twitter/Serial

Twitter Serial

中文文档

Download

Grab the latest version via Gradle from Maven Central:

repositories {
  mavenCentral()
}

dependencies {
  implementation 'com.twitter.serial:serial:0.1.6'
}

Overview

Twitter Serial is a custom serialization implementation that's intended to improve performance and increase developer visibility into and control over an object's serialization.

This framework uses Serializers to explicitly define how a class should be serialized. Some of the major advantages of this approach include:

• more efficient serialization avoiding reflection - preliminary metrics for a large object showed
  • more than 3x faster for roundtrip serialization (5x faster to serialize, 2.5x to deserialize)
  • around 5x smaller in byte array size
• greater control over what's serialized for an object - all serialization is defined explicitly
• better debugging capabilities (see debugging)

Basic Structure

To serialize an object to a byte array, use:

final Serial serial = new ByteBufferSerial();
final byte[] serializedData = serial.toByteArray(object, ExampleObject.SERIALIZER)

To deserialize from a byte array back to an object, use:

final ExampleObject object = serial.fromByteArray(ExampleObject.SERIALIZER)

Defining Serializers

• Instead of implementing Serializable, define a serializer for every object that needs to be serialized
• Serializers explicitly write and read each field of the object by using read/write for primitives or recursively calling serializers for other objects
• Serializers handle null objects for you, as does read/writeString; primitive read/write methods do not
• Serializers are stateless, so they are written as static inner classes of the object and accessed as a static instance variable SERIALIZER

For most classes, you can create a subclass of ObjectSerializer and implement serializeObject and deserializeObject

public static class ExampleObject {
    public static final ObjectSerializer<ExampleObject> SERIALIZER = new ExampleObjectSerializer();

    public final int num;
    public final SubObject obj;

    public ExampleObject(int num, @NotNull SubObject obj) {
        this.num = num;
        this.obj = obj;
    }

    ...

    private static final class ExampleObjectSerializer extends ObjectSerializer<ExampleObject> {
        @Override
        protected void serializeObject(@NotNull SerializationContext context, @NotNull SerializerOutput output,
                @NotNull ExampleObject object) throws IOException {
            output
                .writeInt(object.num) // first field
                .writeObject(object.obj, SubObject.SERIALIZER); // second field
        }

        @Override
        @NotNull
        protected ExampleObject deserializeObject(@NotNull SerializationContext context, @NotNull SerializerInput input,
                int versionNumber) throws IOException, ClassNotFoundException {
            final int num = input.readInt(); // first field
            final SubObject obj = input.readObject(SubObject.SERIALIZER); // second field
            return new ExampleObject(num, obj);
        }
    }
}

For classes that are constructed using builders, or have optional fields added (see updating-serializers), you can use a BuilderSerializer, in which you implement the methods createBuilder (which just returns a new builder object for that class) and deserializeToBuilder (where you populate the builder with the deserialized fields)

public static class ExampleObject {
    ...

    public ExampleObject(@NotNull Builder builder) {
        this.num = builder.mNum;
        this.obj = builder.mObj;
    }

    ...

    public static class Builder extends ModelBuilder<ExampleObject> {
        ...
    }

    private static final class ExampleObjectSerializer extends BuilderSerializer<ExampleObject, Builder> {
        @Override
        @NotNull
        protected Builder createBuilder() {
            return new Builder();
        }

        @Override
        protected void serializeObject(@NotNull SerializationContext context, @NotNull SerializerOutput output,
                @NotNull ExampleObject object) throws IOException {
            output.writeInt(object.num)
                .writeObject(object.obj, SubObject.SERIALIZER);
        }

         @Override
        protected void deserializeToBuilder(@NotNull SerializationContext context, @NotNull SerializerInput input,
                @NotNull Builder builder, int versionNumber) throws IOException, ClassNotFoundException {
            builder.setNum(input.readInt())
                .setObj(input.readObject(SubObject.SERIALIZER));
        }
    }
}

Serialization Utility Methods

• CoreSerializers and CollectionSerializers contain serializers for boxed primitives and have helper methods to serialize objects like collections, enums and comparators.

  • For example, to serialize a list of Strings, you can use:

    CollectionSerializers.getListSerializer(Serializers.STRING);

• In order to serialize an object as its base class, you can construct a base class serializer from the subclass's serializers using the getBaseClassSerializer in Serializers

  • For example, if you have ClassA and ClassB that both extend ClassC, and you want to serialize the objects as ClassC objects, you can create a serializer in ClassC using the serializers of the subclasses:

    final Serializer<ClassC> SERIALIZER = Serializers.getBaseClassSerializer(
        SerializableClass.create(ClassA.class, new ClassA.ClassASerializer()),
        SerializableClass.create(ClassB.class, new ClassB.ClassBSerializer()));

  Note

  You must create new instances of ClassA and B serializers rather than using the static object defined in those classes. Since ClassC is initialized as part of its subclasses, using static objects of its subclasses in its initialization will create a cyclic dependency that will likely lead to a cryptic NPE.

Updating Serializers

If you add or remove a field for an object that's being stored as serialized data, there are a few ways to handle it:

OptionalFieldException

If you add a field to the end of an object, your new serializer will reach the end of an old object when trying to read the new field, which will cause it to throw an OptionalFieldException.

BuilderSerializer handles OptionalFieldExceptions for you by just ignoring that field in the builder, stopping deserialization, and building the rest of the object as is. If you're using a regular Serializer instead, you can explicitly catch the OptionalFieldException and set the remaining field(s) to default values as appropriate.

• Say, for example, you wanted to add a String 'name' to the end of the ExampleObject above

  • For both serializer types, you could simply add .writeString(obj.name) to serializeObject

  • For the BuilderSerializer, to deserialize you would add .setName(input.readString()) to the end of deserializeToBuilder. In the case where an older object without the name field is being deserialized, an OptionalFieldException would be thrown and caught when reading the String, causing the object to be built as is without the name field explicitly set.

  • For the regular Serializer, you would change deserializeObject as follows:

    @Override
    @NotNull
    protected ExampleObject deserializeObject(@NotNull SerializationContext context, @NotNull SerializerInput input,
            int versionNumber) throws IOException, ClassNotFoundException {
        final int num = input.readInt();
        final SubObject obj = input.readObject(SubObject.SERIALIZER);
        final String name;
        try {
            name = input.readString();
        } catch (OptionalFieldException e) {
            name = DEFAULT_NAME;
        }
        return new ExampleObject(num, obj, name);
    }

Version numbers

Another option is to increase the version number of the serializer, and define the deserialization behavior for older versions. To do this, pass the version number into the constructor of the SERIALIZER object, and then in the deserialize method you can specify what to do differently for previous versions.

• To change the above example to use version numbers, do the following:

  final Serializer<ExampleObject> SERIALIZER = new ExampleObjectSerializer(1);
  ...
  
  @Override
  @NotNull
  protected ExampleObject deserializeObject(@NotNull SerializationContext context, @NotNull SerializerInput input, int versionNumber)
          throws IOException, ClassNotFoundException {
      final int num = input.readInt();
      final SubObject obj = input.readObject(SubObject.SERIALIZER);
      final String name;
      if (versionNumber < 1) {
          name = DEFAULT_NAME;
      } else {
          name = input.readString();
      }
      return new ExampleObject(num, obj, name);
  }

If you remove a field from the middle of an object, you need to ignore the whole object during deserialization by using the skipObject method in SerializationUtils. This way you don't need to keep the serializer if you are removing the object all together.

• Say in the above example you also wanted to remove the obj field and delete SubObject:

  @Override
  @NotNull
  protected ExampleObject deserializeObject(@NotNull SerializationContext context, @NotNull SerializerInput input, int versionNumber)
          throws IOException, ClassNotFoundException {
      final int num = input.readInt();
      if (versionNumber < 1) {
          SerializationUtils.skipObject()
          name = DEFAULT_NAME;
      } else {
          name = input.readString();
      }
      return new ExampleObject(num, name);
  }

Another option is to call input.peekType(), which allows you to check the type of the next field before reading the object. This is especially helpful if you hadn't updated the version before making a change and don't want to wipe the database, since it allows you to differentiate between the two versions without a version number. Note that this only works if the two types are different.

@Override
@NotNull
protected ExampleObject deserializeObject(@NotNull SerializationContext context, @NotNull SerializerInput input, int versionNumber)
        throws IOException, ClassNotFoundException {
    final int num = input.readInt();
    if (input.peekType() == SerializerDefs.TYPE_START_OBJECT) {
        SerializationUtils.skipObject();
        name = DEFAULT_NAME;
    } else {
        name = input.readString();
    }
    return new ExampleObject(num, name);
}

Value Serializers

Some objects are so simple that do not require support for versioning: Integer, String, Size, Rect... Using an ObjectSerializer with these objects adds an envelope of 2-3 bytes around the serialized data, which can add significant overhead. When versioning is not required, ValueSerializer is a better choice:

public static final Serializer<Boolean> BOOLEAN = new ValueSerializer<Boolean>() {
    @Override
    protected void serializeValue(@NotNull SerializationContext context, @NotNull SerializerOutput output, @NotNull Boolean object) throws IOException {
        output.writeBoolean(object);
    }

    @NotNull
    @Override
    protected Boolean deserializeValue(@NotNull SerializationContext context, @NotNull SerializerInput input) throws IOException {
        return input.readBoolean();
    }
};

This is just a simpler version of ObjectSerializer that handles null, otherwise, just writes the values into the stream.

Debugging

serial also contains methods to help with debugging:

• dumpSerializedData will create a string log of the data in the serialized byte array
• validateSerializedData ensures that the serialized object has a valid structure (e.g. every object start header has a matching end header)

Exceptions now contain more information about the serialization failure, specifically information about the expected type to be deserialized and the type that was found, based on headers written for each value.