POJOs - Plain Old Java Objects

The 3.5 release of the driver adds POJO support via the PojoCodec, which allows for direct serialization of POJOs and Java Beans to and from BSON. Internally, each PojoCodec utilizes a ClassModel instance to store metadata about how the POJO should be serialized.

A ClassModel for a POJO includes:

• The class of the POJO.
• A new instance factory. Handling the creation of new instances of the POJO. By default it requires the POJO to have an empty constructor.
• Property information, a list of PropertyModel instances that contain all the property metadata. By default this includes; any public getter methods with any corresponding setter methods and any public fields.
• An optional IdProperty (see BsonId annotation). By default the _id or id property in the POJO.
• Type data for the POJO and its fields to work around type erasure.
• An optional discriminator value. The discriminator is the value used to represent the POJO class being stored.
• An optional discriminator key. The document key name for the discriminator.
• The use discriminator flag. This determines if the discriminator should be serialized. By default it is off.
  

Each PropertyModel includes:

• The property name.
• The read name, the name of the property to use as the key when serializing into BSON.
• The write name, the name of the property to use as the key when deserializing from BSON.
• Type data, to work around type erasure.
• An optional Codec for the property. The codec allows for fine grained control over how the property is encoded and decoded.
• A serialization checker. This checks if the value should be serialized. By default, null values are not serialized.
• A property accessor. Used to access the property values from the POJO instance.
• Use discriminator flag, only used when serializing other POJOs. By default it is off. When on the PojoCodecProvider copies the ClassModel for the field’s type and turns on the use discriminator flag. The corresponding ClassModel must be configured with a discriminator key and value.

ClassModels are built using the ClassModelBuilder which can be accessed via the ClassModel.builder(clazz) method. The builder initially uses reflection to create the required metadata.

PojoCodec instances are created by the PojoCodecProvider which is a CodecProvider. CodecProviders are used by the CodecRegistry to find the correct Codec for any given class.

POJO support

Automatic POJO support can be provided by setting PojoCodecProvider.Builder#automatic(true), once built the PojoCodecProvider will automatically create a PojoCodec for any class that contains at least one serializable or deserializable property.

The entry point for customisable POJO support is the PojoCodecProvider. New instances can be created via the PojoCodecProvider.builder() method. The builder allows users to register any combination of:

• Individual POJO classes.
• Package names containing POJO classes.
• ClassModel instances which allow fine grained control over how a POJO is encoded and decoded.
  

The builder also allows the user to register default Conventions for any POJOs that are automatically mapped, either the individual POJO classes or POJOs found from registered packages. The PojoCodecProvider will lookup PojoCodecs and return the first that matches the POJO class:

• Registered ClassModels
• Registered POJO classes
• Registered POJO classes contained in one of the registered packages

Once the PojoCodecProvider has been built, by calling builder.build(), it can be combined with an existing CodecRegistry to create a new registry that will also support the registered POJOs. The following example registers the package org.example.pojos and creates a new CodecRegistry.

import org.bson.codecs.configuration.CodecProvider;
import org.bson.codecs.configuration.CodecRegistry;
import static org.bson.codecs.configuration.CodecRegistries.fromRegistries;
import static org.bson.codecs.configuration.CodecRegistries.fromProviders;

// Create a CodecRegistry containing the PojoCodecProvider instance.
CodecProvider pojoCodecProvider = PojoCodecProvider.builder().register("org.example.pojos").build();
CodecRegistry pojoCodecRegistry = fromRegistries(defaultCodecRegistry, fromProviders(pojoCodecProvider));

Default configuration

By default the PojoCodec will not store null values or a discriminator when converting a POJO to BSON.

Take the following Person class:

public class Person {
    private String firstName;
    private String lastName;
    private Address address = null;
 
    public Person() { }
    
    public Person(final String firstName, final String lastName) {  /* Set values... */ }

    public String getFirstName() {
        return firstName;
    }
    
    public void setFirstName(final String firstName) {
        this.firstName = firstName;
    }

    public String getLastName() {
        return lastName;
    }
    
    public void setLastName(final String lastName) {
        this.lastName = lastName;
    }

    public String getAddress() {
        return address;
    }
    
    public void setAddress(final Address address) {
        this.address = address;
    }
    
}

The instance of new Person("Ada", "Lovelace"); would be serialized to the equivalent of { firstName: "Ada", lastName: "Lovelace"}.

Notice the address property is omitted because it hasn’t been set and has a null value. If the person instance contained an address, it would be stored as a sub document and use the CodecRegistry to look up the Codec for the Address class and use that to encode and decode the address value.

Generics support

Generics are fully supported. During the creation of a ClassModelBuilder type parameters are inspected and saved to work around type erasure. The only requirement is the top level POJO cannot contain any type parameters.

Take the following classes:

public class GenericClass<T> {
    T genericField;
    // Rest of implementation
}

public class GenericTree<A, B> {
    GenericTree<A, B> left;
    GenericTree<A, B> right;
    // Rest of implementation
}

public final class Tree extends GenericTree<Integer, String> {
    GenericClass<Long> genericClass;
    // Rest of implementation
}

The Tree POJO is serializable because it doesn’t have any unknown type parameters. The left, right and genericClass properties are all serializable because they are bound to the concrete types Integer, String and Long.

On their own, instances of GenericTree or GenericClass are not serializable by the PojoCodec. This is because the runtime type parameter information is erased by the JVM, and the type parameters cannot be specialized accurately.

The 3.6 release of the driver further improves generic support with the addition of PropertyCodecProviders. The PropertyCodecProvider API allows type-safe support of container types by providing concrete type parameters for the generic types as declared in the POJO.

A great use of the PropertyCodecProvider API could be to add support for Guava’s Optional class. The following example creates a OptionalPropertyCodecProvider:

public class OptionalPropertyCodecProvider implements PropertyCodecProvider {

    @Override
    @SuppressWarnings({"rawtypes", "unchecked"})
    public <T> Codec<T> get(final TypeWithTypeParameters<T> type, final PropertyCodecRegistry registry) {
        // Check the main type and number of generic parameters
        if (Optional.class.isAssignableFrom(type.getType()) && type.getTypeParameters().size() == 1) {
            // Get the codec for the concrete type of the Optional, as its declared in the POJO.
            Codec<?> valueCodec = registry.get(type.getTypeParameters().get(0));
            return new OptionalCodec(type.getType(), valueCodec);
        } else {
            return null;
        }
    }

    private static final class OptionalCodec<T> implements Codec<Optional<T>> {
        private final Class<Optional<T>> encoderClass;
        private final Codec<T> codec;

        private OptionalCodec(final Class<Optional<T>> encoderClass, final Codec<T> codec) {
            this.encoderClass = encoderClass;
            this.codec = codec;
        }

        @Override
        public void encode(final BsonWriter writer, final Optional<T> optionalValue, final EncoderContext encoderContext) {
            if (optionalValue != null && optionalValue.isPresent()) {
                codec.encode(writer, optionalValue.get(), encoderContext);
            } else {
                writer.writeNull();
            }
        }

        @Override
        public Optional<T> decode(final BsonReader reader, final DecoderContext context) {
            return Optional.of(codec.decode(reader, context));
        }

        @Override
        public Class<Optional<T>> getEncoderClass() {
            return encoderClass;
        }
    }
}

The OptionalPropertyCodecProvider can be registered via the PojoCodecProvider.builder().register method:

CodecProvider pojoCodecProvider = PojoCodecProvider.builder()
        .register("org.example.pojos")
        .register(new OptionalPropertyCodecProvider())
        .build();

Then Optional classes are fully supported in the POJOs. In the following example a Person has an Optional address and an Optional membership id:

public class Person {
    ... 
    private Optional<Address> optionalAddress;
    private Optional<Integer>  optionalMembershipId;
}

Enum support

Enums are fully supported. The PojoCodec uses the name of the enum constant as the property value. This is then converted back into an Enum value by the codec using the static Enum.valueOf method.

Take the following example:

public enum Membership {
    UNREGISTERED,
    SUBSCRIBER,
    PREMIUM
}

public class Person {
    private String firstName;
    private String lastName;
    private Member membership = Member.UNREGISTERED;

    public Person() { }

    public Person(final String firstName, final String lastName, final Membership membership) { }

    // Rest of implementation
}

The instance of new Person("Bryan", "May", SUBSCRIBER); would be serialized to the equivalent of { firstName: "Bryan", lastName: "May", membership: "SUBSCRIBER"}.

If you require an alternative representation of the Enum, you can override how a Enum is stored by registering a custom Codec for the Enum in the CodecRegistry.

Conventions

The Convention interface provides a mechanism for ClassModelBuilder instances to be configured during the build stage and the creation of the ClassModel.

The following Conventions are available from the Conventions class:

• The ANNOTATION_CONVENTION. Applies all the default annotations.
• The CLASS_AND_PROPERTY_CONVENTION. Sets the discriminator key if not set to _t and the discriminator value if not set to the ClassModels simple type name. Also, configures the PropertyModels. If the idProperty isn’t set and there is a property named _id or id then it will be marked as the idProperty.
• The SET_PRIVATE_FIELDS_CONVENTION. Enables private fields to be set directly using reflection, without the need of a setter method. Note this convention is not enabled by default.
• The USE_GETTERS_FOR_SETTERS. Allows getters to be used for Map and Collection properties without setters, the collection/map is then mutated. Note this convention is not enabled by default.
• The DEFAULT_CONVENTIONS, a list containing the ANNOTATION_CONVENTION and the CLASS_AND_PROPERTY_CONVENTION.
• The NO_CONVENTIONS an empty list.

Custom Conventions can either be set globally via the PojoCodecProvider.Builder.conventions(conventions) method, or via the ClassModelBuilder.conventions(conventions) method.

Annotations

Annotations require the ANNOTATION_CONVENTION and provide an easy way to configure how POJOs are serialized.

The following annotations are available from the org.bson.codecs.pojo.annotations package:

• BsonCreator, marks a public constructor or a public static method as the creator for new instances of the class. Can be combined with the BsonProperty annotation to link the parameters with properties.
• BsonDiscriminator, enables using a discriminator. Also allows for setting a custom discriminator key and value.
• BsonId, marks a property to be serialized as the _id property.
• BsonIgnore, marks a property to be ignored. Can be used to configure if a property is serialized and / or deserialized.
• BsonProperty. Allows for an alternative document key name when converting the POJO field to BSON. Also, allows a field to turn on using a discriminator when storing a POJO value.

Annotations can be applied to read and / or write contexts by configuring the getter / setter methods. Any annotations applied to a field are applied to both read and write contexts.

Take the following Person class:

import org.bson.codecs.pojo.annotations.*;

@BsonDiscriminator
public final class Person {

    @BsonId
    public String personId;
    public String firstName;

    @BsonProperty("surname")
    public String lastName;

    @BsonIgnore
    public String password;

    @BsonProperty(useDiscriminator = true)
    public Address addr;

    public Person(){
    }

}

The Person POJO Will produce BSON similar to:

{ "_id": "1234567890", "_t": "Person", "firstName": "Alan", "surname": "Turing",
  "address": { "_t": "Address", "address": "The Mansion", "street": "Sherwood Drive", 
               "town": "Bletchley", "postcode": "MK3 6EB" } }

The getter methods are used to configure the document keys used when storing the data in MongoDB. The _id document key maps to the POJO’s personId property. The _t field contains the discriminator and the address field also contains a discriminator.

Advanced configuration

For most scenarios there is no need for further configuration. However, there are some scenarios where custom configuration is required.

Properties with abstract or interface types.

If a POJO contains a field that has an abstract type or has an interface as its type, then a discriminator is required. The type and all subtypes / implementations need to be registered with the PojoCodecProvider so that values can be encoded and decoded correctly.

The easiest way to enable a discriminator is to annotate the abstract class with the Discriminator annotation. Alternatively, the ClassModelBuilder.enableDiscriminator(true) method can be used to enable the use of a discriminator.

The following example creates a CodecRegistry with discriminators enabled for a User interface and its concrete FreeUser and SubscriberUser implementations:

ClassModel<User> userModel = ClassModel.builder(User.class).enableDiscriminator(true).build();
ClassModel<FreeUser> freeUserModel = ClassModel.builder(FreeUser.class).enableDiscriminator(true).build();
ClassModel<SubscriberUser> subscriberUserModel = ClassModel.builder(SubscriberUser.class).enableDiscriminator(true).build();

PojoCodecProvider pojoCodecProvider = PojoCodecProvider.builder().register(userModel, freeUserModel, subscriberUserModel).build();

CodecRegistry pojoCodecRegistry = fromRegistries(defaultCodecRegistry, fromProviders(pojoCodecProvider));

Supporting POJOs without no args constructors

By default PojoCodecs work with POJOs that have an empty, no arguments, constructor. POJOs with alternative constructors are supported via the ANNOTATION_CONVENTION and the @BsonCreator annotation. Any parameters for the creator constructor should be annotated with the @BsonProperty annotation. Below is an example of a Person POJO that contains final fields that are set via the annotated constructor:

import org.bson.codecs.pojo.annotations.*;

@BsonDiscriminator
public final class Person {
    private final String pid;
    private final String fName;
    private final String lName;
    
    @BsonProperty(useDiscriminator = true)
    private final Address addr;
    
    @BsonCreator
    public Person(@BsonProperty("personId") final String personId, @BsonProperty("firstName") final String firstName,
                  @BsonProperty("lastName") final String lastName, @BsonProperty("address") final Address address) {
        this.pid = personId;
        this.fName = firstName;
        this.lName = lastName;
        this.addr = address;
    }

    @BsonId
    public String getPersonId() {
        return pid;
    }

    public String getFirstName() {
        return fName;
    }

    public String getLastName() {
        return lName;
    }

    public String getAddress() {
        return addr;
    }
}

Changing what is serialized

By default null values aren’t serialized. This is controlled by the default implementation of the FieldSerialization interface. Custom implementations can be set on the PropertyModelBuilder which is available from the ClassModelBuilder.

The BsonIgnore can be used along with the DEFAULT_CONVENTIONS to mark a property to be ignored when serializing and or deserializing.