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
Summary:
After D19565499, the `LongLivedObjectCollection` will be cleared on the JS thread when the jsi::Runtime is deleted. This diff makes it so that we never hold strong references to `CallbackWrapper`s in our Android TurboModules infra. Therefore, we can leverage the changes in D19565499 to ensure that our `jsi::Function`s are deleted before the `jsi::Runtime`.
## Caveat
If you delete a TurboModule by itself, it's jsi::Functions that haven't been invoked won't be released. This is also the case for iOS. I plan to fix this for both iOS and Android at a later point in time.
Changelog:
[Android][Fixed] - Refactor jsi::Function cleanup in TurboModules
Reviewed By: mdvacca
Differential Revision: D19589151
fbshipit-source-id: efa3cc6c83634014159ac7500dcf6bef9c925762
Summary:
This is an incomplete effort to migrate from libfb to libfbjni. This is needed to restore the compatibility with Flipper and other FB Android projects that make use of FBJNI. Effectively, the outcome is that `fbjni` no longer has a checked-in copy here, but instead relies on the public artifacts published at github.com/facebookincubator/fbjni that can be deduplicated at build-time.
**A non-exhaustive list of tasks:**
* [X] Gradle builds the SDK and RNTester for Android.
* [X] Buck build for rntester works in OSS.
* [ ] Move from `java-only` release to full `fbjni` release. This requires finding a solution for stripping out `.so` files that the old `Android.mk` insists on including in the final artifacts and will clash with the full distribution.
* [ ] Import this and fix potential internal build issues.
* [ ] Verify that the changes made to the Hermes integration don't have any unintended consequences.
## Changelog
[Android] [Changed] - Migrated from libfb to libfbjni for JNI calls
Pull Request resolved: https://github.com/facebook/react-native/pull/27729
Test Plan:
- CI is already passing again for Gradle and Buck in OSS.
- After applying the following patch, RNTester builds and works with the latest Flipper SDK:
```
diff --git a/RNTester/android/app/build.gradle b/RNTester/android/app/build.gradle
index b8a6437d7..eac942104 100644
--- a/RNTester/android/app/build.gradle
+++ b/RNTester/android/app/build.gradle
@@ -170,10 +170,19 @@ dependencies {
debugImplementation files(hermesPath + "hermes-debug.aar")
releaseImplementation files(hermesPath + "hermes-release.aar")
- debugImplementation("com.facebook.flipper:flipper:0.23.4") {
+ debugImplementation("com.facebook.flipper🐬+") {
exclude group:'com.facebook.yoga'
- exclude group:'com.facebook.flipper', module: 'fbjni'
- exclude group:'com.facebook.litho', module: 'litho-annotations'
+ exclude group:'com.facebook.fbjni'
+ }
+
+ debugImplementation("com.facebook.flipper:flipper-network-plugin:+") {
+ exclude group:'com.facebook.yoga'
+ exclude group:'com.facebook.fbjni'
+ }
+
+ debugImplementation("com.facebook.flipper:flipper-fresco-plugin:+") {
+ exclude group:'com.facebook.yoga'
+ exclude group:'com.facebook.fbjni'
}
if (useIntlJsc) {
```
Reviewed By: mdvacca
Differential Revision: D19345270
Pulled By: passy
fbshipit-source-id: 33811e7f97f44f2ec5999e1c35339909dc4fd3b1
Summary: I was trying to debug an issue with a method signature mismatch last week, but was having trouble figuring out the problem because the error I was getting was unrelated - apparently JNI will sometimes swallow the error and just fail mysteriously later on. Ramanpreet showed me this macro that will throw any pending exceptions, so let's do that after we try to lookup the Java method in case it fails.
Reviewed By: RSNara
Differential Revision: D17680121
fbshipit-source-id: 1f23e49014f7cc1616e111386d440637e6a74677
Summary: Now that all the plumbing is done, this diff finally implements async method dispatch on the NativeModule thread.
Reviewed By: mdvacca
Differential Revision: D17480605
fbshipit-source-id: 992aab99954c488a0327144d84a1668a2b158d04
Summary:
Self explanatory.
1. **Existing:** We create the NativeModules thread in CatalystInstanceImpl.cpp.
2. D17422165: We wrap this thread in a `BridgeNativeCallInvoker`.
3. D17422164: We use `CatalystInstanceImpl::getNativeCallInvokerHolder()` to get a hold of the `BridgeNitiveCallInvoker` in Java.
4. D17422163: From Java, we pass this `CallInvokerHolder` to `TurboModuleManager`'s constructor.
5. **This diff:** `TurboModuleManager` then unwraps the `CallInvoker` from `CallInvokerHolder`, and passes it to all TurboModules in their constructor.
Reviewed By: PeteTheHeat
Differential Revision: D17422160
fbshipit-source-id: c0a76dfe5fdedac2e0e21f7a562bc7588dc190fb
Summary:
## Motivation
The concept behind JSCallInvoker doesn't necessarily have to apply only to the JS thread. On Android, we need to re-use this abstraction to allow execution of async method calls on the NativeModules thread.
Reviewed By: PeteTheHeat
Differential Revision: D17377313
fbshipit-source-id: 3d9075cbfce0b908d800a366947cfd16a3013d1c
Summary:
This diff improves on D17244061 by moving the `Jni::JniLocalScope` to earlier on in `JavaTurboModule::invokeJavaMethod`. The problem with D17244061 was two-fold:
1. It completely ignored the local references created by just calling `getConstants()`. So, you could spam `getConstants()` and still get an overflow.
2. The capacity was incorrect. Basically, two things happen when you call `PushLocalFrame`. One, we push an actual Jni frame onto a stack somewhere in the dvm: https://fburl.com/f16wfrxi, https://fburl.com/jhpha563. Popping off this stack is how the local references are cleared when we call `PopLocalFrame`. Two, we do a check to make sure that we can support at least `capacity` more local references: https://fburl.com/jucvew8j. The max is 512. The problem before was that, I was just using the argument count for the capacity. This was inaccurate because we create many `jclass` objects, and several misc. intermediate java objects when we invoke `JavaTurboModule::invokeJavaMethod`. So, I've refined the calculation of capacity to be a bit more accurate. This should make sure that capacity check works, which should help us fail faster if our local reference count approaches 512.
Reviewed By: shergin
Differential Revision: D17337354
fbshipit-source-id: 45574bae4748c52d8f919c1480b9a0936d970628
Summary: Whenever we invoke a method, we convert JS arguments into JNI objects. These JNI objects are all local references that need to be manually destroyed after the method call happens. Therefore, I'm using `JniLocalScope` to automatically do this whenever the stack is cleared after the call to `JavaTurboModule::invokeJavaMethod`. This should hopefully get rid of the JNI table overflow we're seeing in T52817336.
Reviewed By: mdvacca
Differential Revision: D17244061
fbshipit-source-id: 92ca78cdb23ad8dfe2425db46e086c10c9662fe2
Summary:
## The Problem
1. `CatalystInstanceImpl` indirectly holds on to the `jsi::Runtime`. When you destroy `CatalystInstanceImpl`, you destroy the `jsi::Runtime`. As a part of reloading React Native, we destroy and re-create `CatalystInstanceImpl`, which destroys and re-creates the `jsi::Runtime`.
2. When JS passes in a callback to a TurboModule method, we take that callback (a `jsi::Function`) and wrap it in a Java `Callback` (implemented by `JCxxCallbackImpl`). This Java `Callback`, when executed, schedules the `jsi::Function` to be invoked on a Java thread at a later point in time. **Note:** The Java NativeModule can hold on to the Java `Callback` (and, by transitivity, the `jsi::Function`) for potentially forever.
3. It is a requirement of `jsi::Runtime` that all objects associated with the Runtime (ex: `jsi::Function`) must be destroyed before the Runtime itself is destroyed. See: https://fburl.com/m3mqk6wt
### jsi.h
```
/// .................................................... In addition, to
/// make shutdown safe, destruction of objects associated with the Runtime
/// must be destroyed before the Runtime is destroyed, or from the
/// destructor of a managed HostObject or HostFunction. Informally, this
/// means that the main source of unsafe behavior is to hold a jsi object
/// in a non-Runtime-managed object, and not clean it up before the Runtime
/// is shut down. If your lifecycle is such that avoiding this is hard,
/// you will probably need to do use your own locks.
class Runtime {
public:
virtual ~Runtime();
```
Therefore, when you delete `CatalystInstanceImpl`, you could end up with a situation where the `jsi::Runtime` is destroyed before all `jsi::Function`s are destroyed. In dev, this leads the program to crash when you reload the app after having used a TurboModule method that uses callbacks.
## The Solution
If the only reference to a `HostObject` or a `HostFunction` is in the JS Heap, then the `HostObject` and `HostFunction` destructors can destroy JSI objects. The TurboModule cache is the only thing, aside from the JS Heap, that holds a reference to all C++ TurboModules. But that cache (and the entire native side of `TurboModuleManager`) is destroyed when we call `mHybridData.resetNative()` in `TurboModuleManager.onCatalystInstanceDestroy()` in D16552730. (I verified this by commenting out `mHybridData.resetNative()` and placing a breakpoint in the destructor of `JavaTurboModule`). So, when we're cleaning up `TurboModuleManager`, the only reference to a Java TurboModule is the JS Heap. Therefore, it's safe and correct for us to destroy all `jsi::Function`s created by the Java TurboModule in `~JavaTurboModule`. So, in this diff, I keep a set of all `CallbackWrappers`, and explicitly call `destroy()` on them in the `JavaTurboModule` destructor. Note that since `~JavaTurboModule` accesses `callbackWrappers_`, it must be executed on the JS Thread, since `createJavaCallbackFromJSIFunction` also accesses `callbackWrappers_` on the JS Thread.
For additional safety, I also eagerly destroyed the `jsi::Function` after it's been invoked once. I'm not yet sure if we only want JS callbacks to only ever be invoked once. So, I've created a Task to document this work: T48128233.
Reviewed By: mdvacca
Differential Revision: D16623340
fbshipit-source-id: 3a4c3efc70b9b3c8d329f19fdf4b4423c489695b
Summary: When you create a TurboModule from the JS side, we instantiate its Java class and simply make this `javaobject` a `jni::global_ref` in C++. But the reason why we need to make this a global ref is because `JavaTurboModule` needs it to be a global reference for method calls. Making this a `jni::global_ref` from the perspective to TurboModuleManager doesn't really make any sense. So, this diff refactors that bit of code.
Differential Revision: D16622133
fbshipit-source-id: 6a5c20bb405b945c06378a3423d5e7eb38ef244c
Summary:
## The Problem
1. `CatalystInstanceImpl` indirectly holds on to the `jsi::Runtime`. When you destroy `CatalystInstanceImpl`, you destroy the `jsi::Runtime`. As a part of reloading React Native, we destroy and re-create `CatalystInstanceImpl`, which destroys and re-creates the `jsi::Runtime`.
2. When JS passes in a callback to a TurboModule method, we take that callback (a `jsi::Function`) and wrap it in a Java `Callback` (implemented by `JCxxCallbackImpl`). This Java `Callback`, when executed, schedules the `jsi::Function` to be invoked on a Java thread at a later point in time. **Note:** The Java NativeModule can hold on to the Java `Callback` (and, by transitivity, the `jsi::Function`) for potentially forever.
3. It is a requirement of `jsi::Runtime` that all objects associated with the Runtime (ex: `jsi::Function`) must be destroyed before the Runtime itself is destroyed. See: https://fburl.com/m3mqk6wt
### jsi.h
```
/// .................................................... In addition, to
/// make shutdown safe, destruction of objects associated with the Runtime
/// must be destroyed before the Runtime is destroyed, or from the
/// destructor of a managed HostObject or HostFunction. Informally, this
/// means that the main source of unsafe behavior is to hold a jsi object
/// in a non-Runtime-managed object, and not clean it up before the Runtime
/// is shut down. If your lifecycle is such that avoiding this is hard,
/// you will probably need to do use your own locks.
class Runtime {
public:
virtual ~Runtime();
```
Therefore, when you delete `CatalystInstanceImpl`, you could end up with a situation where the `jsi::Runtime` is destroyed before all `jsi::Function`s are destroyed. In dev, this leads the program to crash when you reload the app after having used a TurboModule method that uses callbacks.
## The Solution
If the only reference to a `HostObject` or a `HostFunction` is in the JS Heap, then the `HostObject` and `HostFunction` destructors can destroy JSI objects. The TurboModule cache is the only thing, aside from the JS Heap, that holds a reference to all C++ TurboModules. But that cache (and the entire native side of `TurboModuleManager`) is destroyed when we call `mHybridData.resetNative()` in `TurboModuleManager.onCatalystInstanceDestroy()` in D16552730. (I verified this by commenting out `mHybridData.resetNative()` and placing a breakpoint in the destructor of `JavaTurboModule`). So, when we're cleaning up `TurboModuleManager`, the only reference to a Java TurboModule is the JS Heap. Therefore, it's safe and correct for us to destroy all `jsi::Function`s created by the Java TurboModule in `~JavaTurboModule`. So, in this diff, I keep a set of all `CallbackWrappers`, and explicitly call `destroy()` on them in the `JavaTurboModule` destructor. Note that since `~JavaTurboModule` accesses `callbackWrappers_`, it must be executed on the JS Thread, since `createJavaCallbackFromJSIFunction` also accesses `callbackWrappers_` on the JS Thread.
For additional safety, I also eagerly destroyed the `jsi::Function` after it's been invoked once. I'm not yet sure if we only want JS callbacks to only ever be invoked once. So, I've created a Task to document this work: T48128233.
Reviewed By: mhorowitz
Differential Revision: D16589168
fbshipit-source-id: a1c0786999c22bef55d416beb0fc40261447a807
Summary: When you create a TurboModule from the JS side, we instantiate its Java class and simply make this `javaobject` a `jni::global_ref` in C++. But the reason why we need to make this a global ref is because `JavaTurboModule` needs it to be a global reference for method calls. Making this a `jni::global_ref` from the perspective to TurboModuleManager doesn't really make any sense. So, this diff refactors that bit of code.
Reviewed By: mdvacca
Differential Revision: D16555673
fbshipit-source-id: 2778fc5a372c41847e8296c2e22bb9a8826fcc52
Summary:
When you call a TurboModule method with JS arguments, we necessarily convert these JS args to Java objects/primitives before passing them to the Java implementation of the method. The problem is that we let the type of the JS arg dictate the type of the Java object/primitive to convert said arg to. This means that if a JS developer passes in an `number` to a function that expects an `Array`, we'll convert the number to a `double` and try to call the Java method with that `double`, when it actually expects a `ReadableArray`. This will trigger a JNI error, and crash the program. Ideally, we should be able to catch these type mismatches early on.
In this diff, on every TurboModule method call, I parse the method signature to determine the Java type of each JS argument. Then, for any argument, if the JS arg and the Java arg types aren't compatible, I raise an exception, which gets displayed as a RedBox in development. This diff also implements support for `?number` and `?boolean` argument and return types in TurboModules.
Reviewed By: mdvacca
Differential Revision: D16214814
fbshipit-source-id: 4399bb88c5344cff50aa8fe8d54eb2000990a852
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: `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:
If you call into a Java method (from C++ using JNI) that raises an exception, the JNI call won't actually raise a C++ error. Instead, the `JNIEnv` will record the pending Java exception and the C++ will continue executing as normal. This is bad because the next time you call into JNI, the app will actually crash, unless you explicitly cleared the exception using `JNIEnv::ExceptionClear()` before the JNI call.
With respect to TurboModules, we need to make sure that RedBoxes show up whenever a native methods raise an exception. We also don't want the app to crash when a JNI method call fails because of a raised exception. Therefore, in this diff, I raise a C++ exception if `JNIEnv::ExceptionCheck()` is true.
Reviewed By: mdvacca
Differential Revision: D14738540
fbshipit-source-id: 4c3063aa93ae7aef025bd2dab6b45059bb8fb409
Summary:
If the return type of a TurboModule method is `Promise`, the infra should create a `com.facebook.react.bridge.Promise` object and pass it as the final argument of the TurboModule Java method call. The Java TurboModule method can then do some work asynchronously and either resolve or reject the promise at some point in time.
**Note:** I stacked a diff for error handling on top of this one.
Reviewed By: mdvacca
Differential Revision: D13653156
fbshipit-source-id: 4c30c3223ad8f47c6ba7f1236527aaced01c8ae8
Summary:
This commit moves all the turbo module files for Android to Github.
Note that gradle build is not yet enabled.
Sample Turbo Modules will be added in a later commit.
Other missing features
- Support for `CxxModule`
- Remove usage of folly::dynamic for arguments and result conversion
- Support for Promise return types.
Reviewed By: mdvacca
Differential Revision: D13647438
fbshipit-source-id: 5f1188556d6c64bfa2b2fd2146ac72b0fb456891
Ramanpreet Nara
Pascal Hartig
Andres Suarez
Emily Janzer
Eric Williamson
Kevin Gozali
Ram N