Builder pattern: nested objects created through other builders - java

Let's assume I have two objects, both created through the builder pattern and one is nested into other:
class Parent {
private final Child child;
private Parent(Child child) {
this.child = child;
}
public static class Builder {
private Child child;
public Builder() {}
public Builder child(Child child) {
this.child = child;
return this;
}
public Parent build() {
return new Parent(child);
}
}
}
class Child {
private final long id;
private Child(Builder builder) {
this.id = builder.id;
}
public static class Builder {
private long id;
public Builder() {}
public Builder id(long id) {
this.id = id;
return this;
}
public Parent build() {
return new Child(this);
}
}
}
So, the obvious usage is quite simple:
Person.Builder parentBuilder = new Person.Builder().child(new Child.Builder().id(10).build());
Is it quite common to make
public static class Builder {
private ChildBuilder child;
public Builder() {}
public Builder child(ChildBuilder child) {
this.child = child;
return this;
}
public Builder resetChildId() {
child.id(0);
return this;
}
public Parent build() {
Child childToPass = child.build();
return new Parent(childToPass);
}
}
That way it is still possible to update the child#id later, however due to late binding the errors are thrown lately during Parent.Builder#build() method.

I would pass a Child instance to Parent rather than a ChildBuilder instance.
If you wish to change Child properties afterwards then you can simply construct a new ChildBuilder from parentBuilder.child().
However, I'm concerned about the design when I see all those builders. DDD is all about the ubiquitous language and "builder" is certainly not part of it. Sometimes you have no choice to introduce technical concepts in the design, but I believe that you may be forgetting about other DDD building blocks that may help.
I have builders everywhere because I have to do validation for each
domain entity in the app. For example name for Parent not longer than
255, but for child not more than 1000. - Tahar Bakir (from the comments)
The rules you describe above may be encapsulated and enforce upon construction in domain concepts such as ParentName and ChildName that can be implemented as value objects.
Your Parent and Child classes can then work with those concepts rather than strings.

Hope this helps
the example on how to use it is in the main method, this will print
10
0
The parent class:
public class Parent {
private final Child child;
private Parent(Child child) {
this.child = child;
}
public Child getChild(){
return this.child;
}
public static class Builder {
private Child.Builder childBuilder;
public Builder() {}
public Builder child(Child.Builder childBuilder) {
this.childBuilder = childBuilder;
return this;
}
public void resetChildId() {
childBuilder = childBuilder.id(0);
}
public Parent build() {
return new Parent(childBuilder.build());
}
}
public static void main (String[] args){
Parent.Builder parentBuilder = new Parent.Builder().child(new Child.Builder().id(10));
System.out.println(parentBuilder.build().getChild().getId());
//Reset the sucker
parentBuilder.resetChildId();
System.out.println(parentBuilder.build().getChild().getId());
}
}
The child class:
class Child {
private final long id;
private Child(Builder builder) {
this.id = builder.id;
}
public long getId(){
return this.id;
}
public static class Builder {
private long id;
public Builder() {}
public Builder id(long id) {
this.id = id;
return this;
}
public Child build() {
return new Child(this);
}
}
}

Related

How do I get an unique ID per class extention?

I have a class with many extended subclasses:
class FirstImplementation extends Mother { [...]
class SecondImplementation extends Mother { [...]
class ThirdImplementation extends Mother { [...]
What I am trying to do is a simple and light way to know if two instances of Mother class have the same implementation:
Mother a = new FirstImplementation();
Mother b = new SecondImplementation();
Mother c = new FirstImplementation();
a.sameKindOf(b); // return false;
a.sameKindOf(c); // return true;
My idea is to set an integer ID field in each Mother instance, and just compare it in sameKindOf function:
public class Mother {
private final int ID;
protected Mother(int ID) {
this.ID = ID;
}
public int getID() {
return this.ID;
}
public boolean sameKindOf(Mother other) {
return this.ID == other.getID();
}
}
Every extension of Mother shall call Mother's constructor with a precise ID.
My question is: is there a way to automatically give a different ID each time I create a new extension, or do I have to do it myself, giving a different number in each constructor class?
If not, is there a simpler way to accomplish what I am trying to do?
If you are interested only in ID-style solution... Try to use the following mechanism:
In your Mother class declare protected static int childClassesNumber;. It will store the number of all unique childs were loaded:
class Mother {
protected static int childClassesNumber = 0;
private final int ID;
protected Mother(int ID) {
this.ID = ID;
}
public int getID() {
return this.ID;
}
public boolean sameKindOf(Mother other) {
return this.ID == other.getID();
}
}
Then, to assure each child gets unique ID, you should use something like this in each child (and this is not good):
class ChildOne extends Mother {
public static final int ID;
static {
ID = ++Mother.childClassesNumber;
}
public ChildOne() {
super(ID);
}
}
The ID will be given only at class loading stage (only one time)
And (for example) ChildTwo:
class ChildTwo extends Mother {
public static final int ID;
static {
ID = ++Mother.childClassesNumber;
}
public ChildTwo() {
super(ID);
}
}
After that, the following code
System.out.println(new ChildOne().sameKindOf(new ChildOne()));
System.out.println(new ChildOne().sameKindOf(new ChildTwo()));
gets:
true
false
This mechanism has a huge drawback - you should put the static initialization in each child. The boilerplate code and so on and so forth... so I would recommend you use #Ash solution )
Wouldnt
public boolean sameKindOf(Mother other) {
return this.getClass().equals(other.getClass());
}
do the job ?
Take a look at java.util.UUID class and its static factory method public static UUID nameUUIDFromBytes(byte[] name). Is that what you are looking for?

How to return an object from parent class with type of Impl class?

I want to provide all implementation classes a default method that initializes the parent fields with default values. So I don't have to redefine those initialization within each Impl class:
public abstract class Parent {
private String name;
public static Parent defaultParent() {
Parent parent = new Parent();
//fill default field values
parent.name = "default name";
return parent;
}
}
public class Child extends Parent {
}
Now I can call Child.defaultParent(). But that would return me an object of type Parent. Is it possible this way to having the object being a Child object implicit? So that the defaultParent method always returns a type of the implementation class?
Maybe with java 8, functional interfaces or similar?
Here's one way to do it with Java 8:
public abstract class Parent {
private String name;
public static <T extends Parent> T defaultParent(Supplier<T> constructor) {
T parent = constructor.get();
//fill default field values
parent.name = "default name";
return parent;
}
}
public class Child extends Parent {
}
You can call the static method like this:
Child c = Parent.defaultParent(Child::new);
Or you can just use a constructor overload instead of a static factory method:
public abstract class Parent {
private String name;
public Parent(boolean useDefaults) {
if (useDefaults) {
//fill default field values
this.name = "default name";
}
}
}
public class Child extends Parent {
public Child(boolean useDefaults) {
super(useDefaults);
}
public Child() {
super(false);
}
}
This is what constructors were made for.
abstract class Parent {
private String name;
protected Parent(String name) {
this.name = name;
}
protected Parent() {
this("default name");
}
}
class DefaultChild extends Parent {
public DefaultChild() { super(); }
}
class CustomChild extends Parent {
public CustomChild(String value) { super(value); }
}
Everything else is just a fancy init method which is inferior to constructors because you can for example no longer assign values to final fields.
abstract class Parent {
private String name;
protected void init() {
this.name= "default name";
}
// or
protected static void init(Parent parent) {
parent.name = "default name";
}
}

is my implementation of builder violates mutability

I'd like to know whether my implementation of QuestionBuilder violates mutability.
public class Question<T extends Serializable> implements Serializable {
private QuestionHolder<T> questionHolder;
private Question(QuestionHolder<T> questionHolder) {
this.questionHolder = questionHolder;
}
public String getId() {
return questionHolder.id;
}
public int getOrder() {
return questionHolder.order;
}
public QuestionType getType() {
return questionHolder.type;
}
public boolean isImmediate() {
return questionHolder.immediate;
}
public boolean isMandatory() {
return questionHolder.mandatory;
}
public List<T> getSelectedValues() {
return questionHolder.selectedValues;
}
public List<T> getPossibleValues() {
return questionHolder.possibleValues;
}
private static final class QuestionHolder<T extends Serializable> {
private String id;
private int order = 0;
private QuestionType type;
private boolean immediate;
private boolean mandatory;
private List<T> selectedValues;
private List<T> possibleValues;
}
public static final class QuestionBuilder<T extends Serializable> implements Builder<Question<T>> {
private QuestionHolder<T> questionHolder;
public QuestionBuilder(String id) {
questionHolder = new QuestionHolder<>();
questionHolder.id = id;
}
public QuestionBuilder withOrder(int order) {
questionHolder.order = order;
return this;
}
public QuestionBuilder withType(QuestionType questionType) {
questionHolder.type = questionType;
return this;
}
public QuestionBuilder withImmediate(boolean immediate) {
questionHolder.immediate = immediate;
return this;
}
public QuestionBuilder withMandatory(boolean mandatory) {
questionHolder.mandatory = mandatory;
return this;
}
public QuestionBuilder withSelectedValues(List<T> selectedValues) {
questionHolder.selectedValues = selectedValues;
return this;
}
public QuestionBuilder withPossibleValues(List<T> possibleValues) {
questionHolder.possibleValues = possibleValues;
return this;
}
public Question<T> build() {
Question<T> question = new Question<>(questionHolder);
questionHolder = null;
return question;
}
}
}
Or what should I adjust in order to resolve mutability issue. Any suggestions?
If you're worried about thread safety, then your code here is not necessarily thread safe.
It is possible that one thread calls build() and returns a Question pointing to a QuestionHolder. Even though build() sets the holder to null, another thread might not see that null, but instead see the old value of the field. If that other thread called any of your setters, it would potentially mutate the Holder that the Question class had already accessed.
In a single threaded application you would be fine.
As far as I can see, you are mutating the QuestionHolder with each builder call.
What I would do is:
1) Make all properties inside QuestionHolder private and don't create any setters at all.
2) Store each property inside the builder instance and create a new instance of QuestionHolder in the build method of the builder.
For example:
public Question<T> build() {
// DO ALL THE VALIDATIONS NEEDED
QuestionHolder holder = new QuestionHolder(id, order, type, inmediate, mandatory, selectedValues, possibleValues);
return new Question<>(questionHolder);
}
With these approach, you will be mutating the Builder, but that's ok for the Builder Pattern. You will obviously need to create a new Builder instance each time you want to create a Question. If you want to use the same Builder over and over again you will probably need to store some kind of structure inside it (a Map identified by Id, for example).

Using Builder Constructor on Extended Class?

I'm implementing a Builder constructor as documented in Joshua Bloch's "Effective Java 2nd Edition. However, I'm running into a few complications when I try to extend the class and its builder. Essentially, the extended Builder in the extended child class has set field methods that return the parent Builder type, not the child builder type.
Of course, I can cast back to the ChildBuilder in the property build chain (as shown in my main method) but it is not seamless which defeats the purpose of the Builder, and it also forces me to segregate the parent setters and child setters.
I tried to use generics but it ended up becoming more verbose than the cast.
Is there a way I can consistently make the set methods on the builders return the builder type that was actually instantiated?
public class ParentObj {
public static void main(String[] args) {
ChildObj childObj = ((ChildObj.ChildBuilder) (new ChildObj.ChildBuilder())
.prop1(11)
.prop2(21)
.prop3(14))
.prop4(12)
.prop5(33)
.build();
}
private int prop1;
private int prop2;
private int prop3;
protected ParentObj(Builder builder) {
this.prop1 = builder.prop1;
this.prop2 = builder.prop2;
this.prop3 = builder.prop3;
}
public class Builder {
private int prop1;
private int prop2;
private int prop3;
public Builder prop1(int prop1) { this.prop1 = prop1; return this; }
public Builder prop2(int prop2) { this.prop2 = prop2; return this; }
public Builder prop3(int prop3) { this.prop3 = prop3; return this; }
public ParentObj build()
{
return new ParentObj(this);
}
}
}
private class ChildObj extends ParentObj {
private final int prop4;
private final int prop5;
private ChildObj(ChildBuilder childBuilder) {
super(childBuilder);
}
public class ChildBuilder extends Builder {
private int prop4;
private int prop5;
public ChildBuilder prop4(int prop4) { this.prop4 = prop4; return this; }
public ChildBuilder prop5(int prop5) { this.prop5 = prop5; return this; }
public ChildObj build() {
return new ChildObj(this);
}
}
}
Probably the best way would be to Override the parent builder methods.
class ChildBuilder {
public ChildBuilder prop1(int prop1){
return (ChildBuilder) super.prop1(prop1);
}
}
While this isn't exactly clean it will work for what you're trying to do.

Builder Pattern and Inheritance

I have an object hierarchy that increases in complexity as the inheritance tree deepens. None of these are abstract, hence, all of their instances serve a, more or less sophisticated, purpose.
As the number of parameters is quite high, I would want to use the Builder Pattern to set properties rather than code several constructors. As I need to cater to all permutations, leaf classes in my inheritance tree would have telescoping constructors.
I have browsed for an answer here when I hit some problems during my design. First of, let me give you a simple, shallow example to illustrate the problem.
public class Rabbit
{
public String sex;
public String name;
public Rabbit(Builder builder)
{
sex = builder.sex;
name = builder.name;
}
public static class Builder
{
protected String sex;
protected String name;
public Builder() { }
public Builder sex(String sex)
{
this.sex = sex;
return this;
}
public Builder name(String name)
{
this.name = name;
return this;
}
public Rabbit build()
{
return new Rabbit(this);
}
}
}
public class Lop extends Rabbit
{
public float earLength;
public String furColour;
public Lop(LopBuilder builder)
{
super(builder);
this.earLength = builder.earLength;
this.furColour = builder.furColour;
}
public static class LopBuilder extends Rabbit.Builder
{
protected float earLength;
protected String furColour;
public LopBuilder() { }
public Builder earLength(float length)
{
this.earLength = length;
return this;
}
public Builder furColour(String colour)
{
this.furColour = colour;
return this;
}
public Lop build()
{
return new Lop(this);
}
}
}
Now that we have some code to go on, imaging I want to build a Lop:
Lop lop = new Lop.LopBuilder().furColour("Gray").name("Rabbit").earLength(4.6f);
This call will not compile as the last chained call cannot be resolved, Builder not defining the method earLength. So this way requires that all calls be chained in a specific order which is very impractical, especially with a deep hierarchy tree.
Now, during my search for an answer, I came across Subclassing a Java Builder class which suggests using the Curiously Recursive Generic Pattern. However, as my hierarchy does not contain an abstract class, this solution will not work for me. But the approach relies on abstraction and polymorphism to function which is why I don't believe I can adapt it to my needs.
An approach I have currently settled with is to override all methods of the superclass Builder in the hierarchy and simply do the following:
public ConcreteBuilder someOverridenMethod(Object someParameter)
{
super(someParameter);
return this;
}
With this approach I can assure I am being returned an instance I can issue chain calls on. While this is not as worse as the Telescoping Anti-pattern, it is a close second and I consider it a bit "hacky".
Is there another solution to my problem that I am not aware of? Preferably a solution consistent with the design pattern. Thank you!
This is certainly possible with the recursive bound, but the subtype builders need to also be generic, and you need a few interim abstract classes. It's a little bit cumbersome, but it's still easier than the non-generic version.
/**
* Extend this for Mammal subtype builders.
*/
abstract class GenericMammalBuilder<B extends GenericMammalBuilder<B>> {
String sex;
String name;
B sex(String sex) {
this.sex = sex;
return self();
}
B name(String name) {
this.name = name;
return self();
}
abstract Mammal build();
#SuppressWarnings("unchecked")
final B self() {
return (B) this;
}
}
/**
* Use this to actually build new Mammal instances.
*/
final class MammalBuilder extends GenericMammalBuilder<MammalBuilder> {
#Override
Mammal build() {
return new Mammal(this);
}
}
/**
* Extend this for Rabbit subtype builders, e.g. LopBuilder.
*/
abstract class GenericRabbitBuilder<B extends GenericRabbitBuilder<B>>
extends GenericMammalBuilder<B> {
Color furColor;
B furColor(Color furColor) {
this.furColor = furColor;
return self();
}
#Override
abstract Rabbit build();
}
/**
* Use this to actually build new Rabbit instances.
*/
final class RabbitBuilder extends GenericRabbitBuilder<RabbitBuilder> {
#Override
Rabbit build() {
return new Rabbit(this);
}
}
There's a way to avoid having the "concrete" leaf classes, where if we had this:
class MammalBuilder<B extends MammalBuilder<B>> {
...
}
class RabbitBuilder<B extends RabbitBuilder<B>>
extends MammalBuilder<B> {
...
}
Then you need to create new instances with a diamond, and use wildcards in the reference type:
static RabbitBuilder<?> builder() {
return new RabbitBuilder<>();
}
That works because the bound on the type variable ensures that all the methods of e.g. RabbitBuilder have a return type with RabbitBuilder, even when the type argument is just a wildcard.
I'm not much of a fan of that, though, because you need to use wildcards everywhere, and you can only create a new instance using the diamond or a raw type. I suppose you end up with a little awkwardness either way.
And by the way, about this:
#SuppressWarnings("unchecked")
final B self() {
return (B) this;
}
There's a way to avoid that unchecked cast, which is to make the method abstract:
abstract B self();
And then override it in the leaf subclass:
#Override
RabbitBuilder self() { return this; }
The issue with doing it that way is that although it's more type-safe, the subclass can return something other than this. Basically, either way, the subclass has the opportunity to do something wrong, so I don't really see much of a reason to prefer one of those approaches over the other.
Confronted with the same issue, I used the solution proposed by emcmanus at: https://community.oracle.com/blogs/emcmanus/2010/10/24/using-builder-pattern-subclasses
I'm just recopying his/her preferred solution here. Let say we have two classes, Shape and Rectangle. Rectangle inherits from Shape.
public class Shape {
private final double opacity;
public double getOpacity() {
return opacity;
}
protected static abstract class Init<T extends Init<T>> {
private double opacity;
protected abstract T self();
public T opacity(double opacity) {
this.opacity = opacity;
return self();
}
public Shape build() {
return new Shape(this);
}
}
public static class Builder extends Init<Builder> {
#Override
protected Builder self() {
return this;
}
}
protected Shape(Init<?> init) {
this.opacity = init.opacity;
}
}
There is the Init inner class, which is abstract, and the Builder inner class, that is an actual implementation. Will be useful when implementing the Rectangle:
public class Rectangle extends Shape {
private final double height;
public double getHeight() {
return height;
}
protected static abstract class Init<T extends Init<T>> extends Shape.Init<T> {
private double height;
public T height(double height) {
this.height = height;
return self();
}
public Rectangle build() {
return new Rectangle(this);
}
}
public static class Builder extends Init<Builder> {
#Override
protected Builder self() {
return this;
}
}
protected Rectangle(Init<?> init) {
super(init);
this.height = init.height;
}
}
To instantiate the Rectangle:
new Rectangle.Builder().opacity(1.0D).height(1.0D).build();
Again, an abstract Init class, inheriting from Shape.Init, and a Build that is the actual implementation. Each Builder class implement the self method, which is responsible to return a correctly cast version of itself.
Shape.Init <-- Shape.Builder
^
|
|
Rectangle.Init <-- Rectangle.Builder
If anyone still bumped into the same problem, I suggest the following solution, that conforms "Prefer composition over inheritance" design pattern.
Parent class
The main element of it is the interface that parent class Builder must implement:
public interface RabbitBuilder<T> {
public T sex(String sex);
public T name(String name);
}
Here is the changed parent class with the change:
public class Rabbit {
public String sex;
public String name;
public Rabbit(Builder builder) {
sex = builder.sex;
name = builder.name;
}
public static class Builder implements RabbitBuilder<Builder> {
protected String sex;
protected String name;
public Builder() {}
public Rabbit build() {
return new Rabbit(this);
}
#Override
public Builder sex(String sex) {
this.sex = sex;
return this;
}
#Override
public Builder name(String name) {
this.name = name;
return this;
}
}
}
The child class
The child class Builder must implement the same interface (with different generic type):
public static class LopBuilder implements RabbitBuilder<LopBuilder>
Inside the child class Builder the field referencing parentBuilder:
private Rabbit.Builder baseBuilder;
this ensures that parent Builder methods are called in the child, however, their implementation is different:
#Override
public LopBuilder sex(String sex) {
baseBuilder.sex(sex);
return this;
}
#Override
public LopBuilder name(String name) {
baseBuilder.name(name);
return this;
}
public Rabbit build() {
return new Lop(this);
}
The constructor of Builder:
public LopBuilder() {
baseBuilder = new Rabbit.Builder();
}
The constructor of builded child class:
public Lop(LopBuilder builder) {
super(builder.baseBuilder);
}
I have adopted the following guidelines when creating object hierarchies with builders:
Make the constructor of the class at least protected and use it as copy constructor, thus pass it an instance of the class itself.
Make the fields non-final private and use getters to access them.
Add package private setters for the builders, which is also nice for object serialization frameworks.
Make a generic builder for each class that will have a subclass builder. This builder will already contain the setter methods for the current class, but we create also a second non generic builder for the class that contains the constructor and build method.
The builders will not have any fields. Instead the generic builder that is on top of the hierarchy will contain a generic field for the concrete object to be build.
The Rabbit will look like this:
public class Rabbit {
// private non-final fields
private String sex;
private String name;
// copy constructor
Rabbit(Rabbit rabbit) {
sex = rabbit.sex;
name = rabbit.name;
}
// no-arg constructor for serialization and builder
Rabbit() {}
// getter methods
public final String getSex() {
return sex;
}
public final String getName() {
return name;
}
// package private setter methods, good for serialization frameworks
final void setSex(String sex) {
this.sex = sex;
}
final void setName(String name) {
this.name = name;
}
// create a generic builder for builders that have subclass builders
abstract static class RBuilder<R extends Rabbit, B extends RBuilder<R, B>> {
// the builder creates the rabbit
final R rabbit;
// here we pass the concrete subclass that will be constructed
RBuilder(R rabbit) {
this.rabbit = rabbit;
}
public final B sex(String sex) {
rabbit.setSex(sex);
return self();
}
public final B name(String name) {
rabbit.setName(name);
return self();
}
#SuppressWarnings("unchecked")
final B self() {
return (B) this;
}
}
// the builder that creates the rabbits
public static final class Builder extends RBuilder<Rabbit, Builder> {
// creates a new rabbit builder
public Builder() {
super(new Rabbit());
}
// we could provide a public copy constructor to support modifying rabbits
public Builder(Rabbit rabbit) {
super(new Rabbit(rabbit));
}
// create the final rabbit
public Rabbit build() {
// maybe make a validate method call before?
return new Rabbit(rabbit);
}
}
}
and our Lop:
public final class Lop extends Rabbit {
// private non-final fields
private float earLength;
private String furColour;
// copy constructor
private Lop(Lop lop) {
super(lop);
this.earLength = lop.earLength;
this.furColour = lop.furColour;
}
// no-arg constructor for serialization and builder
Lop() {}
// getter methods
public final float getEarLength() {
return earLength;
}
public final String getFurColour() {
return furColour;
}
// package private setter methods, good for serialization frameworks
final void setEarLength(float earLength) {
this.earLength = earLength;
}
final void setFurColour(String furColour) {
this.furColour = furColour;
}
// the builder that creates lops
public static final class Builder extends RBuilder<Lop, Builder> {
public Builder() {
super(new Lop());
}
// we could provide a public copy constructor to support modifying lops
public Builder(Lop lop) {
super(new Lop(lop));
}
public final Builder earLength(float length) {
rabbit.setEarLength(length);
return self(); // this works also here
}
public final Builder furColour(String colour) {
rabbit.setFurColour(colour);
return self();
}
public Lop build() {
return new Lop(rabbit);
}
}
}
Pros:
The builders will exactly mirror the object hierarchy of your classes with a single derivative for each generic builder to build the objects of the current class. No need to create artificial parents.
The class does not have a dependency to its builder. All it needs is an instance of itself to copy the fields, which might be useful for alternative factories.
The classes work very well with serialization frameworks like JSON or Hibernate, since they most often need getters and setters to be present. E.g. Jackson works fine with package private setters.
No need to duplicate fields in the builder. The builder contains the object to be constructed.
No need to override setter methods in the subtype builders since the direct parent class is generic.
Build-in support for copy constructors to allow creating a modified version of an instance, making the objects 'kind of immutable'.
Cons:
Requires at least one additional generic builder.
Fields are not final, thus it's not safe to make them public.
The class itself needs additional setter methods to be called from the builders.
Let's create some rabbits..
#Test
void test() {
// creating a rabbit
Rabbit rabbit = new Rabbit.Builder() //
.sex("M")
.name("Rogger")
.build();
assertEquals("M", rabbit.getSex());
// create a lop
Lop lop = new Lop.Builder() //
.furColour("Gray")
.name("Rabbit")
.earLength(4.6f)
.build();
// modify only the name of the lop
lop = new Lop.Builder(lop) //
.name("Lop")
.build();
assertEquals("Gray", lop.getFurColour());
assertEquals("Lop", lop.getName());
}
This form seems to nearly work. It is not very tidy but it looks like it avoids your issues:
class Rabbit<B extends Rabbit.Builder<B>> {
String name;
public Rabbit(Builder<B> builder) {
this.name = builder.colour;
}
public static class Builder<B extends Rabbit.Builder<B>> {
protected String colour;
public B colour(String colour) {
this.colour = colour;
return (B)this;
}
public Rabbit<B> build () {
return new Rabbit<>(this);
}
}
}
class Lop<B extends Lop.Builder<B>> extends Rabbit<B> {
float earLength;
public Lop(Builder<B> builder) {
super(builder);
this.earLength = builder.earLength;
}
public static class Builder<B extends Lop.Builder<B>> extends Rabbit.Builder<B> {
protected float earLength;
public B earLength(float earLength) {
this.earLength = earLength;
return (B)this;
}
#Override
public Lop<B> build () {
return new Lop<>(this);
}
}
}
public class Test {
public void test() {
Rabbit rabbit = new Rabbit.Builder<>().colour("White").build();
Lop lop1 = new Lop.Builder<>().earLength(1.4F).colour("Brown").build();
Lop lop2 = new Lop.Builder<>().colour("Brown").earLength(1.4F).build();
//Lop.Builder<Lop, Lop.Builder> builder = new Lop.Builder<>();
}
public static void main(String args[]) {
try {
new Test().test();
} catch (Throwable t) {
t.printStackTrace(System.err);
}
}
}
Although I have successfully built Rabbit and Lop (in both forms) I cannot at this stage work out how to actually instantiate one of the Builder objects with it's full type.
The essence of this method relies on the cast to (B) in the Builder methods. This allow you to define the type of object and the type of the Builder and retain that within the object while it is constructed.
If anyone could work out the correct syntax for this (which is wrong) I would appreciate it.
Lop.Builder<Lop.Builder> builder = new Lop.Builder<>();
I did some experimenting and I found this to work quite nicely for me.
Note that I prefer to create the actual instance at the start and the call all the setters on that instance. This is just a preference.
The main differences with the accepted answer is that
I pass a parameter that indicated the return type
There is no need for an Abstract... and a final builder.
I create a 'newBuilder' convenience method.
The code:
public class MySuper {
private int superProperty;
public MySuper() { }
public void setSuperProperty(int superProperty) {
this.superProperty = superProperty;
}
public static SuperBuilder<? extends MySuper, ? extends SuperBuilder> newBuilder() {
return new SuperBuilder<>(new MySuper());
}
public static class SuperBuilder<R extends MySuper, B extends SuperBuilder<R, B>> {
private final R mySuper;
public SuperBuilder(R mySuper) {
this.mySuper = mySuper;
}
public B withSuper(int value) {
mySuper.setSuperProperty(value);
return (B) this;
}
public R build() {
return mySuper;
}
}
}
and then a subclass look like this:
public class MySub extends MySuper {
int subProperty;
public MySub() {
}
public void setSubProperty(int subProperty) {
this.subProperty = subProperty;
}
public static SubBuilder<? extends MySub, ? extends SubBuilder> newBuilder() {
return new SubBuilder(new MySub());
}
public static class SubBuilder<R extends MySub, B extends SubBuilder<R, B>>
extends SuperBuilder<R, B> {
private final R mySub;
public SubBuilder(R mySub) {
super(mySub);
this.mySub = mySub;
}
public B withSub(int value) {
mySub.setSubProperty(value);
return (B) this;
}
}
}
and a subsub class
public class MySubSub extends MySub {
private int subSubProperty;
public MySubSub() {
}
public void setSubSubProperty(int subProperty) {
this.subSubProperty = subProperty;
}
public static SubSubBuilder<? extends MySubSub, ? extends SubSubBuilder> newBuilder() {
return new SubSubBuilder<>(new MySubSub());
}
public static class SubSubBuilder<R extends MySubSub, B extends SubSubBuilder<R, B>>
extends SubBuilder<R, B> {
private final R mySubSub;
public SubSubBuilder(R mySub) {
super(mySub);
this.mySubSub = mySub;
}
public B withSubSub(int value) {
mySubSub.setSubSubProperty(value);
return (B)this;
}
}
}
To verify it fully works I used this test:
MySubSub subSub = MySubSub
.newBuilder()
.withSuper (1)
.withSub (2)
.withSubSub(3)
.withSub (2)
.withSuper (1)
.withSubSub(3)
.withSuper (1)
.withSub (2)
.build();
The following IEEE conference contribution Refined Fluent Builder in Java gives a comprehensive solution to the problem.
It dissects the original question into two sub-problems of inheritance deficiency and quasi invariance and shows how a solution to these two sub-problems opens for inheritance support with code reuse in the classical builder pattern in Java.
As you cannot use generics, now probably the main task is to somehow loosen typing.
I don't know how you process those properties afterwards, but what if you used a HashMap for storing them as key-value pairs? So there will be just one set(key, value) wrapper method in the builder (or builder might not be necessary any more).
The downside would be additional type castings while processing the stored data.
If this case is too loose, then you could keep the existing properties, but have a general set method, which uses reflection and searches for setter method on the basis of 'key' name. Although I think reflection would be an overkill.

Categories

Resources