Aug 17, 2016
May 19, 2018
Google (google)
Tobias Preuss (johnjohndoe)
AMIT SHEKHAR (amitshekhariitbhu)
Segun Famisa (segunfamisa)
Source code


DISCLAIMER: This is not an official Google product. Google's production dependency injection frameworks are Dagger and Guice.

Tiger - The fastest java dependency injection framework


Tiger is highly inspired by Dagger, Kudos to those great guys.


If you find a faster DI framework, let me know. I will drop the title as long as it is proved.

Why Tiger?

It is the fastest! Not faster, but the fastest! I have tried it on a big project with ~200 modules. While it takes hundreds of milliseconds to instantiate dagger components, on the same hardware, it only takes a few milliseconds to instantiate Tiger injectors. Minimal amount of code to write therefore easy to maintain, if you need to maintain it at all. You don't need to write components like in Dagger. You don't need to split a module into several modules one for each scope that used by bindings provided by the module. You will feel it is some easy to change the scope of a binding. Just change it. Way to go, isn’t it?

Build up your knowledge

If you are here, you must already be familiar with DI(Dependency Injection) and its advantage. Otherwise wiki will be your friend. DI has been evolving for long time in the different form. But the concept is not really changed much. This document will not repeat these concepts. If you find some concept not explained, google it. Also Guice probably has explained those concept very well.


Tiger is an annotation process. Therefore just build the jar and use it the way that annotation processors are supposed to be used. All environment should work. The sample uses gradle.


Before diving into details, it will be helpful, very helpful, to understand the intended usage of tiger. Tiger is designed to let the machine do as much as possible and let human do as little as possible. It requires minimal information from the developer. From these information, scoped injectors are generated. Application can instances these injectors and use them to inject classes. To achieve this, tiger has distilled the information needed to generate injectors. Here are they, with related annotation.


Usually application has at least one scope, singleton. Even if there is no scoped binding, it is harmless to have a singleton scope. Therefore, tiger requires there always be a scope tree. The root is usually singleton, but not necessary. Details will be shown later in the sample. Tiger generates one injector class for each scope.


It specifies the dependencies between scopes. All the dependency information form a scope tree.


It provides binding information through @Provides annotated methods with optional scope. Now(2016/08/10) we just reuse dagger.Module. In future, dagger.Module will be copied into tiger.Module so that tiger does not need to depend on dagger.

@javax.inject.Inject on ctor

It provides binding information with optional scope.

@tiger.MembersInjector with mandatory scope

It specifies interfaces which includes class that has fields or methods to be injected. Injectors will implement all these interfaces.

@javax.inject.Inject on fields, methods

It specifies the injection points from which injector fans out when injecting an object.


The package for the generated injectors.


It specify the names of the injectors.


This the annotation that triggers the generation processor. @Module, @Inject and @MembersInjector are naturally scattered around the code. For the others, i.e., @ScopeDependency, @PackageForGenerated, @ScopedComponentNames and @GenerationTriggerAnnotation, we suggest to put them into a dedicated java file as the central configuration information for the app.

Here is the depicted code the sample(with some modification)

public class Scopes {

    @ScopedComponentNames(scope = Singleton.class, name = "ApplicationInjector" )
    public static class ForApplication {}

    @ScopedComponentNames(scope = sample.ActivityScoped.class, name = "ActivityInjector" )
    @ScopeDependency(scope = sample.ActivityScoped.class, parent = Singleton.class)
    public static class ForActivityScoped {}

Here is how ApplicationInjector is instantiated and used.

ApplicationInjector applicationInjector = new ApplicationInjector.Builder().build();
PseudoApplication application = new PseudoApplication();

Here is how ActivityInjector is instantiated and used.

ActivityInjector activityInjector = new ActivityInjector.Builder()

The injectors guarantee that scoped bindings will be instantiated at most once within a scope. The application needs to create related injectors for scope objects, e.g., in android, a context scoped injector for each Activity, a singleton scoped injector for the Application.

For Dagger users

As you can see, Tiger reuses annotations from dagger like dagger.Module, dagger.Provides, etc. You can find the javadoc here. We are not going to repeat them here. Of course, Component and Subcomponent is not needed any longer. Producer related stuff is also irrelevant to injection. There is one nice feature from Tiger. You don't need to split modules according to different scopes. Yes, you can put bindings of different scopes into one module. This way you have less modules. And, if you want to change the scope of a binding, just change it, easy.


Inspecting the generate code will help you. If you want more, there is source code. Enjoy injection!