Roundtrip engineering with Hibernate is possible using a set of commandline tools maintained as part of the Hibernate project, along with Hibernate support built into XDoclet, Middlegen and AndroMDA. The Hibernate main package comes bundled with the most important tool (it can even be used from "inside" Hibernate on-the-fly): DDL schema generation from a mapping file (aka SchemaExport, hbm2ddl) Other tools directly provided by the Hibernate project are delivered with a separate package, Hibernate Extensions. This package includes tools for the following tasks: Java source generation from a mapping file (aka CodeGenerator, hbm2java) mapping file generation from compiled Java classes or from Java source with XDoclet markup (aka MapGenerator, class2hbm) There's actually another utitily living in Hibernate Extensions: ddl2hbm. It is considered deprecated and will no longer be maintained, Middlegen does a better job for the same task. Third party tools with Hibernate support are: Middlegen (mapping file generation from an existing database schema) AndroMDA (MDA (Model-Driven Architecture) approach generating code for persistent classes from UML diagrams and their XML/XMI representation) These 3rd party tools are not documented in this reference. Please refer to the Hibernate website for up-to-date information (a snapshot of the site is included in the Hibernate main package). DDL may be generated from your mapping files by a command line utility. A batch file is located in the hibernate-x.x.x/bin directory of the core Hibernate package. The generated schema include referential integrity constraints (primary and foreign keys) for entity and collection tables. Tables and sequences are also created for mapped identifier generators. You must specify a SQL Dialect via the hibernate.dialect property when using this tool. Many Hibernate mapping elements define an optional attribute named length. You may set the length of a column with this attribute. (Or, for numeric/decimal data types, the precision.) Some tags also accept a not-null attribute (for generating a NOT NULL constraint on table columns) and a unique attribute (for generating UNIQUE constraint on table columns). Some tags accept an index attribute for specifying the name of an index for that column. A unique-key attribute can be used to group columns in a single unit key constraint. Currently, the specified value of the unique-key attribute is not used to name the constraint, only to group the columns in the mapping file. Examples: ]]> Alternatively, these elements also accept a child <column> element. This is particularly useful for multi-column types: ]]> The sql-type attribute allows the user to override the default mapping of Hibernate type to SQL datatype. The check attribute allows you to specify a check constraint. ... ]]> Attribute Values Interpretation length number column length/decimal precision not-null true|false specfies that the column should be non-nullable unique true|false specifies that the column should have a unique constraint index index_name specifies the name of a (multi-column) index unique-key unique_key_name specifies the name of a multi-column unique constraint foreign-key foreign_key_name specifies the name of the foreign key constraint generated for an association sql-type column_type overrides the default column type (attribute of <column> element only) check SQL expression create an SQL check constraint on either column or table The SchemaExport tool writes a DDL script to standard out and/or executes the DDL statements. java -cp hibernate_classpaths net.sf.hibernate.tool.hbm2ddl.SchemaExport options mapping_files Option Description --quiet don't output the script to stdout --drop only drop the tables --text don't export to the database --output=my_schema.ddl output the ddl script to a file --config=hibernate.cfg.xml read Hibernate configuration from an XML file --properties=hibernate.properties read database properties from a file --format format the generated SQL nicely in the script --delimiter=x set an end of line delimiter for the script You may even embed SchemaExport in your application: Database properties may be specified as system properties with -D<property> in hibernate.properties in a named properties file with --properties The needed properties are: Property Name Description hibernate.connection.driver_class jdbc driver class hibernate.connection.url jdbc url hibernate.connection.username database user hibernate.connection.password user password hibernate.dialect dialect You can call SchemaExport from your Ant build script: ]]> The SchemaUpdate tool will update an existing schema with "incremental" changes. Note that SchemaUpdate depends heavily upon the JDBC metadata API, so it will not work with all JDBC drivers. java -cp hibernate_classpaths net.sf.hibernate.tool.hbm2ddl.SchemaUpdate options mapping_files Option Description --quiet don't output the script to stdout --properties=hibernate.properties read database properties from a file You may embed SchemaUpdate in your application: You can call SchemaUpdate from the Ant script: ]]> The Hibernate code generator may be used to generate skeletal Java implementation classes from a Hibernate mapping file. This tool is included in the Hibernate Extensions package (a seperate download). hbm2java parses the mapping files and generates fully working Java source files from these. Thus with hbm2java one could "just" provide the .hbm files, and then don't worry about hand-writing/coding the Java files. java -cp hibernate_classpaths net.sf.hibernate.tool.hbm2java.CodeGenerator options mapping_files Option Description --output=output_dir root directory for generated code --config=config_file optional file for configuring hbm2java The config file provides for a way to specify multiple "renderers" for the source code and to declare <meta> attributes that is "global" in scope. See more about this in the <meta> attribute section. codegen.test.IAuditable ]]> This config file declares a global meta attribute "implements" and specify two renderers, the default one (BasicRenderer) and a renderer that generates Finder's (See more in "Basic Finder generation" below). The second renderer is provided with a package and suffix attribute. The package attribute specifies that the generated source files from this renderer should be placed here instead of the package scope specified in the .hbm files. The suffix attribute specifies the suffix for generated files. E.g. here a file named Foo.java would be FooFinder.java instead. It is also possible to send down arbitrary parameters to the renders by adding <param> attributes to the <generate> elements. hbm2java currently has support for one such parameter, namely generate-concrete-empty-classes which informs the BasicRenderer to only generate empty concrete classes that extends a base class for all your classes. The following config.xml example illustrate this feature true Base ]]> Notice that this config.xml configure 2 (two) renderers. One that generates the Base classes, and a second one that just generates empty concrete classes. The <meta> tag is a simple way of annotating the hbm.xml with information, so tools have a natural place to store/read information that is not directly related to the Hibernate core. You can use the <meta> tag to tell hbm2java to only generate "protected" setters, have classes always implement a certain set of interfaces or even have them extend a certain base class and even more. The following example: Javadoc for the Person class @author Frodo IAuditable protected The name of the person ]]> will produce something like the following (code shortened for better understanding). Notice the Javadoc comment and the protected set methods: Attribute Description class-description inserted into the javadoc for classes field-description inserted into the javadoc for fields/properties interface If true an interface is generated instead of an class. implements interface the class should implement extends class the class should extend (ignored for subclasses) generated-class overrule the name of the actual class generated scope-class scope for class scope-set scope for setter method scope-get scope for getter method scope-field scope for actual field use-in-tostring include this property in the toString() implement-equals include a equals() and hashCode() method in this class. use-in-equals include this property in the equals() and hashCode() method. bound add propertyChangeListener support for a property constrained bound + vetoChangeListener support for a property gen-property property will not be generated if false (use with care) property-type Overrides the default type of property. Use this with any tag's to specify the concrete type instead of just Object. class-code Extra code that will inserted at the end of the class extra-import Extra import that will inserted at the end of all other imports finder-method see "Basic finder generator" below session-method see "Basic finder generator" below Attributes declared via the <meta> tag are per default "inherited" inside an hbm.xml file. What does that mean? It means that if you e.g want to have all your classes implement IAuditable then you just add an <meta attribute="implements">IAuditable</meta> in the top of the hbm.xml file, just after <hibernate-mapping>. Now all classes defined in that hbm.xml file will implement IAuditable! (Except if a class also has an "implements" meta attribute, because local specified meta tags always overrules/replaces any inherited meta tags). Note: This applies to all <meta>-tags. Thus it can also e.g. be used to specify that all fields should be declare protected, instead of the default private. This is done by adding <meta attribute="scope-field">protected</meta> at e.g. just under the <class> tag and all fields of that class will be protected. To avoid having a <meta>-tag inherited then you can simply specify inherit="false" for the attribute, e.g. <meta attribute="scope-class" inherit="false">public abstract</meta> will restrict the "class-scope" to the current class, not the subclasses. It is now possible to have hbm2java generate basic finders for Hibernate properties. This requires two things in the hbm.xml files. The first is an indication of which fields you want to generate finders for. You indicate that with a meta block inside a property tag such as: findByName ]]> The finder method name will be the text enclosed in the meta tags. The second is to create a config file for hbm2java of the format: ]]> And then use the param to hbm2java --config=xxx.xml where xxx.xml is the config file you just created. An optional parameter is meta tag at the class level of the format: com.whatever.SessionTable.getSessionTable().getSession(); ]]> Which would be the way in which you get sessions if you use the Thread Local Session pattern (documented in the Design Patterns area of the Hibernate website). It is now possible to use velocity as an alternative rendering mechanism. The follwing config.xml shows how to configure hbm2java to use its velocity renderer. pojo.vm ]]> The parameter named template is a resource path to the velocity macro file you want to use. This file must be available via the classpath for hbm2java. Thus remember to add the directory where pojo.vm is located to your ant task or shell script. (The default location is ./tools/src/velocity) Be aware that the current pojo.vm generates only the most basic parts of the java beans. It is not as complete and feature rich as the default renderer - primarily a lot of the meta tags are not supported. A skeletal mapping file may be generated from compiled persistent classes using a command line utility called MapGenerator. This utility is part of the Hibernate Extensions package. The Hibernate mapping generator provides a mechanism to produce mappings from compiled classes. It uses Java reflection to find properties and uses heuristics to guess an appropriate mapping from the property type. The generated mapping is intended to be a starting point only. There is no way to produce a full Hibernate mapping without extra input from the user. However, the tool does take away some of the repetitive "grunt" work involved in producing a mapping. Classes are added to the mapping one at a time. The tool will reject classes that it judges are are not Hibernate persistable. To be Hibernate persistable a class must not be a primitive type must not be an array must not be an interface must not be a nested class must have a default (zero argument) constructor. Note that interfaces and nested classes actually are persistable by Hibernate, but this would not usually be intended by the user. MapGenerator will climb the superclass chain of all added classes attempting to add as many Hibernate persistable superclasses as possible to the same database table. The search stops as soon as a property is found that has a name appearing on a list of candidate UID names. The default list of candidate UID property names is: uid, UID, id, ID, key, KEY, pk, PK. Properties are discovered when there are two methods in the class, a setter and a getter, where the type of the setter's single argument is the same as the return type of the zero argument getter, and the setter returns void. Furthermore, the setter's name must start with the string set and either the getter's name starts with get or the getter's name starts with is and the type of the property is boolean. In either case, the remainder of their names must match. This matching portion is the name of the property, except that the initial character of the property name is made lower case if the second letter is lower case. The rules for determining the database type of each property are as follows: If the Java type is Hibernate.basic(), then the property is a simple column of that type. For hibernate.type.Type custom types and PersistentEnum a simple column is used as well. If the property type is an array, then a Hibernate array is used, and MapGenerator attempts to reflect on the array element type. If the property has type java.util.List, java.util.Map, or java.util.Set, then the corresponding Hibernate types are used, but MapGenerator cannot further process the insides of these types. If the property's type is any other class, MapGenerator defers the decision on the database representation until all classes have been processed. At this point, if the class was discovered through the superclass search described above, then the property is an many-to-one association. If the class has any properties, then it is a component. Otherwise it is serializable, or not persistable. The tool writes XML mappings to standard out and/or to a file. When invoking the tool you must place your compiled classes on the classpath. java -cp hibernate_and_your_class_classpaths net.sf.hibernate.tool.class2hbm.MapGenerator options and classnames There are two modes of operation: command line or interactive. The interactive mode is selected by providing the single command line argument --interact. This mode provides a prompt response console. Using it you can set the UID property name for each class using the uid=XXX command where XXX is the UID property name. Other command alternatives are simply a fully qualified class name, or the command done which emits the XML and terminates. In command line mode the arguments are the options below interspersed with fully qualified class names of the classes to be processed. Most of the options are meant to be used multiple times; each use affects subsequently added classes. Option Description --quiet don't output the O-R Mapping to stdout --setUID=uid set the list of candidate UIDs to the singleton uid --addUID=uid add uid to the front of the list of candidate UIDs --select=mode mode use select mode mode(e.g., distinct or all) for subsequently added classes --depth=<small-int> limit the depth of component data recursion for subsequently added classes --output=my_mapping.xml output the O-R Mapping to a file full.class.Name add the class to the mapping --abstract=full.class.Name see below The abstract switch directs the map generator tool to ignore specific super classes so that classes with common inheritance are not mapped to one large table. For instance, consider these class hierarchies: Animal-->Mammal-->Human Animal-->Mammal-->Marsupial-->Kangaroo If the --abstractswitch is not used, all classes will be mapped as subclasses of Animal, resulting in one large table containing all the properties of all the classes plus a discriminator column to indicate which subclass is actually stored. If Mammal is marked as abstract, Human and Marsupial will be mapped to separate <class> declarations and stored in separate tables. Kangaroo will still be a subclass of Marsupial unless Marsupial is also marked as abstract.