arrow left
Back to Developer Education

Dependency Injection in Flutter

Dependency Injection in Flutter

Dependency injection is an object-oriented technique that sends the dependencies of another object to an object. Using dependency injection, we can also move the creation and restriction of dependent objects outside the classes. This concept brings a more significant level of adaptability, decoupling, and simpler testing. <!--more--> The provider package wraps Inherited Widgets in an easy-to-use wrapper. It makes it easy to manage and utilize, and it also includes a state management mechanism for managing data within the app.

In this article, we will learn how to implement dependency injection, get rid of flutter packages, and why we need dependency injection in a Flutter application.

Table of content

Dependency injection in Flutter

Although dependency injection is a simple example, libraries are frequently used to abstract it from the designer. Many of these libraries take advantage of reflection (mirrors in Dart). There are issues in a flutter, however:

  1. Mirrors are crippled for execution reasons.
  2. The settled idea of widgets makes it unfeasible to pass conditions many levels down the tree.

The injection library, on the other hand, can be used to solve similar issues. Annotations used by the injection library are listed below.

  1. @Injector - An inversion of control container developed from a bunch of modules.
  2. @provides and @module - Define classes and methods providing dependencies.
  3. @component - In performing an injection.

Functionalities of dependency injection

Concrete class injection

The concrete class injection can be as shown below:

//inject/inject.dart’ is an auto generated file that is imported
import ‘package:inject/inject.dart’;
//defining a class
@provide
class ExampleExa {
   // implementation
}

Interface injection

First, create an abstract class with some implementation, then provide dependencies in the module. Abstract class:

abstract class NewExa{
  
}class OldExa implements NewExa {
   @override
    {
   // implementation
   }
}

Afterward, we can add dependencies to our module. Final module:

//inject/inject.dart’ is an auto generated file that is imported
import ‘package:inject/inject.dart’;
//defining the class 
@module
class ExampleExa {
//providing dependencies
@provide
NewExa newExa() => OldExa();
}

Providers

Suppose we do not require an instance of a class injected, but rather a provider that will provide us with a fresh instance of the class every time we need to resolve a dependency. Instead of getting a concrete instance in the constructor, we can use a function that returns the appropriate instance and will do the injection correctly. We can define a helper and use it as shown below.

typedef Provider<T> = T Function();

Usage:

@provide
class NewWidget extends StatelessWidget {
   final Provider<StepService> _service;
NewWidget(this._service);
void _someFunction() {
   final service = _service();
   // use service
   }
}

Assisted injection

Since there is no implicit benefit to injecting objects that require only runtime contentions, we can use the standard factory. For example, create a factory class that takes all the compile-time dependencies in the constructor, injects it, and gives a factory technique runtime contention that will make a crucial case.

We need now to construct an injector for everything to work.

import ‘main.inject.dart’ as g;
@Injector(const [ExampleExa, NewExa])
abstract class Main {
   @provide
   NewApp get app;
   static Future<Main> create(
     ExampleExa exampleModule,
     NewExa newModule,
   ) async {
     return await g.Main$Injector.create(
       exampleModule,
       newModule,
     );
   }
}

In this example, ExampleExa and NewExa are already defined, NewApp is our root widget, and main.inject.dart is a file that Flutter creates automatically. As a result, the primary function may be as follows:

void main() async {
var container = await Main.create(
ExampleExa(),
NewExa(),
);
runApp(container.app);
}

We recommend using either the build runner or the watch command to keep the source code in sync throughout code creation.

flutter pub run build_runner build

Alternatively:

flutter pub run build_runner watch

The code will be saved in the cache folder, which Flutter does not yet support. As a result, the following text must be included in the inject_generator.build.yaml file:

builders:
 inject_generator:
 target: “:inject_generator”
 import: “package:inject_generator/inject_generator.dart”
 builder_factories:
 — “summarizeBuilder”
 — “generateBuilder”
 build_extensions:
 “.dart”:
 — “.inject.summary”
 — “.inject.dart”
 auto_apply: dependents
 build_to: source

We can now run the build runner, which will generate the required code (with error warnings if any dependencies cannot be resolved), and then execute Flutter build as usual.

Reason why we need dependency injection

As explained earlier, dependency injection in Flutter is an object-oriented technique that sends the dependencies of another object to an object. The objective of the dependency injection strategy is to eliminate this dependency by isolating the use from the formation of the item.

This diminishes the measure of required standard code and further develops adaptability.

How to get rid of flutter package

  1. Run flutter clean.
  2. Remove the unwanted dependencies in pubspec.yaml.
  3. Delete the pubspec.lock.

Reload the project and again perform 'pub get' and run the project.

Conclusion

Dependency injection permits a customer to eliminate all information on actual execution. It permits us to make more adaptable code and change one execution for another without changing the layer.


Peer Review Contributions by: Jerim Kaura

Published on: Sep 1, 2021
Updated on: Jul 15, 2024
CTA

Start your journey with Cloudzilla

With Cloudzilla, apps freely roam across a global cloud with unbeatable simplicity and cost efficiency
Get Started for Free