Scala traits implementation and interoperability. Part I: Basics

Traits in Scala are similar to interfaces, but much more powerful. They allow implementations of some of the methods, fields, stacking, etc. But have you ever wondered how are they implemented on top of JVM? How is it possible to extend multiple traits and where the implementations go to? In this article, based on my StackOverflow answer, I will give several trait examples and explain how scalac implements them, what are the drawbacks and what you get for free. Often we will look at compiled classes and decompile them to Java. It’s not essential knowledge, but I hope you’ll enjoy it. All examples are compiled against Scala 2.10.1.

Simple trait with no method implementations

The following trait:

trait Solver {
    def answer(question: String): Int

compiles down to the most boring Java interface:

public interface Solver {
    int answer(String);

There is really nothing special in Scala traits. That also means that if you ship Solver.class as part of your library, users can safely implement such interface from Java code. As far as java/javac is concerned, this is an ordinary Java compiled interface.

Traits having some methods implemented

OK, but what if trait actually has some method bodies?

trait Solver {
    def answer(s: String): Int
    def ultimateAnswer = answer("Answer to the Ultimate Question of Life, the Universe, and Everything")

Here, ultimateAnswer method actually has some implementation (the fact that it calls abstract answer() method is irrelevant) while answer() remains unimplemented. What will scalac produce?

public interface Solver {
    int answer(java.lang.String);
    int ultimateAnswer();

Well, it’s still an interface and the implementation is gone. If the implementation is gone, what happens when we extend such a trait?

class DummySolver extends Solver {
    override def answer(s: String) = 42

We need to implement answer() but ultimateAnswer is already available via Solver. Anxious to see how it looks under the hood? DummySolver.class:

public class DummySolver implements Solver {
    public DummySolver() {
    public int ultimateAnswer() {
        return Solver$class.ultimateAnswer(this);
    public int answer(String s) {
        return 42;

That… is… weird… Apparently we missed one new class file, Solver$class.class. Yes, the class is named Solver$class, this is valid even in Java. This is not Solver.class expression which returns Class[Solver], it’s apparently an ordinary Java class named Solver$class with a bunch of static methods. Here is how it looks like:

public abstract class Solver$class {
    public static int ultimateAnswer(Solver $this) {
        return $this.answer("Answer to the Ultimate Question of Life, the Universe, and Everything");
    public static void $init$(Solver solver) {}

Do you see the trick? Scala created a helper Solver$class and methods that are implemented inside trait are actually placed in that hidden class. BTW this is not a companion object, it’s just a helper class invisible to you. But the real trick is $this parameter, invoked as Solver$class.ultimateAnswer(this). Since we are technically in another object, we must somehow get a handle of a real Solver instance. It is like OOP but done manually. So we learned that method implementations from traits are extracted to a special helper class. This class is referenced every time we call such a method. This way we don’t copy method body over and over into every single class extending given trait.

Extending multiple traits with implementations

Imagine extending multiple traits, each implementing distinct set of methods:

trait Foo {
    def foo = "Foo"
trait Bar {
    def bar = "Bar"
class Buzz extends Foo with Bar

By analogy you probably know already how Foo.class and Bar.class look like:

public interface Foo {
    String foo();
public interface Bar {
    String bar();

Implementations are hidden in mysterious ...$class.class files just as before:

public abstract class Foo$class {
    public static String foo(Foo) {
        return "Foo";
    public static void $init$(Foo) {}
public abstract class Bar$class {
    public static String bar(Bar) {
        return "Bar";
    public static void $init$(Bar) {}

Notice that static methods in Foo$class accept instances of Foo while Bar$class require reference to Bar. This works because Buzz implements both Foo and Bar:

public class Buzz implements Foo, Bar {
    public Buzz() {
    public String bar() {
        return Bar$;
    public String foo() {
        return Foo$;

Both foo() and bar() pass this reference, but this is fine.

Traits with fields

Fields are another interesting feature of traits. This time let’s use a real-world example from Spring Data project. As you know interfaces can require certain methods to be implemented. However they can’t force you to provide certain fields (or provide fields on their own). This limitation becomes painful when working with Auditable<U, ID> interface extending Persistable<ID>. While these interfaces merely exist to make sure certain fields are present on your @Entity class, namely createdBy, createdDate, lastModifiedBy, lastModifiedDate and id, this cannot be expressed cleanly. Instead you have to implement the following methods in every single entity extending Auditable:

  U getCreatedBy();
    void setCreatedBy(final U createdBy);
DateTime getCreatedDate();
    void setCreatedDate(final DateTime creationDate);
       U getLastModifiedBy();
    void setLastModifiedBy(final U lastModifiedBy);
DateTime getLastModifiedDate();
    void setLastModifiedDate(final DateTime lastModifiedDate);
      ID getId();
 boolean isNew();

Moreover, every single class must of course define fields highlighted above. Doesn’t feel DRY at all. Luckily traits can help us a lot. In trait we can define what fields should be created in every class extending this trait:

trait IAmAuditable[ID <:] extends Auditable[User, ID] {
    var createdBy: User = _
    def getCreatedBy = createdBy
    def setCreatedBy(createdBy: User) {
        this.createdBy = createdBy
    var createdDate: DateTime = _
    def getCreatedDate = createdDate
    def setCreatedDate(creationDate: DateTime) {
        this.createdDate = creationDate
    var lastModifiedBy: User = _
    def getLastModifiedBy = lastModifiedBy
    def setLastModifiedBy(lastModifiedBy: User) {
        this.lastModifiedBy = lastModifiedBy
    var lastModifiedDate: DateTime = _
    def getLastModifiedDate = lastModifiedDate
    def setLastModifiedDate(lastModifiedDate: DateTime) {
        this.lastModifiedDate = lastModifiedDate
    var id: ID = _
    def getId = id
    def isNew = id == null

But wait, Scala has built-in support for POJO-style getters/setters! So we can shorten this to:

class IAmAuditable[ID <:] extends Auditable[User, ID] {
    @BeanProperty var createdBy: User = _
    @BeanProperty var createdDate: DateTime = _
    @BeanProperty var lastModifiedBy: User = _
    @BeanProperty var lastModifiedDate: DateTime = _
    @BeanProperty var id: ID = _
    def isNew = id == null

Compiler-generated getters and setters implement interface automatically. From now on, every class willing to provide auditing capabilities can extend this trait:

class Person extends IAmAuditable[String] {

All the fields, getters and setters are there. But how is it implemented? Let’s look at the generated Person.class:

public class Person implements IAmAuditable<java.lang.String> {
    private User createdBy;
    private DateTime createdDate;
    private User lastModifiedBy;
    private DateTime lastModifiedDate;
    private id;
    public User createdBy() //...
    public void createdBy_$eq(User) //...
    public User getCreatedBy() //...
    public void setCreatedBy(User) //...
    public DateTime createdDate() //...
    public void createdDate_$eq(DateTime) //...
    public DateTime getCreatedDate() //...
    public void setCreatedDate(DateTime) //...
    public boolean isNew();

There is actually much more, but you get the idea. So fields aren’t refactored into a separate class, but that was to be expected. Instead fields are copied into every single class extending this particular trait. In Java we could have used abstract base class for that, but it’s nice to reserve inheritance for real is-a relationships and do not use it for dummy field holders. In Scala trait is such a holder, grouping common fields that we can reuse across several classes.

What about Java interoperability? Well, IAmAuditable is compiled down to Java interface, thus it doesn’t have fields at all. If you implement it from Java, you gain nothing special. We covered basic use cases for traits in Scala and their implementation. In the next article we will explore how mixins and stackable modifications work.

Reference: Scala traits implementation and interoperability. Part I: Basics from our JCG partner Tomasz Nurkiewicz at the Java and neighbourhood blog.

Tomasz Nurkiewicz

Java EE developer, Scala enthusiast. Enjoying data analysis and visualization. Strongly believes in the power of testing and automation.
Notify of

This site uses Akismet to reduce spam. Learn how your comment data is processed.

1 Comment
Newest Most Voted
Inline Feedbacks
View all comments
Aleskey Lisikh
Aleskey Lisikh
10 years ago

Nice article, thx! :)

Back to top button