Getting started with Kodein DI
Kodein (for KOtlin DEpendency INjection) is a DI framework. It allows you to bind your business unit interfaces with their implementation and thus having each business unit being independent.
This guide assumes a JVM target. Kodein works exactly the same on all targets that are supported by Kotlin (JVM, Android, JS, Native). |
Choose & Install
Flavour
The type erasure problem
Type erasure is the process in which all generic parameters are suppressed when compiled to bytecode.
In essence, in JVM bytecode, List<String>
& List<Date>
are the same object: List
.
Generic parameters are for type safety only, they only exist at the source code level and their purpose is to make sure that you are not putting a Date
in a List<String>
.
As this verification is done at compile time, all generic information is removed.
This means that you cannot have List<String>::class
, only List<*>::class
.
For feature parity and simplicity, Kotlin/JS & Kotlin/Native work the same way. |
This leads to a problem, if you are using generic types in the DI container, because the compiler erases generic type information when using Kodein api.
Kodein flavours
Now, I may have lied a bit: not all generic information is lost, most of it. There is still a way to access generic type information, but it is costly, and it only works on the JVM & Android.
Kodein offers two different flavours: kodein-generic
& kodein-erased
.
Type erasure | Optimized | Non-generic bindings | Generic bindings | |
---|---|---|---|---|
generic |
immune |
no |
simple |
simple |
erased |
subject |
yes |
simple |
complex |
Which to use should, of course, be selected for your needs. However, here is a simple decision tree that I believe should help:
-
Will the application target other platform then the JVM?
-
Yes → Use
kodein-erased
as it is the only option available for non-JVM platforms & multi-platform. -
No, only the JVM → Will object creation & DI be performance critical?
-
No, standard stuff → use
kodein-generic
as it allows for easier and more consistent code. -
Yes → use
kodein-erased
, but be aware that your experience might be degraded ;)
-
-
Keep in mind that it should be quite rare for DI to be performance critical. DI happens only once per injected object, so to measure how critical DI is, ask yourself how much time injected objects will be created, and if these creations are themselves performance critical. |
Install
With Maven
Add the JCenter repository:
<repositories>
<repository>
<id>jcenter</id>
<url>https://jcenter.bintray.com</url>
</repository>
</repositories>
Then add the dependency:
<dependencies>
<dependency>
<groupId>org.kodein.di</groupId>
<artifactId>kodein-di-generic-jvm</artifactId>
<version>6.5.5</version>
</dependency>
</dependencies>
Use kodein-generic-jvm or kodein-erased-jvm .
|
Bindings
Definition
In DI, each business unit will have dependencies. Those dependencies should (nearly almost) always be interfaces. This allows:
-
Loose coupling: the business unit knows what it needs, not how those needs are fulfilled.
-
Unit testing: You can unit test the business unit by mocking its dependencies.
-
Separation: Different people can work on different units / dependencies.
Business units are very often themselves dependencies to other business units. |
Each business unit and dependency need to be managed.
Some dependencies need to be created on demand, while other will need to exist only once.
For example, a Random
object may needs to be re-created every time one is needed, while a Database
object should exist only once in the application.
Have a look at these two sentences:
-
"I want to bind the
Random
type to a provider that creates aSecureRandom
implementation. -
"I want to bind the
Database
type to a singleton that contains aSQLiteDatabase
implementation.
In DI, you bind a type (often an interface) to a binding that manages an implementation. A binding is responsible for returning the implementation when asked. In this example, we have seen to different bindings:
-
The provider always returns a new implementation instance.
-
The singleton creates only one implementation instance, and always returns that same instance.
Declaration
In Kodein, bindings are declared in a Kodein Block. The syntax is quite simple:
val kodein = Kodein {
bind<Random>() with provider { SecureRandom() }
bind<Database>() with singleton { SQLiteDatabase() }
}
As you can see, Kodein offers a DSL (Domain Specific Language) that allows to very easily declare a binding.
Kodein offers many bindings that can manage implementations: provider
, singleton
, factory
, multiton
, instance
, and more, which you can read about in the bindings section of the core documentation.
Most of the time, the type of the interface of the dependency is enough to identify it.
There is only one Database
in the application, so if I’m asking for a Database
, there is no question of which Database
I need: I need the database.
Same goes for Random
. There is only one Random
implementation that I am going to use.
If I am asking for a Random
implementation, I always want the same type of random: SecureRandom
.
There are times, however, where the type of the dependency is not enough to identify it.
For example, you may have two Database
in a mobile application: one being local, and another being a proxy to a distant Database.
For cases like this, Kodein allows you to "tag" a binding: add an additional information to tag it.
val kodein = Kodein {
bind<Database>(tag = "local") with singleton { SQLiteDatabase() }
bind<Database>(tag = "remote") with provider { DatabaseProxy() }
}
Separation
When building large applications, there are often different modules, built by their own team, each defining their own business units.
Kodein allows you to define binding modules that can later be imported in a Kodein container:
val module = Kodein.Module {
bind<Database>(tag = "local") with singleton { SQLiteDatabase() }
bind<Database>(tag = "remote") with provider { DatabaseProxy() }
}
val kodein = Kodein {
import(module)
}
Injection & Retrieval
The container
All declarations are made in the constructor of a Kodein container.
Think of the Kodein container as the glue that allows you to ask for dependency.
Whatever dependency you need, if it was declared in the Kodein container constructor, you can get it by asking Kodein.
This means that you always need to be within reach of the Kodein
object.
There are multiple ways of doing so:
-
You can pass the
Kodein
object around, as a parameter to created objects. -
You can have the
Kodein
object being statically available (in Android, for example, it is common to use a property of theApplication
object)
Injection vs Retrieval
Kodein supports two different methods to allow a business unit to access its dependencies: injection and retrieval.
-
When dependencies are injected, the class is provided its dependencies at construction.
-
When dependencies are retrieved, the class is responsible for getting its own dependencies.
Dependency injection is more pure in the sense that an injected class will have its dependency passed at construction and therefore not know anything about Kodein.
It is however more cumbersome, and does not provide a lot of features.
At contrario, dependency retrieval is easier and feature full, but it requires the class to know about Kodein.
In the end, it boils down to that question: Do you need this class to be Kodein independant? If you are building a library that will be used in multiple architecture, you probably do. If you are building an application, you probably don’t.
Injection
If you want your class to be injected, then you need to declare its dependencies at construction:
class Presenter(private val db: Database, private val rnd: Random) {
}
Now you need to be able to create a new instance of this Presenter
class.
With Kodein, this is very easy:
val presenter by kodein.newInstance { Presenter(instance(), instance()) }
For each argument of the Presenter
constructor, you can simply use the instance()
function, and Kodein will actually pass the correct dependency.
Retrieval
When using retrieval, the class needs to be available to access a Kodein instance, either statically or by argument. In these examples, we’ll use the argument.
class Presenter(val kodein: Kodein) {
private val db: Database by kodein.instance()
private val rnd: Random by kodein.instance()
}
Note the usage of the by
keyword.
When using dependency retrieval, properties are retrieved lazily.
You can go a bit further and use the KodeinAware
interface, which unlocks a lot of features:
class Presenter(override val kodein: Kodein): KodeinAware {
private val db: Database by instance()
private val rnd: Random by instance()
}
Note that because everything is lazy by default, the access to the Kodein
object in a KodeinAware
class can itself be lazy:
class Presenter(): KodeinAware {
override val kodein by lazy { getApplicationContext().kodein }
private val db: Database by instance()
private val rnd: Random by instance()
}
Direct
If you don’t want everything to be lazy (as it is by default), Kodein has you covered. Head to the Retrieval: Direct section of the core documentation.
Transitive dependencies
Let’s say we want to declare the Provider
in a binding.
It has its own dependencies.
Dependencies of dependencies are transitive dependencies.
Handling those dependencies is actually very easy.
If you are using injection, you can pass the argument the exact same way:
val kodein = Kodein {
bind<Presenter>() with singleton { Presenter(instance(), instance()) }
}
If you are using retrieval, simply pass the kodein property:
val kodein = Kodein {
bind<Presenter>() with singleton { Presenter(kodein) }
}
Where to go next
Kodein offers a lot of features. All of them are documented, you can find them here: http://kodein.org/Kodein-DI/?6.5.
If you are using Kodein on Android, you should read the Kodein on Android documentation.