The road to Kotlin Symbol Processing

There’s a long back story to Java annotations. Introduced in 2004 for Java 5 and supported by the javac compiler since Java 6 in 2002, they may be thought of as an industry standard approach to
metaprogramming. However, as mostly used in the Android ecosystem, it is not that popular technique in backend development. I had an
opportunity to dig into the subject while developing Krush, which is based on Kotlin annotation processing (KAPT). In this article I’ll try to show you the different approaches to annotation processing, starting from pure Java solutions, then moving to KAPT and finally to its
successor – Kotlin Symbol Processing.

Java annotation processors

Lombok *

Lombok is one of the first projects that comes into Java devs mind when thinking about annotation processing. You just add a single dependency into your pom.xml / add a plugin into Intellij Idea and some kind of magic turns your classes annotated with @Value into a functional immutable data structure. But in fact, Lombok is not a 100% pure example of an annotation processor… If you just run your debugger and step into generated toString / equals method, you’ll see no generated code:

lombok debug

Lombok starts as an usual annotation processor, but during its run it modifies a compiler abstract syntax tree to insert desired methods to your classes, which is not an intended use-case for annotation processors. In some references this technique is even called a hack:

Lombok … uses annotation processing as a bootstrapping mechanism to include itself into the compilation process and modify the AST via some internal compiler APIs. This hacky technique has nothing to do with the intended purpose of annotation processing

[source]

Apart from not being the clearest solution, using Lombok for some time broke other annotation-processing based libraries configured to run on the same source code.

AutoValue

A similar library which uses annotation processing in a clear way is AutoValue. It processes the @AutoValue and @AutoValue.Builder annotations to generate immutable classes which allow you to safely step into using a debugger. Consider following example:

@AutoValue
abstract class Book {

   static Builder builder() {
       return new AutoValue_Book.Builder();
   }

   @AutoValue.Builder
   interface Builder {
       Builder title(String title);
       Builder author(String author);
       Book build();
   }

   abstract String title();
   abstract String author();
}

Then, if you create an instance of the class using a Builder, you can see generated toString / equals / hashCode methods in the debugger:

autovalue debug

By quickly looking at the above code you can see some characteristics of using an annotation processor in your project:

  • there is a minimal framework/convention that must be built into your classes (abstract getters, static builder method)
  • you are using “third party” code (classes in this example) even by
    directly referring it in your source code or in the runtime (like in @AutoValue example)
  • there is a need to run partial compilation before being able to use
    generated code

Other examples of Java annotation processor include:

You can find other examples on this annotation processing list.

KAPT

I wasn’t aware how much annotation processing is used by libraries from the Android ecosystem. Butterknife, Room, Moshi, Hilt… these names are not quite familiar if you’re a backend developer working on the JVM. As Kotlin started gaining popularity in the Android community, it was crucial for it to support existing ecosystem libraries made for Java. That’s why KAPT was introduced – a Kotlin Annotation Processing Tool. The idea behind it was very simple:

  • make a minimal step to transform Kotlin code to Java
  • run annotation processing on Java sources, just as in any Java project

Krush

One of the examples of using KAPT is Krush – our lightweight persistence layer for Kotlin based on Exposed SQL DSL. Krush interprets standard JPA annotations on the entity classes to generate both Exposed DSL mappings and convenient methods transfer data from / to entity classes.

Consider following entity:

@Entity
data class Reservation(

   @Id
   val uid: UUID = UUID.randomUUID(),
   @Enumerated(EnumType.STRING)

   val status: Status = Status.FREE
)

By adding Krush we will have following mapping generated:

// generated
public object ReservationTable : Table("reservation") {
 public val uid: Column = uuid("uid")
 public override val primaryKey: Table.PrimaryKey = PrimaryKey(uid)
 public val status: Column = 
    enumerationByName("status", 255, pl.touk.krush.Status::class)
}

However, during Krush implementation we observed some drawbacks of using annotation processing in 100% Kotlin codebase:

  • no direct support for code generation. Java annotation API contains limited support for code generation using a Filer interface, that just lets you generate new files, with no distinction between source code, metadata, docs etc. A partial solution to this is using a third-party code generation library, like KotlinPoet, which contains a feature-rich DSL for generating Kotlin classes, properties etc.
  • missing Kotlin-specific information in the API – annotation processing API contains information about the structure of the code during compilation using javax.lang.model package. However, this is Java-specific, so you don’t have access to top-level functions, file annotations etc. However, some of additional Kotlin metadata can be retrieved by using kotlinpoet-metadata integration.
  • a need to generate Java stubs before running the annotation processing itself reduces the overall performance of project build process, some resources estimate that stub generation takes ⅓ time of the whole kotlinc run, which can be painful for large codebases.

KSP

The downsides of using KAPT in pure Kotlin projects (especially Android-based), triggered Google to develop a Kotlin-specific approach, called Kotlin Symbol Processing. It is now a preferred way to implement annotation processing in Kotlin, since its release KAPT has been put into maintenance mode. It provides a separate from javax.lang.model, Kotlin-specific API representing your source code, no need to generate any Java stubs and a better integration with code generator libraries. Let’s look at basic features of KSP by doing a hands-on example.

Implementing @Slf4j

Let’s look at how to use KSP by writing a simple processor – an equivalent of Lombok’s @Slf4j.
We should start with an implementation of our annotation, SymbolProcessor and a provider for it:

annotation class Slf4j

class Slf4jProcessor(val env: SymbolProcessorEnvironment) : SymbolProcessor {

   override fun process(resolver: Resolver): List {
       val symbols = resolver.getSymbolsWithAnnotation(Slf4j::class.java.name)
       val ret = symbols.filter { !it.validate() }.toList()
       symbols
           .filter { it is KSClassDeclaration && it.validate() }
           .forEach { it.accept(Slf4jProcessorVisitor(), Unit) }
       return ret
   }

   inner class Slf4jProcessorVisitor : KSVisitorVoid()
  
}

class Slf4jProcessorProvider : SymbolProcessorProvider {
   override fun create(environment: SymbolProcessorEnvironment) = Slf4jProcessor(environment)
}

Apart from a bunch of bootstrapping code, you may notice that the main part of the implementation would be Slf4jProcessorVisitor – which is how KSP is traversing your source code – by using the visitor pattern. So what you have to do is write appropriate visitXXX method(s) to implement your processor functionality:

override fun visitClassDeclaration(classDeclaration: KSClassDeclaration, data: Unit) {
   val packageName = classDeclaration.packageName.asString()
   val ksType = classDeclaration.asType(emptyList())

    val fileSpec = FileSpec.builder(
        packageName = packageName,
        fileName = classDeclaration.simpleName.asString() + "Ext"
    ).apply {
        val className = ksType.toClassName()
        val loggerName = "_${className.simpleName.replaceFirstChar { it.lowercase() }}Logger"
        addProperty(
            PropertySpec.builder(loggerName, Logger::class.java)
                .addModifiers(KModifier.PRIVATE)
                .initializer("%T.getLogger(%T::class.java)", LoggerFactory::class.java, className)
                .build()
        )
        addProperty(
            PropertySpec.builder("logger", Logger::class.java)
                .receiver(className)
                .getter(
                    FunSpec.getterBuilder()
                        .addStatement("return $loggerName")
                        .build()
                )
                .build()
        )
    }.build()

    fileSpec.writeTo(codeGenerator = env.codeGenerator, aggregating = false)
}

So, I’m using KotlinPoet here, which is quite nice integrated with KSP by this writeTo method – so if you want to generate a new file you just build a FileSpec and then write it to appropriate folder configured by the KSP plugin by calling writeTo method.

In short, for each class annotated with @Slf4j annotation we generate a file with a _serviceLogger property which is initialized with standard LoggerFactory.getLogger call. KotlinPoet comes with a nice templating system, which allows you to just pass class declarations from the KSP model instead of resolving them manually, with imports etc.. The second property is an extension of our annotated class, which we express by using a receiver block, we can also make a custom getter by using another KotlinPoet call.

If we did everything right this how the generated code should look for annotated class:

import org.slf4j.Logger
import org.slf4j.LoggerFactory

private val _serviceLogger: Logger = LoggerFactory.getLogger(Service::class.java)

public val Service.logger: Logger
  get() = _serviceLogger

Which should allow us to use the logger in our class:

@Slf4j
class Service {
   fun test() {
       logger.info("Hello from KSP!")
   }
}

Summary

I hope this article helped you learn the history of annotation processing and motivation behind the KSP project. In fact KAPT is now in maintenance mode, so KSP should be the default library to use in new, pure Kotlin projects. However, if you think of migrating your existing library based on KAPT, you should be aware of some complications when java.lang.model is too tightly coupled to your model. For example the Dagger project has a rough road of supporting KSP, they must first introduce some common model based on XProcessing to support both KSP and traditional annotation processing.

The code for example @Slf4j processor can be found here.

You May Also Like

Hibernate hbm2ddl won’t create schema before creating tables

Situation I have a local H2 in memory database for integration tests and an Oracle db for production. I do not control the Oracle DB model. The in memory H2 database is created automatically by adding <prop key="hibernate.hbm2ddl.auto">update&l...Situation I have a local H2 in memory database for integration tests and an Oracle db for production. I do not control the Oracle DB model. The in memory H2 database is created automatically by adding <prop key="hibernate.hbm2ddl.auto">update&l...

Sample for lift-ng: Micro-burn 1.0.0 released

During a last few evenings in my free time I've worked on mini-application called micro-burn. The idea of it appear from work with Agile Jira in our commercial project. This is a great tool for agile projects management. It has inline tasks edition, drag & drop board, reports and many more, but it also have a few drawbacks that turn down our team motivation.

Motivation

From time to time our sprints scope is changing. It is not a big deal because we are trying to be agile :-) but Jira's burndowchart in this situation draw a peek. Because in fact that chart shows scope changes not a real burndown. It means, that chart cannot break down an x-axis if we really do more than we were planned – it always stop on at most zero.

Also for better progress monitoring we've started to split our user stories to technical tasks and estimating them. Original burndowchart doesn't show points from technical tasks. I can find motivation of this – user story almost finished isn't finished at all until user can use it. But in the other hand, if we know which tasks is problematic we can do some teamwork to move it on.

So I realize that it is a good opportunity to try some new approaches and tools.

Tools

I've started with lift framework. In the World of Single Page Applications, this framework has more than simple interface for serving REST services. It comes with awesome Comet support. Comet is a replacement for WebSockets that run on all browsers. It supports long polling and transparent fallback to short polling if limit of client connections exceed. In backend you can handle pushes in CometActor. For further reading take a look at Roundtrip promises

But lift framework is also a kind of framework of frameworks. You can handle own abstraction of CometActors and push to client javascript that shorten up your way from server to client. So it was the trigger for author of lift-ng to make a lift with Angular integration that is build on top of lift. It provides AngularActors from which you can emit/broadcast events to scope of controller. NgModelBinders that synchronize your backend model with client scope in a few lines! I've used them to send project state (all sprints and thier details) to client and notify him about scrum board changes. My actor doing all of this hard work looks pretty small:

Lift-ng also provides factories for creating of Angular services. Services could respond with futures that are transformed to Angular promises in-fly. This is all what was need to serve sprint history:

And on the client side - use of service:


In my opinion this two frameworks gives a huge boost in developing of web applications. You have the power of strongly typing with Scala, you can design your domain on Actors and all of this with simplicity of node.js – lack of json trasforming boilerplate and dynamic application reload.

DDD + Event Sourcing

I've also tried a few fresh approaches to DDD. I've organize domain objects in actors. There are SprintActors with encapsulate sprint aggregate root. Task changes are stored as events which are computed as a difference between two boards states. When it should be provided a history of sprint, next board states are computed from initial state and sequence of events. So I realize that the best way to keep this kind of event sourcing approach tested is to make random tests. This is a test doing random changes at board, calculating events and checking if initial state + events is equals to previously created state:



First look

Screenshot of first version:


If you want to look at this closer, check the source code or download ready to run fatjar on github.During a last few evenings in my free time I've worked on mini-application called micro-burn. The idea of it appear from work with Agile Jira in our commercial project. This is a great tool for agile projects management. It has inline tasks edition, drag & drop board, reports and many more, but it also have a few drawbacks that turn down our team motivation.