Summary:
It was just adding unnecessary complexity. Users should just use standard React perf best practices, like `PureComponent` and `shouldComponentUpdate`.
This should be backwards compatible - existing `shouldItemUpdate` usage will just be ignored and should consider migrating to this pattern:
```
class MyItem extends React.PureComponent {
_onPress = () => {
this.props.onPressItem(this.props.id);
};
render() {
return (
<SomeOtherWidget title={this.props.title} onPress={this._onPress} />
)
}
}
...
_renderItem = ({item}) => (
<MyItem onPressItem={this._onPressItem} title={item.title} id={item.id} />
);
```
Which will automatically prevent re-renders unless the relavent data changes.
Reviewed By: yungsters
Differential Revision: D4730599
fbshipit-source-id: 0f61efe96eb4d95bb3b7c4ec889e3e0e34436e56
Summary:
This adds support for both automagical sticky section headers in
`SectionList` as well as the more free-form `stickyHeaderIndices` on
`FlatList` or `VirtualizedList`.
The basic concept is to take the initial `stickySectionHeaders` and remap them
to the indices corresponding to the mounted subset in the render window. The
main trick here is that the currently stuck header might itself be outside of
the render window, so we need to search the gap to see if that's the case and
render it (with spacers above and below it instead of one big spacer).
In the `SectionList` we simply pre-compute the sticky headers at the same time
as when we scan the sections to determine the flattened length and pass those
to `VirtualizedList`.
This also requires some updates to `ScrollView` to work in the churny
environment of `VirtualizedList`. We propogate the keys on the children to the
animated wrappers so that as items are removed and the indices of the
remaining items change, react can keep proper track of them. We also fix the
scroll back case where new headers are rendered from the top down and aren't
updated with the `setNextLayoutY` callback because the `onLayout` call for the
next header happened before it was mounted. This is done by just tracking all
the layout values in a map and providing them to the sticky components at
render time. This might also improve perf a little by property configuring the
animations syncronously instead of waiting for the `onLayout` callback. We
also need to protect against stale onLayout callbacks and other fun stuff.
== Test Plan ==
https://www.facebook.com/groups/react.native.community/permalink/940332509435661/
Scroll a lot with and without debug mode on. Make sure spinner
still spins and there are no crashes (lots of crashes during development due
to the animated configuration being non-monotonic if anything stale values get
through). Also made sure that tapping a row to change it's height would
properly update the animation configurations so the collision point would
still be correct.
Reviewed By: yungsters
Differential Revision: D4695065
fbshipit-source-id: 855c4e31c8f8b450d32150dbdb2e07f1a9f9f98e
Summary: People might be tempted to try and scrollTo an index that hasn't been rendered yet, which is broken, so instead of jank let's throw.
Reviewed By: yungsters
Differential Revision: D4727402
fbshipit-source-id: b6f9fd5b70b6f076c30141d00b2b9e2a51b14e87
Summary:
We recommend using `react-navigation` over `Navigator`. Adds a link to the new `native-navigation` component as well.
Did not test website generation, this is a comments only edit that should work fine.
Closes https://github.com/facebook/react-native/pull/12963
Differential Revision: D4749072
Pulled By: hramos
fbshipit-source-id: 4506630306c44b24b95c4f5d5a42c1caa9e2cd4e
Summary:
Bug in Android https://code.google.com/p/android/issues/detail?id=33868 causes the RN catalyst instrumentation test to fail with
```
java.lang.ArrayIndexOutOfBoundsException: length=253; index=-1
at android.text.StaticLayout.calculateEllipsis(StaticLayout.java:667)
at android.text.StaticLayout.out(StaticLayout.java:631)
at android.text.StaticLayout.generate(StaticLayout.java:423)
...
```
The fix is to set singleLine to true when there is only one line of text
Reviewed By: AaaChiuuu
Differential Revision: D4562000
fbshipit-source-id: 84248e3982063b767e8b0465effe2321b54a7fa2
Summary: Only pulls in `EventValidator` for development mode, as warnings about invalid events are pointless in production builds.
Reviewed By: javache
Differential Revision: D4745852
fbshipit-source-id: dbab1026df35d54a82e1e620fac08304c58fbeae
Summary: Finally adding some unit test to increase confidence in the correctness of that piece of code.
Reviewed By: davidaurelio
Differential Revision: D4721543
fbshipit-source-id: 56776290d61f2b51c69d7eeae09663e3bc892b50
Summary: `invariant` is only available in open source because we install it as a transitive dependency into node_modules
Reviewed By: jeanlauliac
Differential Revision: D4745582
fbshipit-source-id: 27c49b576254c8d1d667dea7097d16cdd1205daf
Summary:
Motivation: Fixes Xcode warning `Ivar '_websocketExecutorName' which backs the property is not referenced in this property's accessor` which shows up because this property has no setter (and is never set anywhere).
Closes https://github.com/facebook/react-native/pull/12639
Differential Revision: D4745437
Pulled By: javache
fbshipit-source-id: 3ab4b0df62f90adc2b62d891197dc783e07da4e3
Summary:
Fix `NullPointerException` in `ReactShadowNode.toString`
Simplified `ReactShadowNode.hasNewLayout` since I was already in there
It seems to me unlikely that this bug impacts anything but the debugging experience, so no biggie.
Closes https://github.com/facebook/react-native/pull/12953
Differential Revision: D4739215
fbshipit-source-id: 94955cc783216fdb8868fc8d08010e0d8a238052
Summary:
While working with `RCTEventEmitter` I noticed that if an event is emitted before `_listenerCount` is updated, it will not go through because the listeners count hasn't been updated. Moving the count update before the invokation of `startObserving` and `stopObserving` fixes the issue. Same way if you remove the last listener and an event is fired before the count is updated (while it shouldn't be fired).
**Test plan (required)**
An easy test to demonstrate it is to implement `startObserving` to synchronously fire an event. Without the change, a warning is thrown, with the change, the event is fired. Not very strong on Obj-C here and I didn't know how to mock out the native stuff. Would be glad to write a failing unit test tho :)
Closes https://github.com/facebook/react-native/pull/11907
Differential Revision: D4738965
Pulled By: javache
fbshipit-source-id: cf175051be5b9c5de761d3dcd290560e1639b05e
Summary:
- [x] Explain the **motivation** for making this change.
- [x] Provide a **test plan** demonstrating that the code is solid.
- [x] Match the **code formatting** of the rest of the codebase.
- [x] Target the `master` branch, NOT a "stable" branch.
The PR #11613 (0.43) removed this missing `toggleElementInspector` event send when `jsLoaded` in DevMenu (Now is DevSettings), it should open the inspector if `isElementInspectorShown` is true when we reload JS.
The dev menu text `Show / Hide Inspector` is dependent on `isElementInspectorShown` bool value.
([This behavior in 0.42](https://github.com/facebook/react-native/blob/0.42-stable/React/Modules/RCTDevMenu.mm#L436-L442))
Manual testing in UIExplorer:
* Open the dev menu and click `Show Inspector`
* Open the dev menu and click `Reload JS`
* The built-in inspector should keep open (dev menu text: `Hide Inspector`)
Closes https://github.com/facebook/react-native/pull/12999
Differential Revision: D4738959
Pulled By: javache
fbshipit-source-id: b3f584db51aa0e1b463c52003967b00bcd81bc99
Summary:
The middleware for automatically converting Systrace traces to HTML and popping the browser hasn't worked properly for a while, since the version on Homebrew generates some code that uses `Object.observe`, which was deleted from Chrome ages ago. People have complained about it, but fixing it properly has proven to be harder than expected, so I suggest we simply update the message with instructions for people to load it on Chrome, which is what all of us have been doing anyway (AFAIK).
Closes https://github.com/facebook/react-native/pull/12445
Reviewed By: javache
Differential Revision: D4700153
Pulled By: gaearon
fbshipit-source-id: 0c33099babed93b3c70d36ae9dfc7d82460c8269
Summary: When requiring a module that has previously errored, the implementation of `require` only used the numerical module ID. In this diff, we enable usage of the verbose module name if present.
Reviewed By: bestander
Differential Revision: D4737723
fbshipit-source-id: 1c2d3906435a637f3e440e57f904489d84495bd2
Summary: The logic of the `inline` babel transform regarded identifiers as global if no binding exists for them. We extend that logic to also accept flow-declared variables.
Reviewed By: arcanis
Differential Revision: D4737620
fbshipit-source-id: e71cfdf77c7b7751265cfa4412430b4f29e9e853
Summary: Fixes the messed-up indentation of `polyfills/require.js`. Over half of the lines were indented with an odd number of spaces.
Reviewed By: arcanis, bestander
Differential Revision: D4737435
fbshipit-source-id: a5b9baf0a27f236a4d3d6b6c1c5a92f52859f62c
Summary:
Several things:
* The mess with insets was fixed. Previously we tried to compensate the insets difference with `UITextField` by adjusting `textContainerInset` property, moreover we delegated negative part of this compensation to the view inset. That was terrible because it breaks `contentSize` computation, complicates whole insets consept, complicates everything; it just was not right. Now we are fixing the top and left inset differences in different places. We disable left and right 5pt margin by setting `_textView.textContainer.lineFragmentPadding = 0` and we introduce top 5px inset as a DEFAULT value for top inset for common multiline <TextInput> (this value can be easilly overwritten in Javascript).
* Internal layout and contentSize computations were unified and simplified.
* Now we report `intrinsicContentSize` value to Yoga, one step before auto-expandable TextInput.
Depends on D4640207.
Reviewed By: mmmulani
Differential Revision: D4645921
fbshipit-source-id: da5988ebac50be967caecd71e780c014f6eb257a
Summary: `autoresizingMask` is supposed to be set outside self class, this is UIKit convention.
Reviewed By: mmmulani
Differential Revision: D4697098
fbshipit-source-id: 7e0aa5d3032184de980b3cecafebbc4ce8ef9ada
Summary:
The Math Strikes Back
Several related things:
* When we specify `scale: 0;` style for some view it ends up with calling `CATransform3DScale` with zero scale parameter.
* In this case `CATransform3DScale` returns transform matrix full of zeros. It actually depends on representation and matrix-type (2d or 3d) but in UIView debugger it appears as [0, 0, 0, 0, ...]. And probably it is correct result.
* By default, for hit-testing, UIKit uses specially optimized logic based on GPU/CALayer infrastructure under the hood. And the transform matrix full of zeros breaks this algorithm. I guess, it happens because zero-matrix doesn't quite make sense.
So, `scale: 0;` is a weird edge case, and in this diff, we are trying to illuminate it by replacing with epsilon value.
Related SO issues:
http://stackoverflow.com/questions/25964224/cgaffinetransformscale-not-working-with-zero-scalehttp://stackoverflow.com/questions/7937369/animate-uiview-scale-to-zero
Reviewed By: blairvanderhoof
Differential Revision: D4734475
fbshipit-source-id: 7241cdffa86c05a6552860a25789e2281588ba23
Summary:
Some projects define multiple targets, including app extensions, which are built with a “.appex” extension. This fix prevents the buildProject method from selecting any app extension (e.g. a Today.appex today-widget extension) as the product name.
Thanks for submitting a PR! Please read these instructions carefully:
- [X] Explain the **motivation** for making this change.
- [X] Provide a **test plan** demonstrating that the code is solid.
- [X] Match the **code formatting** of the rest of the codebase.
- [X] Target the `master` branch, NOT a "stable" branch.
When building our workspace, ReactNative was failing to install the app to the simulator because it calculated an incorrect path to the app itself. It was attempting to install "Today.app" when it should have been installing "Remitly.app". I discovered that ReactNative parses the build output to identify the generated app name, and that this was broken when the build also generated an app extension. The f
Closes https://github.com/facebook/react-native/pull/13001
Differential Revision: D4735360
fbshipit-source-id: afeeb2073ccd65c95916b153fcde574b5343af8c
Summary:
Thanks for submitting a PR! Please read these instructions carefully:
- [x] Explain the **motivation** for making this change.
- [x] Provide a **test plan** demonstrating that the code is solid.
- [x] Match the **code formatting** of the rest of the codebase.
- [x] Target the `master` branch, NOT a "stable" branch.
There was a typo in the FlatList docs for numColumns affecting the formatting of its description.
N/A
Sign the [CLA][2], if you haven't already.
Small pull requests are much easier to review and more likely to get merged. Make sure the PR does only one thing, otherwise please split it.
Make sure all **tests pass** on both [Travis][3] and [Circle CI][4]. PRs that break tests are unlikely to be merged.
For more info, see the ["Pull Requests"][5] section of our "Contributing" guidelines.
[1]: https://medium.com/martinkonicek/what-is-a-test-plan-8bfc840ec171#.y9lcuqqi9
[2]: https://code.facebook.com/cla
Closes https://github.com/facebook/react-native/pull/13002
Differential Revision: D4735343
fbshipit-source-id: f781c50c892d64e69f61aec980614e948a48f48b
Spencer Ahrens
Marc Horowitz
Jeremi Stadler
Aaron Chiu
Héctor Ramos
Andrew Y. Chen
Valentin Shergin
Pieter De Baets
Jean Lauliac
David Aurelio
Ben Roth
Brian Vaughn
Petter Hesselberg
Daniele Conti
Jhen
Jeff Thomas
Dan Abramov
Pieter De Baets
Martin Konicek
Adam Ernst
Nick
Arman Dezfuli-Arjomandi