Summary:
On certain devices (in my case, any iPad Pro model), listening to `DeviceEventEmitter.didUpdateDimensions` would call back *before* the interface change takes places (i.e. calling `Dimensions.get()` in this callback would return wrong values). Turns out that we were listening for the wrong native event.
Edit to add: now using `[RCTSharedApplication() statusBarOrientation]` to get the current orientation. Not yet sure why, but the `userInfo` provided by the notification was returning the wrong orientation *only* on the first time you rotate the device.
This fixes the open issue: https://github.com/facebook/react-native/issues/7340
Closes https://github.com/facebook/react-native/pull/11350
Differential Revision: D4348186
Pulled By: javache
fbshipit-source-id: cb2cfb9cccfc459ad4b46a5af2bec4c973132ae8
Summary:
drawRect was sometimes being called with NaN values, which caused
very strange ui behavior on some devices. This patch fixes the problem by
ensuring that we use a default value when the value is NaN.
Reviewed By: AaaChiuuu
Differential Revision: D4338453
Summary: Update package name of java code to refer to yoga instead of csslayout. Still need to change the name of the folder where this code resides but that requires update github sync scripts etc so it is safer and easier to split these diffs apart.
Differential Revision: D4271420
Summary: Remove deprecated java code and make use of CSSEdge instead of the now removed Spacing class.
Reviewed By: AaaChiuuu
Differential Revision: D4233198
Summary:
@public
Virtual shadow nodes (e.g. text) don't use CSSNodes so we don't need to create them. This shows large savings in CSSNodes allocated, depending on the app.
This could be breaking if:
- You have virtual nodes that still set and get CSS properties. The setters now no-op for virtual nodes (I unfortunately couldn't remove them completely -- see the comment on LayoutShadowNode), but the getters will NPE. If you see these NPE's, you should almost definitely be using your own datastructure instead of a CSSNode as virtual nodes will not participate in the layout process (and the CSSNode is then behaving just as a POJO for you).
I do not anticipate this to be breaking for anyone, but am including breaking in the commit message since this is a change in API contract.
Reviewed By: emilsjolander
Differential Revision: D4220204
Summary:
@public
Adds a pool to recycle CSSNodes within UIManager. A follow-up diff will hook this up to a memory pressure listener to drop the pool on memory pressure.
Reviewed By: emilsjolander
Differential Revision: D4189532
Summary: setPadding already calls dirty, and we should only be calling dirty on nodes that have measure functions.
Reviewed By: emilsjolander
Differential Revision: D4205148
Summary:
@public
This diff makes it so ReactShadowNode holds a CSSNode instead of extending one. This will enable us to pool and re-use CSSNodes and will allow us to keep from breaking the CSSNode api assumption that nodes that have measure functions don't have children (right now, text nodes have measure functions, but they also have raw text children).
BREAKING
This diff makes ReactShadowNode no longer extend CSSNodeDEPRECATED. If you have code that depended on that, e.g. via instanceof checks, that will no longer work as expected. Subclasses that override getChildAt/addChildAt/etc will need to update your method signatures. There should be no runtime behavior changes.
Reviewed By: emilsjolander
Differential Revision: D4153818
Summary: This is an API breaking change done to allow us to avoid an allocation during measurement. Instead we do the same trick as is done when passing measure results to C, we path them into a long.
Reviewed By: splhack
Differential Revision: D4081037
Summary:
In Nodes, we sometimes get crashes when we drop an already dropped
view. For now, this catches the exception to allow things to be handled
gracefully (until we can identify the actual root cause). #accept2ship
Reviewed By: sriramramani
Differential Revision: D4059333
Summary: The current implementation was made out of simplicity and to keep the same API as before. Now that the java version of csslayout is deprecated it is time to change the API to make the calls more efficient for the JNI version. This diff with reduce allocations as well as reduce the number of JNI calls done.
Differential Revision: D4050773
Summary:
From task:
in Nodes today, styles (dashed and dotted) only work on borders if the view has rounded corners. this is incorrect as they should work even when we're drawing rectangular borders.
Looking at the current non-nodes code (https://fburl.com/474407319) we can see that whenever there is a pathstyle effect the non-nodes code treats the border as if it was rounded (note that mBorderStyle == null || mBorderStyle == BorderStyle.SOLID means NO path effect is applied).
To bring the Nodes code in line with the non-Nodes code we can simply do the same thing: if there is a path style effect draw it as if it was rounded.
Reviewed By: ahmedre
Differential Revision: D4054722
Summary:
From task:
In some cases, however, drawPath is the more correct thing to do, and this ticket is one such example - if we draw the left border and top border with different colors, Nodes draws rectangularly, whereas non-Nodes makes the edges triangular.
To solve the issue in Nodes, we use drawPath instead of drawRect only when necessary (borders intersect with different colors).
Reviewed By: ahmedre
Differential Revision: D4048685
Summary:
Implemented 2 TODOs from ReactART for Android:
- TODO(7255985): Use TextureView and pass Surface from the view to draw on it asynchronously instead of passing the bitmap (which is inefficient especially in terms of memory usage)
- TODO(6352067): Support dashes in ARTShape
We use ReactNativeART in our Android project.
1. Our app crashes sometimes on large screen smartphones with OutOfMemoryError. Crashes happen in ARTSurfaceShadowNode where TODO(7255985) was suggested in a comment in order to use memory more efficiently.
2. We needed dashes for drawing on ARTSurface.
**Test plan (required)**
I attach a screenshot of our app which shows dashed-lines and two ARTSurfaces on top of each other rendering exactly the same as in the pervious implementation of ARTSurface.
Closes https://github.com/facebook/react-native/pull/9486
Differential Revision: D4021303
Pulled By: foghina
Summary:
In Nodes, we added logic when we dropped a view to also pass the
parent, so we could tell the parent that \"hey, this view is now dropped\"
so that it can be released. While this is fine, there are some crashes due
to the fact that the root node is not being found when we drop the child.
I've spent a lot of time thinking about how this could happen, and the only
plausible explanation I can come up with is that we first detach all views
from the root, then drop the root, and then drop one of the children that
was detached. I can't think of a good way why this would happen.
In the interest of protecting from this crash, this patch adds a check as to
whether or not the id of the parent is that of a root id, and, if so, it
doesn't run the logic that tells this view that this view was removed.
This should be safe, because the root most view should not have clipping
enabled (since it's a vanilla view group as opposed to a scrolling view).
Reviewed By: sriramramani
Differential Revision: D3991682
Summary:
This way `UIImplementation` can hold on to it and use it outside of calls from the `UIManagerModule`.
@public
Reviewed By: lexs
Differential Revision: D3899774
Summary:
@public
The hack for the status bar height is not necessary any longer, so we can remove
all code related to it
Reviewed By: lexs
Differential Revision: D3943770
Summary:
Modals were doing nothing (and sometimes crashing) when they were
being closed. The reason for this was due to the fact that the parent being
removed was not necessarily the view's parent. Consequently, trying to inform
said parent that its child was removed failed, because said parent wasn't a
view, and therefore had no record in mViewsToTags.
Reviewed By: sriramramani
Differential Revision: D3928850
Summary:
@public
This fixes measuring of items in the main axis of a container. Previously items were in a lot of cases measured with UNSPECIFIED instead of AT_MOST. This was to support scrolling containers. The correct way to handle scrolling containers is to instead provide them with their own overflow value to activate this behavior. This is also similar to how the web works.
This is a breaking change. Most of your layouts will continue to function as before however some of them might not. Typically this is due to having a `flex: 1` style where it is currently a no-op due to being measured with an undefined size but after this change it may collapse your component to take zero size due to the implicit `flexBasis: 0` now being correctly treated. Removing the bad `flex: 1` style or changing it to `flexGrow: 1` should solve most if not all layout issues your see after this diff.
Reviewed By: majak
Differential Revision: D3876927
Summary:
With Nodes, we want to support recursive clipping of subviews.
Without this, surfaces like Marketplace won't properly handle subviews.
Reviewed By: sriramramani
Differential Revision: D3904721
Summary:
This fixes a crash for the case when we try to drop a view that has already been dropped.
**The Problem**
We got reports of a crash (t12912526) that occurs when the resolveViewManager method can't resolve a ViewManager for a View being dropped.
Investigating this, one thing in common between all the stack traces for this is that dropView is called from line 210 of FlatNativeViewHierarchyManager. This part of the code is specifically the part we added to remove strong references to any clipped children (from views that have subview clipping enabled).
So this is a problem specifically with Nodes and clipSubviews, which brings up some questions:
**when can this happen?**
The only situation this can possibly happen is when we drop a child (which is clipped) followed by dropping its parent in the same cycle. Consider a tree where each view only has one child, such as: A - B - C - D. This crash would happen if D is clipped, and we removed it, followed by removing any of its parents in the same frame.
**if the removes happen in different frames, does this bug occur?**
No - the reason is that before we execute the DropView operations, we run through StateBuilder, which traverses the shadow tree and marks updates, thus removing the view going away (such that the delete in the next frame doesn't try to re-delete it).
So why doesn't this happen when we're dropping in the same frame? The reason is that manageChildren (where this all starts) asks to remove some views. We handle this by removing said Nodes and their children from the shadow tree. Consequently, when StateBuilder iterates over the shadow tree, it can't do the right thing because said nodes no longer exist.
As a more concrete example, consider A - B - C - D again, and consider that both D and B are removed. StateBuilder only sees A, and realizes that it now has 0 children (whereas before it has 1), so it removes B from its children. However, this process isn't recursive, so C never gets cleaned up.
**why doesn't this happen with Nodes without clipping containers?**
The answer to this is that NativeViewHierarchyManager's dropView method checks the existance of each child before deeply dropping that child and its subtree. So in this case, we drop D and all its children, and when we come to drop B, we try to drop C (which exists) and then its children (D, which doesn't exist because we already dropped it, so we ignore it).
**why doesn't this happen with non-Nodes?**
The reason is that non-Nodes handles removes differently - every remove is enqueued in a call to NativeViewHierarchy's manageChildren, which explicitly asks the parent to remove said child. Consequently, we never try to remove a child that is already removed.
**Fix**
The initial fix was to check whether or not the view exists, but this updated patch just does the right thing at drop time - i.e. whenever a view is dropped, we notify the parent of this fact so that it can clear the reference from clipped views.
**One last Note**
There are two reasons for switching `super.dropView` to `dropView` - first, the comment is only partially correct - calling `super.dropView` will avoid looking at clipped children (as an aside, that could cause a leak in the case of nested clipping subviews), but will look at clipped grandchildren, because of the super class's iteration across the set of children.
Reviewed By: astreet
Differential Revision: D3815485
Summary:
This should probably be two separate diffs, sorry. It takes forever to test these things on fb4a though.
The nodes GK was turned up in fb4a, so I had to make a few changes to make the existing integration work.
Reviewed By: lexs, emilsjolander
Differential Revision: D3863226
Summary:
@public
Introduce `overflow:scroll` so that scrolling can be implemented without the current overflow:visible hackiness. Currently we use AT_MOST to measure in the cross axis but not in the main axis. This was done to enable scrolling containers where children are not constraint in the main axis by their parent. This caused problems for non-scrolling containers though as it meant that their children cannot be measured correctly in the main axis. Introducing `overflow:scroll` fixes this.
Reviewed By: astreet
Differential Revision: D3855801
Summary:
@public
Introduce `overflow:scroll` so that scrolling can be implemented without the current overflow:visible hackiness. Currently we use AT_MOST to measure in the cross axis but not in the main axis. This was done to enable scrolling containers where children are not constraint in the main axis by their parent. This caused problems for non-scrolling containers though as it meant that their children cannot be measured correctly in the main axis. Introducing `overflow:scroll` fixes this.
Reviewed By: astreet
Differential Revision: D3855801
Summary:
@public
This is to be able to depend on ReactClippingViewGroup from BaseViewManager. Devs using ReactClippingViewGroup may need to update their imports when updating past this commit.
Reviewed By: lexs
Differential Revision: D3835328
Summary:
We were always getting LEFT explicitly, and, due to RTL support, we
should be asking for START instead.
Reviewed By: sriramramani
Differential Revision: D3836816
Summary:
@public
Setting the line height with the help of Android-provided StaticLayout is incorrect. A
simple example app will display the following when `setLineSpacing(50.f, 0.f)`
is set: {F62987699}. You'll notice that the height of the first line is a few
pixels shorter than the other lines.
So we use a custom LineHeightSpan instead, which needs to be applied to the text
itself, and no height-related attributes need to be set on the TextView itself.
Reviewed By: lexs
Differential Revision: D3841658
Summary:
Fix TextInput padding on Nodes. We used to not call super and used
to manually do the setting of padding. This stops us from running the same
logic that non-Nodes runs (we bypassed it), so this fixes it.
Reviewed By: astreet
Differential Revision: D3825227
Summary:
Due to the RTL implementation, the ViewProps spacing array has START
and END, whereas Nodes should deal with RIGHT and LEFT directly (just like
non-Nodes does). This is the same implementation in use by non-Nodes.
Reviewed By: astreet
Differential Revision: D3809028
Summary:
Nodes historically had two image implementations -
DrawImageWithDrawee and DrawImageWithPipeline. The drawee implementation
was the default (per request of the Fresco team). At this point, there is
no point of having two (especially since updates to one need to be made to
the other), so this patch removes pipeline.
Reviewed By: sriramramani
Differential Revision: D3755523
Summary:
Nodes would typically clip its images, and then Fresco would then
re-clip as part of ScaleTypeDrawable - in addition to being unnecessary,
it's also incorrect, beacuse it causes the image to be smaller than it
should be.
Reviewed By: sriramramani
Differential Revision: D3754778
Summary:
@public
Setting the line height with the help of Android-provided StaticLayout is incorrect. A
simple example app will display the following when `setLineSpacing(50.f, 0.f)`
is set: {F62987699}. You'll notice that the height of the first line is a few
pixels shorter than the other lines.
So we use a custom LineHeightSpan instead, which needs to be applied to the text
itself, and no height-related attributes need to be set on the TextView itself.
Reviewed By: lexs
Differential Revision: D3751097
Summary:
@public
Setting the line height with the help of Android-provided StaticLayout is incorrect. A
simple example app will display the following when `setLineSpacing(50.f, 0.f)`
is set: {F62987699}. You'll notice that the height of the first line is a few
pixels shorter than the other lines.
So we use a custom LineHeightSpan instead, which needs to be applied to the text
itself, and no height-related attributes need to be set on the TextView itself.
Reviewed By: lexs
Differential Revision: D3751097
Summary:
This is just a minor cleanup, use constants for the LEFT
and RIGHT alignments, since they are hide.
Reviewed By: sriramramani
Differential Revision: D3746019
Summary: Adds a flag that can be set in FlatViewGroup to see known performance issues in Nodes objects. This is mostly useful in internal development of Nodes, and will be a dead code path when not set.
Reviewed By: ahmedre
Differential Revision: D3732675
Summary: Node region bounds are assumed to equal the underlying node bounds. In the case of hit slop, these need to be abstracted.
Reviewed By: ahmedre
Differential Revision: D3713430
Mani Ghasemlou
Mehdi Mulani
Ahmed El-Helw
Emil Sjolander
Andy Street
Sriram Ramasubramanian
Jonathan Lawlor
Dmitry Petukhov
Felix Oghina
Andrei Coman
Andrew Y. Chen
Seth Kirby
