IT Consultant – Java, J2EE, IBM WebSphere Portal, Lotus Notes/Domino
RSS icon Home icon
  • Configuring EMF Teneo with Hibernate, Commons DBCP, Spring Hibernate Transaction Manager, and the OpenSessionInViewFilter

    (6 votes) 1 Star2 Stars3 Stars4 Stars5 Stars
    Posted on 2 October 2010 Sebastian Thomschke*/?> No comments

    While trying to get an application working with , Hibernate, Commons DBCP, Spring Hibernate Transaction Manager and the OpenSessionInViewFilter I encountered several unexpected issues. Here is an example of a working configuration for this software stack. Read the comments below the XML file for some explanations.

    package de.sebthom.util;
    import java.sql.*;
    import java.util.Properties;
    import javax.sql.DataSource;
    import org.eclipse.emf.teneo.PersistenceOptions;
    import org.eclipse.emf.teneo.hibernate.HbSessionDataStore;
    import org.hibernate.HibernateException;
    import org.hibernate.cfg.*;
    import org.hibernate.connection.ConnectionProvider;
     * @author Sebastian Thomschke
    public class HbDataStoreWithDataSource extends HbSessionDataStore {
      public final static class CustomConnectionProvider implements ConnectionProvider {
        private DataSource ds;
        public void close() throws HibernateException { }
        public void closeConnection(final Connection conn) throws SQLException {
        public void configure(final Properties props) throws HibernateException {
          ds = _CONFIG_TIME_DS_HOLDER.get();
        public Connection getConnection() throws SQLException {
          return ds.getConnection();
       public boolean supportsAggressiveRelease() {
         return false;
      private static final long serialVersionUID = 1L;
      private static HbDataStoreWithDataSource INSTANCE;
       * the data source holder acts as a thread safe bridge between the DSConnectionProvider which is instantiated by Hibernate internally
       * and the data source injected into the session factory during spring configuration time. 
      private static final ThreadLocal _CONFIG_TIME_DS_HOLDER = new ThreadLocal();
      public static HbDataStoreWithDataSource getInstance() {
    		return INSTANCE;
      private final Properties props = new Properties();
      public HbDataStoreWithDataSource() {
        INSTANCE = this;
        props.setProperty(PersistenceOptions.ADD_INDEX_FOR_FOREIGN_KEY, "true");
        props.setProperty(PersistenceOptions.FETCH_CONTAINMENT_EAGERLY, "false");
        props.setProperty(PersistenceOptions.FETCH_ASSOCIATION_EXTRA_LAZY, "false");
        // tell Teneo to not change the table names, we do this using Hibernate's ImprovedNamingStrategy
        props.setProperty(PersistenceOptions.SQL_CASE_STRATEGY, "none");
        props.setProperty(PersistenceOptions.MAXIMUM_SQL_NAME_LENGTH, "-1");
      protected Configuration createConfiguration() {
        Configuration cfg = super.createConfiguration();
        // only if a DS is injected we register our connection provider
        if (_CONFIG_TIME_DS_HOLDER.get() != null)
          cfg.setProperty(Environment.CONNECTION_PROVIDER, CustomConnectionProvider.class.getName());
        return cfg;
      public void setDataSource(final DataSource ds) {
      public final void setHibernateProperties(final Properties hibernateProperties) {
        throw new UnsupportedOperationException();
      public final void setPersistenceProperties(final Properties persistenceOptions) {
        throw new UnsupportedOperationException();
      public void setProperties(final Properties props) {






    1. VERY IMPORTANT:The HibernateTransactionManager does NOT get the teneoSessionFactory(HbSessionDataStore) object passed in as sessionFactory, but the underlying Hibernate SessionFactory that can be retrieved from the HbSessionDataStore via the getSessionFactory() method. Not doing so will result in duplicate Hibernate Sessions completely breaking the transaction managment. This happens because in that case the TransactionSynchronizationManager will sometimes internally bind the hibernate session to the current thread based on different keys and thus does not find it again. Here are some stacktraces showing the problem:
      TransactionSynchronizationManager.bindResource(Object, Object) line: 170	
      at OpenSessionInViewFilter.doFilterInternal(HttpServletRequest, HttpServletResponse, FilterChain) line: 183
      => results in TransactionSynchronizationManager.bindResource(HbSessionDataStore, SessionHolder) => as key the Teneo SessionFactory is used
      TransactionSynchronizationManager.bindResource(Object, Object) line: 170	
      at SessionFactoryUtils.doGetSession(SessionFactory, Interceptor, SQLExceptionTranslator, boolean) line: 339	
      at SessionFactoryUtils.doGetSession(SessionFactory, boolean) line: 256	
      at SpringSessionContext.currentSession() line: 60	
      at SessionFactoryImpl.getCurrentSession() line: 700	
      at HbSessionDataStore(HbBaseSessionDataStore).getCurrentSession() line: 161	
      => results in TransactionSynchronizationManager.bindResource(SessionFactoryImpl, SessionHolder) => as key the Hibernate SessionFactory is used
      TransactionSynchronizationManager.bindResource(Object, Object) line: 170	
      at HibernateTransactionManager.doBegin(Object, TransactionDefinition) line: 577	
      at HibernateTransactionManager(AbstractPlatformTransactionManager).getTransaction(TransactionDefinition) line: 371
      at TransactionInterceptor(TransactionAspectSupport).createTransactionIfNecessary(PlatformTransactionManager, TransactionAttribute, String) line: 316	
      at TransactionInterceptor.invoke(MethodInvocation) line: 105	
      at ReflectiveMethodInvocation.proceed() line: 172
      => results in TransactionSynchronizationManager.bindResource(HbSessionDataStore, SessionHolder) => as key the Teneo SessionFactory is used
    2. defaultAutoCommit is set to false to ensure that Spring has full control over the transaction and that the connection pool is not committing statements automatically based on it’s own strategies.
    3. The example uses an extended version of the org.eclipse.emf.teneo.hibernate.HbSessionDataStore class that allows the usage of a DataSource object and that configures Teneo to not alter the table, column names but let’s Hibernate handle this via its ImprovedNamingStrategy.
    4. hibernate.transaction.auto_close_session is set to false to avoid “SessionException: Session is closed!” in conjunction with the OpenSessionInViewListener.
    5. hibernate.current_session_context_class and hibernate.transaction.factory_class are set to special Spring implementations to avoid “org.hibernate.HibernateException: No CurrentSessionContext configured!” when not using Spring Hibernate Template via HibernateDAOSupport. It is not recommended to use the Hibernate Template anymore, see
  • Using EMF ECore model objects with Wicket components

    (2 votes) 1 Star2 Stars3 Stars4 Stars5 Stars
    Posted on 8 September 2010 Sebastian Thomschke*/?> 1 comment

    Apache Wicket uses so called model objects to bind data objects to Wicket components. The framework provides a number of model implementations to access data objects and their properties in various ways. The PropertyModel implementation for example is used to access and set the value of a Java object using reflection.


    If you are working with EMF (Eclipse Modeling Framework) generated model classes you can also use PropertyModels to bind structural features (properties in EMF terminology) of an EObject to a Wicket component. E.g. if your EObject has an attribute called “comment” you could instantiate a model like this

    IModel model = new PropertyModel(myEObject, "comment");

    This approach has several disadvantages. First, the name of the attribute is declared as a String and not as a compile-time reference to the actual attribute. This means the application could break during runtime if you have renamed the attribute at one point and forgot to also change the String value. Second, this property model uses runtime reflection which is a slow alternative to compile time based access.
    Java classes generated from EMF Models by default extend the EObject class which provides a compile-time reflection API. Using the generic eGet and eSet methods any public property of the EObject can be accessed. The EMF Generator also generates a Package class for each EMF package that holds a Literals interface listing all available attributes and references of all generated EClasses of the given package. If you lets say have an EClass “Ticket” with the attribute “comment” in the package “com.acme.service” you could utilize the compile time reflection API as follows:

    EAttribute commentAttribute = com.acme.service.ServicePackage.Literals.TICKET__COMMENT;
    Ticket newTicket = com.acme.service.ServiceFactory.eINSTANCE.create(Ticket.class);
    newTicket.eSet(commentAttribute, "my comment");

    Using the EMF compile-time reflection API I created a generic EMF ECore Feature Model implementation that leverages the compile-time references to structural features. Most of the code is actually related to working around the fact that EAttributes and EReferences are not serializable but which they should to be part safely referenced in a model that potentially is serialized by Wicket between requests.

    import org.apache.wicket.model.IDetachable;
    import org.apache.wicket.model.IObjectClassAwareModel;
    import org.eclipse.emf.ecore.EClass;
    import org.eclipse.emf.ecore.EClassifier;
    import org.eclipse.emf.ecore.EObject;
    import org.eclipse.emf.ecore.EPackage;
    import org.eclipse.emf.ecore.EStructuralFeature;
     * A property model to bind an attribute of an EMF ECore object
     * @author Sebastian Thomschke
    public class EFeatureModel<T> implements IObjectClassAwareModel<T>
    	private static final long serialVersionUID = 1L;
    	private transient EStructuralFeature feat;
    	private String featContainerClassName;
    	private String featName;
    	private String featPackageNsURI;
    	private EObject target;
    	public EFeatureModel(EObject target, EStructuralFeature efeature)
    	{ = target;
    		this.feat = efeature;
    		this.featName = efeature.getName();
    		this.featContainerClassName = efeature.getContainerClass().getName();
    		this.featPackageNsURI = ((EPackage) efeature.eContainer().eContainer()).getNsURI();
    	public void detach()
    		if (target instanceof IDetachable) ((IDetachable) target).detach();
    	public EStructuralFeature getFeature()
    		if (feat == null)
    			EPackage ePackage = EPackage.Registry.INSTANCE.getEPackage(featPackageNsURI);
    			for (Object eClassifierItem : ePackage.getEClassifiers())
    				EClassifier eClassifier = (EClassifier) eClassifierItem;
    				if (eClassifier.getInstanceClass().getName().equals(featContainerClassName))
    					feat = ((EClass) eClassifier).getEStructuralFeature(featName);
    		return feat;
    	public T getObject()
    		return (T) target.eGet(getFeature());
    	public Class<T> getObjectClass()
    		return getFeature().getEType().getInstanceClass();
    	public void setObject(T object)
    		// do nothing if the object is the same instance that is returned by the getter
    		// this is esp. important for collection properties
    		if (target.eGet(getFeature()) == object) return;
    		target.eSet(getFeature(), object);
    	public String toString()
    		return new StringBuilder("Model:classname=[").append(getClass().getName()).append("]").append(":target=[")

    Using this new model implementation is as simple as using the PropertyModel provided by Wicket:

    IModel model = new EFeatureModel(myTicket, com.acme.service.ServicePackage.Literals.TICKET__COMMENT);

    You can now fully enjoy the compile-time safeness of your EMF model.

    Making EObjects serializable

    Wicket usually serializes components and their associated models between HTTP requests. If you are using the provided EPropertyModel your EObject classes must implement the Serializable interface. To achieve this, simply extend the EObject interface with the Serializables interface and set “Root Extends Interface” property in your genmodel configuration to this interface.

     * @author Sebastian Thomschke
    public interface SerializableEObject extends EObject, Serializable {

    When you now regenerate the model classes they will extend the SerializableEObject interface and Wicket will stop complaining about objects in the model being not serializable.

  • Installing Tomcat 6 on Debian Squeeze

    (10 votes) 1 Star2 Stars3 Stars4 Stars5 Stars
    Posted on 13 March 2010 Sebastian Thomschke*/?> 10 comments

    This post describes how to setup Tomcat 6 on Debian Squeeze. The configured Tomcat serves requests on port 80 without the need of an additional web server. This is especially good for virtual servers (VPS) providing limit memory. It also has multiple virtual hosts configured, each with it’s own webapp with context root / and optional support for PHP via the Quercus PHP implementation.

    Installing Sun Java 6

    Ensure the non-free section is enabled for the APT repository configuration in /etc/apt/sources.list, e.g. “deb testing main contrib non-free”

    apt-get update
    apt-get install sun-java6-jdk
    echo 'JAVA_HOME="/usr/lib/jvm/java-6-sun"' >> /etc/environment
    echo 'JRE_HOME="/usr/lib/jvm/java-6-sun/jre"' >> /etc/environment

    Installing Tomcat 6

    apt-get install tomcat6 tomcat6-admin
    /etc/init.d/tomcat6 stop

    Creating standard Tomcat directory layout (optional)

    mkdir /opt/tomcat
    cd /opt/tomcat
    ln -s /etc/tomcat6/ conf
    ln -s /usr/share/tomcat6/bin/ bin
    ln -s /usr/share/tomcat6/lib/ lib
    ln -s /var/lib/tomcat6/webapps webapps
    ln -s /var/log/tomcat6/ logs

    Creating a Tomcat admin user

    In /opt/tomcat/conf/tomcat-users.xml add an entry like:

    <user name="ADMIN_USERNAME" password="ADMIN_PASSWORD" roles="admin,manager" />

    Setting up virtual hosts

    For each virtual host execute the following command. Replace “” with the desired virtual host name, but omit the “www.” part.

    mkdir -p /opt/tomcat/

    In the <Engine> tag of “/opt/tomcat/conf/server.xml” add one host entry for each virtual host.

    <Host name="" appBase="/opt/tomcat/">
        <Valve className="org.apache.catalina.valves.AccessLogValve" prefix="mydomain_access_log." suffix=".txt" pattern="common"/>

    The <Alias> tag tells Tomcat to redirect from to
    The <Valve> tag enables access logging in the standard logging format.

    Using xinetd to configure port 80 for Tomcat

    Binding a service on port 80 requires root permissions. Thus we use port forwarding to “bind” Tomcat to port 80. My VPS does not support the use of “iptables -j REDIRECT” therefore I am using xinetd as a web proxy.
    Ensure that no other service is listening on port 80/443:

    netstat -pan | grep ":80\|:443"

    Register the required xinetd services:

    echo echo "
    service www
            socket_type     = stream
            protocol        = tcp
            user            = root
            wait            = no
            bind            =
            port            = 80
            redirect        = localhost 8080
            disable         = no
            flags           = REUSE
            log_type        = FILE /var/log/wwwaccess.log
            log_on_success  -= PID HOST DURATION EXIT
            per_source      = UNLIMITED
            instances       = UNLIMITED
    service https
            socket_type     = stream
            protocol        = tcp
            user            = root
            wait            = no
            bind            =
            port            = 443
            redirect        = localhost 8443
            disable         = no
            flags           = REUSE
            log_type        = FILE /var/log/httpsaccess.log
            log_on_success  -= PID HOST DURATION EXIT
            per_source      = UNLIMITED
            instances       = UNLIMITED
    " > /etc/xinetd.d/tomcat
    /etc/init.d/xinetd restart

    If you want to use a different service name, e.g. “tomcat” instead of “www” you must add this service to /var/services, e.g. “tomcat 80/tcp”
    In /opt/tomcat/conf/server.xml modify the <Connector> as follows:

    <Connector port="8080" protocol="HTTP/1.1"
                   redirectPort="8443" proxyPort="80" address="" />

    This binds Tomcat to localhost. It also tells Tomcat that port 80 is the proxy port which is necessary for correct URL generation.
    From now on the Tomcat admin applications are only accessible via localhost. You can use SSH port forwarding to still access the applications from your workstation’s web browser. E.g. if you are using PuTTY you can use this command line option “-L 8080:localhost:8080” to forward the server’s local 8080 port to your workstation’s local 8080 port. On your workstation’s browser you then simply enter http://localhost:8080/manager/html and are connected to the server’s Tomcat admin application.

    Enabling PHP support (optional)

    Download Quercus.

    mkdir -p /opt/downloads
    wget -o /opt/downloads/quercus-4.0.3.war

    Install Quercus as a shared library.

    unzip -j /opt/downloads/quercus-4.0.3.war \*.jar -d /opt/tomcat/lib

    Enable PHP support for the virtual hosts by installing the quercus web.xml. For each virtual host execute:

    unzip /opt/downloads/quercus-4.0.3.war *web.xml -d /opt/tomcat/

    Starting Tomcat

    /etc/init.d/tomcat start


  • Leveraging PyDev’s auto-completion for indirectly created objects

    (1 votes) 1 Star2 Stars3 Stars4 Stars5 Stars
    Posted on 20 February 2010 Sebastian Thomschke*/?> 5 comments

    This is just a quick tip how to enable auto completion in PyDev for indirectly created objects.

    By default, PyDev has no problems in presenting you the possible object attributes and methods on a variable if that variable gets a new object instance assigned directly in the code.

    For example, when you first type …

    mybook = Book()

    … and in the next line you enter …


    … PyDev will dutifully present you the statically declared features for the Book class in a popup.

    If you however get that Book instance as the result of another method, e.g. …

    mybook = bookstore.findBook("foobar")

    … PyDev currently seems to be helpless.

    Because of Python’s dynamic nature it is of course a bit harder for an IDE to figure out what kind of objects may a method. But PyDev could for example honor the @return / @rtype PyDoc annotations (which you can add to the findBook’s method declaration) but currently it just does not.

    To still have auto-completion you have two options:

    1. Temporarily instantiate a Book object in the code.
      mybook = Book()
      mybook = bookstore.findBook("foobar")
      mybook. #--> now you have auto completion

      This has the drawback, that if you forget to remove the first line before you finish coding an unneccessary Book instance will be created on every execution during runtime.

    2. Use an assertion to ensure that the mybook variable contains an object of type Book
      mybook = bookstore.findBook("foobar")
      assert isinstance(mybook, Book)
      mybook. #--> now you have auto completion
  • OVal 1.40 released

    1 Star2 Stars3 Stars4 Stars5 Stars
    Posted on 28 September 2009 Sebastian Thomschke*/?> No comments

    I am happy to announce the immediate availability of Version 1.40 OVal the Object Validation Framework for Java. This release fixes some minor issues and provides the following two new features:

    1. Enabling/disabling constraints based on the object state:
    Using the newly introduced when attribute for constraints it is possible to specify under which circumstances the constraint will be considered during object validation. The when attribute holds a formula in one of the supported scripting languages. If the formula returns true, the constraint will be activated for the current validation cycle otherwise it is ignored. In the following example the alias attribute must have a value only when the name attribute is not set.

    public class User {
       private String name;
       @NotNull(when = " == null")
       private String alias;
       // . . .

    2. Fine grain control over how constraints are applied on arrays, maps, collections and their elements
    The newly introduced attribute appliesTo allows you to specify if and how a constraint declared for a map, a collection or an array is applied to their elements as well or exclusively. See the following example how this works:

    public class BusinessObject {
       // the ids field may be null, but if it is not null 
       // none of it's items must be null
       @NotNull(appliesTo = { ConstraintTarget.VALUES })
       private String[] ids;
       // only the field itself must not be null,
       // the list can contain null values
       @NotNull(appliesTo = { ConstraintTarget.CONTAINER })
       private List items;
       // the field must not be null as well as all keys 
       // and values of the map
       @NotNull(appliesTo = { ConstraintTarget.CONTAINER, ConstraintTarget.KEYS, ConstraintTarget.VALUES })
       private Map itemsByGroup;
       // . . .

    You can get the latest binaries and source files of OVal from here:

  • Comparing version numbers in Jython / Python

    1 Star2 Stars3 Stars4 Stars5 Stars
    Posted on 14 March 2009 Sebastian Thomschke*/?> No comments

    Here comes a function to compare version numbers of e.g. Maven artifacts in Jython / Python.

    import re
    def cmpver(vA, vB):
        Compares two version number strings
        @param vA: first version string to compare
        @param vB: second version string to compare
        @author Sebastian Thomschke
        @return negative if vA < vB, zero if vA == vB, positive if vA > vB.
        >>> cmpver("0", "1")
        >>> cmpver("1", "0")
        >>> cmpver("1", "1")
        >>> cmpver("1.0", "1.0")
        >>> cmpver("1.0", "1")
        >>> cmpver("1", "1.0")
        >>> cmpver("1.1.0", "1.0.1")
        >>> cmpver("1.0.1", "1.1.1")
        >>> cmpver("0.3-SNAPSHOT", "0.3")
        >>> cmpver("0.3", "0.3-SNAPSHOT")
        >>> cmpver("1.3b", "1.3c")
        >>> cmpver("1.14b", "1.3c")
        if vA == vB: return 0
        def num(s): 
            if s.isdigit(): return int(s)
            return s
        seqA = map(num, re.findall('\d+|\w+', vA.replace('-SNAPSHOT', '')))
        seqB = map(num, re.findall('\d+|\w+', vB.replace('-SNAPSHOT', '')))
        # this is to ensure that 1.0 == 1.0.0 in cmp(..)
        lenA, lenB = len(seqA), len(seqB)
        for i in range(lenA, lenB): seqA += (0,) 
        for i in range(lenB, lenA): seqB += (0,)
        rc = cmp(seqA, seqB)
        if rc == 0:
            if vA.endswith('-SNAPSHOT'): return -1
            if vB.endswith('-SNAPSHOT'): return 1
        return rc

    Theoretically lines 43 – 49 could be written in a more compact way but using the short circuit evaluations (as below) lead to wrong results – at least in Jython 2.1:

        seqa = map(lambda s: s.isdigit() and int(s) or s, re.findall('\d+|\w+', vA.replace('-SNAPSHOT', '')))
        seqb = map(lambda s: s.isdigit() and int(s) or s, re.findall('\d+|\w+', vB.replace('-SNAPSHOT', '')))