Intro to Clean Architecture in Android. Example with independent features.

Intro

There are lots of articles about Clean Architecture:

• Google guide
• Google search
• Google Gemini he, he..

So, I don't wanna copy those guides and my advice is read them carefully. But I have to write about domain layer for excluding confusion of terms.

Domain layer

There is something wrong with understanding domain layer in the Android development context. Here's the quote from the Google guide:

Note: The term "domain layer" is used in other software architectures, such as "clean" architecture, and has a different meaning there. Don't confuse the definition of "domain layer" defined in the Android official architecture guidance with other definitions you may have read elsewhere. There may be subtle but important differences.

The common pattern of software architecture says that the domain layer is business-logic. But what is business-logic in Android development context? - It's not about Enterprise rules, it's about how transform data for showing on the smartphone's screen. There is no complex rules or something else.

Of course, you may have not simple rules for showing data. You may have caching, paging, aggregation, complex calculation, rendering etc. But the core questions are:

1. How we show data for user?
2. How we collect user's input?

Okay, in the examples below I'll use domain layer as layer with reusable logic with UseCases as in the recommendations.

On the other side, I try to make domain layer as independent layer regardless of Google diagram where we see dependency to data layer. This is a turn to a classic Clean Architecture.

So, the domain layer consists from interfaces of UseCases, data classes of models, independent (from other libraries) clear Kotlin logic that is need in many features.

UseCases

Another point I wanna write about UseCases. There is a common pattern to make UseCases as single function, such as:

class SomeUseCaseImpl @Inject constructor(
    private val someRepo: SomeRepository
): SomeUseCase {
    override suspend fun invoke() = someRepo.someAction()
}

I don't like this approach. I see in this more complex solution that have to be. You may say: "Bro, it's a Single responsibility principle from SOLID". But in general this is just sophistry, example functions: getMoney(), saveMoney(), changeMoney() etc. Do each of these functions have Single responsibility, or we can say that the MoneyUseCase with those all functions has Single responsibility of working with money?

Got an idea? Credo: Less boilerplate, more code elegance

---

In the multi-modules projects the Clean Architecture is the crucial point for maintain in future. So, I'll concentrate in this kind of projects in my examples.

Example 1

Requirements

1. Platform is only Android (not CMP/KMP)
2. Our project is multi-module
3. The :app module have a navigation to each screen
4. The features have their own screens
5. The features are independent on each other
6. The feature can't navigate to another feature directly
7. There's some pieces of data that have to be in many screens

Structure

Generally structure

Domain Core module

• Has data classes and interfaces of repositories.
• It's an independent module.
• Often it's only Kotlin library, but not always.

I think I have to move attention to library type. The common way is making domain module as Kotlin library. And it's correct way, the best solution... ideal...
But sometimes we have to move away from this pattern. And the main problem is in the resources and annotations:

• Annotations @StringRes, @DrawableRes etc. is needs in androidx.annotation:annotation library.
• Also we wanna add some resources as params, such as

enum CustomErrors(val messageResId: Int) {
    ERROR_1(R.string.err_1_msg),
    ERROR_2(R.string.err_2_msg),
    //etc
}

So, what can I say? - Do it. Make module as Android library. It's wrong and it's not a canonical way... but it's much more simple for maintain in future.

Domain Shared module

Here are the single-line UseCases and their implementations. The module depends on only :domain:core.

Feature module

Depends on:

• :core:ui - common UI components
• :domain:core - models and repositories
• :domain:shared - common UseCases

Here you are packages:

• navigation Here are navigation routes, part of navigation graph. The only module that has public modifier. Other have internal.
• presentation Composable functions, ViewModels, UiStates... as usual... One point - in ViewModel we also have to inject shared UseCases from :domain:shared we need in.
• domain Feature UseCases
• data Implementations of feature UseCases

Summary

So, I tried to make an optimal architecture tailored to project's set of requirements.

• data module as SSOT
• domain module as a model
• feature module as a UI
• app module as a glue

---

You know, there are no universal scheme for organizing modules and creating architecture. It depends on requirements. In text above was certain example with certain requirements, and that solution can not be applied for another type of project. Do not copy without thinking.

Okay, it's all for now. Thanks for reading!