We will see some of the Dagger 2 generated code more specifically on how it uses SingleCheck<T> & DoubleCheck<T> implementations of javax.inject.Provider<T> internally.
Whenever you create a @Component or @Subcomponent, dagger creates an implementation extending your type with some fields (from your modules & submodules), constructor & an initialize method.
@Singleton // this is just a scope, we can use a custom one if we want.
@Component(modules = [AppModule::class])
interface AppComponent {
...
@Component.Factory
interface Factory {
// BindInstance ties parameter within the component heirarchy so it
// can be accessed/used in modules, subcomponents, etc.
fun create(@BindInstance application Application) : AppComponent
}
fun inject(app: CustomApplication) // `CustomApplication` is my application class.
}
// Dagger generated code
public final class DaggerAppComponent implements AppComponent {
private final AppModule appModule;
private final Application application;
...
// `AppModule` contents are omitted for brevity.
public DaggerAppComponent(AppModule appModuleParam, Application applicationParam) {
this.appModule = appModuleParam;
this.application = applicationParam;
initialize(appModuleParam, applicationParam);
}
private void initialize(final AppModule appModuleParam, final Application applicationParam) {
...
}
@Override
public void inject(CustomApplication app) { // used to inject in my Application class.
...
}
}
As you can see the Component.Factory or Component.Builder method parameters will become the arguments for the constructor & are stored in the private fields of the class. If @BindInstance is not be applied to the create method, we won’t be seeing private final Application application; as dagger assumes it is only needed to construct the component and not to been used in any modules or submodules. If you try to use it within Module or component heirarchy then dagger will throw a compile time error as the field does not exist.
AppModule
Let’s take a real example from an app which has AppModule as follows.
class NetworkUtils @Inject constructor (...) {
fun getOkHttpBuilder() : OkHttpClient.Builder { ... }
fun getRetrofitBuilder() : Retrofit.Builder { ... }
}
@Module
class AppModule {
@Provides
fun provideMyApi(networkUtils: NetworkUtils) : IpApi {
return networkUtils.getRetrofitBuilder()
...
.create(...)
}
}
Whenever you add @javax.inject.Inject annotation, dagger will create a factory class eg: For NetworkUtils, NetworkUtils_Factory.java will be created implementing dagger.internal.Factory<T> which itself is a Provider<T>. It will also create factories for all the @Provide methods in your module eg: AppModule_ProvideMyApiFactory.
A Provider<T> contains single method get() : T which generate or provide an instance of the T. The optimization happens for this get() : T method, whether to provide a new instance or reuse existing instance (which was cached after initialization for the first time).
For this we can use scope annotations which dagger provides to create a custom scope or use an existing one (@Reusable, @Singleton in our case, we will see this later).
For now, the dependencies injected in my application class would generate the follow dagger code.
fun CustomApplication : Application() {
val appComponent: AppComponent = DaggerAppComponent.create(this)
@Inject lateinit var myApi : MyApi
@Inject lateinit var networkUtils : NetworkUtils
fun onCreate() {
appComponent.inject(this) // using the inject method defined in `DaggerAppComponent`
...
}
}
public final class DaggerAppComponent implements AppComponent {
... // from above
private Provider<MyApi> provideMyApiProvider; // Factory for `provideMyApi` method
private void initialize(...) {
// `create` or return as existing factory class.
// Remember `create` method will return a factory & `get` will return an instance depending on how
// many times it has been wrapped with a Provider<T>.
this.provideMyApiProvider = AppModule_ProvideMyApiFactory.create(..., NetworkUtils_Factory.create());
}
@Override
public void inject(CustomApplication app) {
injectApp(app);
}
private App injectApp(CustomApplication instance) {
// internally `provideMyApi` call will delegate the call to `AppModule.provideIpApi(...)`
instance.myApi = AppModule_ProvideMyApiFactory.provideMyApi(appModule, new NetworkUtils(...));
instance.networkUtils = new NetworkUtils(...);
}
}
The inject method actually sets the value for the fields in the CustomApplication. But we observe that we are always providing a new instance of NetworkUtils whenever requested. Let’s see how can we customize this behavior.
Note: Similar results will be produced if we try to
@ProvideNetworkUtilsvia a module.
Scoped annotations
-
@Reusable
From the docs,
@Reusableis useful when you want to limit the number of provisions of a type, but there is no specific lifetime over which there must be only one instance.
Hmm, so it says there is no guarantee that the cached instance could be a singleton & there is a chance that it could be reinitialized (in an edge case, maybe?).
Let’s see how the generated code looks.
@Reusable
class NetworkUtils @Inject constructor (...) {
...
}
// Dagger generated code
public final class DaggerAppComponent implements AppComponent {
...
private Provider<NetworkUtils> provideNetworkUtilsProvider;
...
private void initialize(...) {
this.networkUtilsProvider = SingleCheck.provider(NetworkUtils_Factory.create(provideApplicationContextProvider));
}
}
We see it wraps our factory provider using SingleCheck<T>. But what does it do?
From the SingleCheck<T> docs,
A Provider implementation that memoizes the result of another Provider using simple lazy initialization, not the double-checked lock pattern.
So it caches the result using “simple” lazy initialization & is able to provide the cached instance when requested. But it could break under certain circumstances, I’ll answer the “when” question later.
-
@Singleton
There are no restrictions on only using @Singleton annotation, you can create a custom one but don’t forget to annotate your @Component or @Subcomponent with it. In my case, I annotated AppComponent with @Singleton which means any dependency annotated with @Singleton will bind to AppComponent & will live as long as AppComponent lives (i.e tied to the Application).
Lets see how the generated code looks like when we annotate NetworkUtils with it.
@Singleton
class NetworkUtils @Inject constructor (...) {
...
}
// Dagger generated code
public final class DaggerAppComponent implements AppComponent {
...
private Provider<NetworkUtils> provideNetworkUtilsProvider;
...
private void initialize(...) {
this.networkUtilsProvider = DoubleCheck.provider(NetworkUtils_Factory.create(provideApplicationContextProvider));
}
}
Instead of SingleCheck<T> we see the use of DoubleCheck<T> provider.
From the DoubleCheck<T> docs,
A Lazy and Provider implementation that memoizes the value returned from a delegate using the double-check idiom described in Item 71 of Effective Java 2.
From this statement I know that I need to read the Item 71 of Effective Java 2 to understand lazy initialization in more depth.
Note: If you inject your dependency using
dagger.Lazy<T>provider then you will seeDoubleCheck.lazy(...)in the generated code where you have to manually call.get()to retrieve the dependency in the calling code. The working is same but the dependency is only initialized when.get()is called for the first time.
SingleCheck<T> vs DoubleCheck<T>
From the docs, we know both of these idioms are used for lazy initializations & after reading Item 71 of Effective Java 2 I’m sure DoubleCheck<T> is thread safe. Lets explore them,
-
Simple lazy initialization (Single check idiom)
Here I’ve created a simple copy-paste-run sample that initializes the field lazily (no locking).
class Box
// Single check idiom used by `SingleCheck<T>`
class Container {
companion object {
@Volatile // writes to this field are immediately made visible to other threads
private var box: Box? = null
fun getBox() : Box {
var field = box
if (field == null) {
box = Box()
field = box
}
return field!!
}
}
}
fun thread1() {
Thread {
val field = Container.getField()
println("Value from thread1: $field")
}.start()
}
fun thread2() {
Thread {
val field = Container.getField()
println("Value from thread2: $field")
}.start()
}
fun main() {
thread1()
thread2()
/* Output
* ------
* Value from thread 2: Box@6ae36b9b
* Value from thread 1: Box@667c6a02
*/
}
The order of println is not guaranteed but we see both the accessor of Box in these two threads are different. This is because the SingleCheck<T> idiom is not thread safe & could cause repeated initialization when we try to access the field from different threads at the same time.
-
Double check idiom (locking pattern)
The way this idiom is implemented is first we check the field if it is uninitialized without locking, then we lock the object & again check the field if uninitialized. If it appears to be uninitialized the second time we initialize the field & return it in both the cases.
We modify the Container class to follow this idiom & run the main function again.
// Double check idiom (thread safe) used by `DoubleCheck<T>`
class Container {
companion object {
@Volatile
private var box: Box? = null
fun getField() : Box {
var field = box
if (field == null) {
synchronized(this) {
field = box
if (field == null) {
box = Box()
field = box
}
}
}
return field!!
}
}
}
fun main() {
thread1()
thread2()
/* Output
* ------
* Value from thread 2: Box@6b925f95
* Value from thread 1: Box@6b925f95
*/
}
We’ve achieved thread safety as both the threads have the same instance of Box.
Conclusion
- The optimization happens only on
Provider<T>.get(T)which could be recursive, hence some checks likeDoubleCheck<T>.reentrantCheckare made to avoid transitivity. - If we use
@Reusablethen dagger caches the factory provider usingSingleCheck<T>idiom which is not thread safe & can cause repeated initialization if the field is accessed from different threads at the same time. Hence, the docs says “limit the number of provisions of a type” not a single instance. - Using
@Singletonor injecting viadagger.Lazy<T>caches the factory provider usingDoubleCheck<T>idiom which is thread safe.