Frequently Asked Questions

1. Compatibility
2. Serialization
3. XML specifics
4. JSON specifics
5. Security aspects
6. Comparison to other products
7. Scalability
8. Uses of XStream

Compatibility

Which Java runtime is required to run XStream?

XStream 1.4.x requires Java 1.4 or later. Note, that the XStream libraries contains class files targeting different Java runtime versions or Java features. These classes are loaded by reflection and only used if XStream is running on an appropriate runtime environment.

Environments that load all class files of a Java archive can fail with this approach, the Android runtime is such an example. You can build your own version of XStream with a lower JDK then.

For Java 9 and later you will currently have to permit the now illegal access for XStream to operate.

Which JDK is required to build XStream?

XStream 1.4.x can be build still with JDK 1.4 (see BUILD.txt). However, to support the latest features it requires currently a JDK of Java 8. Otherwise the resulting jar file will miss some class files not available on earlier runtimes. Depending on the target environment this can be useful (e.g. for Android or GAE).

Note, that such Java archives will fail on higher Java runtimes then.

Which dependencies are required to run XStream?

All dependencies are optional, XStream uses since version 1.4.1 by default xpp3:xpp3_min and xmlpull:xmlpull. However it depends on the use case. XStream will run without dependencies using the DOM driver on all Java runtimes or the StAX driver in combination with Java 6 or greater. See the list of optional dependencies.

Note, that XStream's manifest contains OSGi entries that declare all dependencies as optional.

Does XStream behave differently across different JVMs?

XStream has two modes of operation: Pure Java and Enhanced. In pure Java mode, XStream behaves in the same way across different JVMs, however its features are limited to what reflection allows, meaning it cannot serialize certain classes or fields. In enhanced mode, XStream does not have these limitations, however this mode of operation is not available to all JVMs.

Which JVMs allow XStream to operate in enhanced mode?

XStream checks since version 1.4.5 automatically for a working enhanced mode - based on undocumented internal Java runtime classes. This enhanced mode is known to be working on the Oracle/Sun, Apple, HP, IBM and Blackdown 1.4 JVMs and onwards, for IcedTea 6 and onwards, for Hitachi, SAP and Diablo from 1.5 and onwards, for BEA JRockit starting with R25.1.0. Generally it works for all modern Java runtimes based on OpenJDK. Android basically supports the enhanced mode as well as the Google Application Engine, but the latter's security model limits the types that can be handled. Note, that an active SecurityManager might prevent the usage of the enhanced mode also.

Since Java 9 it is required to permit the now illegal access. For Java 17 see below.

What are the advantages of using enhanced mode over pure Java mode?

Currently it is not possible to recreate every instance of a type using the official Java API only. The enhanced mode uses some undocumented, but wide-spread available functionality to recreate such instances nevertheless. However, in a secured secured environment, older Java run times or a limited Java environment might prevent the usage of the enhanced mode and XStream uses the plain Java API as fallback. This mode has some restrictions though:

Java runtime warns me about an illegal reflective access by XStream!

Yes, this is normal. A big part of XStream is reflection based and there is currently no replacement for the complete required functionality. You will have to permit this access currently, otherwise XStream will not work.

XStream fails since Java 17, because types in modules cannot be accessed from the unnamed module!

Again, this is normal. The reflection stuff is required to get all required information to recreate an instance of a Java type at unmarshalling time. However, since Java 17 it is no longer possible to allow this access with a single runtime option. You have to open all packages of the individual modules for the unnamed module with the option --add-opens, where XStream requires access, e.g. --add-opens java.base/java.util=ALL-UNNAMED

Why does XStream not even declare an automated module name?

Such a declaration would move XStream automatically into the module class path. However, in this environment a lot of functionality does no longer work. Therefore it is on purpose that XStream stays currently in the unnamed module.

Will XStream support the Java Platform Module System (JPMS)?

At some point definitely. However, you will have to accept a limited functionality only, comparable to the pure Java mode. The access model is very restrictive and XStream will no longer be able to marshal all types of the Java runtime like now.

Why is my application not able to create a XmlPullParser with the XppDriver since XStream 1.4?

The XML Pull Parser API defines an own mechanism to load the factory for the available XPP implementation. XStream's XppDriver never used this lookup mechanism automatically before version 1.4, now it will. Therefore you will have to add a dependency to xmlpull if the XPP implementation does not deliver the classes on its own. This dependency is necessary for Xpp3 in contrast to kXML2 that contains the classes. Use the Xpp3Driver or the KXml2Driver if you want to select one of the directly supported XPP implementation on your own without using the XPP factory. Note, that the minimal version of kXML2 does not support the XPP factory, but can be used by the KXml2Driver.

Can I use XStream in an Android application?

XStream does work in Android 1.0, but is reported to have limited capabilities. Since XStream 1.4 Android is treated at least as JDK 5 platform, but it e.g. does not include the java.beans package. Therefore you cannot use the JavaBeanConverter. Note, that Android provides an XML Pull Parser, therefore XStream can work without additional dependencies.

XStream contains class files targeting different Java runtime versions. Depending on the target version of Android, it is not possible to use the original XStream library directly, because it tries to convert all classes of XStream to the Dalvik runtime. You might have to build a custom version of XStream (see BUILD.txt) with a JDK that is equivalent to the Java level supported by the target version of Android.

Since XStream 1.4.10 an additional artifact is deployed to the Central Maven Repository with -java7 appended to the version that explicitly does not contain any Java 8 related stuff. Note that this version will fail on higher runtimes.

Which limits exists for XStream in Google's Application Engine (GAE)?

Starting with XStream 1.4.6 it is possible to instantiate an XStream instance in a GAE environment. Nevertheless does GAE set some severe restrictions for XStream and therefore XStream will behave differently. Actually a reflection-based converter cannot handle any type from the JDK itself. Nor is it possible to create an ObjectInputStream or an ObjectOutputStream. It is not possible to define a field alias for any type within the JDK. XStream will typically work as general rule, if you process your own objects.

Why does XStream fail on Apache Harmony?

Since JDK 5 it is possible according the Java specification to write into final fields using reflection. This is not yet supported by Harmony and therefore the PureJavaReflectionProvider fails. We have also already investigated into enhanced mode in Harmony, but the Harmony JVM crashed running the unit tests. However, Harmony has been retired, we will no longer make any efforts in this direction.

Are there plans to provide enhanced mode support to other JVMs?

Yes. Let us know which JVM you would like supported.

When should I use XStream not in enhanced mode?

Running XStream in a secured environment can prevent XStream from running in enhanced mode. This is especially true when running XStream in an applet. You may also try to use the JavaBeanConverter as alternative to the ReflectionConverter running in enhanced or pure Java mode.

Which permissions does XStream need when running with an active SecurityManager?

This depends on the mode XStream is running in. Refer to the SecurityManagerTest for details. Actually XStream's converters try to check since version 1.4.6 any critical operation, before they claim to be able to handle a type. As consequence XStream can behave differently running under a SecurityManager. E.g. if the SecurityManager does not permit to create an instance for a derived class of ObjectOutputStream, the SerializationConverter will not handle any type and the ReflectionConverter will take over (as long it has proper rights for its own reflection-based operations).

Why does XStream 1.2 no longer read XML generated with XStream 1.1.x?

The architecture in XStream has slightly changed. Starting with XStream 1.2 the HierarchicalStreamDriver implementation is responsible to ensure that XML tags and attributes are valid names in XML, in XStream 1.1.x this responsibility was part of the ClassMapper implementations. Under some rare circumstances this will result in an unreadable XML due to the different processing order in the workflow of such problematic tag names.

You can run XStream in 1.1 compatibility mode though:

XStream xstream = new XStream(new XppDriver(new XStream11XmlFriendlyReplacer())) {
    protected boolean useXStream11XmlFriendlyMapper() {
        return true;
    }
};

XStream 1.3 ignores suddenly annotated converters (@XStreamConverter and @XStreamConverters)?

XStream treats now all annotations the same and therefore it no longer auto-detects any annotation by default. You can configure XStream to run in auto-detection mode, but be aware if the implications. As alternative you might register the deprecated AnnotationReflectionConverter, that was used for XStream pre 1.3.x, but as drawback the functionality to register a local converter with XStream.registerLocalConverter will no longer work.

XStream 1.3 suddenly has a different field order?

Yes. This was announced with the last 1.2.x release and was done to support the type inheritance of XML schemas. However, XStream is delivered with the XStream12FieldKeySorter that can be used to sort the fields according XStream 1.2.2.

WebSphere 8 can no longer use XStream 1.4.x?

XStream has a long history to support types from recent JDKs without dropping backward compatibility. Therefore it contains class files targeting different JDKs. However, WebShpere scans by default all JAR files in its classpath for annotations to support CDI independent of the presence of a beans.xml file in META-INF. This scanning fails for class files targeting a higher JDK runtime as currently used by WebSphere. Please, consult your WebSphere documentation, how to turn off the scanning for individual files by providing an amm.filter.proeprties file.

Serialization

Which class types can be serialized by XStream?

In contrast to the JDK XStream is not tied to a marker interface to serialize a class. XStream ships with some specialized converters, but will use reflection by default for "unknown" classes to examine, read and write the class' data. Therefore XStream can handle quite any class, especially the ones referred as POJO (Plain Old Java Object).

However, some types of classes exist with typical characteristics, that cannot be handled - at least not out of the box:

1. Objects that are based on threads or thread local data: Thread, Timer, ThreadLocal and so on. These classes keep different data for different threads and there's no possibility to recreate a thread in a generic way nor recreating thread specific data. There might be special use cases, but this will always involve a custom converter where threads can be recreated in a specific way tied to that use case.
2. Class types that are based on generated classes. Such types have often names that are unique to the current process and will have no meaning in a different process. A custom converter might help to write the appropriate data into the serializing stream to be able to recreate a equivalent class at deserialization time.
3. Types that keep and use system resources like file handles, sockets, pipes and so on. ClassLoader, FileInputStream, FileOutputStream, Socket and so on. To deserialize such a class the converter must be able to claim the appropriate resource from the system again. With the help of a custom converter this might be possible, but with the reflection converter the deserialized class might refer a system resource that is no longer valid or belongs to somebody else. Behavior is undefined then.
4. A very special case of such allocated system resources are those classes that keep handles to system memory directly, because they are partly implemented native. It is known to be true for the Linux version of Sun's JDK that the BufferedImage references some system specific types of the JDK that themselves have member fields with such memory handles. While it is possible at first sight to serialize and deserialize a BufferedImage, the reflection converter will also duplicate the memory handle. As a result the JVM might crash easily because of freeing unallocated memory or freeing the same memory twice. It might be possible to create a custom converter, but the data structure is really complex in this area and nobody has been investigating so far to such an extent. However, do not use the reflection converter for these types! You have been warned!
5. Inner class types of the JDK can often vary in implementation details between JDK versions and vendors and are therefore only compatible for the same JDK. This includes collection types returned by the methods of the Collections class that wrap another one (like unmodifiableList) or the collections that are returned by the different Map implementations for the keySet(), entrySet() and values() methods.
6. Non-serializable lambda expressions cannot be deserialized at all and serializable lambda expression contain compiler and vendor specific information that might cause deserialization to fail.

How do I specify that a field should not be serialized?

Make it transient, specify it with XStream.omitField() or annotate it with @XStreamOmitField

How do I initialize a transient field at deserialization?

XStream uses the same mechanism as the JDK serialization. Example:

class ThreadAwareComponent {
  private transient ThreadLocal component;
  // ...
  private Object readResolve() {
    component = new ThreadLocal();
    return this;
  }
}

or in case your type implements java.io.Serializable, you have an alternative:

class ThreadAwareComponent implements Serializable{
  private transient ThreadLocal component;
  // ...
  private void readObject(ObjectInputStream in) throws IOException, ClassNotFoundException {
    in.defaultReadObject();
    component = new ThreadLocal();
  }
}

Use the latter in class hierarchies, readResolve is not called for base classes.

XStream is not calling the default constructor during deserialization.

This is, in fact, the same case as above. XStream uses the same mechanism as the JDK serialization. When using the enhanced mode with the optimized reflection API, it does not invoke the default constructor. The solution is to implement the readResolve or readObject as demonstrated with the last question.

What do serialized collections look like?

See example for the CollectionConverter.

Note, that it is possible to configure XStream to omit the container element toys using implicit collections.

Why do serialized types, fields or methods do not use aliasing for the names?

XStream normally has no to separate between a primitive and its boxed type. The complete reflection API works always with the boxed types and converts to primitives types on the fly. However, for method and field type signatures the difference is essential. Nevertheless it is possible to register derived versions of the converters that are able to respect the aliasing with some minor effort. Following lines are taken from the AliasTest in the acceptance tests:

XStream xstream = new XStream();
Mapper mapper = new MapperWrapper(xstream.getMapper().lookupMapperOfType(ArrayMapper.class)) {
  public Class realClass(String elementName) {
    Class primitiveType = Primitives.primitiveType(elementName);
    return primitiveType != null ? primitiveType : super.realClass(elementName);
  }
};
SingleValueConverter javaClassConverter = new JavaClassConverter(mapper) {};
xstream.registerConverter(javaClassConverter);
xstream.registerConverter(new JavaMethodConverter(javaClassConverter){});
xstream.registerConverter(new JavaFieldConverter(javaClassConverter, mapper){});

My implicit collection is suddenly null after deserialization instead of empty!

By declaring a collection as implicit, the result will have no direct representation of the collection container itself anymore. Therefore, if the collection was empty at serialization time, the serialized result does not contain a trace of the collection anymore. At deserialization time it will therefore not know anything about the collection and will not initialize it. XStream cannot decide anyway at deserialization time, if the collection was empty or null.

The type of my implicit collection is different after deserialization.

By declaring a collection as implicit, the result will have no direct representation of the collection container itself anymore. Therefore XStream cannot track the original type of the collection. At deserialization time it will therefore look at the declaration type of the field that holds the collection and use this type's default implementation, e.g. for a List this is by default an ArrayList.

Beware, that this also means that collections with additional information (e.g. a TreeSet with a Comparator) cannot be restored, since the comparator was already omitted at serialization time.

Do my classes have to implement Serializable if XStream is to serialize them?

No (except for lambda expressions), but XStream respects the Java serialization methods even for types not declared as Serializable.

Can dynamic proxies be serialized?

Yes.

Can immutable types be referenced?

Adding a type as immutable implies that a referencing marshaller strategy will write each occurrence of the same instance separately into the stream without referencing it, i.e. instance of those types are normally never referenced at deserialization time. Such referenced can exist though for persisted streams when a type is added as immutable in later versions. In such cases the type can be added as immutable, but still referenceable.

Any immutable type could be dereferenced before version 1.4.9, but only at the cost of a large memory footprint. Since version 1.4.9 this is only possible if the immutable type has been explicitly declared as referenceable too. This should be done only if backward compatibility is required and a persisted stream may contain a reference of such an instance at all. Any type that has been declared as immutable by XStream itself before version 1.4 will not be referenceable now by default (all primitive types and their boxed counterparts, java.lang.Class, java.lang.String, java.math.BigInteger, java.math.BigDecimal, java.io.File, java.net.URL, and java.awt.font.TextAttribute). All other immutable types (java.util.Currency, java.util.UUID, java.net.URI, java.nio.charset.Charset and the empty collection types) will currently still be referenceable at deserialization time for compatibility reasons. This support by default will be dropped with the next major version of XStream. You can always overwrite the default by adding the same type again as immutable with a different value for the referenceable flag.

My lambda expression is serialized to null!

Non-serializable lambda expressions to not contain any information at all to recreate the instance at a later time again. These instances are treated as temporary objects and as such XStream has no other possibility as to serialize null instead.

Suddenly the deserialization of my (serializable) lambda expression fails!

Serializable lambda expressions contain information that is specific for compiler and vendor. Even worse, the compiler is free to add information related to the location of the lambda expression in the source i.e. you may not be able to deserialize a lambda expression after source code changes. XStream has no control over this information and how it is used by native functionality in the JDK. Therefore Oracle strongly discourages the usage of serializable lambda expressions in the JDK documentation.

Can CGLIB proxies be serialized?

Only limitedly. A proxy generated with the CGLIB Enhancer is supported, if the proxy uses either a factory or only one callback. Then it is possible to recreate the proxy instance at unmarshalling time. Starting with XStream 1.3.1 CGLIB support is no longer automatically installed because of possible classloader problems and side-effects, because of incompatible ASM versions. You can enable CGLIB support with:

XStream xstream = new XStream() {
    protected MapperWrapper wrapMapper(MapperWrapper next) {
        return new CGLIBMapper(next);
    }
};
xstream.registerConverter(new CGLIBEnhancedConverter(xstream.getMapper(), xstream.getReflectionProvider()));

CGLIBEnhancedConverter fails at initialization with ExceptionInInitializerError

This is not a problem of XStream. You have incompatible ASM versions in your classpath. CGLIB 2.1.x and below is based on ASM 1.5.x which is incompatible to newer versions that are used by common packages like Hibernate, Groovy or Guice. Check your dependencies and ensure that you are using either using cglib-nodep-2.x.jar instead of cglib-2.x.jar or update to cglib-2.2.x that depends on ASM 3.1. However, the nodep version contains a copy of the ASM classes with private packages and will therefore not raise class incompatibilities at all.

Serialization fails with NoSuchMethodError: net.sf.cglib.proxy.Enhancer.isEnhanced(Ljava/lang/Class;)Z

XStream uses this method to detect a CGLIB-enhanced proxy. Unfortunately the method is not available in the cglib-2.0 version. Since this version is many years old and the method is available starting with cglib-2.0.1, please consider an upgrade of the dependency, it works usually smoothly.

How do I use XStream's Hibernate package to serialize my objects?

Support of Hibernate enhanced collections and proxied types. To drop the internals of Hibernate when marshalling such objects to XStream, all converters and the mapper has to be registered for the XStream instance:

final XStream xstream = new XStream() {
  protected MapperWrapper wrapMapper(final MapperWrapper next) {
    return new HibernateMapper(next);
  }
};
xstream.registerConverter(new HibernateProxyConverter());
xstream.registerConverter(new HibernatePersistentCollectionConverter(xstream.getMapper()));
xstream.registerConverter(new HibernatePersistentMapConverter(xstream.getMapper()));
xstream.registerConverter(new HibernatePersistentSortedMapConverter(xstream.getMapper()));
xstream.registerConverter(new HibernatePersistentSortedSetConverter(xstream.getMapper()));

Does XStream's Hibernate package support Envers?

Yes. Hibernate Envers is an optional dependency for XStream and it is automatically supported by XStream's Hibernate package when the proxy collection types of Envers are available on the classpath.

My attributes are interpreted by XStream itself and cause unexpected behavior

XStream's generic converters and the marshalling strategies use a number of attributes on their own. Especially the attributes named id, class and reference are likely to cause such collisions. Main reason is XStream's history, because originally user defined attributes were not supported and all attribute were system generated. Starting with XStream 1.3.1 you can redefine those attributes to allow the names to be used for your own ones. The following snippet defines XStream to use different system attributes for id and class while the field id of YourClass is written into the attribute class:

XStream xstream = new XStream() {
xstream.useAttributeFor(YourClass.class, "id");
xstream.aliasAttribute("class", "id");
xstream.aliasSystemAttribute("type", "class");
xstream.aliasSystemAttribute("refid", "id");

Can I select the field order in which XStream serializes objects?

Yes. XStream's ReflectionConverter uses the defined field order by default. You can override it by using an specific FieldKeySorter:

SortableFieldKeySorter sorter = new SortableFieldKeySorter();
sorter.registerFieldOrder(MyType.class, new String[] { "firstToSerialize", "secondToSerialize", "thirdToSerialize" });
xstream = new XStream(new Sun14ReflectionProvider(new FieldDictionary(sorter)));

How does XStream deal with newer versions of classes?

• If a new field is added to the class, deserializing an old version will leave the field uninitialized.
• If a field is removed from the class, deserializing an old version that contains the field will cause an exception. Leaving the field in place but declaring it as transient will avoid the exception, but XStream will not try to deserialize it.
• If a class is renamed, aliases can be used to create an abstraction between the name used in XML and the real class name.
• If a field is renamed, this should be treated as adding a field and removing a field.

For more advanced class migrations, you may

• have to do custom pre-processing of the XML before sending it to XStream (for example, with XSLT or DOM manipulations)
• declare new fields as transient
• implement your own converter, that can handle the situation
• add a readResolve() method to your class, that initializes the object accordingly
• implement a custom mapper to ignore unknown fields automatically (see acceptance test CustomMapperTest.testCanBeUsedToOmitUnexpectedElements())

Future versions of XStream will include features to make these type of migrations easier.

How does XStream cope with isolated class loaders?

Serializing an object graph is never a problem, even if the classes of those objects have been loaded by a different class loader. The situation changes completely at deserialization time. In this case you must set the class loader to use with:

xstream.setClassLoader(yourClassLoader);

Although XStream caches a lot of type related information to gain speed, it keeps those information in tables with weak references that should be cleaned by the garbage collector when the class loader is freed.

Note, that this call should be made quite immediately after creating the XStream and before any other configuration is done. Otherwise configuration based on special types might refer classes loaded with the wrong classloader.

XML specifics

Why does XStream not respect the encoding in the XML declaration?

XStream architecture is based on IO Readers and Writers, while the XML declaration is the responsibility of XML parsers. All HierarchicalStreamDriver implementations respect the encoding since version 1.3, but only if you provide an InputStream. If XStream consumes a Reader you have to initialize the reader with the appropriate encoding yourself, since it is now the reader's task to perform the encoding and no XML parser can change the encoding of a Reader and any encoding definition in the XML header will be ignored.

Why does XStream not write an XML declaration?

XStream is designed to write XML snippets, so you can embed its output into an existing stream or string. You can write the XML declaration yourself into the Writer before using it to call XStream.toXML(writer).

Why does XStream not write XML in UTF-8?

XStream does no character encoding by itself, it relies on the configuration of the underlying XML writer. By default it uses its own PrettyPrintWriter which writes into the default encoding of the current locale. To write UTF-8 you have to provide a Writer with the appropriate encoding yourself.

Why do field names suddenly have double underscores in the generated XML?

XStream maps Java class names and field names to XML tags or attributes. Unfortunately this mapping cannot be 1:1, since some characters used for identifiers in Java are invalid in XML names. Therefore XStream uses an XmlFriendlyNameCoder to replace these characters with a replacement. By default this NameCoder uses an underscore as escape character and has therefore to escape the underscore itself also. You may provide a different configured instance of the XmlFriendlyNameCoder or a complete different implementation like the NoNameCoder to prevent name coding at all. However it is your responsibility then to ensure, that the resulting names are valid for XML.

XStream fails to unmarshal my given XML and I do not know why?

By default XStream is written for persistence i.e. it will read the XML it can write. If you have to transform a given XML into an object graph, you should go the other way round. Use XStream to transfer your objects into XML. If the written XML matches your schema, XStream is also able to read it. This way is much easier, since you can spot the differences in the XML much more easy than to interpret the exceptions XStream will throw if it cannot match the XML into your objects.

My parser claims the � character to be invalid, but it was written with XStream!

Your parser is basically right! A character of value 0 is not valid as part of XML according the XML specification (see version 1.0 or 1.1), neither directly nor as character entity nor within CDATA. But not every parser respects this part of the specification (e.g. Xpp3 will ignore it and read character entities). If you expect such characters in your strings and you do not use the Xpp3 parser, you should consider to use a converter that writes the string as byte array in Base64 code. As alternative you may force the PrettyPrintWriter or derived writers to be XML 1.0 or 1.1. compliant, i.e. in this mode a StreamException is thrown.

My parser claims a control character to be invalid, but it was written with XStream!

Your parser is probably right! Control characters are only valid as part of XML 1.1. You should add an XML header declaring this version or use a parser that does not care about this part of the specification (e.g. Xpp3 parser).

Why is my element not written as XML attribute although I have configured it?

You can only write types as attributes that are represented as a single String value and are handled therefore by SingleValueConverter implementations. If your type is handled by a Converter implementation, the configuration of XStream to write an attribute (using XStream.useAttributeFor() or @XStreamAsAttribute) is simply ignored.

Why are whitespace characters wrong in my attribute values after deserialization?

This is part of the XML specification and a required functionality for any XML parser called attribute value normalization. It cannot be influenced by XStream. A compliant XML parser will replace by default real tab, carriage return and line feed characters with normal spaces. If you want to keep these characters you will have to encode them with entities.

Why does XStream not have any namespace support?

Not every XML parser supports namespaces and not every XML parser that supports namespaces can be configured within XStream to use those. Basically namespaces must be supported individually for the different XML parsers and the only support for namespaces that has currently been implemented in XStream is for the StAX paser. Therefore use and configure the StaxDriver of XStream to use namespaces.

My XML contains XPath expressions in the references, but they seem not to work?

XStream generates only XPath compliant expressions. These have a very limited syntax and they are the only ones that can be interpreted at deserialization again, since XStream does not use an XPath interpreter. Therefore there is no support for attribute selectors, qualified element access with axis names or functions. For real XPath support you will have to implement your own MarshallingStrategy.

The XPath expressions in the references do select a list, but not a single node!

Yes, this is right. However, the result type of an XPath expression evaluation can be defined. A node result from a node list is the lists first node, therefore the XPath of XStream is compliant. Since XStream does not use a real XPath engine, you do not have to worry about memory consumption or wasted evaluation time, XStream will always operate on a single node anyway. Since XStream 1.4 you can force XStream to write XPath expressions that select explicit the single node by using the new modes XStream.SINGLE_NODE_XPATH_ABSOLUTE_REFERENCES or SINGLE_NODE_XPATH_RELATIVE_REFERENCES. Instead of generating a path like "/doc/list/elem/field" XStream will then generate "/doc[1]/list[1]/elem[1]/field[1]". The two notations are transparent at deserialization time.

Does XStream support entities?

Entity support is completely dependent on the XML parser. XStream uses by default the Xpp3 parser that does not support entities at all (like the kXML2 parser). Other parsers support entities, but they might have been turned off to avoid XXE vulnerability. To enable the entities again, you have to overload the individual method of the HierarchicalStreamDriver implementation that generated the parser factory.

JSON specifics

Why are there two JSON driver implementations?

As always, first for historical reasons! Main difference is that the JettisonMappedXmlDriver is a thin wrapper around Jettison in combination with the StaxDriver, while the JsonHierarchicalStreamDriver uses an own more flexible implementation, but can only be used to generate JSON, deserialization is not implemented.

Which versions of Jettison are supported?

Users of Java 5 or higher can use Jettison 1.2, users of Java 1.4.2 have to use Jettison 1.0.1. Jettison 1.1 nor Jettison 1.3 or higher is supported.

Why is it not possible to deserialize a JSON string starting with an array?

XStream's implementation to deserialize JSON is based on Jettison and StAX. Jettison implements a XMLStreamReader of StaX and transforms the processed JSON virtually into XML first. However, if the JSON string starts with an array it is not possible for Jettison to create a valid root element, since it has no name.

XStream fails to unmarshal my JSON string and I do not know why?

Deserialization of JSON is currently done by Jettison, that transforms the JSON string into a StAX stream. XStream itself does nothing know about the JSON format here. If your JSON string reaches some kind of complexity and you do not know how to design your Java objects and configure XStream to match those, you should have a look at the intermediate XML that is processed by XStream in the end. This might help to identify the problematic spots. Also consider then marshalling your Java objects into XML first. You can use following code to generate the XML:

String json = "{\"string\": \"foo\"}";
HierarchicalStreamDriver driver = new JettisonMappedXmlDriver();
StringReader reader = new StringReader(json);
HierarchicalStreamReader hsr = driver.createReader(reader);
StringWriter writer = new StringWriter();
new HierarchicalStreamCopier().copy(hsr, new PrettyPrintWriter(writer));
writer.close();
System.out.println(writer.toString());

What limitations has XStream's JSON support?

JSON represents a very simple data model for easy data transfer. Especially it has no equivalent for XML attributes. Those are written with a leading "@" character, but this is not always possible without violating the syntax (e.g. for array types). Those may silently dropped (and makes it therefore difficult to implement deserialization). References are another issue in the serialized object graph, since JSON has no possibility to express such a construct. You should therefore always set the NO_REFERENCES mode of XStream. Additionally you cannot use implicit collections, since the properties in a JSON object must have unique names.

Why are my Long values incorrect in JavaScript?

JavaScript does not know about integer values. All numbers are represented with double precition floats using 64 bits (IEEE 754). These types cannot represent technically the complete value range of 64-bit integers like Java's Long. With the JsonWriter you have the possibility since XStream 1.4.5 to set IEEE_754_MODE to force any long value that is not representable as JavaScript number to be written as string value in JSON. With the Jettison-based JettisonMappedXmlDriver you may either set a different TypeConverter or force the default converter to write integer values out of the range of 32-bit always as string setting the system property jettison.mapped.typeconverter.enforce_32bit_integer to true (not available for Jettison 1.0.1 and Java 1.4).

Why are there invalid characters in my JSON representation?

The JSON spec requires any JSON string to be in UTF-8 encoding. However, XStream ensures this only if you provide an InputStream or an OutputStream. If you provide a Reader or Writer you have to ensure this requirement on your own.

The generated JSON is invalid, it contains a dash in the label!

Well, no, the JSON is valid! Please check yourself with the JSON syntax checker. However, some JavaScript libraries silently assume that the JSON labels are valid JavaScript identifiers, because JavaScript supports a convenient way to address an element, if the label is a valid JavaScript identifier:

var json = {"label": "foo", "label-with-dash": "bar"};
var fooVar = json.label; // works for labels that are JavaScript identifiers
var barVar = json["label-with-dash"]; // using an array index works always

As alternative you may wrap the JsonWriter and replace any dash with an underscore:

HierarchicalStreamDriver driver = new JsonHierarchicalStreamDriver() {
    public HierarchicalStreamWriter createWriter(Writer out) {
        return new WriterWrapper(super.createWriter(out)) {
            public void startNode(String name) {
                startNode(name, null);
            }
            public void startNode(String name, Class clazz) {
                wrapped.startNode(name.replace('-', '_'), clazz);
            }
        }
    }
};
XStream xstream = new XStream(driver);

Security Aspects

Why does XStream not convert an java.beans.EventHandler?

Since XStream version 1.4.7 it does no longer handle an EventHandler automatically. Such an instance can be used to initiate calls on arbitrary instances at deserialization time e.g. ProcessBuilder.start() You can register a ReflectionConverter instance explicitly for the EventHandler if you need support for such instances.

XStream deserializes arbitrary objects!

Yes, XStream is designed to convert any object form Java to XML and back out of the box. In consequence it is possible to adjust the processed XML manually to inject arbitrary objects into the deserialized object graph. To avoid such a behavior, you have several options:

• Prevent the usage of the reflection-based converters. Register an own converter with priority LOW that claims to handle any type and throw a ConversionException in the marshal and unmarshal methods.
• Overload XStream.setupConverters() and register only converters for the types that are allowed in your object graph.
• Provide own implementations for ConverterLookup and ConverterRegistry constructing the XStream. Your implementation can then select the appropriate converter at lookup time on its own or prevent the registration of specific converters.
• Since XStream 1.4.7 you have the possibility to setup rules for the class types that are allowed to be deserialized.

XStream suddenly emits a vulnerability warning!

If the security framework has not been initialized, XStream will emit such a warning since version 1.4.10. You use the security rules to setup a white list of class types that will be processed.

For XStream 1.4.x you may call the setupDefaultSecurity method to initialize a XStream instance in the same way as it will be done for XStream 1.5.x. This means also a compatibility preparation for XStream 1.5.x.

Is XStream XXE vulnerable (CVE-2016-3674)?

XStream does not contain an own XML parser, therefore it depends on the parser selected with the HierarchicalStreamDriver if the current XStream instance is XXE vulnerable at deserialization time. However, XStream tries to deactivate the processing of external entities by default. Status for the different supported XML parsers:

Note: Only a HierarchicalStreamReader created with the HierarchicalStreamDriver is setup to avoid the XXE vulnerability. If you create such driver instances on your own, it is also your task to setup the XML parser instance properly.

Comparison to other products

How does XStream compare to java.beans.XMLEncoder?

XStream is designed for serializing objects using internal fields, whereas XMLEncoder is designed for serializing JavaBeans using public API methods (typically in the form of getXXX(), setXXX(), addXXX() and removeXXX() methods.

How does XStream compare to JAXB (Java API for XML Binding)?

JAXB is a Java binding tool. It generates Java code from a schema and you are able to transform from those classes into XML matching the processed schema and back. Note, that you cannot use your own objects, you have to use what is generated.

Scalability

Is XStream thread safe?

Yes. Once the XStream instance has been created and configured, it may be shared across multiple threads allowing objects to be serialized/deserialized concurrently (unless you enable the auto-detection to process annotations on-the-fly). Actually the creation and initialization of XStream is quite expensive, therefore it is recommended to keep the XStream instance itself. If you absolutely have to rely on annotation processing on the fly, you will have to use separate XStream instances for each thread - either by using every time a new instance or by a shared pool.

How much memory does XStream consume?

This cannot be answered in general, but following topics have impact on the memory:

1. XML parser technology in use: You should use a streaming parser like Xpp3 or StAX. DOM-based parsers process the complete XML and create their document model in memory before the first converter of XStream is called.
2. Your object model: Is it necessary to keep the complete object graph in memory at once? As alternative you might use object streams or write custom converters that can load and save objects of your object model on the fly without adding them to the object graph physically. As example see the implementation of the XmlArrayList in combination with the FileStreamStrategy from XStream's persistence package to keep parts of the object graph separate.
3. References: By default XStream supports references to the same object in an object graph. This implies that XStream keeps track of all serialized and deserialized objects internally. These references are kept with WeakReferences, so that the memory can be freed as soon as nobody references these objects anymore.
4. XML values: Any tag and attribute value that is converted into a Java String in the object graph will use by default the same String instance unless it exceeds 38 characters (length of a UUID string representation).
5. XStream caches: To increase performance XStream caches quite a lot like classes, converters to use, aliasing, tag names. All those caches make usage of WeakReferences or will exist only while marshalling one object graph resp. unmarshalling one input stream.

Can the performance of XStream be increased?

XStream is a generalizing library, it inspects and handles your types on the fly. Therefore it will normally be slower than a piece of optimized Java code generated out of a schema. However, it is possible to increase the performance anyway:

• Write custom converters for those of your types that occur very often in your XML.
• Keep a configured XStream instance for multiple usage. Creation and initialization is quite expensive compared to the overhead of XStream when calling marshall or unmarshal.
• Use Xpp3 or StAX parsers.

Note, you should never try to optimize code for performance simply because you believe that you have detected a bottle neck. Always use proper tools like a profiler to verify where your hotspots are and whether your optimization was really successful or not.

Uses of XStream

Is XStream a data binding tool?

No. It is a serialization tool.

Can XStream generate classes from XSD?

No. For this kind of work data binding tools such as JAXB or XMLBeans are appropriate.

Why is there no SaxReader?

XStream works on a stream-based parser model, while SAX is event-based. The stream based model implies, that the caller consumes the individual tokens from the XML parser on demand, while in an event-based model the parser controls the application flow on its own and will use callbacks to support client processing. The different architecture makes it therefore impossible for XStream to use an event-driven XML parser.