This is a simple Service Locator for Dart and Flutter projects with some additional goodies highly inspired by Splat. It can be used instead of InheritedWidget or Provider to access objects e.g. from your UI.
Typical usage:
- Accessing service objects like REST API clients or databases so that they easily can be mocked.
- Accessing View/AppModels/Managers/BLoCs from Flutter Views
V5.0 has some breaking changes Check please check the release notes to see what's new.
Breaking Change with V4.0.0 The principle on how to synchronize your registered instances creation has been rethought and improved 😃. Please see Synchronizing asynchronous initializations of Singletons.
As your App grows, at some point you will need to put your app's logic in classes that are separated from your Widgets. Keeping your widgets from having direct dependencies makes your code better organized and easier to test and maintain. But now you need a way to access these objects from your UI code. When I came to Flutter from the .Net world, the only way to do this was the use of InheritedWidgets. I found the way to use them by wrapping them in a StatefulWidget; quite cumbersome and has problems working consistently. Also:
- I missed the ability to easily switch the implementation for a mocked version without changing the UI.
- The fact that you need a
BuildContextto access your objects made it inaccessible from the Business layer.
Accessing an object from anywhere in an App can be done by other ways, but:
- If you use a Singleton you can't easily switch the implementation out for a mock version in tests
- IoC containers for Dependency Injections offers similar functionality, but with the cost of slow start-up time and less readability because you don't know where the magically injected object come from. Most IoC libs rely on reflection they cannot be ported to Flutter.
As I was used to use the Service Locator Splat from .Net, I decided to port it to Dart. Since then, more features have been added.
If you are not familiar with the concept of Service Locators, it's a way to decouple the interface (abstract base class) from a concrete implementation, and at the same time allows to access the concrete implementation from everywhere in your App over the interface. I can only highly recommend to read this classic article by from Martin Fowler Inversion of Control Containers and the Dependency Injection pattern.
GetIt is:
- Extremely fast (O(1))
- Easy to learn/use
- Doesn't clutter your UI tree with special Widgets to access your data like provider or Redux does.
GetIt isn't a state management solution! It's a locator for your objects so you need some other way to notify your UI about changes like Streams or ValueNotifiers.
At your start-up you register all the objects you want to access later like this:
final getIt = GetIt.instance;
void setup() {
getIt.registerSingleton<AppModel>(AppModel());
// Alternatively you could write it if you don't like global variables
GetIt.I.registerSingleton<AppModel>(AppModel());
}After that you can access your AppModel class from anywhere like this:
MaterialButton(
child: Text("Update"),
onPressed: getIt<AppModel>().update // given that your AppModel has a method update
),You can find here a detailed blog post on how to use GetIt
As Dart supports global (or euphemistic ambient) variables I often assign my GetIt instance to a global variable to make the access to it as easy as possible
Although the approach with a global variable worked well, it has its limitations if you want to use GetIt across multiple packages. Therefore GetIt itself is a singleton and the default way to access an instance of GetIt is to call:
GetIt getIt = GetIt.instance;
//There is also a shortcut (if you don't like it just ignore it):
GetIt getIt = GetIt.I;Through this any call to instance in any package of a project will get the same instance of GetIt. I still recommend just to assign the instance to a global variable in your project as it is more convenient and doesn't harm (Also it allows you to give your service locator your own name).
GetIt getIt = GetIt.instance;You can use any name you want which makes Brian 😃 happy like (
sl, backend, services...) ;-)
Before you can access your objects you have to register them within GetIt typically direct in your start-up code.
getIt.registerSingleton<AppModel>(AppModelImplementation());
getIt.registerLazySingleton<RESTAPI>(() =>RestAPIImplementation());
// if you want to work just with the singleton:
GetIt.instance.registerSingleton<AppModel>(AppModelImplementation());
GetIt.I.registerLazySingleton<RESTAPI>(() =>RestAPIImplementation());
/// `AppModel` and `RESTAPI` are both abstract base classes in this exampleTo access the registered objects call get<Type>() on your GetIt instance
var myAppModel = getIt.get<AppModel>();Alternatively as GetIt is a callable class depending on the name you choose for your GetIt instance you can use the shorter version:
var myAppModel = getIt<AppModel>();
// as Singleton:
var myAppModel = GetIt.instance<AppModel>();
var myAppModel = GetIt.I<AppModel>();GetIt offers different ways how objects are registered that effect the lifetime of this objects.
void registerFactory<T>(FactoryFunc<T> func)You have to pass a factory function func that returns an NEW instance of an implementation of T. Each time you call get<T>() you will get a new instance returned. How to pass parameters to a factory you can find here
Although I always would recommend using an abstract base class as registration type so that you can vary the implementations you don't have to do this. You can also register concrete types.
void registerSingleton<T>(T instance)You have to pass an instance of T or a derived class of T that you will always get returned on a call to get<T>().
As creating this instance can be time consuming at app start-up you can shift the creation to the time the object is the first time requested with:
void registerLazySingleton<T>(FactoryFunc<T> func)You have to pass a factory function func that returns an instance of an implementation of T. Only the first time you call get<T>() this factory function will be called to create a new instance. After that you will always get the same instance returned.
If you try to register a type more than once you will fail with an assertion in debug mode because normally this is not needed and probably a bug.
If you really have to overwrite a registration, then you can by setting the property allowReassignment==true.
You can check if a certain Type or instance is already registered in GetIt with:
/// Tests if an [instance] of an object or aType [T] or a name [instanceName]
/// is registered inside GetIt
bool isRegistered<T>({Object instance, String instanceName});If you need to you can also unregister your registered singletons and factories and pass a optional disposingFunction for clean-up.
/// Unregister an [instance] of an object or a factory/singleton by Type [T] or by name [instanceName]
/// if you need to dispose some resources before the reset, you can
/// provide a [disposingFunction]. This function overrides the disposing
/// you might have provided when registering.
void unregister<T>({Object instance,String instanceName, void Function(T) disposingFunction})In some cases you might not want to unregister a LazySingleton but instead to reset its instance so that it gets newly created on the next access to it.
/// Clears the instance of a lazy singleton,
/// being able to call the factory function on the next call
/// of [get] on that type again.
/// you select the lazy Singleton you want to reset by either providing
/// an [instance], its registered type [T] or its registration name.
/// if you need to dispose some resources before the reset, you can
/// provide a [disposingFunction]. This function overrides the disposing
/// you might have provided when registering.
void resetLazySingleton<T>({Object instance,
String instanceName,
void Function(T) disposingFunction})/// Clears all registered types. Handy when writing unit tests
/// If you provided dispose function when registering they will be called
/// [dispose] if `false` it only resets without calling any dispose
/// functions
/// As dispose funcions can be async, you should await this function.
Future<void> reset({bool dispose = true});With V5.0 of GetIt it now supports hierarchical scoping of registration. What does this mean?
You can push a new registration scope like you push a new page on the Navigator. Any registration after that will be registered in this new scope. When accessing an object with get GetIt first checks the topmost scope for an registration and then the ones below. This means you can register the same type that was already registered in a lower scope again in a scope above and you will always get the latest registered object.
Imagine an app that can be used with or without a login. On App start-up a DefaultUser object is registered with the abstract type User as singleton. As soon as the user logs in, a new scope is pushed and a new LoggedInUser object again with the User type is registered that allows more functions. For the rest of the App nothing has changed as it still accesses User objects through GetIt.
As soon as the user Logs off all you have to do is pop the Scope and automatically the DefaultUser is used again.
Another example could be a shopping basket where you want to ensure that not a cart from a previous session is used again. So at the beginning of a new session you push a new scope and register a new cart object. At the end of the session you pop this scope again.
From V5.0 on you can pass a dispose function when registering any Singletons. For this the registration functions have a optional parameter:
DisposingFunc<T> disposewhere DisposingFunc is defined as
typedef DisposingFunc<T> = FutureOr Function(T param);So you can pass simple and async functions as this parameter. This function is called when you pop or reset the scope or when you reset GetIt completely.
When you push a new scope you can also pass a dispose function that is called when a scope is popped or reset but before the dispose functions of the registered objects is called which mean it can still access the objects that were registered in that scope.
/// Creates a new registration scope. If you register types after creating
/// a new scope they will hide any previous registration of the same type.
/// Scopes allow you to manage different live times of your Objects.
/// [scopeName] if you name a scope you can pop all scopes above the named one
/// by using the name.
/// [dispose] function that will be called when you pop this scope. The scope
/// is still valied while it is executed
void pushNewScope({String scopeName, ScopeDisposeFunc dispose});
/// Disposes all factories/Singletons that have ben registered in this scope
/// and pops (destroys) the scope so that the previous scope gets active again.
/// if you provided dispose functions on registration, they will be called.
/// if you passed a dispose function when you pushed this scope it will be
/// calles before the scope is popped.
/// As dispose funcions can be async, you should await this function.
Future<void> popScope();
/// if you have a lot of scopes with names you can pop (see [popScope]) all
/// scopes above the scope with [name] including that scope
/// Scopes are poped in order from the top
/// As dispose funcions can be async, you should await this function.
/// it no scope with [name] exists, nothing is popped and `false` is returned
Future<bool> popScopesTill(String name);
/// Clears all registered types for the current scope
/// If you provided dispose function when registering they will be called
/// [dispose] if `false` it only resets without calling any dispose
/// functions
/// As dispose funcions can be async, you should await this function.
Future<void> resetScope({bool dispose = true});If a factory needs to call an async function you can use registerFactoryAsync()
/// [T] type to register
/// [func] factory function for this type
/// [instanceName] if you provide a value here your factory gets registered with that
/// name instead of a type. This should only be necessary if you need to register more
/// than one instance of one type. Its highly not recommended
void registerFactoryAsync<T>(FactoryFuncAsync<T> func, {String instanceName});To access instances created by such a factory you can't use get() but you have to use getAsync() so that
you can await the creation of the requested new instance.
/// Returns an Future of an instance that is created by an async factory or a Singleton that is
/// not ready with its initialization.
Future<T> getAsync<T>([String instanceName]);Additionally you can register asynchronous Singletons which means Singletons that have an initialization that requires async function calls. To be able to control such asynchronous start-up behaviour GetIt supports mechanisms to ensure the correct initialization sequence.
You create an Singleton with an asynchronous creation function
void registerSingletonAsync<T>(FactoryFuncAsync<T> factoryfunc,
{String instanceName,
Iterable<Type> dependsOn,
bool signalsReady = false});The difference to a normal Singleton is that you don't pass an existing instance but provide an factory function
that returns a Future that completes at the end of factoryFunc and signals that the Singleton is ready to use unless true is passed for signalsReady. (see next chapter)
To synchronize with other "async Singletons" you can pass a list of Types in dependsOn that have to be ready before the passed factory is executed.
There are two ways to signal the system that an instance is ready.
Often your registered services need to do asynchronous initialization work before they can be used from the rest of the app. As this is such a common task, and its closely related to registration/initialization GetIt supports you here too.
GetIt has the function allReady which returns Future<void> that can be used e.g. with a Flutter FutureBuilder to await that all asynchronous initialization is finished.
Future<void> allReady({Duration timeout, bool ignorePendingAsyncCreation = false});There are different approaches how the returned Future can be completed:
If you register any async Singletons allReady will complete only after all of them have completed their factory functions. Like:
class RestService {
Future<RestService> init() async {
Future.delayed(Duration(seconds: 1));
return this;
}
}
final getIt = GetIt.instance;
/// in your setup function:
getIt.registerSingletonAsync<ConfigService>(() async {
final configService = ConfigService();
await configService.init();
return configService;
});
getIt.registerSingletonAsync<RestService>(() async => RestService().init());
// here we asume an async factory function `createDbServiceAsync`
getIt.registerSingletonAsync<DbService>(createDbServiceAsync);
/// ... in your startup page:
return FutureBuilder(
future: getIt.allReady(),
builder: (BuildContext context, AsyncSnapshot snapshot) {
if (snapshot.hasData) {
return Scaffold(
body: Center(
child: Text('The first real Page of your App'),
),
);
} else {
return CircularProgressIndicator();
}
});The above example shows you different ways to register async Singletons. The start-up page will display a CircularProgressIndicator until all services have been created.
In case that this services have to be initialized in a certain order because they depend on that other services are already ready to be used you can use the dependsOn parameter of registerFactoryAsync. If you have a non async Singleton that depends on other Singletons, there is registerSingletonWithDependencies. In the following example, DbService depends on ConfigService, and AppModel depends on ConfigService and RestService
getIt.registerSingletonAsync<ConfigService>(() async {
final configService = ConfigService();
await configService.init();
return configService;
});
getIt.registerSingletonAsync<RestService>(() async => RestService().init());
getIt.registerSingletonAsync<DbService>(createDbServiceAsync,
dependsOn: [ConfigService]);
getIt.registerSingletonWithDependencies<AppModel>(
() => AppModelImplmentation(),
dependsOn: [ConfigService, DbService, RestService]);When using dependsOn you ensure that the registration waits with creating its singleton on the completion of the type defined in dependsOn.
Sometimes the mechanism of dependsOn might not give you enough control. For this case you can use isReady to wait for a certain singleton:
/// Returns a Future that completes if the instance of an Singleton, defined by Type [T] or
/// by name [instanceName] or by passing the an existing [instance], is ready
/// If you pass a [timeout], an [WaitingTimeOutException] will be thrown if the instance
/// is not ready in the given time. The Exception contains details on which Singletons are
/// not ready at that time.
/// [callee] optional parameter which makes debugging easier. Pass `this` in here.
Future<void> isReady<T>({
Object instance,
String instanceName,
Duration timeout,
Object callee,
});To signal that a singleton is ready it can use signalReady, provided you have set the optional signalsReady parameter when registering it OR make your registration type implement the empty abstract class WillSignalReady. Otherwise, allReady will wait on a call to signalsReady. No automatic signaling will happen in that case.
/// Typically this is used in this way inside the registered objects init
/// method `GetIt.instance.signalReady(this);`
void signalReady(Object instance);You can use this to initialize your Singletons without async registration by using a fire and forget async function from your constructors like so:
class ConfigService {
ConfigService()
{
init();
}
Future init() async {
// do your async initialisation...
GetIt.instance.signalReady(this);
}
}Even if you already have awaited allReady, the moment you register new async singletons or singletons with dependencies you can use allReady again. This makes especially sense if you uses scopes where every scope needs to get initialized.
By calling signalReady(null) on your GetIt instance the Future you can get from allReady will be completed.
This is the most basic way to synchronize your start-up. If you want to do that don't use signalsReady or async Singletons!!!
I recommend using one of the other ways because they are more flexible and express your intention more clear.
You can find here a detailed blog post on async factories and startup synchronization
In some cases its handy if you could pass changing values to factories when calling get(). For that there are two variants for registering factories:
/// registers a type so that a new instance will be created on each call of [get] on that type based on
/// up to two parameters provided to [get()]
/// [T] type to register
/// [P1] type of param1
/// [P2] type of param2
/// if you use only one parameter pass void here
/// [factoryfunc] factory function for this type that accepts two parameters
/// [instanceName] if you provide a value here your factory gets registered with that
/// name instead of a type. This should only be necessary if you need to register more
/// than one instance of one type. Its highly not recommended
///
/// example:
/// getIt.registerFactoryParam<TestClassParam,String,int>((s,i)
/// => TestClassParam(param1:s, param2: i));
///
/// if you only use one parameter:
///
/// getIt.registerFactoryParam<TestClassParam,String,void>((s,_)
/// => TestClassParam(param1:s);
void registerFactoryParam<T,P1,P2>(FactoryFuncParam<T,P1,P2> factoryfunc, {String instanceName});
and
void registerFactoryParamAsync<T,P1,P2>(FactoryFuncParamAsync<T,P1,P2> factoryfunc, {String instanceName});The reason why I settled to use two parameters is that I can imagine some scenarios where you might want to register a builder function for Flutter Widgets that need to get passed a BuildContext and some data object.
When accessing these factories you pass the parameters a optional arguments to get():
var instance = getIt<TestClassParam>(param1: 'abc',param2:3);These parameters are passed as dynamics (otherwise I would have had add more generic parameters to get()), but they are checked at runtime to be the correct types.
When you are writing unit tests with GetIt in your App you have two possibilities:
- Register all the Objects you need inside your unit Tests so that GetIt can provide its objects to the objects that you are testing.
- Pass your dependent objects into the constructor of your test objects like:
GetIt getIt = GetIt.instance;
class UserManager {
AppModel appModel;
DbService dbService;
UserManager({AppModel appModel, DbService dbService}) {
this.appModel = appModel ?? getIt.get<AppModel>();
this.dbService = dbService ?? getIt.get<DbService>();
}
}This way you don't need to pass them in the AppModel and dbService inside your App but you can pass them(or a mocked version) in your Unit tests.
If you have a mocked version of a Service, you can easily switch between that and the real one based on a flag:
if (testing) {
getIt.registerSingleton<AppModel>(AppModelMock());
} else {
getIt.registerSingleton<AppModel>(AppModelImplementation());
}Ok you have been warned! All registration functions have an optional named parameter instanceName. Providing a name with factory/singleton here registers that instance with that name and a type. Consequently get() has also an optional parameter instanceName to access
factories/singletons that were registered by name.
IMPORTANT: Each name must be unique per type.
While not recommended, you can create your own independent instance of GetItif you don't want to share your locator with some
other package or because the physics of your planet demands it :-)
/// To make sure you really know what you are doing
/// you have to first enable this feature:
GetIt myOwnInstance = GetIt.asNewInstance();This new instance does not share any registrations with the singleton instance.
Many thanks to the insightful discussions on the API with Brian Egan and Simon Lightfoot