Dependency injection in a multi module project

What we learned from introducing a DI framework to Plaid

Ben Weiss
Ben Weiss
Jan 9, 2019 · 6 min read
Illustrated by Virginia Poltrack

This is not an article about dependency injection in general or about why we picked library X over Y.
Instead this post covers key takeaways of our efforts to modularize Plaid from a dependency injection perspective.

Our setup

In a previous post I wrote about the overall modularization story of Plaid.

Let me quickly recap what Plaid looks like from a bird’s eye view.

We have an app module, which contains the main launcher activity. Also there are several dynamic feature modules (DFM) which depend on the app module. Each DFM contains at least one activity, code and resources related to only the feature in question.

The app module depends on a core module which contains shared code and resources as well as third party libraries.

Plaid’s module dependency graph

Before we started modularizing and introducing Dagger as main actor for dependency injection, Plaid’s code had a couple of classes and functions like this:

class DesignerNewsInjector {    fun providesApi(...): DesignerNewsService { ... }}

While this is a perfectly fine solution, we were left with writing a lot of boilerplate and plumbing code by hand.

Wherever anything was required from the injector, we had to call the underlying function at the right point, in many cases either object initialization or onCreate.

A very brief intro to dependency injection

Dependency injection basically means that you don’t create objects in the place you need them but rather create them somewhere else. Then references to these objects can get passed into classes where they are required.

This can be done either manually or with one of the many libraries out there. We chose Dagger 2. Thanks to Dagger, all we have to do to get a hold of an initialized service that’s ready to use is this:

@Inject lateinit var service: DesignerNewsService

All the dependencies of the service can be passed into the provides function as parameters. Having chosen Dagger for our dependency injection needs means that our dependency graphs are created at compile time. Bear this in mind for the following sections.

Our approach to introducing Dagger to Plaid

At the time we decided to introduce Dagger into Plaid we had already learned a valuable lesson that is particularly true for modularization.

Don’t try to cover too much ground at once.

This means that it’s worthwhile to spend some time figuring out the smallest scope necessary to implement a new feature. This MVP we then discussed within the team to see whether we’re moving in the right direction. Adhering to this practice prevents us from running off with changes that are too large to efficiently work with. This also allows us to gradually roll out changes throughout our code base while everyone else continued working on their tasks.

Within Plaid we used the already proven about feature module as playground for Dagger. Here we could add Dagger without interfering with other modules or workload. You can find the initial commit here.

Dependency graphs

When introducing a dependency injection library to a monolithic application usually there’s one single dependency graph for the whole of the application.

Classic simplified dependency graph in a monolithic project

This enables sharing dependencies between components. In some libraries dependencies can be scoped in order to avoid conflicts or provide a specific implementation to an injection target.

Modular oddities

For a modularized app, especially for one using dynamic feature modules this doesn’t work though. Let’s take a closer look at how application and dynamic feature modules’ depend on one another. A dynamic feature module knows that an application module exists. The application module kind of knows that the dynamic feature module exists, but can’t directly execute code from within that module. For dependency injection, this means that the graph has to be broken into pieces.

For a modularized app the simplified dependency graph usually looks kind of like this.

Modules have clear boundaries and are encapsulated within a DFM’s dependency graph

More concrete, within Plaid the component landscape looks like this.

Plaid’s component landscape

Each DFM has its own component named after the feature module it sits in. As does the app module through HomeComponent.

There’s also a component containing shared dependencies. It sits within core and is called CoreComponent. The main idea behind CoreComponent is to provide objects that can be used throughout the app. It combines a couple of Dagger modules which sit within the core library and can be re-used throughout the app.

Also, since the graphs are directed there’s only one way to share Dagger components:
A DFM can access Dagger components from the application module. The application module can access components from libraries it depends on. But not the other way around.

Sharing components across module boundaries

In order to share Dagger components, they need to be made accessible throughout the application. Within Plaid we decided to make our CoreComponent accessible via the Application class.

class PlaidApplication : Application() {  private val coreComponent: CoreComponent by lazy {
companion object { @JvmStatic fun coreComponent(context: Context) =
(context.applicationContext as PlaidApplication).coreComponent

The instantiated core component can now be accessed from anywhere within the app where there’s a context available by calling PlaidApplication.coreComponent(context).

Using an extension function makes access to this even sweeter:

fun Activity.coreComponent() = PlaidApplication.coreComponent(this)

Components in Components

To incorporate CoreComponent in another component it is necessary to provide it during component creation. Let’s see how this works within SearchComponent:

@Component(modules = [...], dependencies = [CoreComponent::class])
interface SearchComponent {
interface Builder {
fun coreComponent(coreComponent: CoreComponent): Builder
// modules

During initialization of the generated DaggerSearchComponent we set CoreComponent like this:

// modules

The trick here is to set CoreComponent as a dependency of SearchComponent.

modules = [SearchModule::class],
dependencies = [CoreComponent::class]
interface SearchComponent : BaseActivityComponent<SearchActivity>

CoreComponent is a dependency of SearchComponent. When CoreComponent is included as a component dependency of SearchComponent like above, all of CoreComponent’s methods can be used in SearchComponent or other Dagger components as if they were @Provides methods.

Component dependencies with their respective modules (in green) providing implementations to SearchActivity

A benefit of this approach is that @Modules don’t have to be repeated throughout the feature graphs but can be transparently provided through CoreComponent or modules bound by it.

For example, CoreDataModule is bound in CoreComponent and provides Retrofit amongst others. That Retrofit instance can now be accessed in any component whereCoreComponent is incorporated.

What’s next

After reading this article you have seen that modularizing your app also has to take dependency injection into account. The introduced feature module boundaries are reflected in DI through separated dependency graphs. Being aware of these restrictions enables finding the right place for shared components.

You can dive into the code to see how we solved dependency injection using Dagger in Plaid.

CoreComponent is a good starting point, as is AboutComponent since it doesn’t have many external dependencies.

Android Developers

The official Android Developers publication on Medium

Thanks to Zac Sweers and Jake Wharton

Ben Weiss

Written by

Ben Weiss

#Android Developer Relations @ Google

Android Developers

The official Android Developers publication on Medium

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade