arbor
Modular and compile-time safe DI for Dart without fragility and magic.
For Flutter integration, see flutter_arbor.
Index
About
arbor
offers a dependency injection solution for Dart that is compile-time safe and does not rely on magic – no code generation, no runtime type checking, no singletons, and no dynamic mutable Maps
. arbor
abstracts out a tree-like structure of dependencies, which can be resolved at compile-time. This allows for a more flexible and modular approach to dependency injection.
arbor
is partially inspired by Needle and common Dart approach for dependency storages.
Motivation
Currently, there are two main approaches to dependency injection in Dart: constructor injection and singleton injection. Both approaches have their pros and cons. Constructor injection is a good fit for small projects, but it quickly becomes cumbersome to manage dependencies in larger projects. Singleton injection relieves the burden of managing dependencies, but it is not compile-time safe and relies on magic.
arbor
is a middle ground between the two approaches. It is compile-time safe and does not rely on magic, but it is not as cumbersome as constructor injection to manually manage dependencies. arbor
is also more flexible than singleton injection, as it allows for a more modular approach to dependency injection.
Install
Add arbor
to your pubspec.yaml
file:
dependencies:
arbor: "current version"
Or do it via CLI.
For Flutter projects:
$ flutter pub add arbor
For Dart projects:
$ dart pub add arbor
Usage
arbor
can be accustomed to both small projects that require a simple dependency injection solution and large projects that require a more modular approach to dependency injection. The following examples will demonstrate how to use arbor
in both scenarios.
Starting point
Every arbor
project requires a Tree
to be created. A Tree
is a tree-like structure of dependencies that can be resolved at compile-time. A Tree
can be created by using the BaseTree
class:
class Tree extends BaseTree<Tree> {
Tree({super.observer});
}
Every other Tree
class should extend the BaseTree
class. The Tree
class should also be passed as a type parameter to the BaseTree
class. The Tree
class should also have a named constructor that takes in an optional ArborObserver
as a parameter. The Tree
class should also have a named constructor that takes in an optional ArborObserver
as a parameter. The ArborObserver
is used to observe the state of the Tree
and can be used for debugging purposes.
Dependency nodes
The created Tree
class is a shared starting point for every node of dependencies, which can be created by using either module
or child
methods that accept a factory of ModuleNode
or ChildNode
respectively.
Both dependency nodes can have their descendants and can create a shared object that is bound to a specific lifecycle point which differs between the two dependency nodes via the shared
method and creates new instances of objects via the instance
method.
Both node types have the same set of lifecycle methods that can be overridden to perform actions at specific lifecycle points and both have their parent, passed as a type parameter.
Module
A module is a stateless node that acts as a namespace for dependencies and descendant dependency nodes.
Its lifecycle and any stateful operations, such as creating shared dependencies, are strictly bound and are delegated to the nearest stateful parent, which can be either a ChildNode
or a BaseTree
.
Usually, a module is used to group dependencies that are related to a specific feature or a specific part of the application, or a module can represent a set of features as a whole. Modules are lightweight and are cached in memory for the duration of the application's lifecycle in the root Tree
node, so they can be used as needed.
Child
On the other hand, the child is a stateful node that acts as a container for dependencies and descendant dependency nodes. The child
method returns an ObjectFactory<Child>
, which essentially is a thunk – Child Function()
, which must be called to create a new instance of the Child
class.
Child nodes are used to describe a node with some ephemeral state bound to it which must be recreated several times through the lifecycle of the application.
Lifecycle
All nodes implement a set of lifecycle methods that can be overridden to perform actions at specific lifecycle points. The lifecycle methods are init
and dispose
, which are represented by the Lifecycle
interface.
Nodes are created lazily, which means that the init
method is called only when the node is resolved for the first time. The dispose
method can be called from the outside, either by hand or by integrations, such as flutter_arbor
.
Dependency injection
To actually perform dependency injection, a dependency node implements consumers dependency interfaces, utilizing the dependency inversion.
// consumer.dart
abstract class ConsumerDependencies {
StreamController<String> get messagesController;
}
class Consumer {
final ConsumerDependencies _dependencies;
Consumer(this._dependencies);
}
// di.dart
class SomeFeatureNode
extends BaseChildNode<SomeFeatureNode, SomeFeatureParent>
implements ConsumerDependencies {
SomeFeatureNode(super.parent);
@override
StreamController<String> get messagesController => shared(
StreamController.broadcast,
dispose: (controller) => controller.close(),
);
}
// main.dart
void main() {
final appDependencies = AppDependencies();
final node = appDependencies.features.feature();
final consumer = Consumer(node);
}
Three things are happening here:
- The
Consumer
class depends on theConsumerDependencies
interface. It declares the dependencies it needs to function itself, utilizing the dependency inversion, and does not care about how the dependencies are resolved. - The
SomeFeatureNode
class implements theConsumerDependencies
interface. It provides the dependencies that theConsumer
class needs to function itself. - The
Consumer
class is created by passing theSomeFeatureNode
instance to its constructor. TheConsumer
class does not care about how the dependencies are resolved, it only cares that they are resolved, and depends only on the dependencies that it needs to function itself.
Observer
The Tree
class can be observed by passing an ArborObserver
to its constructor. The ArborObserver
is a base class that can be extended to override specific methods that are called in specific lifecycle points of the Tree
and its nodes.
An example that logs the lifecycle of the Tree
and its nodes:
class PrintObserver extends ArborObserver {
@override
void onInit<A extends Lifecycle>() {
super.onInit();
print('Init $A');
}
@override
void onDisposed<A extends Disposable>() {
super.onDisposed();
print('Disposed $A');
}
}
To view the full list of methods that can be overridden, see the ArborObserver
class.
Modularity
As discussed previously, arbor
can be used in both small and large projects with consideration to tradeoffs between modularity and simplicity.
Simpler
A simple approach could look something like that - concrete parents and concrete children:
abstract class StringConsumerDependencies {
String get veryImportantString;
String get anotherImportantString;
String get yetAnotherImportantString;
}
class AppDependencies extends BaseTree<AppDependencies> {
String get veryImportantString => shared(() => 'hello');
ExampleChildModule get exampleModule => module(ExampleChildModule.new);
}
class ExampleChildModule
extends BaseChildNode<ExampleChildModule, AppDependencies> {
ExampleChildModule(super.parent);
String get anotherImportantString => shared(() => 'world');
ObjectFactory<ExampleChildNode> get exampleChild => child(ExampleChildNode.new);
}
class ExampleChildNode
extends BaseChildNode<ExampleChildNode, ExampleChildModule>
implements StringConsumerDependencies {
ExampleChildNode(super.parent);
@override
String get veryImportantString => parent.parent.veryImportantString;
@override
String get anotherImportantString => parent.anotherImportantString;
@override
String get yetAnotherImportantString => shared(() => '!');
}
This approach is pretty straightforward – each node declares its parent as a Type parameter, and each node has access to the whole tree since parents are concrete classes that are connected.
It relieves the developer from having to think about the structure of the tree, since the tree is a concrete class that is connected, and the developer can just focus on the dependencies themselves. The most common use case for this approach is a small project with a small number of dependencies.
More modular
One could spot that this approach has some drawbacks. The most obvious one is that the ExampleChildModule
has access to the whole tree, which is not always desirable and breaks a few rules of "good code". Secondly, the children are hard-attached to their position in the tree, which makes it hard to reuse them in different parts of the tree. A more modular approach could look something like that - abstract parents and interfaces for children.
Firstly, to abstract the parents, an interface for said parents should be created. The interfaces will implement a set of common dependencies that are shared across all children:
abstract class ImportantStringDependency {
String get veryImportantString;
}
abstract class AnotherImportantStringDependency {
String get anotherImportantString;
}
abstract class StringConsumerDependencies
implements ImportantStringDependency, AnotherImportantStringDependency {
String get yetAnotherImportantString;
}
abstract class SharedParent<N extends SharedParent<N>>
implements
Node<N>,
AnotherImportantStringDependency,
ImportantStringDependency {}
The SharedParent
interface is a common interface for all parents, and it implements the Node
interface that makes it eligible as a parent for a child node.
After that, the children that previously declared their concrete parents as Type parameters, now declare their parents as vague implementers of the SharedParent
interface:
class AppDependencies extends BaseTree<AppDependencies>
implements SharedParent<AppDependencies> {
@override
String get veryImportantString => shared(() => 'hello');
@override
String get anotherImportantString => exampleModule.anotherImportantString;
ExampleChildModule<AppDependencies> get exampleModule => module(
ExampleChildModule.new,
);
}
class ExampleChildModule<P extends SharedParent<P>>
extends BaseModule<ExampleChildModule<P>, P>
implements SharedParent<ExampleChildModule<P>> {
ExampleChildModule(super.parent);
@override
String get veryImportantString => parent.veryImportantString;
@override
String get anotherImportantString => shared(() => 'world');
ObjectFactory<ExampleChildNode<ExampleChildModule<P>>>
get exampleModule => child(ExampleChildNode.new);
}
class ExampleChildNode<P extends SharedParent<P>>
extends BaseChildNode<ExampleChildNode<P>, P>
implements
SharedParent<ExampleChildNode<P>>,
StringConsumerDependencies {
ExampleChildNode(super.parent);
@override
String get veryImportantString => parent.veryImportantString;
@override
String get anotherImportantString => parent.anotherImportantString;
@override
String get yetAnotherImportantString => shared(() => '!');
}
The naïve usage of SharedParent has resolved the main issues, now the children are not hard-attached to their position in the tree, and they have access only to the dependencies that they need. It may take some squinting to fight through type parameters, as this approach uses recursive generics.
But two new problems have emerged. Firstly, the code is not DRY anymore, since the SharedParent
interface is implemented by both the AppDependencies
and the ExampleChildModule
classes, and secondly, new type parameter info leaked even more implementation details.
To solve the first problem, a mixin can be created that will retrieve the dependencies from the parent, and implement the SharedParent
interface:
mixin SharedParentMixin<C extends SharedParentMixin<C, P>,
P extends SharedParent<P>> on HasParent<P> implements SharedParent<C> {
@override
String get veryImportantString => parent.veryImportantString;
@override
String get anotherImportantString => parent.anotherImportantString;
}
class AppDependencies extends BaseTree<AppDependencies>
implements SharedParent<AppDependencies> {
@override
String get veryImportantString => shared(() => 'hello');
@override
String get anotherImportantString => exampleModule.anotherImportantString;
ExampleChildModule<AppDependencies> get exampleModule => module(
ExampleChildModule.new,
);
}
class ExampleChildModule<P extends SharedParent<P>>
extends BaseModule<ExampleChildModule<P>, P>
with SharedParentMixin<ExampleChildModule<P>, P> {
ExampleChildModule(super.parent);
@override
String get anotherImportantString => shared(() => 'world');
ObjectFactory<ExampleChildNode<ExampleChildModule<P>>>
get exampleChild => child(ExampleChildNode.new);
}
class ExampleChildNode<P extends SharedParent<P>>
extends BaseChildNode<ExampleChildNode<P>, P>
with SharedParentMixin<ExampleChildNode<P>, P>
implements StringConsumerDependencies {
ExampleChildNode(super.parent);
@override
String get yetAnotherImportantString => shared(() => '!');
}
The mixin works as intended, retrieving the dependencies from the parent for us and implementing the SharedParent
interface.
The second problem can also be easily solved by using good old interfaces that will describe the tree of dependencies:
abstract class IDependencies {
IExampleModule get exampleModule;
}
abstract class IExampleModule {
ObjectFactory<IExampleChild> get exampleChild;
}
abstract class IExampleChild implements StringConsumerDependencies {}
class AppDependencies extends BaseTree<AppDependencies>
implements SharedParent<AppDependencies>, IDependencies {
@override
String get veryImportantString => shared(() => 'hello');
@override
String get anotherImportantString => exampleModule.anotherImportantString;
@override
ExampleChildModule<AppDependencies> get exampleModule => module(
ExampleChildModule.new,
);
}
class ExampleChildModule<P extends SharedParent<P>>
extends BaseModule<ExampleChildModule<P>, P>
with SharedParentMixin<ExampleChildModule<P>, P>
implements IExampleModule {
ExampleChildModule(super.parent);
@override
String get anotherImportantString => shared(() => 'world');
@override
ObjectFactory<IExampleChild> get exampleChild =>
child<ExampleChildNode<ExampleChildModule<P>>>(
ExampleChildNode.new,
);
}
class ExampleChildNode<P extends SharedParent<P>>
extends BaseChildNode<ExampleChildNode<P>, P>
with SharedParentMixin<ExampleChildNode<P>, P>
implements IExampleChild {
ExampleChildNode(super.parent);
@override
String get yetAnotherImportantString => shared(() => '!');
}
For a complete example, see the example project