The app architecture has three layers: a [data layer](https://developer.android.com/jetpack/guide/data-layer), a [domain layer](https://developer.android.com/jetpack/guide/domain-layer) and a [UI layer](https://developer.android.com/jetpack/guide/ui-layer).
The architecture follows a reactive programming model with [unidirectional data flow](https://developer.android.com/jetpack/guide/ui-layer#udf). With the data layer at the bottom, the key concepts are:
When the app is first run it will attempt to load a list of news resources from a remote server (when the `prod` build flavor is selected, `demo` builds will use local data). Once loaded, these are shown to the user based on the interests they choose.
![Diagram showing how news resources are displayed on the For You screen](images/architecture-2-example.png "Diagram showing how news resources are displayed on the For You screen")
Here's what's happening in each step. The easiest way to find the associated code is to load the project into Android Studio and search for the text in the Code column (handy shortcut: tap <kbd>⇧ SHIFT</kbd> twice).
<td>On app startup, a <ahref="https://developer.android.com/topic/libraries/architecture/workmanager">WorkManager</a> job to sync all repositories is enqueued.
<td>The <code>ForYouViewModel</code> calls <code>GetSaveableNewsResourcesUseCase</code> to obtain a stream of news resources with their bookmarked/saved state. No items will be emitted into this stream until both the user and news repositories emit an item. While waiting, the feed state is set to <code>Loading</code>.
<td><code>OfflineFirstNewsRepository</code> calls <code>RetrofitNiaNetwork</code> to execute the actual API request using <ahref="https://square.github.io/retrofit/">Retrofit</a>.
<td><code>OfflineFirstNewsRepository</code> syncs the remote data with <code>NewsResourceDao</code> by inserting, updating or deleting data in a local <ahref="https://developer.android.com/training/data-storage/room">Room database</a>.
<td>When data changes in <code>NewsResourceDao</code> it is emitted into the news resources data stream (which is a <ahref="https://developer.android.com/kotlin/flow">Flow</a>).
<td><code>OfflineFirstNewsRepository</code> acts as an <ahref="https://developer.android.com/kotlin/flow#modify">intermediate operator</a> on this stream, transforming the incoming <code>PopulatedNewsResource</code> (a database model, internal to the data layer) to the public <code>NewsResource</code> model which is consumed by other layers.
<td><code>GetSaveableNewsResourcesUseCase</code> combines the list of news resources with the user data to emit a list of <code>SaveableNewsResource</code>s.
Repositories are the public API for other layers, they provide the _only_ way to access the app data. The repositories typically offer one or more methods for reading and writing data.
Data is exposed as data streams. This means each client of the repository must be prepared to react to data changes. Data is not exposed as a snapshot (e.g. `getModel`) because there's no guarantee that it will still be valid by the time it is used.
Reads are performed from local storage as the source of truth, therefore errors are not expected when reading from `Repository` instances. However, errors may occur when trying to reconcile data in local storage with remote sources. For more on error reconciliation, check the data synchronization section below.
<td>Persistent unstructured data associated with user preferences, specifically which Topics the user is interested in. This is defined and modeled in a .proto file, using the protobuf syntax.
Repositories are responsible for reconciling data in local storage with remote sources. Once data is obtained from a remote data source it is immediately written to local storage. The updated data is emitted from local storage (Room) into the relevant data stream and received by any listening clients.
This approach ensures that the read and write concerns of the app are separate and do not interfere with each other.
In the case of errors during data synchronization, an exponential backoff strategy is employed. This is delegated to `WorkManager` via the `SyncWorker`, an implementation of the `Synchronizer` interface.
See the `OfflineFirstNewsRepository.syncWith` for an example of data synchronization.
The [domain layer](https://developer.android.com/topic/architecture/domain-layer) contains use cases. These are classes which have a single invocable method (`operator fun invoke`) containing business logic.
These use cases are used to simplify and remove duplicate logic from ViewModels. They typically combine and transform data from repositories.
For example, `GetSaveableNewsResourcesUseCase` combines a stream (implemented using `Flow`) of `NewsResource`s from a `NewsRepository` with a stream of `UserData` objects from a `UserDataRepository` to create a stream of `SaveableNewsResource`s. This stream is used by various ViewModels to display news resources on screen with their bookmarked state.
Notably, the domain layer in Now in Android _does not_ (for now) contain any use cases for event handling. Events are handled by the UI layer calling methods on repositories directly.
The ViewModels receive streams of data from use cases and repositories, and transforms them into UI state. The UI elements reflect this state, and provide ways for the user to interact with the app. These interactions are passed as events to the ViewModel where they are processed.
![Diagram showing the UI layer architecture](images/architecture-4-ui-layer.png "Diagram showing the UI layer architecture")
### Modeling UI state
UI state is modeled as a sealed hierarchy using interfaces and immutable data classes. State objects are only ever emitted through the transform of data streams. This approach ensures that:
* the UI state always represents the underlying app data - the app data is the source-of-truth.
The feed (a list) of news resources on the For You screen is modeled using `NewsFeedUiState`. This is a sealed interface which creates a hierarchy of two possible states:
ViewModels receive streams of data as cold [flows](https://kotlin.github.io/kotlinx.coroutines/kotlinx-coroutines-core/kotlinx.coroutines.flow/-flow/index.html) from one or more use cases or repositories. These are [combined](https://kotlin.github.io/kotlinx.coroutines/kotlinx-coroutines-core/kotlinx.coroutines.flow/combine.html) together, or simply [mapped](https://kotlinlang.org/api/kotlinx.coroutines/kotlinx-coroutines-core/kotlinx.coroutines.flow/map.html), to produce a single flow of UI state. This single flow is then converted to a hot flow using [stateIn](https://kotlin.github.io/kotlinx.coroutines/kotlinx-coroutines-core/kotlinx.coroutines.flow/state-in.html). The conversion to a state flow enables UI elements to read the last known state from the flow.
The `InterestsViewModel` exposes `uiState` as a `StateFlow<InterestsUiState>`. This hot flow is created by obtaining the cold flow of `List<FollowableTopic>` provided by `GetFollowableTopicsUseCase`. Each time a new list is emitted, it is converted into an `InterestsUiState.Interests` state which is exposed to the UI.
User actions are communicated from UI elements to ViewModels using regular method invocations. These methods are passed to the UI elements as lambda expressions.
The `InterestsScreen` takes a lambda expression named `followTopic` which is supplied from `InterestsViewModel.followTopic`. Each time the user taps on a topic to follow this method is called. The ViewModel then processes this action by informing the topics repository.