app_state 0.6.6 app_state: ^0.6.6 copied to clipboard
A BLoC-based state management solution on top of Router API for larger apps.
A bloc-based state management solution on top of Router API for larger apps.
This is a routing library. See how it compares to auto_route and go_router.
See tons of runnable examples here.
- Unique Features
- Architecture
- The Bare Minimal App
- Pushing and Popping Pages
- Web Architecture
- Multiple Tabs with Independent Stacks
- Advanced Ways to Return Result
- Help is Wanted
Unique Features #
Blocs First #
In the core there is a stack of Page Blocs instead of routes or widgets.
These Blocs are then wrapped into widgets to form the navigator's stack.
This simplifies state management: no need for providers,
stateful widgets, manual bloc creation and disposal, or BuildContext
.
Push-pop Type Safety at Compile Time #
Page Blocs are typed and can only pop what is allowed. And the Page entry you push to the stack is also typed, so your awaited type is also guaranteed.
Stack Recovery from URL #
When a URL is typed in, the page stack is recovered by the rules you define. If you navigate to
/books/123/rate
, you may get a stack of 3 pages: The Book List at the bottom, the Book Details,
and Rate dialog at the top, all can be popped by the back button.
Unlike with beamer, the rules can be any, and not only derived from
path segments.
Dialog Awaiting Survives the App Restart #
Blocs listen to events from other blocs above them. When the top bloc pops, its result is passed
to didPopNext
method of a bloc under it. So you get the result without awaiting a future.
If you type in /books/123/rate
and close the dialog by rating a book, the book details bloc
will get the rate without any future. See and run
the license dialog example.
Architecture #
The main state unit is PageStackBloc
. You normally create it as a global object
available to all code, or you provide it with get_it.
It contains the stack at runtime:
For each page, there are 3 main units:
- PageBloc.
This is the heart of a page during its life cycle.
If your screen was a
StatefulWidget
in the traditional architecture, thenPageBloc
plays the role of itsState
. It contains everything to preserve so the screen itself can be a stateless widget. However,PageBloc
is richer than aState
. It can produce a stream of output as the purpose of BLoC suggests. It is also aware of the page stacking and can handle events related to it. - Screen.
Most often this is just a stateless widget with
PageBloc
as the single argument. - Page.
Navigator
in Flutter accepts the list ofPage
objects to maintain the stack of routes that are displayed. SoPage
is a necessary adapter for the pair of bloc and screen to show in the app.
These three together are collectively referred to as a 'page' (lowercase) to distinguish it from Page
class.
Zooming in, this is how they interact:
The Bare Minimal App #
This app has one screen and no navigation. It even has no page bloc, since it has no state and the blocs are optional. See and run the example project.
import 'package:app_state/app_state.dart';
import 'package:flutter/material.dart';
final pageStackBloc = PageStackBloc(bottomPage: HomePage());
final _routerDelegate = PageStackRouterDelegate(pageStackBloc);
void main() => runApp(MyApp());
class HomePage extends StatelessMaterialPage {
HomePage() : super(key: const ValueKey('Home'), child: HomeScreen());
}
class HomeScreen extends StatelessWidget {
@override
Widget build(BuildContext context) {
return Scaffold(
appBar: AppBar(title: const Text('Hello World with app_state!')),
);
}
}
class MyApp extends StatelessWidget {
@override
Widget build(BuildContext context) {
return MaterialApp.router(
routerDelegate: _routerDelegate,
routeInformationParser: const PageStackRouteInformationParser(),
);
}
}
Pushing and Popping Pages #
Pushing #
To push a screen, you create its Page
object and push
it into a PageStackBloc
:
final result = await pageStacksBloc.push(
BookDetailsPage(bookId: id),
);
You can do this from anywhere and don't need BuildContext
.
The Android back button and the default BackButton
in Scaffold
just work.
See and run this example with a book list. It has no URLs so far, we will get to them soon.
Programmatic Popping #
To pop a page, call pop()
on its bloc with an optional return value:
onPressed: () => bloc.pop(result);
Overriding the Back Button #
In your bloc, override onBackPressed()
:
@override
Future<BackPressedResult> onBackPressed() {
if (!_saved) {
return Future.value(BackPressedResult.keep);
}
return Future.value(BackPressedResult.close);
}
Page Keys #
Each page in a stack must have a unique key. If you push a page with an existing key,
the onDuplicateKey
argument of pop
determines what to do. It can:
- Bring the older page up and drop the new page (the default).
- Drop the older page and show the new one.
- Raise an exception.
Pages of the same class may have different keys. In a social app, you may have a user page with the user ID in the key. This allows you to show multiple user profiles in the stack, but if a duplicate user profile is about to be shown, you can bring the old one above to save memory and to not bore your user with many back taps later.
Web Architecture #
All the above examples work in web, but the URL always stays /
. Here is how to support URLs.
Parsing URLs #
PagePath
class is an object representation of a URL. You subclass it for every
page you support. For example, BookListPath
will likely have no arguments,
but a BookDetailsPath
will likely have final int id;
You then create a URL parser that is called by Flutter on start-up
and also on back and forward navigation.
This parser chooses the particular PagePath
.
It is easiest to maintain if made of one-liners like this:
class MyRouteInformationParser extends PageStackRouteInformationParser {
@override
Future<PagePath> parsePagePath(RouteInformation ri) async {
return
BookDetailsPath.tryParse(ri) ??
const BookListPath(); // The default page if nothing worked.
}
}
You use this parser instead of the ordinary PageStackRouteInformationParser
in your app constructor.
See and run this example
that adds URL support to the earlier book list example.
Recommended PagePath Structure #
This is the class from the above example.
class BookDetailsPath extends PagePath {
final int bookId;
static final _regExp = RegExp(r'^/books/(\d+)$');
BookDetailsPagePath({
required this.bookId,
}) : super(
key: BookDetailsPage.formatKey(bookId: bookId),
factoryKey: BookDetailsPage.classFactoryKey,
state: {'bookId': bookId},
);
@override
String get location => '/books/$bookId';
static BookDetailsPath? tryParse(RouteInformation ri) {
final matches = _regExp.firstMatch(ri.location ?? '');
if (matches == null) return null;
final bookId = int.tryParse(matches[1] ?? '');
if (bookId == null) {
return null; // Will never get here with present _regExp.
}
return BookDetailsPath(
bookId: bookId,
);
}
@override
get defaultStackPaths => [
const BookListPath(),
this,
];
}
tryParse
is the recommended static method in each of yourPagePath
classes. You then call them in a chain to return the one that worked. Here it applies the regular expression to getbookId
.location
getter is the reverse. It returns the path of the URL to put in the address bar in case the location changes programmatically in the app and not as a result of the address bar change. If not present, the URL will be/
.defaultStackPaths
returns a list ofPagePath
objects to pre-populate the stack with, bottom to top. In this case, the user has typed in a particular book's URL, so we compose the initial stack of two pages. At the bottom is the book list page, and above it is this one we parsed. So when this stack is loaded, the back button would lead the user to the book list page. If this getter is not present, the stack defaults to this single page.super.key
is to diff the pages in the stack to see if after the URL change some pages need to be kicked out of the stack, updated, or added.super.state
is the map of all fields. It is required because sometimes this object gets serialized into the browser history and then recovered.super.factoryKey
will be described in the next section.
This was the most complicated example you will see with this package. Everything else will be simpler.
Creating the Page Objects #
Create a global method like this:
AbstractPage? createPage(
String factoryKey,
Map<String, dynamic> state,
) {
switch (factoryKey) {
case BookDetailsPage.classFactoryKey: return BookDetailsPage(bookId: state['bookId']);
case BookListPage.classFactoryKey: return BookListPage();
}
return null;
}
Then pass it to your PageStackBloc
:
final pageStackBloc = PageStackBloc(
bottomPage: BookListPage(),
createPage: createPage,
);
Pages also must be upgraded to support the factory:
class BookDetailsPage extends StatelessMaterialPage {
static const classFactoryKey = 'BookDetails';
BookDetailsPage({
required int bookId,
}) : super(
key: ValueKey(formatKey(bookId: bookId)),
child: BookDetailsScreen(book: bookRepository[bookId]),
path: BookDetailsPath(bookId: bookId),
);
static String formatKey({required int bookId}) {
return '${classFactoryKey}_$bookId';
}
}
The new things in this Page
class are:
classFactoryKey
is any unique string among your page classes. This is whatPagePath.factoryKey
is matched to when your factory selects the page class to create.formatKey
is a method to create a runtime key to avoid page duplication in the stack. It is just for convenience because we now need to create these keys in two different places: here and inBookDetailsPath
.super.path
is described next.
Updating the Address Bar #
The address bar is not the only source of PagePath
objects.
When you programmatically create a page and push it to the stack, it may report
a PagePath
object to the framework so it updates the address bar from it.
Page Without Bloc
For a page without bloc, you hardcode a PagePath
like in the snippet above.
super(
// ...
path: BookDetailsPath(bookId: bookId),
);
When this page is pushed, that PagePath.location
ends up in the address bar.
Page With Bloc
A page with bloc delegates this to the bloc for higher flexibility.
Override the PageBloc.path
getter:
@override
BookListPath get path => const BookListPath();
Page Without URL
The address bar content is always taken from the highest page in the stack that has non-null
PagePath
, with or without bloc. For minor dialogs that should not affect the address bar
and should not get to the browser history, just do not introduce any path classes.
Updating the URL Programmatically
Imagine a tree browser that dynamically updates the address bar with all of this happening in a single screen:
What gets updated is its bloc. A bloc can emit a new PagePath
at any time.
For this, call its emitPathChanged()
method.
This will call the path
getter for the actual path to propagate.
Redirecting a URL #
Option 1. Single Class
The URL that your PagePath
returns takes over the one
that was parsed. To redirect /
to /books
, do the following:
- In
BookListPath.tryParse()
, allow both/
and/books
. - Return
/books
inlocation
getter.
Option 2. Multiple Classes
You can have many PagePath
classes for different URLs that all return
BookListPath
in their tryParse
methods, and it in turn will return /books
for location.
This is useful for:
- Redirecting to a page without its knowledge.
- Conditional redirects where you don't want to concentrate the condition logic
in the target
PagePath
.
Browser Back and Forward Buttons #
Back and Forward buttons in the browser work automatically. Unlike the Android back button, these buttons traverse the browser's history. The following is happening on any navigation with browser buttons:
RouteInformationParser
is called with a URL to parsePagePath
object from it. This is exactly like when the app starts.- Unlike the initial parsing, the
RouteInformation
now contains saved states for each page. This is because when the framework emitted thePagePath
earlier, it was smart enough to also save all pages' serialized states. This is why you did passstate
toPagePath
super constructor earlier. This is how History API works in JavaScript, nicely abstracted by Flutter.PageStackRouteInformationParser
detects this and skips the URL parsing. PageStackBloc.setConfiguration()
is called with all pages' states recovered into maps. The current pages in the stack are diffed against that recovered state. Unchanged pages are not affected, unwanted pages are popped, and new pages are created using thecreatePage
factory you provided when creating this bloc. This is whycreatePage
gets a map and not aPagePath
object.PageStckBloc
emits an event to rebuild any navigator listening to it. This is usuallyPageStackBlocNavigator
created for you by the defaultPageStackRouterDelegate
you passed to the app constructor.- The
Navigator
then runs its own diff on pages to update the routes that actually show the screen widgets. This way you get the updated UI.
It all works for you automatically if you correctly set all pages' and PagePath
keys.
Recovering Unsaved Input on Back and Forward Navigation #
Use case: In a multi-screen app, the user enters some text, then closes the screen without saving by ‘Back’ button. They then click ‘Forward’ and expect their input is still there.
Read this tutorial on how this app is made.
Multiple Tabs with Independent Stacks #
Use case: Each tab must have its own navigation stack. The Android back button and the Scaffold's back button should only pop pages on the current stack. Tab switching should not affect the back button history (except the browser's navigation buttons, this is how browsers work).
Read this tutorial on how this app is made.
Advanced Ways to Return Result #
Push and Pop Type Safety at Compile Time #
PageBloc
and BlocMaterialPage
are typed by a return type R
:
PageBloc<R>
BlocMaterialPage<R, B extends PageBloc<R>>
In most examples we ignore those types for faster learning, but in production
you always specify the return type even if it is void
.
If you await the push
call, the result is inferred with that type:
class InputPageBloc extends PageBloc<int> { /* ... */ }
class InputPage extends BlocMaterialPage<int, InputPageBloc> { /* ... */ }
final result = await pageStackBloc.push(InputPage()); // result is inferred as int.
When you pop
the result from the bloc, you can only pop
that exact type,
otherwise it will not build.
And your Page
can only have a bloc of the same return type, otherwise it will not build.
Receiving the Dialog Result after the App Restart #
Use case:
- The user opens the input dialog, copies its URL and restarts the app at that URL.
- The user inputs a value and closes the dialog.
- The underlying page receives the result.
This cannot be done by mere awaiting of the pushed page. When the app restarts, even if we recover the page stack, we create the stack in the factory, so the bottom page has no future to await.
For this, we employ the alternative way to receive the result. In the bottom page bloc,
override didPopNext
:
@override
void didPopNext(AbstractPage page, PageBlocCloseEvent event) {
print('didPopNext: ${event.data}');
}
This way the compiler still makes sure the bloc pops with the right type, but the event.data
is not type-checked by the receiver. This is because a bloc can push pages of different classes that produce different
result types, but they all are collected in this method.
See this tutorial with more details and the link to the full runnable example.
Help is Wanted #
Do you like this package? Do not buy me a coffee, I don't drink it. Here is what you can do:
-
Spread:
- Like this package on https://pub.dev/packages/app_state, because people judge the packages by the like count.
- Spread a word to Flutterers you know.
- Follow me on Medium. It will show the app_state tutorials to more people.
-
Info Help:
- I need a Comparison to auto_route, go_router, beamer, raw Router API implementation, and the old Navigator push-pop. You can write an article or at least provide me with points on how they compare, what they could not do for you, what is too hard with them, or how this package made your life easier. My email is here: https://pub.dev/publishers/ainkin.com/packages
- Send me your project link so I can list it among users to have a richer gallery of examples. Get a free ad from me. If it is a proprietary website or app, people will see what is possible. If it is open-source, it will get more attention here and gain more users.
- Reports an Issue if you find anything wrong.
-
Program:
- Code Generator. This package needs one to generate
Page
andPagePath
classes. You can design or implement it. - Tests. You can cover it.
- Other PRs. Please share your idea in an issue first.
- Code Generator. This package needs one to generate
Here are tons of runnable examples again if you have missed the link in the beginning.