seems totally reasonable). So, seems using the whole `ShadowNodeFragment` for that purpose is a good choise.
Reviewed By: mdvacca
Differential Revision: D15039135
fbshipit-source-id: d9a5f47f971ccf6cdb2f888bd31f7948b37b67ef
Summary:
`ShadowNodeFragment` is very cheap by design because it does not own stuff it contains, so it's great. But... sometimes we need to own the stuff (e.g. to pass it on the other thread), in those cases we can use `ShadowNodeFragment::Value` now.
`ShadowNodeFragment::Value` cannot be used alone, it needs to be constructed from `ShadowNodeFragment` and then used as opaque object and then it can be converted to ``ShadowNodeFragment`.
We will need it soon.
Reviewed By: mdvacca
Differential Revision: D15039136
fbshipit-source-id: d40875cac05f4088358d8d418007d17df9ff14f4
Summary:
Trivial.
We are replacing rootTag with surfaceId according to the plan describing here: https://fb.workplace.com/groups/rn.fabric/permalink/1374002366064519/
Reviewed By: JoshuaGross, mdvacca
Differential Revision: D15039134
fbshipit-source-id: ec8c3044f9f3f23939488bc01c66e9b653e651dd
Summary: QE expired a while ago. Remove the experiment code.
Reviewed By: fkgozali
Differential Revision: D15133830
fbshipit-source-id: 562c3998cc7860497eefa9899dfb6bfcee4fe210
Summary:
@public
Adds `YGStyle::ValueRepr` to make code depending on the actual type easier to write.
So far, we have treated `yoga::detail::CompactValue` as an implementation detail, and that’s what it’s supposed to stay.
React Native Fabric has one value conversion overload that depends on that type, though, and used `decltype(YGStyle{}.margin()[0])` until now.
That’s problematic for two reasons:
- we want to constrain the parameter of `operator[](...)` to enum types, making the `0` unsuitable
- we want to return the non-const overload of the operator to return a custom `Ref` type, which is not the type needed by Fabric.
Making the storage type explicit allows to write more forward-compatible code.
Reviewed By: SidharthGuglani
Differential Revision: D15078960
fbshipit-source-id: 932c27ef2f2cdc6ce965b79894268170f0ccdce5
Summary:
@public
Some `YGNode*` passed as `const YGNode*`, some const refs to sub-objects introduced.
This helps selecting the desired methods in more places, i.e. `const` overloads of accessors on `YGStyle`.
Reviewed By: SidharthGuglani
Differential Revision: D15078963
fbshipit-source-id: 5013721d6edcc68f42f4504f5c331da647a294bd
Summary:
@public
Having binary operators as member functions has disadvantages:
- the left hand side cannot be converted to `YGFloatOptional` implicitly (which we need for `YGStyle` refs)
- Operators are not necessarily commutative.
By moving these operators into free functions, and adding overloads for both variants if one operand is `float`, we get these properties.
Reviewed By: SidharthGuglani
Differential Revision: D15078962
fbshipit-source-id: 2e228a2ef90a8083c91788caa9eedfd4d140677f
Summary:
After upgrading RN from 0.57 to 0.59.4 we've received a lot of crash reports like `Exception in HostObject::get: <unknown>` with no clue what native module caused the crash. This commit adds native module name on crash in this situations. Related to https://github.com/facebook/react-native/issues/24607.
[General] [Added] - Report native module name on crash when native module has failed to load
Pull Request resolved: https://github.com/facebook/react-native/pull/24633
Differential Revision: D15120225
Pulled By: cpojer
fbshipit-source-id: cf8e3e5953548a58f1d010eb70343da5ee946ae8
Summary:
@public
Takes a const reference to the style of the printed node once, instead of using repeated calls to `node->getStyle()`.
Makes the code a bit shorter, and ensures that we are operating on `const YGStyle&`, which helps selecting the correct methods further up the stack.
Reviewed By: SidharthGuglani
Differential Revision: D14999094
fbshipit-source-id: 814f06b7e3179ac8cfb43d79fbec48ee4115d6e3
Summary: In `ObjCTurboModule::getArgumentTypeName`, I replaced all instances of `':'` with `''` to transform the selector into a TurboModule methodName. This transformation works when the method has 0 or 1 argument, however, it breaks when the method has more than 1 argument. In all cases, we just want to get the substring until the first `':'`.
Reviewed By: fkgozali
Differential Revision: D15056937
fbshipit-source-id: 3a7dce1ce62ca9758e46c0af951b269166d68454
Summary:
Trivial.
Apparently, `DEBUG` is non-standard feature and using `assert` with `DEBUG` is practically asking for bugs. So, if your `assert` relies on some variable which is only defined when `DEBUG` is set, it's easy to get invalid code because NDEBUG and DEBUG can be unsync.
So, we have to use clunky double negative `#ifndef NDEBUG` everywhere where we used DEBUG.
Reviewed By: JoshuaGross
Differential Revision: D15031328
fbshipit-source-id: 036f573e68925741ca46384261885766c87db1e3
Summary:
@public
Introduces `YGNodeConstRef` as `const YGNode*`, i.e. a pointer to a constant `YGNode`.
We also use it for all style getters, which will avoid casts to `const YGNode*` in diffs up the stack.
We should use this pointer type for all functions that do not modify the underlying node.
Reviewed By: SidharthGuglani
Differential Revision: D14999095
fbshipit-source-id: 61cc53bb35e787a12ae12e70438d84c0a4983752
Summary:
@public
In order to encapsulate property access on `YGStyle`, as a first measure we wrap all fields with accessors.
This will e.g. enable dynamic property storage and instrumentation in the future.
All accessors have a `const` version that allows direct access via `const&`. For mutation, bit fields are wrapped with a custom reference object.
This style allows for the least amount of changes in client code. Property access simply needs appended parens, eg `style.direction` becomes `style.direction`.
Reviewed By: shergin
Differential Revision: D14999096
fbshipit-source-id: fbf29f7ddab520513d4618f5e70094c4f6330b30
Summary:
@public
Remove unnecessary `const` and `extern` specifiers from `Yoga.h`.
- Function declarations are `extern` by default
- The removed `const` specifiers for pass-by-valye parameters are only meaningful for the *definition* of functions, not for the declaration.
In this specific case, I found `const YGNodeRef` particularly confusing, as it is a `typedef` for a pointer type. `const` does not refer to the pointed-to object, but to the parameter itself, i.e. `const YGNodeRef` is `YGNode * const`, and not `const YGNode *`.
Reviewed By: SidharthGuglani
Differential Revision: D14999097
fbshipit-source-id: 8350870cb67f4a34722f796c4f4a2fc7dde41b99
Summary: `jClassName_` is unnecessary because you can use `JNIEnv::GetObjectClass` to get the TurboModule's Java class.
Reviewed By: fkgozali
Differential Revision: D14937480
fbshipit-source-id: 2c1c9be53217331152270dbac3d13f372a2ed818
Summary:
UAs must adjust border radius values to fit a content box:
>>> Corner curves must not overlap: When the sum of any two adjacent border radii exceeds the size of the border box, UAs must proportionally reduce the used values of all border radii until none of them overlap.
This diff implements that.
Reviewed By: mdvacca
Differential Revision: D15028325
fbshipit-source-id: 368232ffa2fa0409d13759bbbe7fe10f8474c400
Summary: Additional check verifies that requested type matches the type of a stored value.
Reviewed By: mdvacca
Differential Revision: D14944336
fbshipit-source-id: 6d3a1654d9b9a64ced83f553236093f02f2c97c7
Ramanpreet Nara
Valentin Shergin
Joshua Gross
David Aurelio
Kevin Gozali
David Vacca
Dmitry Dushkin
Estevão Lucas