Feather

Additional

Language
Java
Version
1.0 (Oct 27, 2015)
Created
Sep 22, 2015
Updated
Sep 8, 2016
Owner
Zsolt Herpai (zsoltherpai)
Contributors
Vladislav Bauer (vbauer)
Zsolt Herpai (zsoltherpai)
zsoltconax
3
Activity
Badge
Generate
Download
Source code
APK file

Show card

About Feather

Feather is an ultra-lightweight dependency injection ( JSR-330) library for Java and Android. Dependency injection frameworks are often perceived as "magical" and complex. Feather - with just a few hundred lines of code - is probably the easiest, tiniest, most obvious one, and is quite efficient too (see comparison section below).

<dependency>
    <groupId>org.codejargon.feather</groupId>
    <artifactId>feather</artifactId>
    <version>1.0</version>
</dependency>

Javadoc for Feather

Usage - code examples
Create Feather (the injector)
Feather feather = Feather.with();

An application typically needs a single Feather instance.

Instantiating dependencies

Dependencies with @Inject constructor or a default constructor can be injected by Feather without the need for any configuration. Eg:

public class A {
    @Inject
    public A(B b) {
        // ...
    }
}

public class B {
    @Inject
    public B(C c, D d) {
        // ...
    }
}

public class C {}

@Singleton
public class D {
    // something expensive or other reasons for being singleton
}

Creating an instance of A:

A a = feather.instance(A.class);
Providing additional dependencies to Feather

When injecting an interface, a 3rd party class or an object needing custom instantiation, Feather relies on configuration modules providing those dependencies:

public class MyModule {
    @Provides
    @Singleton // an app will probably need a single instance 
    DataSource ds() {
        DataSource dataSource = // instantiate some DataSource
        return dataSource;
    }
}

Setting up Feather with module(s):

Feather feather = Feather.with(new MyModule());

The DataSource dependency will now be available for injection:

public class MyApp {
    @Inject 
    public MyApp(DataSource ds) {
        // ...
    }
}

Feather injects dependencies to @Provides methods aguments. This is particularly useful for binding an implementation to an interface:

public interface Foo {}

public class FooBar implements Foo {
    @Inject
    public FooBar(X x, Y y, Z z) {
        // ...
    }
}

public class MyModule {
    @Provides
    Foo foo(FooBar fooBar) {
        return fooBar;
    }
}

// injecting an instance of Foo interface will work using the MyModule above:
public class A {
    @Inject
    public A(Foo foo) {
        // ...
    }
}

Note that the @Provides method serves just as a binding declaration here, no manual instantiation needed

Qualifiers

Feather supports Qualifiers (@Named or custom qualifiers)

public class MyModule {
    @Provides
    @Named("greeting")
    String greeting() {
        return "hi";
    }

    @Provides
    @SomeQualifier
    Foo some(FooSome fooSome) {
        return fooSome;
    };
}

Injecting:

public class A {
    @Inject
    public A(@SomeQualifier Foo foo, @Named("greeting") String greet) {
        // ...
    }
}

Or directly from feather:

String greet = feather.instance(String.class, "greeting");
Foo foo = feather.instance(Key.of(Foo.class, SomeQualifier.class));
Provider injection

Feather injects Providers to facilitate lazy loading or circular dependencies:

public class A {
    @Inject
    public A(Provider<B> b) {
        B b = b.get(); // fetch a new instance when needed
    }
}

Or getting a Provider directly from Feather:

Provider<B> bProvider = feather.provider(B.class);
Override modules
public class Module {
    @Provides
    DataSource dataSource() {
        // return a mysql datasource
    }

    // other @Provides methods
}

public class TestModule extends Module {
    @Override
    @Provides
    DataSource dataSource() {
        // return a h2 datasource
    }
}
Field injection

Feather supports Constructor injection only when injecting to a dependency graph. It inject fields also if it's explicitly triggered for a target object - eg to facilitate testing. A simple example with a junit test:

public class AUnitTest {
    @Inject
    private Foo foo;
    @Inject
    private Bar bar;

    @Before
    public void setUp() {
        Feather feather = // obtain a Feather instance
        feather.injectFields(this);
    }
}
Method injection

Not supported. The need for it can be generally avoided by a Provider / solid design (favoring immutability, injection via constructor).

Android example
class ExampleApplication extends Application {
    private Feather feather;

    @Override public void onCreate() {
        // ...
        feather = Feather.with( /* modules if needed*/ );
    }

    public Feather feather() {
        return feather;
    }
}

class ExampleActivity extends Activity {
    @Inject
    private Foo foo;
    @Inject
    private Bar bar;

  @Override public void onCreate(Bundle savedState) {
    // ...
    ((ExampleApplication) getApplication())
        .feather()
            .injectFields(this);
  }
}

For best possible performance, dependencies should be immutable and @Singleton. See full example in android-test.

Footprint, performance, comparison

Small footprint and high performance is in Feather's main focus.

  • compared to Guice: 1/50 the library size, ~10x startup speed
  • compared to Dagger: 1/4 the library size (of just Dagger's run-time part), ~2x startup speed

Note: startup means creation of the container and instantiation of an object graph. Executable comparison including Spring, Guice, Dagger, PicoContainer is in 'performance-test' module.

How it works under the hood

Feather is based on optimal use of reflection to provide dependencies. No code generating, classpath scanning, proxying or anything costly involved.

A simple example with some explanation:

class A {
    @Inject
    A(B b) {

    }
}

class B {

}

Without the use of Feather, class A could be instantiated with the following factory methods:

A a() {
    return new A(b());
}

B b() {
    return new B();
}

Most of the information in these factories are redundant and they tend to be hot spots for changes and sources for merge hells. Feather avoids the need for writing such factories - by doing the same thing internally: When an instance of A is injected, Feather calls A's constructor with the necessary arguments - an instance of B. That instance of B is created the same way - a simple recursion, this time with no further dependencies - and the instance of A is created.