When setState is called in a widget's state, the corresponding element in the element tree gets marked as dirty, and the widget gets rebuilt. However, how does it handle descendents? For example, the Text widget below gets rebuilt when its ancestor SampleWidgetState gets rebuilt.
Why?
class SampleWidget extends StatefulWidget {
#override
SampleWidgetState createState() => SampleWidgetState();
}
class SampleWidgetState extends State<SampleWidget> {
String text = "text1";
#override
Widget build(BuildContext context) {
return Column(
children: [
Text(text),
ElevatedButton(
child: Text('call SetState'),
onPressed: () {
setState(() {
text = "text2";
});
},
),
],
);
}
}
from Flutter's official documentation, inside Flutter:
In response to user input (or other stimuli), an element can become dirty, for example if the developer calls setState() on the associated state object. The framework keeps a list of dirty elements and jumps directly to them during the build phase, skipping over clean elements. During the build phase, information flows unidirectionally down the element tree, which means each element is visited at most once during the build phase. Once cleaned, an element cannot become dirty again because, by induction, all its ancestor elements are also clean.
I guess this answer what Flutter does under the hood in the updating process of the widget's descendents.
SampleWidgetState is a state class, when you calling the setState() its mean build() method will reinvoke, everything inside will rebuild. thats how its works.
if you want to prevent the descendents to not rebuild, there is several ways,
use const keyword.
warp the widget you want to change its own state, example use StatefullBuilder
refactor widget to statefulwidget so its have its own state
in your case, Text widget consume SampleWidgetState : String text = "text1";, its mean Text widget is not independent, its dependent on that state.
I am trying to create tests for my Flutter application. Simple example:
class MyWidget extends StatelessWidget {
#override
build(BuildContext context) {
return MySecondWidget();
}
}
I would like to verify that MyWidget is actually calling MySecondWidget without building MySecondWidget.
void main() {
testWidgets('It should call MySecondWidget', (WidgetTester tester) async {
await tester.pumpWidget(MyWidget());
expect(find.byType(MySecondWidget), findsOneWidget);
}
}
In my case this will not work because MySecondWidget needs some specific and complex setup (like an API key, a value in a Provider...). What I would like is to "mock" MySecondWidget to be an empty Container (for example) so it doesn't raise any error during the test.
How can I do something like that ?
There is nothing done out of the box to mock a widget. I'm going to write some examples/ideas on how to "mock"/replace a widget during a test (for example with a SizedBox.shrink().
But first, let me explain why I think this is not a good idea.
In Flutter you are building a widget tree. A specific widget has a parent and usually has one or several children.
Flutter chose a single pass layout algorithm for performance reasons (see this):
Flutter performs one layout per frame, and the layout algorithm works in a single pass. Constraints are passed down the tree by parent objects calling the layout method on each of their children. The children recursively perform their own layout and then return geometry up the tree by returning from their layout method. Importantly, once a render object has returned from its layout method, that render object will not be visited again until the layout for the next frame. This approach combines what might otherwise be separate measure and layout passes into a single pass and, as a result, each render object is visited at most twice during layout: once on the way down the tree, and once on the way up the tree.
From this, we need to understand that a parent needs its children to build to get their sizes and then render itself properly. If you remove its children, it might behave completely differently.
It is better to mock the services if possible. For example, if your child makes an HTTP request, you can mock the HTTP client:
HttpOverrides.runZoned(() {
// Operations will use MyHttpClient instead of the real HttpClient
// implementation whenever HttpClient is used.
}, createHttpClient: (SecurityContext? c) => MyHttpClient(c));
If the child needs a specific provider you can provide a dummy one:
testWidgets('My test', (tester) async {
tester.pumpWidget(
Provider<MyProvider>(
create: (_) => MyDummyProvider(),
child: MyWidget(),
),
);
});
If you still want to change a widget with another one during your tests, here are some ideas:
1. Use Platform.environment.containsKey('FLUTTER_TEST')
You can either import Platform from dart:io (not supported on web) or universal_io (supported on web).
and your build method could be:
#override
Widget build(BuildContext context) {
final isTest = Platform.environment.containsKey('FLUTTER_TEST');
if (isTest) return const SizedBox.shrink();
return // Your real implementation.
}
2. Use the annotation #visibleForTesting
You can annotate a parameter (ex: mockChild) that is only visible/usable in a test file:
class MyWidget extends StatelessWidget {
const MyWidget({
#visibleForTesting this.mockChild,
});
final Widget? child;
#override
Widget build(BuildContext context) {
return mockChild ?? // Your real widget implementation here.
}
}
And in your test:
tester.pumpWidget(
MyWidget(
mockChild: MyMockChild(),
),
);
You can mock MySecondWidget (eg using Mockito) but you do need to change your real code to create a MockMySecondWidget when in test mode, so it's not pretty. Flutter does not support object instantiation based on a Type (except through dart:mirrors but that is not compatible with Flutter), so you cannot 'inject' the type as a dependency. To determine if you are in test mode use Platform.environment.containsKey('FLUTTER_TEST') - best to determine this once upon startup and set the result as a global final variable, which will make any conditional statements quick.
One way to do it, is to wrap the child widget into a function, and pass the function to parent widget's constructor:
class MyWidget extends StatelessWidget {
final Widget Function() buildMySecondWidgetFn;
const MyWidget({
Key? key,
this.buildMySecondWidgetFn = _buildMySecondWidget
}): super(key: key);
#override
build(BuildContext context) {
return buildMySecondWidgetFn();
}
}
Widget _buildMySecondWidget() => MySecondWidget();
Then you can make up your mock widget, pass it thru buildMySecondWidgetFn in test.
Looking at the StatefulWidget usage, I am wondering about the design decision to have circular dependency like this:
class YellowBird extends StatefulWidget {
const YellowBird({ Key key }) : super(key: key);
#override
_YellowBirdState createState() => _YellowBirdState();
}
class _YellowBirdState extends State<YellowBird> {
#override
Widget build(BuildContext context) {
return Container(color: const Color(0xFFFFE306));
}
}
example is taken from Flutter StatefulWidget docs
Any thoughts?
This is (one of the) correct way's to build a stateful widget and is properly in line with the flutter documentation.
This is not a circular dependency. You are defining a Stateful Widget class, and a corresponding State class.
The Widget contains a State object (or rather a sub-class of it), using the principle of composition-over-inheritance.
From the Stateful widget documentation:
StatefulWidget instances themselves are immutable and store their mutable state either in separate State objects that are created by the createState method, or in objects to which that State subscribes, for example Stream or ChangeNotifier objects, to which references are stored in final fields on the StatefulWidget itself.
This way Flutter framework can better manage the manipulation and change of this Widget's data. For example (from the link above):
[...] multiple State objects might be associated with the same StatefulWidget if that widget has been inserted into the tree in multiple places.
This principle of composition-over-inheritance is a core-concept of Flutter framework and most IDE's can create this boilerplate code for you automatically (e.g. typing stful on AndroidStudio suggests the StatefulWidgets class and its State related class) so you don't have to worry about it.
In a flutter app, let say I have this widget class with multiple widgets inside (ie. just a typical single long Widget build() class with multiple widgets inside). Then these are split into multiple children widgets, with their own classes: As an example,
Before:
class Parents extend StatelessWidget{
Widget build(BuildContext context){
//Parent
return Column{
children: <Widget>[
//Child 1
Container('something inside'),
//Child 2
Container('something inside'),
//Child 3
Container('something inside'),
...
]
}
}
}
Now:
class Parents extend StatelessWidget{
Widget build(BuildContext context){
//Parent
return Column{
children: <Widget>[
//Child 1
myContainer('first child'),
//Child 2
myContainer('first child'),
//Child 3
myContainer('first child'),
...
]
}
}
}
class myContainer extend StatelessWidget{
Widget build(BuildContext context){
//child reusable widget
return Container{
'something inside'
}
}
}
The question that I have is this Widget build(BuildContext context).
In the above example, I called the myContainer class three times in the parent class. In my mind, it means that the build widget is called four times (one with parent, 3 times from each child).
I mean, I saw a bunch of examples online that above approach is recommended, but is it really a proper way of doing it? I may not understand the flutter completely yet, but since it is a widget tree, would it be more efficient (in terms of performance wise) to simply pass the parent context down to the children? like below:
class myContainer extend StatelessWidget{
final BuildContext parentContext;
Widget build(parentContext){
//child reusable widget
return Container{
'something inside'
}
}
}
Both approaches seem to work but wanted to see if I am way off with my way of thinking. I don't fully understand the mechanism of Context and any clarification would be super appreciated!
Thanks guys!
From the docs:
Each widget has its own BuildContext, which becomes the parent of the
widget returned by the StatelessWidget.build or State.build function.
(And similarly, the parent of any children for RenderObjectWidgets.)
In particular, this means that within a build method, the build
context of the widget of the build method is not the same as the build
context of the widgets returned by that build method. This can lead to
some tricky cases. For example, Theme.of(context) looks for the
nearest enclosing Theme of the given build context. If a build method
for a widget Q includes a Theme within its returned widget tree, and
attempts to use Theme.of passing its own context, the build method for
Q will not find that Theme object. It will instead find whatever Theme
was an ancestor to the widget Q. If the build context for a subpart of
the returned tree is needed, a Builder widget can be used: the build
context passed to the Builder.builder callback will be that of the
Builder itself.
So, this basically means that the BuildContext inside the build() method is actually that of it's parent. Hence, their is no need to explicitly pass it.
I have realized that it is possible to create widgets using plain functions instead of subclassing StatelessWidget. An example would be this:
Widget function({ String title, VoidCallback callback }) {
return GestureDetector(
onTap: callback,
child: // some widget
);
}
This is interesting because it requires far less code than a full-blown class. Example:
class SomeWidget extends StatelessWidget {
final VoidCallback callback;
final String title;
const SomeWidget({Key key, this.callback, this.title}) : super(key: key);
#override
Widget build(BuildContext context) {
return GestureDetector(
onTap: callback,
child: // some widget
);
}
}
So I've been wondering: Is there any difference besides syntax between functions and classes to create widgets? And is it a good practice to use functions?
Edit: The Flutter team has now taken an official stance on the matter and stated that classes are preferable. See https://www.youtube.com/watch?v=IOyq-eTRhvo
TL;DR: Prefer using classes over functions to make reusable widget-tree.
EDIT: To make up for some misunderstanding:
This is not about functions causing problems, but classes solving some.
Flutter wouldn't have StatelessWidget if a function could do the same thing.
Similarly, it is mainly directed at public widgets, made to be reused. It doesn't matter as much for private functions made to be used only once – although being aware of this behavior is still good.
There is an important difference between using functions instead of classes, that is: The framework is unaware of functions, but can see classes.
Consider the following "widget" function:
Widget functionWidget({ Widget child}) {
return Container(child: child);
}
used this way:
functionWidget(
child: functionWidget(),
);
And it's class equivalent:
class ClassWidget extends StatelessWidget {
final Widget child;
const ClassWidget({Key key, this.child}) : super(key: key);
#override
Widget build(BuildContext context) {
return Container(
child: child,
);
}
}
used like that:
new ClassWidget(
child: new ClassWidget(),
);
On paper, both seem to do exactly the same thing: Create 2 Container, with one nested into the other. But the reality is slightly different.
In the case of functions, the generated widget tree looks like this:
Container
Container
While with classes, the widget tree is:
ClassWidget
Container
ClassWidget
Container
This is important because it changes how the framework behaves when updating a widget.
Why that matters
By using functions to split your widget tree into multiple widgets, you expose yourself to bugs and miss on some performance optimizations.
There is no guarantee that you will have bugs by using functions, but by using classes, you are guaranteed to not face these issues.
Here are a few interactive examples on Dartpad that you can run yourself to better understand the issues:
https://dartpad.dev/?id=bcae5878ccced764b35dd9a659a593db
This example showcases how by splitting your app into functions,
you may accidentally break things like AnimatedSwitcher
https://dartpad.dev/?id=481a2c301c2e4bed6c30ba651d01bacb
This example showcases how classes allow more granular rebuilds of the
widget tree, improving performances
https://dartpad.dev/?id=8bcb85ba535102bed652e5bf1540ac3b
This example showcases how, by using functions, you expose yourself
to misusing BuildContext and facing bugs when using InheritedWidgets (such as Theme or providers)
Conclusion
Here's a curated list of the differences between using functions and classes:
Classes:
allow performance optimization (const constructor, more granular rebuild)
ensure that switching between two different layouts correctly disposes of the resources (functions may reuse some previous state)
ensures that hot-reload works properly (using functions could break hot-reload for showDialogs & similar)
are integrated into the widget inspector.
We see ClassWidget in the widget-tree showed by the devtool, which
helps understanding what is on screen
We can override debugFillProperties to print what the parameters passed to a widget are
better error messages
If an exception happens (like ProviderNotFound), the framework will give you the name of the currently building widget.
If you've split your widget tree only in functions + Builder, your errors won't have a helpful name
can define keys
can use the context API
Functions:
have less code (which can be solved using code-generation functional_widget)
Overall, it is considered a bad practice to use functions over classes for reusing widgets because of these reasons.
You can, but it may bite you in the future.
I've been researching on this issue for the past 2 days. I came to the following conclusion: it is OKAY to break down pieces of the app into functions. It's just ideal that those functions return a StatelessWidget, so optimisations can be made, such as making the StatelessWidget const, so it doesn't rebuild if it doesn't have to.
For example, this piece of code is perfectly valid:
import 'package:flutter/material.dart';
void main() => runApp(MyApp());
class MyApp extends StatelessWidget {
#override
Widget build(BuildContext context) {
return MaterialApp(
title: 'Flutter Demo',
theme: ThemeData(
primarySwatch: Colors.blue,
),
home: MyHomePage(title: 'Flutter Demo Home Page'),
);
}
}
class MyHomePage extends StatefulWidget {
MyHomePage({Key key, this.title}) : super(key: key);
final String title;
#override
_MyHomePageState createState() => _MyHomePageState();
}
class _MyHomePageState extends State<MyHomePage> {
int _counter = 0;
void _incrementCounter() {
setState(() {
++_counter;
});
}
#override
Widget build(BuildContext context) {
return Scaffold(
appBar: AppBar(
title: Text(widget.title),
),
body: Center(
child: Column(
mainAxisAlignment: MainAxisAlignment.center,
children: <Widget>[
Text(
'You have pushed the button this many times:',
),
Text(
'$_counter',
style: Theme.of(context).textTheme.display1,
),
const MyWidgetClass(key: const Key('const')),
MyWidgetClass(key: Key('non-const')),
_buildSomeWidgets(_counter),
],
),
),
floatingActionButton: FloatingActionButton(
onPressed: _incrementCounter,
tooltip: 'Increment',
child: Icon(Icons.add),
), // This trailing comma makes auto-formatting nicer for build methods.
);
}
Widget _buildSomeWidgets(int val) {
print('${DateTime.now()} Rebuild _buildSomeWidgets');
return const MyWidgetClass(key: Key('function'));
// This is bad, because it would rebuild this every time
// return Container(
// child: Text("hi"),
// );
}
}
class MyWidgetClass extends StatelessWidget {
const MyWidgetClass({Key key}) : super(key: key);
#override
Widget build(BuildContext context) {
print('${DateTime.now()} Rebuild MyWidgetClass $key');
return Container(
child: Text("hi"),
);
}
}
The use of function there is perfectly fine, as it returns a const StatelessWidget. Please correct me if I'm wrong.
1 - Most of the time build method (child widget) calls number of synchronous and asynchronous functions.
Ex:
To download network image
get input from users etc.
so build method need to keep in the separate class widget (because all other methods call by build() method can keep in one class)
2 - Using the widget class you can create a number of other classes without writing the same code again and again (** Use Of Inheritance** (extends)).
And also using inheritance(extend) and polymorphism (override) you can create your own custom class.
(Down below example, In there I will customize (Override) the animation by extending MaterialPageRoute (because its default transition I want to customize).👇
class MyCustomRoute<T> extends MaterialPageRoute<T> {
MyCustomRoute({ WidgetBuilder builder, RouteSettings settings })
: super(builder: builder, settings: settings);
#override //Customize transition
Widget buildTransitions(BuildContext context,
Animation<double> animation,
Animation<double> secondaryAnimation,
Widget child) {
if (settings.isInitialRoute)
return child;
// Fades between routes. (If you don't want any animation,
// just return child.)
return new FadeTransition(opacity: animation, child: child);
}
}
3 - Functions cannot add conditions for their parameters, But using the class widget's constructor You can do this.
Down below Code example👇 (this feature is heavily used by framework widgets)
const Scaffold({
Key key,
this.bottomNavigationBar,
this.bottomSheet,
this.backgroundColor,
this.resizeToAvoidBottomPadding,
this.resizeToAvoidBottomInset,
this.primary = true,
this.drawerDragStartBehavior = DragStartBehavior.start,
this.extendBody = false,
this.extendBodyBehindAppBar = false,
this.drawerScrimColor,
this.drawerEdgeDragWidth,
}) : assert(primary != null),
assert(extendBody != null),
assert(extendBodyBehindAppBar != null),
assert(drawerDragStartBehavior != null),
super(key: key);
4 - Functions cannot use const and the Class widget can use the const for their constructors.
(that affect the performance of the main thread)
5 - You can create any number of independent widgets using the same class (instances of a class/objects)
But function cannot create independant widgets(instance), but reusing can.
[each instance has its own instance variable and that is completely independant from other widgets(object), But function's local variable is dependant on each function call* (which means, when you change a value of a local variable it affect for all other parts of the application which use this function)]
There were many Advantages in class over functions..(above are few use cases only)
My Final Thought
So don't use Functions as building blocks of your application, use them only for doing Operations.
Otherwise, it causes many unchangeable problems when your application gets scalable.
Use functions for doing a small portion of the task
Use class as a building block of an application(Managing application)
As Remi has eloquently put repeatedly, it's not the functions by themselves that cause a problem, the problem is us thinking that using a function has a similar benefit to using a new widget.
Unfortunately this advice is evolving into "the act of merely using a function is inefficient", with often incorrect speculations into why this might be.
Using a function is almost the same as using what the function returns in place of that function. So, if you are calling a widget constructor and giving it as a child to another widget, you are not making your code inefficient by moving that constructor call into a function.
//...
child: SomeWidget(),
//...
is not significantly better in terms of efficiency than
//...
child: buildSomeWidget();
//...
Widget buildSomeWidget() => SomeWidget();
It is fine to argue the following about the second one:
It's ugly
It's unnecessary
I don't like it
Function does not appear in Flutter Inspector
Two functions may not work with AnimatedSwitcher et al.
It does not create a new context, so you can't reach the Scaffold above it through context
If you use ChangeNotifier in it, its rebuild is not contained within the function
But it's not correct to argue this:
Using a function is inefficient in terms of performance
Creating a new widget brings these performance benefits:
ChangeNotifier within it does not make its parent rebuild upon changes
Sibling widgets are protected from each other's rebuilds
Creating it with const (if possible) protects it from parent's rebuilds
You are more likely to keep your const constructor if you can isolate the changing children to other widgets
However, if you do not have any of these cases, and your build function is looking more and more like pyramid of doom, it is better to refactor a part of it to a function rather than keeping the pyramid. Especially if you are enforcing 80 character limit, you may find yourself writing code in about 20 character-wide space. I see a lot of newbies falling into this trap. The message to those newbies should be "You should really be creating new widgets here. But if you can't, at least create a function.", not "You have to create a widget or else!". Which is why I think we have to be more specific when we promote widgets over functions and avoid being factually incorrect about efficiency.
For your convenience, I have refactored Remi's code to show that the problem is not simply using functions, but the problem is avoiding creating new widgets. So, if you were to place the widget-creating code in those functions into where the functions are called (refactor-inline) you have the exact same behavior as using functions, but without using functions! So, it's not using functions that's the problem, it's the avoidance of creating new widget classes.
(remember to turn off null safety as the original code is from 2018)
Here are a few interactive examples on Dartpad that you can run
yourself to better understand the issues:
https://dartpad.dev/1870e726d7e04699bc8f9d78ba71da35 This example
showcases how by splitting your app into functions, you may
accidentally break things like AnimatedSwitcher
Non-function version: https://dartpad.dev/?id=ae5686f3f760e7a37b682039f546a784
https://dartpad.dev/a869b21a2ebd2466b876a5997c9cf3f1 This example
showcases how classes allow more granular rebuilds of the widget tree,
improving performances
Non-function version: https://dartpad.dev/?id=795f286791110e3abc1900e4dcd9150b
https://dartpad.dev/06842ae9e4b82fad917acb88da108eee This example
showcases how, by using functions, you expose yourself to misusing
BuildContext and facing bugs when using InheritedWidgets (such as
Theme or providers)
Non-function version: https://dartpad.dev/?id=65f753b633f68503262d5adc22ea27c0
You will find that not having them in a function creates the exact same behavior. So it's adding widgets that gives you the win. It's not adding functions that creates a problem.
So the suggestions should be:
Avoid the pyramid of doom at any cost! You need horizontal space to code. Don't get stuck at the right margin.
Create functions if you need, but do not give parameters to them as it's impossible to find the line that calls the function through Flutter Inspector.
Consider creating new widget classes, it's the better way! Try Refactor->Extract Flutter Widget. You won't be able to if your code is too coupled with the current class. Next time you should plan better.
Try to comment out things that prevent you from extracting a new widget. Most likely they are function calls in the current class (setState, etc.). Extract your widget then, and find ways of adding that stuff in. Passing functions to the constructor may be ok (think onPressed). Using a state management system may be even better.
I hope this can help remind why we prefer widgets over functions and that simply using a function is not a huge problem.
Edit: one point that was missed in this whole discussion: when you widgetize, siblings don't rebuild each other anymore. This Dartpad demonstrates this: https://dartpad.dartlang.org/?id=8d9b6d5bd53a23b441c117cd95524892
When you are calling the Flutter widget make sure you use the const keyword. For example const MyListWidget();
In case this helps anyone passing this way, some things I have in my conceptual model of Flutter developed from this question and working with Flutter in general (caveat: I could still be deeply confused and wrong about this stuff).
A Widget is what you want and the Elements are what you have. It is the job of the rendering engine to reconcile the two as efficiently as possible.
Use Keys, they can help a lot.
A BuildContext is an Element.
Any Thing.of(context) is likely to introduce a build dependency. If Thing changes it will trigger a rebuild from the context element.
In your build() if you access a BuildContext from a nested widget you are acting on the Element at the top of your subtree.
Widget build(BuildContext rootElement) {
return Container(
child:Container(
child:Container(
child:Text(
"Depends on rootElement",
// This introduces a build trigger
// If ThemeData changes a rebuild is triggered
// on rootElement not this Text()-induced element
style:Theme.of(rootElement).textTheme.caption,
),
),
),
);
}
AnimatedSwitcher is a slippery beast - it has to be able to distinguish its children. You can use functions if they return different types or return the same type but with different Keys
If you are authoring a Widget use a class not a Function but feel free to refactor your 1000 line build() method with functions/methods, the outcome is identical*.
* but could be even better to refactor into classes