Summary:
Pull Request resolved: https://github.com/facebook/yoga/pull/1368
X-link: https://github.com/facebook/react-native/pull/39372
These were marked as deprecated as part of the public Yoga 2.0 release, and were alredy emitting deprecation warnings. Remove them.
Reviewed By: javache
Differential Revision: D49131250
fbshipit-source-id: cc1d4e8b179697b9a11a685f4fc4e9d36e1a26a0
Summary:
X-link: https://github.com/facebook/react-native/pull/37243
X-link: https://github.com/facebook/litho/pull/944
Pull Request resolved: https://github.com/facebook/yoga/pull/1279
Java bindings for Yoga rely solely on garbage collection for memory management. Each Java `YogaNode` has references to its children and parent Java Nodes. This means, for a node to be garbage collected, it cannot be reachable from any user accessible node. Each node then has single ownership of a `YGNodeRef`. When the `YogaNode` is garbage collected, a finalizer is run to call `YGNodeFree` and free the underlying native Yoga Node.
This may cause a use-after-free if finalizers are run from multiple threads. This is because `YGNodeFree` does more than just freeing, but instead also interacts with its parent and children nodes to detach itself, and remove any dangling pointers. If multiple threads run finalizers at once, one may traverse and try to mutate a node which another is freeing.
Because we know the entire connected tree is dead, there is no need to remove dangling pointers, so I want to expose a way to just free a Yoga Node, without it mutating the tree as a side effect.
This adds a currently private `YGNodeDeallocate` that frees without traversal. Ideally from naming this is what `YGNodeFree` would do, but we think changing the behavior of that might be too disruptive to OSS. At the same time there may be other memory safety related API changes we would like to eventually make, so this isn't made public beyond the JNI bindings to prevent needing to transition more APIs.
Changelog: [Internal]
Reviewed By: rshest
Differential Revision: D45556206
fbshipit-source-id: 62a1394c6f6bdc2b437b388098ea362a0fbcd0f7
Summary:
Pull Request resolved: https://github.com/facebook/yoga/pull/1269
X-link: https://github.com/facebook/react-native/pull/37127
This prevents targets which include Yoga from using its private APIs.
Instances of this have been mostly cleaned up in the past diffs, with the major exception of RN Fabric. To stage this without blocking on that, I added a `yoga-private-api` target for now to keep using these headers while making it unlikely new usages will show up.
Reviewed By: javache
Differential Revision: D45339425
fbshipit-source-id: eb7ef151ad2467d7c3370cd7c10d47e8db9496a0
Summary:
This removes some unused flags which will cause Yoga to layout every tree twice, then diffing the tree, reporting whether the whole tree is different. This is too expensive to run outside of local experimentation, but we have more nuanced ways to implement the `YGNodeLayoutAffectedByQuirk` I am wanting to add.
Changelog: [Internal]
Reviewed By: lunaleaps
Differential Revision: D42406917
fbshipit-source-id: b415ed02768f6b59de3a6fa90c60c750d56fd4b0
Summary:
X-link: https://github.com/facebook/react-native/pull/35841
X-link: https://github.com/facebook/litho/pull/928
Pull Request resolved: https://github.com/facebook/yoga/pull/1216
The Yoga JNI bindings use Reflection, so we need to let ProGuard know not to strip certain annotated fields.
This is done internally using a single copy of `com.facebook.proguard.annotations` from fbandroid (sometimes), which is then repackaged externally, and published as its own whole Yoga specific package. We never actually inform the stock Gradle project of the rules for the annotations though, so apps must add these manually.
This simplifies the setup, where Yoga has its own self-contained annotations/rules. The rules are exposed for Gradle/Buck dependencies, but RN and Litho both consume Yoga via dirsync + custom Gradle logic, so we need to duplicate the proguard rules to them instead of them being propagated automatically.
Changelog: [Internal]
Reviewed By: rshest
Differential Revision: D42406641
fbshipit-source-id: c2b12fd498f93f144e5651917ca878d2a5050e08
Summary:
## Changelog:
[General] [Yoga] - Use vanilla jni instead of fbjni for all the jni communication
Reviewed By: astreet
Differential Revision: D17808005
fbshipit-source-id: 5e9a1ed73978f519c71c248774a28e5a294e7c7f
Summary: This diff moves methods related to actions on YGNode like free node, reset node etc. to vanilla JNI
Reviewed By: amir-shalem
Differential Revision: D17668008
fbshipit-source-id: 03bfc51ec1fcf06569713400f984d551827e22fe
Summary:
This diffs adds a separate file YGJNIVanilla.cpp to add jni methods which uses vanilla JNI instead of FBJNI.
In this diff only one method has been added to setup the experiment boolean setup.
At the end of this diff stack , we will be able to experiment between fbjni and vanilla jni in yoga and finally get rid of fbjni which saves us around 300Kb per architecture in yoga binary size.
Reviewed By: Andrey-Mishanin
Differential Revision: D17601591
fbshipit-source-id: a88520c625bd8b5d9ffcf8ab5f02fc71dc800081
Summary: Removes config param useBatchingForLayoutOutputs and now we are using batching of layout outputs as float array while passing data from c++ to java
Reviewed By: davidaurelio
Differential Revision: D16221483
fbshipit-source-id: 326c668d4dfd13b2cf031f98a84bfa50b1440513
Summary:
@public
`YGNodeGetInstanceCount` was only ever meant for tests, and caused data races in multi-threaded environments.
It was completely replaced with event-based counting for tests.
Here we remove public API around the counter, and the counter itself.
Reviewed By: SidharthGuglani
Differential Revision: D15174857
fbshipit-source-id: 228e85da565bac9e8485121e956a2e41910b11c8
Summary:
Using a config flag to switch between different implementations of transferring layout outputs
- YogaNodeJNI uses multiple access of java fields to pass all properties like width, height, margin etc...
- YogaNodeJNIBatching uses a float array to pass all the data in one java field access
Reviewed By: davidaurelio
Differential Revision: D14378301
fbshipit-source-id: 0da5b28e6a67ad8fd60eb7efe622d9b2deaf177f
Summary:
@public
Passing primitive data via JNI is more efficient than passing objects.
Here, we avoid creating `YogaValue` (Java) instances via JNI, and rather pass a `long` back to Java. The instance is then created by extracting the necessary bytes on the Java side.
Reviewed By: foghina
Differential Revision: D14576755
fbshipit-source-id: 22d09ad50c3ac6c49b0a797a0dad639ea4829df9
Summary:
@public
Moving all native methods in a single class provides the benefit of not having to load native bindings eagerly when just creating config objects in the startup paths, or setting Java-only values on them.
Loading native bindings triggers additional class loads (`YogaConfig` / `YogaNode`), and can lead to problems in multi-dex scenarions.
Reviewed By: pasqualeanatriello
Differential Revision: D14560658
fbshipit-source-id: 14e31e3c3b560675b5a752a38ae75ab80a565ea1
