Automatic generation of the Builder pattern for Java 1.6+
The Builder pattern is a good choice when designing classes whose constructors or static factories would have more than a handful of parameters. — Effective Java, Second Edition, page 39
- Background
- How to use
@FreeBuilder
- Build tools and IDEs
- Troubleshooting
- Alternatives
- Wait, why "free"?
- License
Implementing the Builder pattern in Java is tedious, error-prone and repetitive. Who hasn't seen a ten-argument constructor, thought cross thoughts about the previous maintainers of the class, then added "just one more"? Even a simple four-field class requires 39 lines of code for the most basic builder API, or 72 lines if you don't use a utility like AutoValue to generate the value boilerplate.
@FreeBuilder
produces all the boilerplate for you, as well as free extras like
JavaDoc, getter methods, mapper methods (Java 8+), collections support,
nested builders, and partial values
(used in testing), which are highly useful, but would very rarely justify
their creation and maintenance burden in hand-crafted code. (We also reserve
the right to add more awesome methods in future!)
[The Builder pattern] is more verbose…so should only be used if there are enough parameters, say, four or more. But keep in mind that you may want to add parameters in the future. If you start out with constructors or static factories, and add a builder when the class evolves to the point where the number of parameters starts to get out of hand, the obsolete constructors or static factories will stick out like a sore thumb. Therefore, it's often better to start with a builder in the first place. — Effective Java, Second Edition, page 39
See Build tools and IDEs for how to add @FreeBuilder
to your project's build and/or IDE.
Create your value type (e.g. Person
) as an interface or abstract class,
containing an abstract accessor method for each desired field. Add the
@FreeBuilder
annotation to your class, and it will automatically generate an
implementing class and a package-visible builder API (Person_Builder
), which
you must subclass. For instance:
import EBuilder;
@FreeBuilder
public interface Person {
/** Returns the person's full (English) name. */
String name();
/** Returns the person's age in years, rounded down. */
int age();
/** Builder of {@link Person} instances. */
class Builder extends Person_Builder { }
}
If you are writing an abstract class, or using Java 8, you may wish to hide the
builder's constructor and manually provide instead a static builder()
method
on the value type (though Effective Java does not do this).
If you write the Person interface shown above, you get:
- A builder class with:
- a no-args constructor
- JavaDoc
- getters (throwing
IllegalStateException
for unset fields) - setters
- lambda-accepting mapper methods (Java 8+)
from
andmergeFrom
methods to copy data from existing values or builders- a
build
method that verifies all fields have been set
- An implementation of
Person
with:toString
equals
andhashCode
- A partial implementation of
Person
for unit tests with:UnsupportedOperationException
-throwing getters for unset fieldstoString
equals
andhashCode
Person person = new Person.Builder()
.name("Phil")
.age(31)
.build();
System.out.println(person); // Person{name=Phil, age=31}
If you prefer your value types to follow the JavaBean naming convention, just prefix your accessor methods with 'get' (or, optionally, 'is' for boolean accessors). FreeBuilder will follow suit, and additionally add 'set' prefixes on setter methods, as well as dropping the prefix from its toString output.
@FreeBuilder
public interface Person {
/** Returns the person's full (English) name. */
String getName();
/** Returns the person's age in years, rounded down. */
int getAge();
/** Builder of {@link Person} instances. */
class Builder extends Person_Builder { }
}
Person person = new Person.Builder()
.setName("Phil")
.setAge(31)
.build();
System.out.println(person); // Person{name=Phil, age=31}
For each property foo
, the builder gets:
Method | Description |
---|---|
A setter method, foo |
Throws a NullPointerException if provided a null. (See the sections on Optional and Nullable for ways to store properties that can be missing.) |
A getter method, foo |
Throws an IllegalStateException if the property value has not yet been set. |
A mapper method, mapFoo |
Java 8+ Takes a UnaryOperator. Replaces the current property value with the result of invoking the unary operator on it. Throws a NullPointerException if the operator, or the value it returns, is null. Throws an IllegalStateException if the property value has not yet been set. |
The mapper methods are very useful when modifying existing values, e.g.
Person olderPerson = Person.Builder.from(person)
.mapAge(age -> age + 1)
.build();
We use method overrides to add customization like default values and constraint checks. For instance:
@FreeBuilder
public interface Person {
/** Returns the person's full (English) name. */
String name();
/** Returns the person's age in years, rounded down. */
int age();
/** Returns a human-readable description of the person. */
String description();
/** Builder class for {@link Person}. */
class Builder extends Person_Builder {
public Builder() {
// Set defaults in the builder constructor.
description("Indescribable");
}
@Override Builder age(int age) {
// Check single-field (argument) constraints in the setter method.
checkArgument(age >= 0);
return super.age(age);
}
@Override public Person build() {
// Check cross-field (state) constraints in the build method.
Person person = super.build();
checkState(!person.description().contains(person.name()));
return person;
}
}
}
If a property is optional—that is, has no reasonable default—then use the Java 8 Optional type (or the Guava Optional type for backwards-compatibility).
/** Returns an optional human-readable description of the person. */
Optional<String> description();
This property will now default to Optional.empty(), and the Builder will gain additional convenience setter methods:
Method | Description |
---|---|
description(String value) |
Sets the property to Optional.of(value) . Throws a NullPointerException if value is null; this avoids users accidentally clearing an optional value in a way that peer review is unlikely to catch. |
clearDescription() |
Sets the property to Optional.empty() . |
description(Optional<String> value) |
Sets the property to value . |
nullableDescription(String value) |
Sets the property to Optional.ofNullable(value) . |
mapDescription(UnaryOperator<String> mapper |
Java 8+ If the property value is not empty, this replaces the value with the result of invoking mapper with the existing value, or clears it if mapper returns null. Throws a NullPointerException if mapper is null. |
Prefer to use explicit defaults where meaningful, as it avoids the need for edge-case code; but prefer Optional to ad-hoc 'not set' defaults, like -1 or the empty string, as it forces the user to think about those edge cases.
As Java currently stands, you should strongly prefer Optional to returning nulls. Using null to represent unset properties is the classic example of Hoare's billion-dollar mistake: a silent source of errors that users won't remember to write test cases for, and which won't be spotted in code reviews. The large "air gap" between the declaration of the getter and the usage is the cause of this problem. Optional uses the compiler to force the call sites to perform explicit null handling, giving reviewers a better chance of seeing mistakes. See also Using and Avoiding Null.
Obviously, greenfield code can trivially adopt Optional, but even existing APIs
can be converted to Optional via a simple refactoring sequence; see below.
However, if you have compelling legacy reasons that mandate using nulls,
you can disable null-checking by marking the getter method @Nullable
. (Any
annotation type named "Nullable" will do, but you may wish to use
javax.annotation.Nullable
, as used in Google Guava.)
/** Returns an optional title to use when addressing the person. */
@Nullable String title();
This property will now default to null, and the Builder's setter methods will change their null-handling behaviour:
Method | Description |
---|---|
title(@Nullable String title) |
Sets the property to title . |
title() |
Returns the current value of the property. May be null. |
mapTitle(UnaryOperator<String> mapper) |
Java 8+ Takes a UnaryOperator. Replaces the current property value, if it is not null, with the result of invoking mapper on it. Throws a NullPointerException if mapper is null. mapper may return a null. |
This is the O(1), non-tedious, non–error-prone way we recomment converting
@Nullable
to Optional:
- Load all your code in Eclipse, or another IDE with support for renaming and inlining.
- [IDE REFACTOR] Rename all your
@Nullable
getters tonullableX()
(orgetNullableX()
if you use JavaBean naming conventions). - Add an Optional-returning
x()
(orgetX()
) - Implement your nullableX methods as:
return x().orElse(null)
(Guava:return x().orNull()
) - [IDE REFACTOR] Inline your
nullableX()
methods
At this point, you have effectively performed an automatic translation of a
@Nullable
method to an Optional-returning one. Of course, your code is not
optimal yet (e.g. if (foo.x().orElse(null) != null)
instead of if (foo.x().isPresent())
). Search-and-replace should get most of these issues.
- [IDE REFACTOR] Rename all your
@Nullable
setters tonullableX
(orsetNullableX
).
Your API is now @FreeBuilder
-compatible :)
@FreeBuilder
has special support for collection and map properties, removing
the foo
accessor method and generating new ones appropriate to the type.
Collection and map properties default to an empty collection/map and cannot hold
nulls.
/** Returns a list of descendents for this person. **/
List<String> descendants();
A List
, Set
or Multiset
property called 'descendants' would generate:
Method | Description |
---|---|
addDescendants(String element) |
Appends element to the collection of descendants. If descendants is a set and the element is already present, it is ignored. Throws a NullPointerException if element is null. |
addDescendants(String... elements) |
Appends all elements to the collection of descendants. If descendants is a set, any elements already present are ignored. Throws a NullPointerException if elements, or any of the values it holds, is null. |
addAllDescendants(Iterable<String> elements) |
Appends all elements to the collection of descendants. If descendants is a set, any elements already present are ignored. Throws a NullPointerException if elements, or any of the values it holds, is null. |
mutateDescendants(Consumer<...<String>> mutator) |
Java 8+ Invokes the Consumer mutator with the collection of descendants. (The mutator takes a list, set or map as appropriate.) Throws a NullPointerException if mutator is null. As mutator is a void consumer, any value returned from a lambda will be ignored, so be careful not to call pure functions like stream expecting the returned collection to replace the existing collection. |
clearDescendants() |
Removes all elements from the collection of descendants, leaving it empty. |
descendants() |
Returns an unmodifiable view of the collection of descendants. Changes to the collection held by the builder will be reflected in the view. |
/** Returns a map of favourite albums by year. **/
Map<Integer, String> albums();
A Map
property called 'albums' would generate:
Method | Description |
---|---|
putAlbums(int key, String value) |
Associates key with value in albums. Throws a NullPointerException if either parameter is null. Replaces any existing entry. |
putAllAlbums(Map<? extends Integer, ? extends String> map) |
Associates all of map 's keys and values in albums. Throws a NullPointerException if the map is null or contains a null key or value. Throws an IllegalArgumentException if any key is already present. |
removeAlbums(int key) |
Removes the mapping for key from albums. Throws a NullPointerException if the parameter is null. Does nothing if the key is not present. |
mutateAlbums(Consumer<Map<Integer, String>> mutator) |
Java 8+ Invokes the Consumer mutator with the map of albums. Throws a NullPointerException if mutator is null. As mutator is a void consumer, any value returned from a lambda will be ignored, so be careful not to call pure functions like stream expecting the returned map to replace the existing map. |
clearAlbums() |
Removes all mappings from albums, leaving it empty. |
albums() |
Returns an unmodifiable view of the map of albums. Changes to the map held by the builder will be reflected in this view. |
/** Returns a multimap of all awards by year. **/
SetMultimap<Integer, String> awards();
A Multimap
property called 'awards' would generate:
Method | Description |
---|---|
putAwards(int key, String value) |
Associates key with value in awards. Throws a NullPointerException if either parameter is null. |
putAllAwards(int key, Iterable<? extends String> values) |
Associates key with every element of values in awards. Throws a NullPointerException if either parameter, or any value, is null. |
putAllAwards(Map<? extends Integer, ? extends String> map) |
Associates all of map 's keys and values in awards. Throws a NullPointerException if the map is null or contains a null key or value. If awards is a map, an IllegalArgumentException will be thrown if any key is already present. |
removeAwards(int key, String value) |
Removes the single pair key -value from awards. If multiple pairs match, which is removed is unspecified. Throws a NullPointerException if either parameter is null. |
removeAllAwards(int key) |
Removes all values associated with key from awards. Throws a NullPointerException if the key is null. |
mutateAwards(Consumer<Map<Integer, String>> mutator) |
Java 8+ Invokes the Consumer mutator with the multimap of awards. Throws a NullPointerException if mutator is null. As mutator is a void consumer, any value returned from a lambda will be ignored, so be careful not to call pure functions like stream expecting the returned multimap to replace the existing multimap. |
clearAwards() |
Removes all mappings from awards, leaving it empty. |
awards() |
Returns an unmodifiable view of the multimap of awards. Changes to the multimap held by the builder will be reflected in this view. |
In all cases, the value type will return immutable objects from its getter.
The mutator methods are useful for invoking methods not directly exposed on the builder, like subList, or methods that take a mutable collection, like sort:
personBuilder
// Delete the fourth and fifth descendants in the list
.mutateDescendants(d -> d.subList(3,5).clear())
// Sort the remaining descendants
.mutateDescendants(Collections::sort);
/** Returns the person responsible for this project. */
Person owner();
@FreeBuilder
has special support for buildable types like protos and other
@FreeBuilder
types. A buildable property called 'owner' would generate:
Method | Description |
---|---|
owner(Person owner) |
Sets the owner. Throws a NullPointerException if provided a null. |
owner(Person.Builder builder) |
Calls build() on builder and sets the owner to the result. Throws a NullPointerException if builder or the result of calling build() is null. |
ownerBuilder() |
Returns a builder for the owner property. Unlike other getter methods in FreeBuilder-generated API, this object is mutable, and modifying it will modify the underlying property. |
mutateOwner(Consumer<Person.Builder> mutator) |
Java 8+ Invokes the Consumer mutator with the builder for the property. Throws a NullPointerException if mutator is null. As mutator is a void consumer, any value returned from a lambda will be ignored. |
The mutate method allows the buildable property to be set up or modified succinctly and readably in Java 8+:
Project project = new Project.Builder()
.mutateOwner(b -> b
.name("Phil")
.department("HR"))
.build();
Effective Java recommends passing required parameters in to the Builder
constructor. While we follow most of the recommendations therein, we explicitly
do not follow this one: while you gain compile-time verification that all
parameters are set, you lose flexibility in client code, as well as opening
yourself back up to the exact same subtle usage bugs as traditional constructors
and factory methods. For the default @FreeBuilder
case, where all parameters
are required, this does not scale.
If you want to follow Effective Java more faithfully in your own types, however, just create the appropriate constructor in your builder subclass:
public Builder(String name, int age) {
// Set all initial values in the builder constructor
name(name);
age(age);
}
Implementation note: in javac, we spot these fields being set in the constructor, and do not check again at runtime.
A partial value is an implementation of the value type which does not conform to the type's state constraints. It may be missing required fields, or it may violate a cross-field constraint.
Person person = new Person.Builder()
.name("Phil")
.buildPartial(); // build() would throw an IllegalStateException here
System.out.println(person); // prints: partial Person{name="Phil"}
person.age(); // throws UnsupportedOperationException
As partials violate the (legitimate) expectations of your program, they must not be created in production code. (They may also affect the performance of your program, as the JVM cannot make as many optimizations.) However, when testing a component which does not rely on the full state restrictions of the value type, partials can reduce the fragility of your test suite, allowing you to add new required fields or other constraints to an existing value type without breaking swathes of test code.
To create types compatible with the Jackson JSON serialization library, use the builder property of @JsonDeserialize to point Jackson at your Builder class. For instance:
// This type can be freely converted to and from JSON with Jackson
@JsonDeserialize(builder = Address.Builder.class)
interface Address {
String city();
String state();
class Builder extends Address_Builder {}
}
@FreeBuilder
will generate appropriate @JsonProperty annotations on the
builder. (If you use Java 8 or Guava types, you may need to include the
relevant Jackson extension modules, jackson-datatype-jdk8 and
jackson-datatype-guava.)
To enable GWT serialization of the generated Value subclass, just add
@GwtCompatible(serializable = true)
to your @FreeBuilder
-annotated type, and
extend/implement Serializable
. This will generate a CustomFieldSerializer,
and ensure all necessary types are whitelisted.
Download the latest FreeBuilder JAR and add it to the classpath (or processorpath, if you supply one) on the command line. If Guava is available, FreeBuilder will use it to generate cleaner, more interoperable implementation code (e.g returning immutable collections).
Add the @FreeBuilder
artifact as an optional dependency to your Maven POM:
<dependencies>
<dependency>
<groupId>org.inferred</groupId>
<artifactId>freebuilder</artifactId>
<version>[current version]</version>
<optional>true</optional>
</dependency>
</dependencies>
If Guava is available, FreeBuilder will use it to generate cleaner, more interoperable implementation code (e.g returning immutable collections).
Add the following lines to your dependencies:
compile 'org.inferred:freebuilder:<current version>'
If Guava is available, FreeBuilder will use it to generate cleaner, more interoperable implementation code (e.g returning immutable collections).
If you use Eclipse or IDEA along with Gradle, consider using the org.inferred.processors plugin to correctly configure code generation in your IDE.
Condensed from Eclipse Indigo's documentation.
Download the latest FreeBuilder JAR and add it to your project. Select it, right-click and choose Build path > Add to Build path.
In your projects properties dialog, go to Java Compiler > Annotation Processing and ensure Enable annotation processing is checked. Next, go to Java Compiler > Annotation Processing > Factory Path, select Add JARs, and select the FreeBuilder JAR.
Condensed from the IntelliJ 14.0.3 documentation and Auto Issue #106.
Download the latest FreeBuilder JAR, add it to your project, right-click it and select Use as Project Library.
In your Settings, go to Build, Execution, Deployment > Compiler > Annotation
Processors and ensure Enable annotation processing is selected, and
Store generated sources relative to is set to Module content root.
(If you have specified a processor path, ensure you add the new JAR to it.
Similarly, if you choose to specify processors explicitly, add
Processor
to the list.)
Run Build > Rebuild Project, then right-click the new generated
folder
(this may have a different name if you have changed the Production sources
directory setting) and select Mark Directory As > Generated Sources Root.
If you make a mistake in your code (e.g. giving your value type a private
constructor), @FreeBuilder
is designed to output a Builder superclass anyway,
with as much of the interface intact as possible, so the only errors you see
are the ones output by the annotation processor.
Unfortunately, javac
has a broken design: if any annotation processor
outputs any error whatsoever, all generated code is discarded, and all
references to that generated code are flagged as broken references. Since
your Builder API is generated, that means every usage of a @FreeBuilder
builder becomes an error. This deluge of false errors swamps the actual
error you need to find and fix. (See also the related issue.)
If you find yourself in this situation, search the output of javac
for the
string "@FreeBuilder type"; nearly all errors include this in their text.
Eclipse manages, somehow, to be worse than javac
here. It will never output
annotation processor errors unless there is another error of some kind; and,
even then, only after an incremental, not clean, compile. In practice, most
mistakes are made while editing the @FreeBuilder
type, which means the
incremental compiler will flag the errors. If you find a generated superclass
appears to be missing after a clean compile, however, try touching the
relevant file to trigger the incremental compiler. (Or run javac.)
- If you find yourself stuck, needing help, wondering whether a given
behaviour is a feature or a bug, or just wanting to discuss
@FreeBuilder
, please join and/or post to the@FreeBuilder
mailing list. - To see a list of open issues, or add a new one, see the
@FreeBuilder
issue tracker. - To submit a bug fix, or land a sweet new feature, read up on how to contribute.
Where is Immutables better than @FreeBuilder
?
Immutables provides many of the same features as @FreeBuilder
, plus a whole host more. Some are optional ways to potentially enhance performance, like derived and lazy attributes, singleton and interned instances, and hash code precomputation. Some are API-changing: strict builders provide a compile-time guarantee that all fields are set (but limit the use of builders as a consequence), while copy methods provide a concise way to clone a value with a single field changed (but require you to reference the generated ImmutableFoo type, not Foo). It provides advanced binary serialization options and GSON support. It lets you easily customize the conventional method prefixes. As Immutables is an active project, this list is likely incomplete.
Where is @FreeBuilder
better than Immutables?
@FreeBuilder
provides some features that are missing from, or not the default in, Immutables:
- Partials.
- Builder getter, mapper and mutation methods.
- Nested builders.
- Jackson serialization is clean and easily customised.
- GWT support
- Proxyable types1.
- Traditional OO customization1.
The first three points are increasingly useful as your interfaces grow in complexity and usage. Partials greatly reduce the fragility of your tests by only setting the values the code being tested actually uses. Your interfaces are liable accumulate more constraints, like new required fields, or cross-field constraints, and while these will be vitally important across the application as a whole, they create a big maintenance burden when they break unit tests for existing code that does not rely on those constraints. Builder getter, mapper and mutation methods and nested builders empower the modify-rebuild pattern, where code changes a small part of an object without affecting the remainder.
The last two points arise because Immutables is type-generating: you write your value type as a prototype, but you always use the generated class. This type is final, meaning you can't proxy it, which unfortunately breaks tools like Mockito and its wonderful smart nulls. The generated builder is hard to customize as you cannot override methods, explaining why Immutables has so many annotations: for instance, the @Value.Check
annotation is unnecessary in @FreeBuilder
, as you can simply override the setters or build method to perform field or cross-field validation.
1: But note that you can write a Builder subclass in your Foo type, enabling FreeBuilder-style override-based customization, though that is not the default approach recommended in the guide, and you will break all your callers if you change between the two.
Immutables is a very active project, frequently adding new features, and future releases may address some or all of these deficiencies.
Why is @FreeBuilder
better than AutoValue?
It’s not! AutoValue provides an implementing class with a package-visible constructor, so you can easily implement the Factory pattern. If you’re writing an immutable type that needs a small number of values to create (Effective Java suggests at most three), and is not likely to require more in future, use the Factory pattern.
How about if you want a builder? AutoValue.Builder lets you explicitly
specify a minimal Builder interface that will then be implemented by generated
code, while @FreeBuilder
provides a generated builder API. AutoValue.Builder
is better if you must have a minimal API—for instance, in an Android
project, where every method is expensive—or strongly prefer a
visible-in-source API at the expense of many useful methods. Otherwise, consider
using @FreeBuilder
to implement the Builder pattern.
I used AutoValue, but now have more than three properties! How
do I migrate to @FreeBuilder
?
- Ensure your getter methods start with 'get' or 'is'.
- Change your annotation to
@FreeBuilder
. - Rewrite your factory method(s) to use the builder API.
- Inline your factory method(s) with a refactoring tool (e.g. Eclipse).
You can always skip step 4 and have both factory and builder methods, if that seems cleaner!
Can I use both AutoValue and @FreeBuilder
?
Not really. You can certainly use both annotations, but you will end up with two different implementing classes that never compare equal, even if they have the same values.
Protocol buffers have provided builders for ages. Why should I
use @FreeBuilder
?
Protocol buffers are cross-platform, backwards- and forwards-compatible, and have a very efficient wire format. Unfortunately, they do not support custom validation logic; nor can you use appropriate Java domain types, such as Instant or Range. Generally, it will be clear which one is appropriate for your use-case.
- Free as in beer: you don't pay the cost of writing or maintaining the builder code.
- Free as in flexible: you should always be able to customize the builder where the defaults don't work for you.
- Free as in liberty: you can always drop
@FreeBuilder
and walk away with the code it generated for you.
Copyright 2014 Google Inc. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.