In Spring it's possible to define bean dependencies in separate modules, which are then resolved via the classpath at runtime. Is it possible to do something similar in Quarkus?
For example, a multi-module setup that looks like this:
- service
- service-test
- service-artifact
In Spring it's possible to define #Configuration in the service module, that resolves concrete dependencies at runtime via the classpath of its current context, either service-test or service-artifact, allowing injection of dummy or test dependencies when under test, and real ones in the production artifact.
For example, a class in service requires an instance of SomeInterface. The implementation of SomeInterface is defined in either the -test or -artifact module. The service module has no direct dependency on either the -test or -artifact modules.
Some code:
In the service module:
#ApplicationScoped
class OrderService(private val repository: OrderRepository) {
fun process(order: Order) {
repository.save(order)
}
}
interface OrderRepository {
fun save(order: Order)
}
In the service-test module:
class InMemoryOrderRepository : OrderRepository {
val orders = mutableListOf<Order>()
override fun save(order: Order) {
orders.add(order)
}
}
class OrderServiceTestConfig {
#ApplicationScoped
fun orderRepository(): OrderRepository {
return InMemoryOrderRepository()
}
}
#QuarkusTest
class OrderServiceTest {
#Inject
private lateinit var service: OrderService
#Test
fun `injected order service with resolved repository dependency`() {
// This builds and runs OK
service.process(Order("some_test_order"))
}
}
Where I have tried to replicate a Spring-style setup as above in Quarkus, ArC validation is failing with UnsatisfiedResolutionException on the build of the service module, even though everywhere it is actually consumed provides the correct dependencies; a test successfully resolves the dependency and passes.
How do I achieve the separation of dependency interface from the implementation, and keep ArC validation happy, with Quarkus?
(Note: this behaviour occurs with Java and Maven also.)
I have included a maven example here. Note that ./mvnw install fails with the UnsatisfiedResolutionException but that it's possible to build and run the test successfully using ./mvnw test.
Build files:
root project build.gradle.kts:
import org.jetbrains.kotlin.gradle.tasks.KotlinCompile
plugins {
kotlin("jvm") version "1.3.72"
kotlin("plugin.allopen") version "1.3.72"
}
allprojects {
group = "my-group"
version = "1.0.0-SNAPSHOT"
repositories {
mavenLocal()
mavenCentral()
}
}
subprojects {
apply {
plugin("kotlin")
plugin("kotlin-allopen")
}
java {
sourceCompatibility = JavaVersion.VERSION_11
targetCompatibility = JavaVersion.VERSION_11
}
allOpen {
annotation("javax.ws.rs.Path")
annotation("javax.enterprise.context.ApplicationScoped")
annotation("io.quarkus.test.junit.QuarkusTest")
}
apply {
plugin("kotlin")
}
dependencies {
implementation("org.jetbrains.kotlin:kotlin-reflect")
implementation("org.jetbrains.kotlin:kotlin-stdlib-jdk8")
}
tasks.withType<KotlinCompile> {
kotlinOptions.jvmTarget = JavaVersion.VERSION_11.toString()
kotlinOptions.javaParameters = true
}
}
build.gradle.kts for service:
import io.quarkus.gradle.tasks.QuarkusDev
plugins {
id("io.quarkus") version "1.9.1.Final"
}
apply {
plugin("io.quarkus")
}
dependencies {
implementation(project(":common:model"))
implementation(enforcedPlatform("io.quarkus:quarkus-universe-bom:1.9.1.Final"))
implementation("io.quarkus:quarkus-kotlin")
}
build.gradle.kts for service-test:
import io.quarkus.gradle.tasks.QuarkusDev
plugins {
id("io.quarkus") version "1.9.1.Final"
}
apply {
plugin("io.quarkus")
}
dependencies {
implementation(project(":service"))
implementation(enforcedPlatform("io.quarkus:quarkus-universe-bom:1.9.1.Final"))
implementation("io.quarkus:quarkus-kotlin")
testImplementation("io.quarkus:quarkus-junit5")
}
Try to use instance injection (java example):
import javax.enterprise.inject.Instance;
...
#Inject
Instance<MyBeanClass> bean;
...
bean.get(); // for a single bean
bean.stream(); // for a collection
Unfortunately, Quarkus has a bit different way of creating and injecting beans as in Spring.
It's using "simplified bean discovery", and that means that the beans are scanned on the classpath during the build time, but only those that have annotations considered as "discovery mode", are taken into the account.
Those would be: #ApplicationScoped, #SessionScoped, #ConversationScoped and #RequestScoped, #Interceptor and #Decorator more described here
In addition to that, beans must not have the visibility boundaries.
In you'd like to use beans from other modules, create configuration within that module annotated with #Dependent and create beans with #Producer annotation and one of the above annotations.
But notice, despite that, some beans are treated by Quarkus in a special way (ex. all beans that have #Path annotation) and those should be annotated preferably with #ApplicationScope, using either constructor, or field injection. Produced by #Producer methods won't allow for all the magic that Quarkus is doing.
If you'd like some more quarkus dependant beans, ex. the bean that bounds a configuration (using #ConfigMapping annotated beans), in addition you need to have either beans.xml in your META-INF directory, or which seems easier to add the jandex index to your build system:
plugins {
id("io.quarkus") version "2.14.1.Final"
id("org.kordamp.gradle.jandex") version "1.0.0"
}
Summary: don't use configuration beans as in spring, only the constructor/field injection, and to have beans discovered from different modules, add the jandex index file using plugin.
Related
Just starting a new Gradle project.
This test passes:
def 'Launcher.main should call App.launch'(){
given:
GroovyMock(Application, global: true)
when:
Launcher.main()
then:
1 * Application.launch( App, null ) >> null
}
... until, to get another test using a (Java) Mock to work, I have to add these dependencies:
testImplementation 'net.bytebuddy:byte-buddy:1.10.8'
testImplementation 'org.objenesis:objenesis:3.1'
(NB I assume these versions are OK for Groovy 3.+, which I'm now using ... both are the most up-to-date available at Maven Repo).
With these dependencies the above test fails:
java.lang.InstantiationError: javafx.application.Application
at org.objenesis.instantiator.sun.SunReflectionFactoryInstantiator.newInstance(SunReflectionFactoryInstantiator.java:48)
at org.objenesis.ObjenesisBase.newInstance(ObjenesisBase.java:73)
at org.objenesis.ObjenesisHelper.newInstance(ObjenesisHelper.java:44)
at org.spockframework.mock.runtime.MockInstantiator$ObjenesisInstantiator.instantiate(MockInstantiator.java:45)
at org.spockframework.mock.runtime.MockInstantiator.instantiate(MockInstantiator.java:31)
at org.spockframework.mock.runtime.GroovyMockFactory.create(GroovyMockFactory.java:57)
at org.spockframework.mock.runtime.CompositeMockFactory.create(CompositeMockFactory.java:42)
at org.spockframework.lang.SpecInternals.createMock(SpecInternals.java:47)
at org.spockframework.lang.SpecInternals.createMockImpl(SpecInternals.java:298)
at org.spockframework.lang.SpecInternals.createMockImpl(SpecInternals.java:288)
at org.spockframework.lang.SpecInternals.GroovyMockImpl(SpecInternals.java:215)
at core.AppSpec.Launcher.main should call App.launch(first_tests.groovy:30)
I confess that I have only the sketchiest notion of what "bytebuddy" and "objenesis" actually do, although I assume it is fiendishly clever. Edit: having just visited their respective home pages my notion is now slightly less sketchy, and yes, it is fiendishly clever.
If an orthodox solution to this is not available, is it by any chance possible to turn off the use of these dependencies for an individual feature (i.e. test)? Possibly using some annotation maybe?
Edit
This is an MCVE:
Specs: Java 11.0.5, OS Linux Mint 18.3.
build.gradle:
plugins {
id 'groovy'
id 'java'
id 'application'
id 'org.openjfx.javafxplugin' version '0.0.8'
}
repositories { mavenCentral() }
javafx {
version = "11.0.2"
modules = [ 'javafx.controls', 'javafx.fxml' ]
}
dependencies {
implementation 'org.codehaus.groovy:groovy:3.+'
testImplementation 'junit:junit:4.12'
testImplementation 'org.spockframework:spock-core:2.0-M2-groovy-3.0'
testImplementation 'net.bytebuddy:byte-buddy:1.10.8'
testImplementation 'org.objenesis:objenesis:3.1'
// in light of kriegaex's comments:
implementation group: 'cglib', name: 'cglib', version: '3.3.0'
}
test { useJUnitPlatform() }
application {
mainClassName = 'core.Launcher'
}
installDist{}
main.groovy:
class Launcher {
static void main(String[] args) {
Application.launch(App, null )
}
}
class App extends Application {
void start(Stage primaryStage) {
}
}
first_tests.groovy:
class AppSpec extends Specification {
def 'Launcher.main should call App.launch'(){
given:
GroovyMock(Application, global: true)
when:
Launcher.main()
then:
1 * Application.launch( App, null ) >> null
}
}
The reason why this project needs something to call the Application subclass is explained here: it's so that it is possible to do an installDist which bundles in JavaFX.
Don't we have to use a global GroovyMock?
If you want to check the interaction, yes. But actually you are testing the JavaFX launcher rather than your application. So I doubt that there is any benefit. I would focus on testing the App class instead. Also imagine for a moment that you would write the classes with main methods in Java instead of Groovy. Groovy mocks would not work when called from Java code, especially not global ones. Then you would end up testing via Powermockito from Spock, which would also work but still you would test the JavaFX launcher rather than your application.
Also isn't it slightly extreme to say any use of Groovy mocks is wrong?
I did not say that. I said: "probably something is wrong with your application design". The reason I said that is because the use of Groovy mocks and things like mocking static methods are test code smells. You can check the smell and then decide it is okay, which IMO in most cases it is not. Besides, instead of application design the problem can also be in the test itself, which in this case I would say it is. But that is arguable, so I am going to present a solution to you further below.
In this case technically the global Application mock is your only way if you do insist to test the JavaFX launcher because even a global mock on App would not work as the launcher uses reflection in order to call the App constructor and that is not intercepted by the mock framework.
you say that Spock spock-core:2.0-M2-groovy-3.0 is a "pre-release". I can't see anything on this page (...) which says that. How do you know?
You found out already by checking out the GitHub repository, but I was just seeing it in the unusual version number containing "M2" like "milestone 2" which is similar to "RC" (or "CR") for release candidates (or candidate releases).
As for the technical problem, you can either not declare Objenesis in your Gradle script because it is an optional dependency, then the test compiles and runs fine, as you already noticed yourself. But assuming you need optional dependencies like Objenesis, CGLIB (actually cglib-nodep), Bytebuddy and ASM for other tests in your suite, you can just tell Spock not to use Objenesis in this case. So assuming you have a Gradle build file like this:
plugins {
id 'groovy'
id 'java'
id 'application'
id 'org.openjfx.javafxplugin' version '0.0.8'
}
repositories { mavenCentral() }
javafx {
version = "11.0.2"
modules = ['javafx.controls', 'javafx.fxml']
}
dependencies {
implementation 'org.codehaus.groovy:groovy:3.+'
testImplementation 'org.spockframework:spock-core:2.0-M2-groovy-3.0'
// Optional Spock dependencies, versions matching the ones listed at
// https://mvnrepository.com/artifact/org.spockframework/spock-core/2.0-M2-groovy-3.0
testImplementation 'net.bytebuddy:byte-buddy:1.9.11'
testImplementation 'org.objenesis:objenesis:3.0.1'
testImplementation 'cglib:cglib-nodep:3.2.10'
testImplementation 'org.ow2.asm:asm:7.1'
}
test { useJUnitPlatform() }
application {
mainClassName = 'de.scrum_master.app.Launcher'
}
installDist {}
My version of your MCVE would looks like this (sorry, I added my own package names and also imports because otherwise it is not really an MCVE):
package de.scrum_master.app
import javafx.application.Application
import javafx.scene.Scene
import javafx.scene.control.Label
import javafx.scene.layout.StackPane
import javafx.stage.Stage
class App extends Application {
#Override
void start(Stage stage) {
def javaVersion = System.getProperty("java.version")
def javafxVersion = System.getProperty("javafx.version")
Label l = new Label("Hello, JavaFX $javafxVersion, running on Java $javaVersion.")
Scene scene = new Scene(new StackPane(l), 640, 480)
stage.setScene(scene)
stage.show()
}
}
package de.scrum_master.app
import javafx.application.Application
class Launcher {
static void main(String[] args) {
Application.launch(App, null)
}
}
package de.scrum_master.app
import javafx.application.Application
import spock.lang.Specification
class AppSpec extends Specification {
def 'Launcher.main should call App.launch'() {
given:
GroovyMock(Application, global: true, useObjenesis: false)
when:
Launcher.main()
then:
1 * Application.launch(App, null)
}
}
The decisive detail here is the useObjenesis: false parameter.
Update: Just for reference, this is how you would do it with a launcher class implemented in Java using PowerMockito.
Attention, this solution needs the Sputnik runner from Spock 1.x which was removed in 2.x. So in Spock 2 this currently does not work because it is based on JUnit 5 and can no longer use #RunWith(PowerMockRunner) and #PowerMockRunnerDelegate(Sputnik) because PowerMock currently does not support JUnit 5. But I tested it with Spock 1.3-groovy-2.5 and Groovy 2.5.8.
package de.scrum_master.app
import javafx.application.Application
import org.junit.runner.RunWith
import org.powermock.core.classloader.annotations.PrepareForTest
import org.powermock.modules.junit4.PowerMockRunner
import org.powermock.modules.junit4.PowerMockRunnerDelegate
import org.spockframework.runtime.Sputnik
import spock.lang.Specification
import static org.mockito.Mockito.*
import static org.powermock.api.mockito.PowerMockito.*
#RunWith(PowerMockRunner)
#PowerMockRunnerDelegate(Sputnik)
#PrepareForTest(Application)
class JavaAppSpec extends Specification {
def 'JavaLauncher.main should launch JavaApp'() {
given:
mockStatic(Application)
when:
JavaLauncher.main()
then:
verifyStatic(Application, times(1))
Application.launch(JavaApp)
}
}
Is there possible method to programmatically shut down caches? I tried inject DefaultCacheConfiguration in test class and set inside setup method set testMode=true and maximumSize=0, but still not working.
#MicronautTest
class CategoriesServiceTest extends Specification {
#Inject
#Shared
DefaultCacheConfiguration cacheConfiguration
def setupSpec() {
cacheConfiguration.setMaximumSize(0)
cacheConfiguration.setTestMode(true)
}
}
I found little hack. I'm not sure that is the best solution, but works.
I created class, which globaly shutdown caching for tests:
#Factory
class TestConfig {
#EachBean(DefaultCacheConfiguration)
CacheConfiguration cacheConfiguration(DefaultCacheConfiguration configuration) {
println "disabling cache ${configuration.cacheName}"
configuration.maximumSize = 0
return configuration
}
}
Starting in 1.3.0.M1 a noop cache implementation was added that you can enable for tests or any other environment. See https://micronaut-projects.github.io/micronaut-cache/latest/guide/#noop
I'm using tapestry 5.4.1. I have a component with a module that requires prototype. I know prototype is available in the core stack. However how do I import this as a dependency.
My module:
define(["prototype"], function(container, link) {
return new Ajax.PeriodicalUpdater(container, link, {
method : 'post', frequency : 5, decay : 1
});
});
I tried adding this to the class but the path cannot be resolved
#Import(library = {"prototype.js"})
public class Update {
Tried injecting the asset and adding it to the environmental javascriptsupport but it somehow looks for it in the wrong location.
#Inject
#Path("classpath:META-INF/assets/tapestry5/prototype.js")
private Asset prototype;
and
javascriptSupport.importJavaScriptLibrary(prototype);
javascriptSupport.require("update").with(container, getLink());
I don't want to hard code the url with the generated hash.
/assets/meta/z67bxxxx/tapestry5/scriptaculous_1_9_0/prototype.js
Anything I am missing here? Any help would be appreciated.
Make sure you define correct infrastructure in your AppModule
#ApplicationDefaults
public static void contributeApplicationDefaults(MappedConfiguration<String, Object> configuration) {
configuration.add(SymbolConstants.JAVASCRIPT_INFRASTRUCTURE_PROVIDER, "prototype");
}
You don't have to specify dependency clearly ["prototype"].
I am wondering how would you use typescript IOC specifically node app.
In case of external module-based architecture there is no any classes in the app. Just pure modules because my app heavily depends on node_modules.
How would I integrate IOC solution in such case? Any thoughts?
Here is my specific case I want to use IOC for:
I have mongoose model:
interface IStuffModel extends IStuff, mongoose.Document { }
var Stuff= mongoose.model<IStuffModel>('Stuff', Schemas.stuffSchema);
export = Stuff;
And related fake class:
export class Stuff implements IStuff {
//do stuff
}
How would I integrate IOC solution in such case
Here is a very popular library that I recommend : https://github.com/inversify/InversifyJS
External modules
Using external modules doesn't change the code at all. Instead of
kernel.bind(new TypeBinding<FooBarInterface>("FooBarInterface", FooBar));
Production
You just have
import {ProdFooBar} from "./prodFooBar";
kernel.bind(new TypeBinding<FooBarInterface>("FooBarInterface", ProdFooBar));
Test
import {MockFooBar} from "./mockFooBar";
kernel.bind(new TypeBinding<FooBarInterface>("FooBarInterface", MockFooBar));
As Basarat indicated in his answer, I have developed an IoC container called InversifyJS with advanced dependency injection features like contextual bindings.
You need to follow 3 basic steps to use it:
1. Add annotations
The annotation API is based on Angular 2.0:
import { injectable, inject } from "inversify";
#injectable()
class Katana implements IKatana {
public hit() {
return "cut!";
}
}
#injectable()
class Shuriken implements IShuriken {
public throw() {
return "hit!";
}
}
#injectable()
class Ninja implements INinja {
private _katana: IKatana;
private _shuriken: IShuriken;
public constructor(
#inject("IKatana") katana: IKatana,
#inject("IShuriken") shuriken: IShuriken
) {
this._katana = katana;
this._shuriken = shuriken;
}
public fight() { return this._katana.hit(); };
public sneak() { return this._shuriken.throw(); };
}
2. Declare bindings
The binding API is based on Ninject:
import { Kernel } from "inversify";
import { Ninja } from "./entities/ninja";
import { Katana } from "./entities/katana";
import { Shuriken} from "./entities/shuriken";
var kernel = new Kernel();
kernel.bind<INinja>("INinja").to(Ninja);
kernel.bind<IKatana>("IKatana").to(Katana);
kernel.bind<IShuriken>("IShuriken").to(Shuriken);
export default kernel;
3. Resolve dependencies
The resolution API is based on Ninject:
import kernel = from "./inversify.config";
var ninja = kernel.get<INinja>("INinja");
expect(ninja.fight()).eql("cut!"); // true
expect(ninja.sneak()).eql("hit!"); // true
The latest release (2.0.0) supports many use cases:
Kernel modules
Kernel middleware
Use classes, string literals or Symbols as dependency identifiers
Injection of constant values
Injection of class constructors
Injection of factories
Auto factory
Injection of providers (async factory)
Activation handlers (used to inject proxies)
Multi injections
Tagged bindings
Custom tag decorators
Named bindings
Contextual bindings
Friendly exceptions (e.g. Circular dependencies)
You can learn more about it at https://github.com/inversify/InversifyJS
In the particular context of Node.js there is a hapi.js example that uses InversifyJS.
I have built a simple dependency injection container that I pass around my classes that need it, everything works and all is good.
My question is that say I have 2 classes such as
class A {
public function __construct() {
}
}
class B {
public function __construct(A $a) {
}
}
Should I enforce the typehinting in the class itself or in the injection container such as;
$di->set('A', function() {
return new A();
});
$di->set('B', function(A $a) {
return new B($a);
});
Should I do both or either/or.
For answers why is it better to use one over the other etc?
Thanks.
I would use the first case, enforce the type hinting in the class itself.
This will make it clear for readers of the code what are the actual dependencies of the class.
If you decide to change the DI container (or hypothetically remove it) or reuse the classes in other project, it is good to have the type hinting in the class itself.
The DI container is there simply to help you manage dependencies.