Traits are a structural construct of the language which allows:

  • composition of behaviors

  • runtime implementation of interfaces

  • behavior overriding

  • compatibility with static type checking/compilation

They can be seen as interfaces carrying both default implementations and state. A trait is defined using the trait keyword:

trait FlyingAbility {                           (1)
        String fly() { "I'm flying!" }          (2)
}
1 declaration of a trait
2 declaration of a method inside a trait

Then it can be used like a normal interface using the implements keyword:

class Bird implements FlyingAbility {}          (1)
def b = new Bird()                              (2)
assert b.fly() == "I'm flying!"                 (3)
1 Adds the trait FlyingAbility to the Bird class capabilities
2 instantiate a new Bird
3 the Bird class automatically gets the behavior of the FlyingAbility trait

Traits allow a wide range of capabilities, from simple composition to testing, which are described thoroughly in this section.

1. Methods

1.1. Public methods

Declaring a method in a trait can be done like any regular method in a class:

trait FlyingAbility {                           (1)
        String fly() { "I'm flying!" }          (2)
}
1 declaration of a trait
2 declaration of a method inside a trait

1.2. Abstract methods

In addition, traits may declare abstract methods too, which therefore need to be implemented in the class implementing the trait:

trait Greetable {
    abstract String name()                              (1)
    String greeting() { "Hello, ${name()}!" }           (2)
}
1 implementing class will have to declare the name method
2 can be mixed with a concrete method

Then the trait can be used like this:

class Person implements Greetable {                     (1)
    String name() { 'Bob' }                             (2)
}

def p = new Person()
assert p.greeting() == 'Hello, Bob!'                    (3)
1 implement the trait Greetable
2 since name was abstract, it is required to implement it
3 then greeting can be called

1.3. Private methods

Traits may also define private methods. Those methods will not appear in the trait contract interface:

trait Greeter {
    private String greetingMessage() {                      (1)
        'Hello from a private method!'
    }
    String greet() {
        def m = greetingMessage()                           (2)
        println m
        m
    }
}
class GreetingMachine implements Greeter {}                 (3)
def g = new GreetingMachine()
assert g.greet() == "Hello from a private method!"          (4)
try {
    assert g.greetingMessage()                              (5)
} catch (MissingMethodException e) {
    println "greetingMessage is private in trait"
}
1 define a private method greetingMessage in the trait
2 the public greet message calls greetingMessage by default
3 create a class implementing the trait
4 greet can be called
5 but not greetingMessage
Traits only support public and private methods. Neither protected nor package private scopes are supported.

2. The meaning of this

this represents the implementing instance. Think of a trait as a superclass. This means that when you write:

trait Introspector {
    def whoAmI() { this }
}
class Foo implements Introspector {}
def foo = new Foo()

then calling:

foo.whoAmI()

will return the same instance:

assert foo.whoAmI().is(foo)

3. Interfaces

Traits may implement interfaces, in which case the interfaces are declared using the implements keyword:

interface Named {                                       (1)
    String name()
}
trait Greetable implements Named {                      (2)
    String greeting() { "Hello, ${name()}!" }
}
class Person implements Greetable {                     (3)
    String name() { 'Bob' }                             (4)
}

def p = new Person()
assert p.greeting() == 'Hello, Bob!'                    (5)
assert p instanceof Named                               (6)
assert p instanceof Greetable                           (7)
1 declaration of a normal interface
2 add Named to the list of implemented interfaces
3 declare a class that implements the Greetable trait
4 implement the missing name method
5 the greeting implementation comes from the trait
6 make sure Person implements the Named interface
7 make sure Person implements the Greetable trait

4. Properties

A trait may define properties, like in the following example:

trait Named {
    String name                             (1)
}
class Person implements Named {}            (2)
def p = new Person(name: 'Bob')             (3)
assert p.name == 'Bob'                      (4)
assert p.getName() == 'Bob'                 (5)
1 declare a property name inside a trait
2 declare a class which implements the trait
3 the property is automatically made visible
4 it can be accessed using the regular property accessor
5 or using the regular getter syntax

5. Fields

5.1. Private fields

Since traits allow the use of private methods, it can also be interesting to use private fields to store state. Traits will let you do that:

trait Counter {
    private int count = 0                   (1)
    int count() { count += 1; count }       (2)
}
class Foo implements Counter {}             (3)
def f = new Foo()
assert f.count() == 1                       (4)
assert f.count() == 2
1 declare a private field count inside a trait
2 declare a public method count that increments the counter and returns it
3 declare a class that implements the Counter trait
4 the count method can use the private field to keep state
This is a major difference with Java 8 virtual extension methods. While virtual extension methods do not carry state, traits can. Moreover, traits in Groovy are supported starting with Java 6, because their implementation does not rely on virtual extension methods. This means that even if a trait can be seen from a Java class as a regular interface, that interface will not have default methods, only abstract ones.

5.2. Public fields

Public fields work the same way as private fields, but in order to avoid the diamond problem, field names are remapped in the implementing class:

trait Named {
    public String name                      (1)
}
class Person implements Named {}            (2)
def p = new Person()                        (3)
p.Named__name = 'Bob'                       (4)
1 declare a public field inside the trait
2 declare a class implementing the trait
3 create an instance of that class
4 the public field is available, but renamed

The name of the field depends on the fully qualified name of the trait. All dots (.) in package are replaced with an underscore (_), and the final name includes a double underscore. So if the type of the field is String, the name of the package is my.package, the name of the trait is Foo and the name of the field is bar, in the implementing class, the public field will appear as:

String my_package_Foo__bar
While traits support public fields, it is not recommended to use them and considered as a bad practice.

6. Composition of behaviors

Traits can be used to implement multiple inheritance in a controlled way. For example, we can have the following traits:

trait FlyingAbility {                           (1)
        String fly() { "I'm flying!" }          (2)
}
trait SpeakingAbility {
    String speak() { "I'm speaking!" }
}

And a class implementing both traits:

class Duck implements FlyingAbility, SpeakingAbility {} (1)

def d = new Duck()                                      (2)
assert d.fly() == "I'm flying!"                         (3)
assert d.speak() == "I'm speaking!"                     (4)
1 the Duck class implements both FlyingAbility and SpeakingAbility
2 creates a new instance of Duck
3 we can call the method fly from FlyingAbility
4 but also the method speak from SpeakingAbility

Traits encourage the reuse of capabilities among objects, and the creation of new classes by the composition of existing behavior.

7. Overriding default methods

Traits provide default implementations for methods, but it is possible to override them in the implementing class. For example, we can slightly change the example above, by having a duck which quacks:

class Duck implements FlyingAbility, SpeakingAbility {
    String quack() { "Quack!" }                         (1)
    String speak() { quack() }                          (2)
}

def d = new Duck()
assert d.fly() == "I'm flying!"                         (3)
assert d.quack() == "Quack!"                            (4)
assert d.speak() == "Quack!"                            (5)
1 define a method specific to Duck, named quack
2 override the default implementation of speak so that we use quack instead
3 the duck is still flying, from the default implementation
4 quack comes from the Duck class
5 speak no longer uses the default implementation from SpeakingAbility

8. Extending traits

8.1. Simple inheritance

Traits may extend another trait, in which case you must use the extends keyword:

trait Named {
    String name                                     (1)
}
trait Polite extends Named {                        (2)
    String introduce() { "Hello, I am $name" }      (3)
}
class Person implements Polite {}
def p = new Person(name: 'Alice')                   (4)
assert p.introduce() == 'Hello, I am Alice'         (5)
1 the Named trait defines a single name property
2 the Polite trait extends the Named trait
3 Polite adds a new method which has access to the name property of the super-trait
4 the name property is visible from the Person class implementing Polite
5 as is the introduce method

8.2. Multiple inheritance

Alternatively, a trait may extend multiple traits. In that case, all super traits must be declared in the implements clause:

trait WithId {                                      (1)
    Long id
}
trait WithName {                                    (2)
    String name
}
trait Identified implements WithId, WithName {}     (3)
1 WithId trait defines the id property
2 WithName trait defines the name property
3 Identified is a trait which inherits both WithId and WithName

9. Duck typing and traits

9.1. Dynamic code

Traits can call any dynamic code, like a normal Groovy class. This means that you can, in the body of a method, call methods which are supposed to exist in an implementing class, without having to explicitly declare them in an interface. This means that traits are fully compatible with duck typing:

trait SpeakingDuck {
    String speak() { quack() }                      (1)
}
class Duck implements SpeakingDuck {
    String methodMissing(String name, args) {
        "${name.capitalize()}!"                     (2)
    }
}
def d = new Duck()
assert d.speak() == 'Quack!'                        (3)
1 the SpeakingDuck expects the quack method to be defined
2 the Duck class does implement the method using methodMissing
3 calling the speak method triggers a call to quack which is handled by methodMissing

9.2. Dynamic methods in a trait

It is also possible for a trait to implement MOP methods like methodMissing or propertyMissing, in which case implementing classes will inherit the behavior from the trait, like in this example:

trait DynamicObject {                               (1)
    private Map props = [:]
    def methodMissing(String name, args) {
        name.toUpperCase()
    }
    def propertyMissing(String prop) {
        props[prop]
    }
    void setProperty(String prop, Object value) {
        props[prop] = value
    }
}

class Dynamic implements DynamicObject {
    String existingProperty = 'ok'                  (2)
    String existingMethod() { 'ok' }                (3)
}
def d = new Dynamic()
assert d.existingProperty == 'ok'                   (4)
assert d.foo == null                                (5)
d.foo = 'bar'                                       (6)
assert d.foo == 'bar'                               (7)
assert d.existingMethod() == 'ok'                   (8)
assert d.someMethod() == 'SOMEMETHOD'               (9)
1 create a trait implementing several MOP methods
2 the Dynamic class defines a property
3 the Dynamic class defines a method
4 calling an existing property will call the method from Dynamic
5 calling an non-existing property will call the method from the trait
6 will call setProperty defined on the trait
7 will call getProperty defined on the trait
8 calling an existing method on Dynamic
9 but calling a non existing method thanks to the trait methodMissing

10. Multiple inheritance conflicts

10.1. Default conflict resolution

It is possible for a class to implement multiple traits. If some trait defines a method with the same signature as a method in another trait, we have a conflict:

trait A {
    String exec() { 'A' }               (1)
}
trait B {
    String exec() { 'B' }               (2)
}
class C implements A,B {}               (3)
1 trait A defines a method named exec returning a String
2 trait B defines the very same method
3 class C implements both traits

In this case, the default behavior is that methods from the last declared trait wins. Here, B is declared after A so the method from B will be picked up:

def c = new C()
assert c.exec() == 'B'

10.2. User conflict resolution

In case this behavior is not the one you want, you can explicitly choose which method to call using the Trait.super.foo syntax. In the example above, we can force to choose the method from trait A, by writing this:

class C implements A,B {
    String exec() { A.super.exec() }    (1)
}
def c = new C()
assert c.exec() == 'A'                  (2)
1 explicit call of exec from the trait A
2 calls the version from A instead of using the default resolution, which would be the one from B

11. Runtime implementation of traits

11.1. Implementing a trait at runtime

Groovy also supports implementing traits dynamically at runtime. It allows you to "decorate" an existing object using a trait. As an example, let’s start with this trait and the following class:

trait Extra {
    String extra() { "I'm an extra method" }            (1)
}
class Something {                                       (2)
    String doSomething() { 'Something' }                (3)
}
1 the Extra trait defines an extra method
2 the Something class does not implement the Extra trait
3 Something only defines a method doSomething

Then if we do:

def s = new Something()
s.extra()

the call to extra would fail because Something is not implementing Extra. It is possible to do it at runtime with the following syntax:

def s = new Something() as Extra                        (1)
s.extra()                                               (2)
s.doSomething()                                         (3)
1 use of the as keyword to coerce an object to a trait at runtime
2 then extra can be called on the object
3 and doSomething is still callable
When coercing an object to a trait, the result of the operation is not the same instance. It is guaranteed that the coerced object will implement both the trait and the interfaces that the original object implements, but the result will not be an instance of the original class.

11.2. Implementing multiple traits at once

Should you need to implement several traits at once, you can use the withTraits method instead of the as keyword:

trait A { void methodFromA() {} }
trait B { void methodFromB() {} }

class C {}

def c = new C()
c.methodFromA()                     (1)
c.methodFromB()                     (2)
def d = c.withTraits A, B           (3)
d.methodFromA()                     (4)
d.methodFromB()                     (5)
1 call to methodFromA will fail because C doesn’t implement A
2 call to methodFromB will fail because C doesn’t implement B
3 withTrait will wrap c into something which implements A and B
4 methodFromA will now pass because d implements A
5 methodFromB will now pass because d also implements B
When coercing an object to multiple traits, the result of the operation is not the same instance. It is guaranteed that the coerced object will implement both the traits and the interfaces that the original object implements, but the result will not be an instance of the original class.

12. Chaining behavior

Groovy supports the concept of stackable traits. The idea is to delegate from one trait to the other if the current trait is not capable of handling a message. To illustrate this, let’s imagine a message handler interface like this:

interface MessageHandler {
    void on(String message, Map payload)
}

Then you can compose a message handler by applying small behaviors. For example, let’s define a default handler in the form of a trait:

trait DefaultHandler implements MessageHandler {
    void on(String message, Map payload) {
        println "Received $message with payload $payload"
    }
}

Then any class can inherit the behavior of the default handler by implementing the trait:

class SimpleHandler implements DefaultHandler {}

Now what if you want to log all messages, in addition to the default handler? One option is to write this:

class SimpleHandlerWithLogging implements DefaultHandler {
    void on(String message, Map payload) {                                  (1)
        println "Seeing $message with payload $payload"                     (2)
        DefaultHandler.super.on(message, payload)                           (3)
    }
}
1 explicitly implement the on method
2 perform logging
3 continue by delegating to the DefaultHandler trait

This works but this approach has drawbacks:

  1. the logging logic is bound to a "concrete" handler

  2. we have an explicit reference to DefaultHandler in the on method, meaning that if we happen to change the trait that our class implements, code will be broken

As an alternative, we can write another trait which responsibility is limited to logging:

trait LoggingHandler implements MessageHandler {                            (1)
    void on(String message, Map payload) {
        println "Seeing $message with payload $payload"                     (2)
        super.on(message, payload)                                          (3)
    }
}
1 the logging handler is itself a handler
2 prints the message it receives
3 then super makes it delegate the call to the next trait in the chain

Then our class can be rewritten as this:

class HandlerWithLogger implements DefaultHandler, LoggingHandler {}
def loggingHandler = new HandlerWithLogger()
loggingHandler.on('test logging', [:])

which will print:

Seeing test logging with payload [:]
Received test logging with payload [:]

As the priority rules imply that LoggerHandler wins because it is declared last, then a call to on will use the implementation from LoggingHandler. But the latter has a call to super, which means the next trait in the chain. Here, the next trait is DefaultHandler so both will be called:

The interest of this approach becomes more evident if we add a third handler, which is responsible for handling messages that start with say:

trait SayHandler implements MessageHandler {
    void on(String message, Map payload) {
        if (message.startsWith("say")) {                                    (1)
            println "I say ${message - 'say'}!"
        } else {
            super.on(message, payload)                                      (2)
        }
    }
}
1 a handler specific precondition
2 if the precondition is not met, pass the message to the next handler in the chain

Then our final handler looks like this:

class Handler implements DefaultHandler, SayHandler, LoggingHandler {}
def h = new Handler()
h.on('foo', [:])
h.on('sayHello', [:])

Which means:

  • messages will first go through the logging handler

  • the logging handler calls super which will delegate to the next handler, which is the SayHandler

  • if the message starts with say, then the handler consumes the message

  • if not, the say handler delegates to the next handler in the chain

This approach is very powerful because it allows you to write handlers that do not know each other and yet let you combine them in the order you want. For example, if we execute the code, it will print:

Seeing foo with payload [:]
Received foo with payload [:]
Seeing sayHello with payload [:]
I say Hello!

but if we move the logging handler to be the second one in the chain, the output is different:

class AlternateHandler implements DefaultHandler, LoggingHandler, SayHandler {}
h = new AlternateHandler()
h.on('foo', [:])
h.on('sayHello', [:])

prints:

Seeing foo with payload [:]
Received foo with payload [:]
I say Hello!

The reason is that now, since the SayHandler consumes the message without calling super, the logging handler is not called anymore.

12.1. Semantics of super inside a trait

If a class implements multiple traits and a call to an unqualified super is found, then:

  1. if the class implements another trait, the call delegates to the next trait in the chain

  2. if there isn’t any trait left in the chain, super refers to the super class of the implementing class (this)

For example, it is possible to decorate final classes thanks to this behavior:

trait Filtering {                                       (1)
    StringBuilder append(String str) {                  (2)
        def subst = str.replace('o','')                 (3)
        super.append(subst)                             (4)
    }
    String toString() { super.toString() }              (5)
}
def sb = new StringBuilder().withTraits Filtering       (6)
sb.append('Groovy')
assert sb.toString() == 'Grvy'                          (7)
1 define a trait named Filtering, supposed to be applied on a StringBuilder at runtime
2 redefine the append method
3 remove all 'o’s from the string
4 then delegate to super
5 in case toString is called, delegate to super.toString
6 runtime implementation of the Filtering trait on a StringBuilder instance
7 the string which has been appended no longer contains the letter o

In this example, when super.append is encountered, there is no other trait implemented by the target object, so the method which is called is the original append method, that is to say the one from StringBuilder. The same trick is used for toString, so that the string representation of the proxy object which is generated delegates to the toString of the StringBuilder instance.

13. Advanced features

13.1. SAM type coercion

If a trait defines a single abstract method, it is candidate for SAM (Single Abstract Method) type coercion. For example, imagine the following trait:

trait Greeter {
    String greet() { "Hello $name" }        (1)
    abstract String getName()               (2)
}
1 the greet method is not abstract and calls the abstract method getName
2 getName is an abstract method

Since getName is the single abstract method in the Greeter trait, you can write:

Greeter greeter = { 'Alice' }               (1)
1 the closure "becomes" the implementation of the getName single abstract method

or even:

void greet(Greeter g) { println g.greet() } (1)
greet { 'Alice' }                           (2)
1 the greet method accepts the SAM type Greeter as parameter
2 we can call it directly with a closure

13.2. Differences with Java 8 default methods

In Java 8, interfaces can have default implementations of methods. If a class implements an interface and does not provide an implementation for a default method, then the implementation from the interface is chosen. Traits behave the same but with a major difference: the implementation from the trait is always used if the class declares the trait in its interface list and that it doesn’t provide an implementation.

This feature can be used to compose behaviors in an very precise way, in case you want to override the behavior of an already implemented method.

To illustrate the concept, let’s start with this simple example:

import groovy.transform.CompileStatic
import org.codehaus.groovy.control.CompilerConfiguration
import org.codehaus.groovy.control.customizers.ASTTransformationCustomizer
import org.codehaus.groovy.control.customizers.ImportCustomizer

class SomeTest extends GroovyTestCase {
    def config
    def shell

    void setup() {
        config = new CompilerConfiguration()
        shell = new GroovyShell(config)
    }
    void testSomething() {
        assert shell.evaluate('1+1') == 2
    }
    void otherTest() { /* ... */ }
}

In this example, we create a simple test case which uses two properties (config and shell) and uses those in multiple test methods. Now imagine that you want to test the same, but with another distinct compiler configuration. One option is to create a subclass of SomeTest:

class AnotherTest extends SomeTest {
    void setup() {
        config = new CompilerConfiguration()
        config.addCompilationCustomizers( ... )
        shell = new GroovyShell(config)
    }
}

It works, but what if you have actually multiple test classes, and that you want to test the new configuration for all those test classes? Then you would have to create a distinct subclass for each test class:

class YetAnotherTest extends SomeTest {
    void setup() {
        config = new CompilerConfiguration()
        config.addCompilationCustomizers( ... )
        shell = new GroovyShell(config)
    }
}

Then what you see is that the setup method of both tests is the same. The idea, then, is to create a trait:

trait MyTestSupport {
    void setup() {
        config = new CompilerConfiguration()
        config.addCompilationCustomizers( new ASTTransformationCustomizer(CompileStatic) )
        shell = new GroovyShell(config)
    }
}

Then use it in the subclasses:

class AnotherTest extends SomeTest implements MyTestSupport {}
class YetAnotherTest extends SomeTest2 implements MyTestSupport {}
...

It would allow us to dramatically reduce the boilerplate code, and reduces the risk of forgetting to change the setup code in case we decide to change it. Even if setup is already implemented in the super class, since the test class declares the trait in its interface list, the behavior will be borrowed from the trait implementation!

This feature is in particular useful when you don’t have access to the super class source code. It can be used to mock methods or force a particular implementation of a method in a subclass. It lets you refactor your code to keep the overridden logic in a single trait and inherit a new behavior just by implementing it. The alternative, of course, is to override the method in every place you would have used the new code.

It’s worth noting that if you use runtime traits, the methods from the trait are always preferred to those of the proxied object: 
class Person {
    String name                                         (1)
}
trait Bob {
    String getName() { 'Bob' }                          (2)
}

def p = new Person(name: 'Alice')
assert p.name == 'Alice'                                (3)
def p2 = p as Bob                                       (4)
assert p2.name == 'Bob'                                 (5)
1 the Person class defines a name property which results in a getName method
2 Bob is a trait which defines getName as returning Bob
3 the default object will return Alice
4 p2 coerces p into Bob at runtime
5 getName returns Bob because getName is taken from the trait
Again, don’t forget that dynamic trait coercion returns a distinct object which only implements the original interfaces, as well as the traits.

14. Differences with mixins

There are several conceptual differences with mixins, as they are available in Groovy. Note that we are talking about runtime mixins, not the @Mixin annotation which is deprecated in favour of traits.

First of all, methods defined in a trait are visible in bytecode:

  • internally, the trait is represented as an interface (without default methods) and several helper classes

  • this means that an object implementing a trait effectively implements an interface

  • those methods are visible from Java

  • they are compatible with type checking and static compilation

Methods added through a mixin are, on the contrary, only visible at runtime:

class A { String methodFromA() { 'A' } }        (1)
class B { String methodFromB() { 'B' } }        (2)
A.metaClass.mixin B                             (3)
def o = new A()
assert o.methodFromA() == 'A'                   (4)
assert o.methodFromB() == 'B'                   (5)
assert o instanceof A                           (6)
assert !(o instanceof B)                        (7)
1 class A defines methodFromA
2 class B defines methodFromB
3 mixin B into A
4 we can call methodFromA
5 we can also call methodFromB
6 the object is an instance of A
7 but it’s not an instanceof B

The last point is actually a very important and illustrates a place where mixins have an advantage over traits: the instances are not modified, so if you mixin some class into another, there isn’t a third class generated, and methods which respond to A will continue responding to A even if mixed in.

15. Static methods, properties and fields

The following instructions are subject to caution. Static member support is work in progress and still experimental. The information below is valid for 2.4.12 only.

It is possible to define static methods in a trait, but it comes with numerous limitations:

  • traits with static methods cannot be compiled statically or type checked. All static methods/properties/field are accessed dynamically (it’s a limitation from the JVM).

  • the trait is interpreted as a template for the implementing class, which means that each implementing class will get its own static methods/properties/methods. So a static member declared on a trait doesn’t belong to the Trait, but to it’s implementing class.

Let’s start with a simple example:

trait TestHelper {
    public static boolean CALLED = false        (1)
    static void init() {                        (2)
        CALLED = true                           (3)
    }
}
class Foo implements TestHelper {}
Foo.init()                                      (4)
assert Foo.TestHelper__CALLED                   (5)
1 the static field is declared in the trait
2 a static method is also declared in the trait
3 the static field is updated within the trait
4 a static method init is made available to the implementing class
5 the static field is remapped to avoid the diamond issue

As usual, it is not recommended to use public fields. Anyway, should you want this, you must understand that the following code would fail:

Foo.CALLED = true

because there is no static field CALLED defined on the trait itself. Likewise, if you have two distinct implementing classes, each one gets a distinct static field:

class Bar implements TestHelper {}              (1)
class Baz implements TestHelper {}              (2)
Bar.init()                                      (3)
assert Bar.TestHelper__CALLED                   (4)
assert !Baz.TestHelper__CALLED                  (5)
1 class Bar implements the trait
2 class Baz also implements the trait
3 init is only called on Bar
4 the static field CALLED on Bar is updated
5 but the static field CALLED on Baz is not, because it is distinct

16. Inheritance of state gotchas

We have seen that traits are stateful. It is possible for a trait to define fields or properties, but when a class implements a trait, it gets those fields/properties on a per-trait basis. So consider the following example:

trait IntCouple {
    int x = 1
    int y = 2
    int sum() { x+y }
}

The trait defines two properties, x and y, as well as a sum method. Now let’s create a class which implements the trait:

class BaseElem implements IntCouple {
    int f() { sum() }
}
def base = new BaseElem()
assert base.f() == 3

The result of calling f is 3, because f delegates to sum in the trait, which has state. But what if we write this instead?

class Elem implements IntCouple {
    int x = 3                                       (1)
    int y = 4                                       (2)
    int f() { sum() }                               (3)
}
def elem = new Elem()
1 Override property x
2 Override property y
3 Call sum from trait

If you call elem.f(), what is the expected output? Actually it is:

assert elem.f() == 3

The reason is that the sum method accesses the fields of the trait. So it is using the x and y values defined in the trait. If you want to use the values from the implementing class, then you need to dereference fields by using getters and setters, like in this last example:

trait IntCouple {
    int x = 1
    int y = 2
    int sum() { getX()+getY() }
}

class Elem implements IntCouple {
    int x = 3
    int y = 4
    int f() { sum() }
}
def elem = new Elem()
assert elem.f() == 7

17. Self types

17.1. Type constraints on traits

Sometimes you will want to write a trait that can only be applied to some type. For example, you may want to apply a trait on a class that extends another class which is beyond your control, and still be able to call those methods. To illustrate this, let’s start with this example:

class CommunicationService {
    static void sendMessage(String from, String to, String message) {       (1)
        println "$from sent [$message] to $to"
    }
}

class Device { String id }                                                  (2)

trait Communicating {
    void sendMessage(Device to, String message) {
        CommunicationService.sendMessage(id, to.id, message)                (3)
    }
}

class MyDevice extends Device implements Communicating {}                   (4)

def bob = new MyDevice(id:'Bob')
def alice = new MyDevice(id:'Alice')
bob.sendMessage(alice,'secret')                                             (5)
1 A Service class, beyond your control (in a library, …​) defines a sendMessage method
2 A Device class, beyond your control (in a library, …​)
3 Defines a communicating trait for devices that can call the service
4 Defines MyDevice as a communicating device
5 The method from the trait is called, and id is resolved

It is clear, here, that the Communicating trait can only apply to Device. However, there’s no explicit contract to tell that, because traits cannot extend classes. However, the code compiles and runs perfectly fine, because id in the trait method will be resolved dynamically. The problem is that there is nothing that prevents the trait from being applied to any class which is not a Device. Any class which has an id would work, while any class that does not have an id property would cause a runtime error.

The problem is even more complex if you want to enable type checking or apply @CompileStatic on the trait: because the trait knows nothing about itself being a Device, the type checker will complain saying that it does not find the id property.

One possibility is to explicitly add a getId method in the trait, but it would not solve all issues. What if a method requires this as a parameter, and actually requires it to be a Device?

class SecurityService {
    static void check(Device d) { if (d.id==null) throw new SecurityException() }
}

If you want to be able to call this in the trait, then you will explicitly need to cast this into a Device. This can quickly become unreadable with explicit casts to this everywhere.

17.2. The @SelfType annotation

In order to make this contract explicit, and to make the type checker aware of the type of itself, Groovy provides a @SelfType annotation that will:

  • let you declare the types that a class that implements this trait must inherit or implement

  • throw a compile time error if those type constraints are not satisfied

So in our previous example, we can fix the trait using the @groovy.transform.SelfType annotation:

@SelfType(Device)
@CompileStatic
trait Communicating {
    void sendMessage(Device to, String message) {
        SecurityService.check(this)
        CommunicationService.sendMessage(id, to.id, message)
    }
}

Now if you try to implement this trait on a class that is not a device, a compile-time error will occur:

class MyDevice implements Communicating {} // forgot to extend Device

The error will be:

class 'MyDevice' implements trait 'Communicating' but does not extend self type class 'Device'

In conclusion, self types are a powerful way of declaring constraints on traits without having to declare the contract directly in the trait or having to use casts everywhere, maintaining separation of concerns as tight as it should be.

18. Limitations

18.1. Compatibility with AST transformations

Traits are not officially compatible with AST transformations. Some of them, like @CompileStatic will be applied on the trait itself (not on implementing classes), while others will apply on both the implementing class and the trait. There is absolutely no guarantee that an AST transformation will run on a trait as it does on a regular class, so use it at your own risk!

18.2. Prefix and postfix operations

Within traits, prefix and postfix operations are not allowed if they update a field of the trait:

trait Counting {
    int x
    void inc() {
        x++                             (1)
    }
    void dec() {
        --x                             (2)
    }
}
class Counter implements Counting {}
def c = new Counter()
c.inc()
1 x is defined within the trait, postfix increment is not allowed
2 x is defined within the trait, prefix decrement is not allowed

A workaround is to use the += operator instead.