Dependency Injection re-designed: simpler, yet more powerful new principles
Extremely lightweight with lightning-fast runtime and no third-party dependencies
Entire runtime code of Injector consists of ~50 lines of code, and all dependencies graph preprocessing is performed at start-up time
Extensive use of Java8+ functional-style programming: for binding definitions, user-defined binding transformations, and binding generators
Even more powerful: support of nested scopes, singletons and transient bindings, modules, optimized multi-threaded and single-threaded injectors
All reflection is completely abstracted out in a separate module and is entirely optional
Completely transparent for introspection of its dependency graph and instances
Even more strict checks of dependency graph during start-up time
Has no dependencies on its own: can be used as a standalone component
DataKernel DI :: Basics
If you are familiar with common DI concepts, you can skip this part and go directly to DI internals
Dependency Injection, Keys, Bindings
Applications consist of components and each component has an inner id, which is called Key.
Key consists of Type type and nullable Name (useful when you
need different implementations of the same Type):
Application components can require some dependencies in order to be created.
DI takes care of supplying application components with these required objects.
In order to do it, DI needs to know what it needs to provide and how to use provided objects.
So, Binding has two corresponding attributes:
a set of needed for creation Set<Dependency> dependencies (POJO with Key)
compiler which will compile required instances
Binding is like a “recipe” of how to create an instance of a component:
dependencies show what ingredients should be used
compiler knows how to cook them together
Now we need something that can use the recipe to cook the component properly, and here comes the Injector.
Provides all required dependencies (injects) for the component
recursively traversing the dependencies graph in a postorder way and creates them first.
Bindings are by default singletons - if an instance was created once, it won’t be recreated from scratch again. If
it is needed for other bindings, Injector will take it from cache. You don’t need to apply any additional
annotations for it.
To provide the requested key, Injector recursively creates all of its dependencies and falls back to injector of its
parent scope if binding in its scope is not found.
Datakernel DI :: Internals
In short - scopes give us “local singletons” which live as long as the scope itself.
In DataKernel DI scopes are a bit different from other DI frameworks:
The internal structure of the Injector is a prefix tree and the prefix is a scope.
The identifiers (or prefixes) of the tree are simple annotations.
Injector can enter the scope. This means you create an Injector and its scope will be set to the one that it’s entering.
This can be done multiple times, so you can have N injectors in certain scope.
If no Multibinder is defined for particular key, exception is thrown
If dependency graph has missing dependencies, they are automatically generated with BindingGenerators:
BindingGenerators are user-defined and exported by Modules
There is an implicit DefaultModule with default BindingGenerator, which automatically provides required dependencies by scanning @Inject annotations of required classes
User-specified modules can also export custom binding generators for special classes
You can opt-out of using DefaultModule and its default BindingGenerators
All bindings are transformed with user-provided BindingTransformers:
To intercept/modify/wrap provided instances
To intercept/modify/wrap the dependencies of provided instances
Multibinders, BindingGenerators and BindingTransformers can be made with clean and extremely simple Java8+ functional DSL
Resulting dependency graph is validated - checked for cyclic and missing dependencies, then compiled into a final scope tree and passed to Injector
It’s trivial to manually implement the Module interface, but it’s even easier to extend AbstractModule, which
supports @Provides method scanning and the DSL for creating/transforming/generating bindings.
We’ve compared DataKernel DI to Guice and Spring in the same scenario, using JMH as the benchmark tool.
We ran benchmarks in AverageTime mode and made 20 measurements.
All measurement results are represented in nanoseconds.
Benchmarks were launched on a machine with the following parameters: Ubuntu 18.04 bionic,
Kernel: x86_64 Linux 4.15.0-55-generic,
CPU: Intel Core i5-8400 @ 6x 4GHz [27.8°C].
To represent the main concepts and features of DataKernel DI, we’ve created an example, which starts
with low-level DI concepts and gradually covers more specific advanced features.
To run the examples, you need to clone DataKernel from GitHub:
$ git clone https://github.com/softindex/datakernel And import it as a Maven project. Check out branch v3.1. Before running the examples, build the project.
This example is located at datakernel -> core-di -> test and named DiFollowUpTest
In this example we have a kitchen, where you can automatically create tasty cookies with our wonderful DI.
Before we get to cooking, there are several POJOs with default constructors marked with @Inject annotation: Kitchen,
Sugar, Butter, Flour, Pastry and Cookie.
Let’s bake a Cookie using DI in a hardcore way.
First of all, we need to provide all the needed ingredients for the cookie which are Sugar, Butter and
Flour. Next, there is a recipe for Pastry, which includes
ingredients we already know how to get. Finally, we can store a recipe of how to bake a Cookie.
It’s baking time now! Just create the injector with all these recipes and ask it for your Cookie instance.
Bind Using ModuleBuilder
This time we will bake a Cookie with a simple DSL.
We will bundle our recipes for Sugar, Butter and Flour in a ‘cookbook’ module.
Instead of creating bindings explicitly and storing them directly in a map, we just bind the recipes in our module and then give it to the injector.
Bind Using @Provides
It’s time for real Cookie business. Instead of making bindings explicitly, we will use the declarative DSL.
Like in the previous example, we create a cookbook module, but this time all bindings for the ingredients will be created
automatically from the provider methods. These methods are marked with the @Provides annotation:
Bind Using Instance or Class Scan
Sometimes it happens that you prepare an injection scheme, but this scheme is not a module. But there is a scan()
method which can help you to make a connection between DI entities and your scheme.
If your class provides a scheme, you can use it easily:
Automatic Bind Using @Inject
When we created our POJOs, we’ve marked their constructors with @Inject annotation:
If a binding depends on a class that has no known binding, injector will try to automatically generate binding for it.
It will search for @Inject annotation on its constructors, static factory methods or the class itself (in this case
the default constructor is used) and use them as a factory in generated binding.
Since nothing depends on the Cookie binding, by default no bindings will be generated at all.
Here we use a plain bind to tell the injector that we want this binding to be present.
Thus the whole tree of bindings it depends on will be generated:
Different Cookie Recipes Using @Named
Let’s be trendy and bake a sugar-free cookie. In order to do so, along with @Provides annotation, we will
also use @Named annotation and provide two different Sugar, Pastry and Cookie factory functions. This
approach allows to use different instances of the same class. Now we can tell our injector, which of the
cookies we need - a normal one or sugar-free.
Also ModuleBuilder usage is also possible, but we want you to use Module.create().scan() method.
Cooking Non-singleton Cookies Using Scopes
Our cookies turned out to be so amazingly tasty, that now there are a lot of people who want to try them. But there is a
small problem, DataKernel DI makes instances singleton by default. And we can’t sell the same cookie to all our
Luckily, there is a solution: we can use a custom ScopeAnnotation@OrderScope to create ORDER_SCOPE scope:
So our cookbook will look as follows:
In this way, only kitchen will remain singleton:
We received 10 orders from our customers, so now we need 10 instances of cookies:
First, we inject an instance of Kitchen. Now this instance is stored in the root scope injector.
Next, we create 10 subinjectors which enter ORDER_SCOPE.
Each subinjector creates only one instance of Cookie and refers to the single Kitchen instance of their parent root scope.
You can configure the process of how your injector gets instances and transform this process. For example, you can
simply add some logging by overriding configure method:
Now you will receive an output which will represent the time when an instance was created and the instance itself.