Object Serialization Enhancements in Previous Releases

Documentation Contents

Support for deserialization of unshared objects (since 1.4)

Serialization now provides extra support for deserialization of objects which are known to be unshared in the data-serialization stream. The new support is provided by the following API additions in package java.io:

ObjectInputStream.readUnshared()

ObjectOutputStream.writeUnshared(Object obj)

ObjectStreamField(String name, Class type, boolean unshared)

Previously, security-conscious programmers had to clone private internal objects after deserializing them to guard against the possibility that outside parties with access to the serialization stream could append spurious back handles to the sensitive objects, resulting in extra references to them during deserialization. However, this solution slows performance and wastes memory -- two objects must be created and a copy operation invoked in order to ensure a unique reference to a single usable object. The new API additions provide a more efficient solution by allowing receivers to invalidate "unshared" objects as they are read in, making it impossible for third parties to dereference back-handles to those objects occurring later in the stream.

Security permissions now required to override putFields, readFields (since 1.4)

Beginning with J2SE 1.4.0, ObjectOutputStream's public one-argument constructor requires the "enableSubclassImplementation" SerializablePermission when invoked (either directly or indirectly) by a subclass which overrides ObjectOutputStream.putFields or ObjectOutputStream.writeUnshared.

Also beginning with J2SE 1.4.0, ObjectInputStream's public one-argument constructor requires the "enableSubclassImplementation" SerializablePermission when invoked (either directly or indirectly) by a subclass which overrides ObjectInputStream.readFields or ObjectInputStream.readUnshared.

These changes will not affect the great majority of applications. However, it will affect any ObjectInputStream/ObjectOutputStream subclasses which override the putFields or readFields methods without also overriding the rest of the serialization infrastructure.

Support for class-defined readObjectNoData method (since 1.4)

In addition to supporting class-defined writeObject() and readObject() methods, serialization now includes support for class-defined readObjectNoData() methods. Each class-defined readObjectNoData() method is required to have the following signature:

private void readObjectNoData() throws ObjectStreamException;

The readObjectNoData() method is analogous to the class-defined readObject() method, except that (if defined) it is called in cases where the class descriptor for a superclass of the object being deserialized (and hence the object data described by that class descriptor) is not present in the serialization stream. More formally:

If object O of class C is being deserialized, and S is a superclass of C in the VM which is deserializing O, then S.readObjectNoData() is invoked by ObjectInputStream during the deserialization of O if and only if the following conditions are true:

1. S implements java.io.Serializable (directly or indirectly).
2. S defines an readObjectNoData() method with the signature listed above.
3. The serialization stream containing O does not include a class descriptor for S among its list of superclass descriptors for C.

Note that readObjectNoData() is never invoked in cases where a class-defined readObject() method could be called, though serializable class implementors can call readObjectNoData() from within readObject() as a means of consolidating initialization code.

See the class description in the API specification of ObjectInputStream for more information.

Bug fix: Deserialization fails for Class object of primitive type (since 1.4)

In previous releases, bug 4171142 caused attempts to deserialize Class objects of primitive types to fail with a ClassNotFoundException. The problem was that ObjectInputStream.resolveClass() did not work for ObjectStreamClass descriptors for primitive types. This bug is fixed in J2SE 1.4.0.

Bug fix: ObjectInputStream.resolveProxyClass can fail for non-public interface cases (since 1.4)

In previous releases, ObjectInputStream.resolveProxyClass would not always select the proper class loader to define the proxy class in if one or more of the proxy interfaces were non-public. In this release, if ObjectInputStream.resolveProxyClass detects a non-public interface, it attempts to define the implementing proxy class in the same class loader as the interface (barring conflicts, in which case an exception is thrown), which is necessary in order for the proxy to implement the interface.

Bug fix: Invalid serialPersistentFields field name causes NullPointerException (since 1.4)

In previous releases, bug 4387368 caused NullPointerExceptions to be thrown when serializing objects which used default serialization but also declared serialPersistentField entries which did not map to actual class fields. Serialization will now throw InvalidClassExceptions in such cases (since it is never necessary to define such "unbacked" serialPersistentFields when using default serialization).

Bug fix: ClassNotFoundException in skipped objects causes serialization to fail (since 1.4)

In previous releases, ClassNotFoundExceptions triggered by "skipped" objects--objects associated with fields not present in the classes loaded by the deserializing party--would cause deserialization of the entire object graph to fail, even though the skipped values would not be included in the graph. This release of serialization addresses this problem by ignoring ClassNotFoundExceptions associated with such skipped objects, thus eliminating a class of unnecessary deserialization errors. Other miscellaneous changes have also been made to improve the overall robustness of serialization with regards to ClassNotFoundExceptions encountered during deserialization.

Strings longer than 64K can now be serialized (since 1.3)

Prior to 1.3, an attempt to serialize a string longer than 64K would result in a java.io.UTFDataFormatException being thrown. In 1.3, the serialization protocol has been enhanced to allow strings longer than 64K to be serialized. Note that if a 1.2 (or earlier) JVM attempts to read a long string written from a 1.3-compatible JVM, the 1.2 (or earlier) JVM will receive a java.io.StreamCorruptedException.

Serialization performance enhancements (since 1.3)

Several changes have been made to serialization to improve overall performance:

• UTF string reads/writes have been optimized to reduce unnecessary memory allocation and synchronization/method call overhead.
• Code for reading and writing primitive data arrays has been streamlined. Float and double array reads/writes have been reimplemented to minimize the number of calls to native methods.
• Internal buffering has been improved.
• Reflective operations for getting/setting primitive field values have been batched to minimize the number of separate native method calls.

Improved exception reporting (since 1.3)

If a class cannot be found during the class resolution process of deserialization, the original java.lang.ClassNotFoundException is thrown instead of a generic one so that more information about the failure is available. Also, deserialization exception reporting now includes maintaining the name of the original class that could not be found instead of reporting a higher-level class that was being deserialized. For example, if (in an RMI call) the stub class can be found but the remote interface class cannot, the serialization mechanism will now report correctly that the interface class was the class that could not be found instead of erroneously reporting that the stub class could not be found.

java.io.ObjectOutputStream.writeClassDescriptor,
java.io.ObjectInputStream.readClassDescriptor (since 1.3)

The writeClassDescriptor and readClassDescriptor methods have been added to provide a means of customizing the serialized representation of java.io.ObjectStreamClass class descriptors. writeClassDescriptor is called when an instance of java.io.ObjectStreamClass needs to be serialized, and is responsible for writing the ObjectStreamClass to the serialization stream. Conversely, readClassDescriptor is called when the ObjectInputStream expects an ObjectStreamClass instance as the next item in the serialization stream. By overriding these methods, subclasses of ObjectOutputStream and ObjectInputStream can transmit class descriptors in an application-specific format. For more information, refer to sections 2.1 and 3.1 of the Java Object Serialization Specification.

java.io.ObjectOutputStream.annotateProxyClass,
java.io.ObjectInputStream.resolveProxyClass (since 1.3)

These methods are similar in purpose to ObjectOutputStream.annotateClass and ObjectInputStream.resolveClass, except that they apply to dynamic proxy classes (see java.lang.reflect.Proxy), as opposed to non-proxy classes. Subclasses of ObjectOutputStream may override annotateProxyClass to store custom data in the stream along with descriptors for dynamic proxy classes. ObjectInputStream subclasses may then override resolveProxyClass to make use of the custom data in selecting a local class to associate with the given proxy class descriptor. For details, see section 4 of the Java Object Serialization Specification.

The javadoc tool tags @serial, @serialField, and @serialData (since 1.2)

The javadoc tags @serial, @serialField, and @serialData have been added to provide a way to document the serialized form of a class. Javadoc generates a serialization specification based on the contents of these tags. For details, refer to section 1.6 of the Java Object Serialization Specification.

Protocol versioning (since 1.2)

Prior to 1.2, object serialization used a protocol that did not support skipping over objects implementing the java.io.Externalizable interface if the classes for those objects were not available. In 1.2, a new protocol version was added which addressed this deficiency. For backwards compatibility, ObjectOutputStream and ObjectInputStream can read and write serialization streams written in either protocol; the protocol version used can be selected by calling the ObjectOutputStream.useProtocolVersion method. For details and a discussion of compatibility issues, see section 6.3 of the Java Object Serialization Specification.

Class-defined writeReplace and readResolve methods (since 1.2)

Since 1.2, classes can define writeReplace and readResolve methods which allow instances of the given classes to nominate replacements for themselves during serialization and deserialization. The required signatures of these methods, along with further details, are described in sections 2.5 and 3.6 of the Java Object Serialization Specification.

java.io.ObjectOutputStream.writeObjectOverride, java.io.ObjectInputStream.readObjectOverride (since 1.2)

Since 1.2, subclasses of ObjectOutputStream and ObjectInputStream can implement a custom serialization protocol by overriding the writeObjectOverride and readObjectOverride methods. Note that these methods will only be called if the ObjectOutputStream/ObjectInputStream subclasses possess the permission java.io.SerializablePermission("enableSubclassImplementation"), and call the no-argument constructors of ObjectOutputStream/ObjectInputStream. See sections 2.1 and 3.1 of the Java Object Serialization Specification for more information.

Security permission checks (since 1.2)

Subclasses of ObjectOutputStream and ObjectInputStream may override inherited methods to obtain "hooks" into certain aspects of the serialization process. Since 1.2, object serialization uses the 1.2 security model to verify that subclasses possess adequate permissions to override certain hooks. The permissions java.io.SerializablePermission("enableSubclassImplementation") and java.io.SerializablePermission("enableSubstitution") govern whether or not the methods ObjectOutputStream.writeObjectOverride, ObjectOutputStream.replaceObject, ObjectInputStream.readObjectOverride, and ObjectInputStream.resolveObject will be called during the course of serialization. See sections 2.1 and 3.1 of the Java Object Serialization Specifications for more information.

Defining serializable fields for a class (since 1.2)

By default, the values of all non-static and non-transient fields of a serializable class are written when an instance of that class is serialized. In 1.2, a new mechanism was introduced to allow classes finer control of this process. By declaring a special field serialPersistentFields, serializable classes can dictate which fields will be written when instances of the class (or subclasses) are serialized. This feature also enables classes to "define" serializable fields which do not correspond directly to actual fields in the class. Used in conjunction with the serializable fields API (described below), this capability allows fields to be added or removed from a class without altering the serialized representation of the class. See sections 1.5 and 1.7 of the Java Object Serialization Specification for details.

Serializable fields API (since 1.2)

Introduced in 1.2, the serializable fields API allows class-defined writeObject/readObject methods to explicitly set and retrieve serializable field values by name and type. This API is particularly useful for classes that need to maintain backwards compatibility with older class versions; in some cases, the older version of the class may have defined a set of serializable fields that cannot be mapped directly to the fields of the current class. In this case, newer versions of the class can define custom writeObject and readObject methods that convert the internal state of a given instance of the (new) class into the "old" serialized form, and vice versa. For more information, see section 1.7 of the Java Object Serialization Specification.

Copyright © 2004 Sun Microsystems, Inc. All Rights Reserved.  Please send comments to: rmi-comments@java.sun.com