Summary:
As we did in the previous diff, here we implemented `EventEmitter`'s ownership model as a `shared_ptr`. This change fixes problem with leaking `WeakObject`s which happens on hot-reload.
So, in short:
* `EventTargetWrapper` object owns `jsi::WeakObject` that can be converted to actual `jsi::Object` that represent event target in JavaScript realm;
* `EventTargetWrapper` and `jsi::WeakObject` objects must be deallocated as soon as native part does not need them anymore;
* `EventEmitter` objects retain `EventTarget` objects;
* `EventEmitter` can loose event target object in case if assosiated `ShadowNode` got unmounted (not deallocated); in this case `EventEmitter` is loosing possibility to dispatch event even if some mounting-layer code is still retaining it.
Reviewed By: mdvacca
Differential Revision: D9762755
fbshipit-source-id: 96e989767a32914db9f4627fce51b044c71f257a
Summary:
Previously, we used special JSI bindings method to release an event handler (`JSIReleaseFabricEventHandler`). Now we expose this ownership model as a regular `std::shared_ptr`, so when the owner got deallocated, the event handler will be released automatically.
Why not use `unique_ptr`? `unique_ptr` is faster (and simpler) indeed, but it does not provide `type erasure` functionality that we need; to use `unique_ptr` we would have to make JSI an explicit Fabric dependency (we will probably end up with it eventually, but I this particular case is not a good reason for that).
All interactions with `eventHandler_` are done in a non-owning manner, so it's as performant as unique_ptr anyway.
(Please ignore all changes in JSCFabricUIManager.h/cpp files, we will delete them soon.)
Reviewed By: mdvacca
Differential Revision: D9756732
fbshipit-source-id: bffdee0c724dc95855ced7c35e7c13cf1554796e
Summary:
The source of truth has already moved, so now we just need to fix references
This diff is mostly the result of running:
```
$ tools/mobile-unification/loadmod --fixup xplat/configurations/buck/apple/flag_defs.bzl tools/build_defs/apple/
```
Then I committed with `hg commit -I xplat/`
The controller you requested could not be found.
Differential Revision: D9772194
fbshipit-source-id: 93d23ae8e1c62440c7876cad965d963bde960db9
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:
@public
Now we simply skip `uiManagerDidFinishTransaction` calls if they refer to unregister surfaces. In the future, after we have proper asynchronous scheduling and sync unmounting (and if we chose to have sync unmounting), we can avoid this situation (and assert in this cases).
Reviewed By: sahrens
Differential Revision: D9652731
fbshipit-source-id: e376ea1ae4f93960a903e6397d843bd7c4b72400
Summary:
@public
We don't need this anymore.
The same functionality is now implemented as `ShadowView::operator==` in much more reasonable way.
Reviewed By: sahrens
Differential Revision: D9649821
fbshipit-source-id: 8cd5f3cb4f583fd10d2d1e060aba914541341b5b
Summary:
@public
Apparently, it's how it should be.
Reviewed By: rsnara
Differential Revision: D9631870
fbshipit-source-id: 46f58270104d699fbc9abe21062c12f791460536
Summary:
@public
Now it's clear that we don't need to store/handle ShadowTree objects as `shared_ptr`s; Scheduler should own it. This diff changes that to using unique_ptr and removes a base class of ShadowTree.
Reviewed By: mdvacca
Differential Revision: D9403567
fbshipit-source-id: 6e411714b632a04233fd5b25c8ab7cdd260105fd
Summary:
@public
Voalá, this small change actually implements view flattening. Obviously, it does not work right now because there are no `ShadowNode` classes which implement `isLayoutOnly`.
Surprisingly, correct implementing of `isLayoutOnly` is quite tricky, we will work on this in coming diffs.
Reviewed By: mdvacca
Differential Revision: D9403565
fbshipit-source-id: 1f16f912cb5c6841405a1fc3cf36aec28698c11f
Summary:
@public
This is quite a big diff but the actual meaningful change is simple: now we use ShadowView class instead of ShadowNode in mutation instructions.
Note:
* In some places (especially during diffing) we have to operate with ShadowNodeViewPair objects (which represents a pair of ShadowNode and ShadowView). The reason for that is that we cannot construct child ShadowViews from parent ShadowViews because they don't have any information about children.
* `ShadowTree::emitLayoutEvents` is now much simpler because ShadowView better represents the specifics of this kind of object.
* The code in RCTMountingManager also became simpler.
This change will allow us to implement more cool tricks soon.
Reviewed By: mdvacca
Differential Revision: D9403564
fbshipit-source-id: dbc7c61af250144d6c7335a01dc30df0005559a2
Summary:
@public
We need some another object like ShadowNode (but not ShadowNode) to represent an instance of the component in the mutation instructions. This is
the main motivation for introducing ShadowView.
Why not use ShadowNode? ShadowNode is designed to represent a node in ShadowTree, not be a part of a mutation instruction.
* ShadowNode exposes some APIs that should not be exposed to the mounting layer;
* ShadowNode is an immutable data structure, so we cannot mutate it in some way which can be meaningful for mounting;
* We should not add to ShadowNode any functionality which is needed only for mounting;
* ShadowNode is a bit more heavy object to share that it needs to be; it's exposed (embedded into Mutation) as a `shared_ptr` which is not optimal from the performance perspective;
* Retaining ShadowNode from mounting code can unnecessarily extend its lifetime which can negatively affect memory usage.
Reviewed By: mdvacca
Differential Revision: D9403562
fbshipit-source-id: 72ad81ed918157a62cd3d1a03261f14447649d0b
Summary: This diff implements the HorizontalScrollView component for Android Fabric C++, as part of this diff I also re-named the components AndroidHorizontalScrollContentView for RCTAndroidHorizontalScrollContentView and AndroidHorizontalScrollView for RCTAndroidHorizontalScrollView. This might sound against our plan of removing the RCT preffix, but it is to make it simpler to map components between current implementation of RN and Fabric (otherwise we don't know when to add the RCT preffix in Android side to find the right View Manager), later we can just remove the preffix from C++, Android, iOS and JS.
Reviewed By: shergin, achen1
Differential Revision: D9122729
fbshipit-source-id: e9299552857c6dd0c18abfa5fa49a3d50e221729
Summary:
@public
Now, it's not just an abstract class, it's a regular class which unifies event delivery priorities using specific event beats and event pipe.
Reviewed By: mdvacca
Differential Revision: D8886232
fbshipit-source-id: c4360511e5fd477ca7407fc3ebbd99ca578e79cc
Summary:
@public
We need that to ensure that we will not deliver events to nodes with invalid state.
Reviewed By: mdvacca
Differential Revision: D8886234
fbshipit-source-id: 1d6ca129c97a5dca0411e85909aea48185f46c54
Summary:
@public
Instead of having two methods it's easier to have just one which can be abstracted as `EventPipe`.
Reviewed By: mdvacca
Differential Revision: D8886231
fbshipit-source-id: af9fd92dc4afa1219a11acce0aa021a85c94d232
Summary: This diff improves the error message that is displayed when a component descriptor is not implemented in C++
Reviewed By: shergin
Differential Revision: D9093562
fbshipit-source-id: 930b381bc66c20af6fa160b09e7484bad4666e28
Summary:
@public
To avoid unnecessary copying of `shared_ptr`s inside ShadowNodeFragment, now we store them as `const &` references.
Reviewed By: mdvacca
Differential Revision: D8988388
fbshipit-source-id: 0b3582e57ce7577b8fa819392bf33f34e1a60b59
Summary:
@public
Now we use same data structure to specify a shape of shadow node as we use in ShadowNode (sub)clases.
Reviewed By: mdvacca
Differential Revision: D8988387
fbshipit-source-id: 475298b2c71ee7ee2b197db009f7b8313b54f5df
Summary:
@public
When we copy-construct ShadowNode, we don't need to retain a source shadow node, so there is no need to pass it as a `shared_ptr`. Passing an argument to constructor as `const &` is also more idiomatic in C++.
Reviewed By: mdvacca
Differential Revision: D8988384
fbshipit-source-id: 1279d9185fa1b4b82fd26e3040bd62fa9495b4d3
Summary:
@public
This diff changes a way how we specify a shape of newly created and/or cloned of ShadowNode. Previously we pass those values as a list of arguments, now those values are coupled into a new data structure called ShadowNodeFragment. All that makes suppose to make code much more easy to read and maintain, this is especially important because we want to add a couple of new entities in this set.
Reviewed By: mdvacca
Differential Revision: D8988389
fbshipit-source-id: 1835f646e1ecc6a1f413feaf1900f3d3ad0ebc05
Summary:
@public
Previously, ContextContainer could store only `shared_ptr`s, but now it wraps all values in own `shared_ptr` container.
I wish we can use `unique_ptr` here, but apparently we cannot because `unique_ptr` does not support type-erasure (`std::unique_ptr<void>` is illigal).
Becasue ContextContainer is not supposed to be used in hot paths, the performance aspect of that does not actually matter.
Reviewed By: mdvacca
Differential Revision: D8853446
fbshipit-source-id: e5d0a5595fe44c59f1395d6ffccf9d3fed923c83
Summary:
@public
We need that because gonna add much more event-related stuff, so it deserves separate buck target.
Reviewed By: mdvacca
Differential Revision: D8831547
fbshipit-source-id: 616581b39b425a49302d5f7f86267e62b0d58389
Summary:
@public
We need this in case when we want to store several intances of the same class in the container.
Reviewed By: mdvacca
Differential Revision: D8814808
fbshipit-source-id: 78ab15d78cf3878d03bf0a45bc42b968d87435e7
Summary:
@public
There is no reason to have it inside View; it deserves that.
Reviewed By: mdvacca
Differential Revision: D8757012
fbshipit-source-id: 881b54008b51614cd203ab97811494fa7c30e4ef
Summary:
@public
Everything is better with C++ templates.
In this cases templates allow us to remove additional parameters and casts on the callsite.
Reviewed By: mdvacca
Differential Revision: D8754523
fbshipit-source-id: 2340b2cd96ab0a60d54d9aa30dea3c072b951a8a
Summary:
@public
* In case of `ShadowTree` we just pass original old node as a `commit` method argument;
* In case of `ConcreteViewShadowNode` we just don't need that because diffing algorithm does not use that information anymore.
Reviewed By: mdvacca
Differential Revision: D8753906
fbshipit-source-id: b8555083c7e72e9b3c0f9a8065745946d4cf44c7
Summary:
@public
Non-null owner pointer in Yoga node indicates that this node is already being used by some other subtree, so it must be cloned in case of possible (re)layout.
Theoretically, this node must/can be cloned by Yoga right before applying a new layout to this node, but Yoga has a special optimization that uses that fact that Yoga always cloning *all* children of a particular node altogether. This is not true for React; to meet React and Yoga worlds we double check the owner pointer in `addChild` and clone node preliminary if needed.
See also the previous diff for more context.
Reviewed By: mdvacca
Differential Revision: D8709952
fbshipit-source-id: 84ef0faa0f1d9cc9a8136b550cf325bc20508d53
Summary:
@public
... and it's as efficient as it was before.
The previous version of the algorithm used `sourceNode` reference to know the previous state of the node to call the algorithm recursively.
That wasn't so good because of several reasons:
- It was fragile because we had two different sources of the truth of the "previous state of the tree": committed tree and source node pointer;
- We had to store weak pointers to source nodes inside cloned nodes. That is not free in terms of performance;
- The old approach introduced a constraint that all previously used and now reinserted nodes must be cloned to update source node (otherwise, the algorithm would regenerate instructions recreating already existing subtrees);
- That cloning required access to `isSealed` flag which is supposed to be a debug-only thing (that actually affects performance and must be compile-out for release builds).
The new approach compares nodes with same react tag and naturally cloning-artifacts resilient.
Yes, the new approach uses a map of inserted nodes, but the previous one already had it (otherwise there is no way to tell which nodes should be "deleted"). And anyway, this is a very little map that exists for a very little period of time.
Reviewed By: mdvacca
Differential Revision: D8709953
fbshipit-source-id: 027abb326cf45f00f7bb0bbd7c4e612578268c66
Summary: Revert the order of "remove mount items", to ensure views are removed from high index to low index.
Reviewed By: shergin
Differential Revision: D8742796
fbshipit-source-id: 6e04c39386d290bf3958ee83256d4fbe23e2c4ca
Summary: We were supposed to pass in proper eventEmitter, but passed in one with null eventTarget instead, causing assertion failures when dispatching event.
Reviewed By: sebmarkbage, shergin
Differential Revision: D8720793
fbshipit-source-id: 891f3b2a2c76a6dd3e40039623c6e86991aad50b
Summary:
Removes the concept of instance handle. Instead we pass the event target
to createNode and don't pass it to subsequent clones.
The life time of the event target is managed by native (the event emitter).
It has to be released manually.
Reviewed By: shergin
Differential Revision: D8688330
fbshipit-source-id: e11b61f147ea9ca4dfb453fe07063ed06f24b7ac
Summary:
@public
Most of them are legit issues which should not be compilable anyways (but Clang tolerates thems).
Reviewed By: mdvacca
Differential Revision: D8655539
fbshipit-source-id: 645729fb9d6a120ce1ab2b07542abcdacd72320d
Summary:
@public
Suddenly, it is not supported on Android.
Luckelly `folly:to<std::string>()` is as good as `std::to_string()`.
Reviewed By: mdvacca
Differential Revision: D8655538
fbshipit-source-id: 2b3b970f6a261253aaa6b22dba8338dc66b7195d
Summary: This is basic impl of <PerformanceLoggerFlag> component without any layout/mounting computation, just TTI.
Reviewed By: shergin, mdvacca
Differential Revision: D8598983
fbshipit-source-id: b938753d6396088735cbbeab26d69c9aaa45608e
Summary:
@public
The current Fabric architecture, in general, does not support "subscribing" for events, so all kinds of events are always delivered no matter have JavaScript components `on`-handlers for them or not.
At this point, we are not sure should it be this way or not. But we are sure that for some extremely noisy events (like onLayout) we have to make an exception right now (otherwise overall performance will suffer).
So, this diff implements that for `onLayout`.
Reviewed By: fkgozali
Differential Revision: D8597408
fbshipit-source-id: 6933b7cb96e24f0660bd7850b625ff27e3146a2b
Summary:
@public
We do preventing cloning in UIManager especially to add a layer to Shadow Node source chain,
so apparently there is no point illuminate that by calling `shallowSourceNode`.
Reviewed By: fkgozali
Differential Revision: D8585163
fbshipit-source-id: 3743edc30bf2183c420fd79ce1e59d68ceaa278b
Summary:
@public
We need this temporary for testing until we support them all.
Reviewed By: mdvacca
Differential Revision: D8552361
fbshipit-source-id: 25f48cebcf5a665a24b92803dd7738f947ca74b2
Summary:
@public
This diff implements basics of cross-platform part of <Image> component.
Known issues:
- Events does not work yet.
- Some quite specific image source parameters (like custom http headers) are not supported yet.
Reviewed By: fkgozali
Differential Revision: D8526575
fbshipit-source-id: ecc97d9fda2b2e65bb1b079af057f8e176a161e5
