Definition of the API for a transactional editing domain.

Package Specification

Creating an Editing Domain

The following snippet illustrates the creation of a transactional editing domain:

TransactionalEditingDomain domain = TransactionalEditingDomain.Factory.INSTANCE.createEditingDomain();
ResourceSet rset = domain.getResourceSet();

// or, create our own resource set and initialize the domain with it
rset = new MyResourceSetImpl();
domain = TransactionalEditingDomain.Factory.INSTANCE.createEditingDomain(rset);

To share a named editing domain with other applications, the editing domain registry can be used to obtain domains by ID, creating them if necessary. Editing domain IDs are configured on an extension point providing the factory implementation that the registry uses to initialize them:

<!-- In the plugin.xml -->
<extension point="org.eclipse.emf.transaction.editingDomains">
    <editingDomain
        id="com.example.MyEditingDomain"
        factory="com.example.MyEditingDomainFactory"/>
</extension>

// in code, access the registered editing domain by:

TransactionalEditingDomain myDomain = TransactionalEditingDomain.Registry.INSTANCE.getEditingDomain(
    "com.example.MyEditingDomain");

Reading a Resource Set

All work in an editing domain is done within the context of a transaction. This applies to both reading the model and writing it, as well as creating, loading, saving, and unloading resources. To load and read a resource:

Runnable read = new RunnableWithResult.Impl() {
        public void run() {
            Resource res = rset.getResource(
                URI.createFileURI("/tmp/my.xmi"),
                true);

            setResult(new Integer(res.getContents().size()));
        }
    };

// execute the read operation as a read-only transaction
int size = ((Integer) domain.runExclusive(read)).intValue();

Modifying a Resource Set

Modifications to a resource require read-write transactions, implemented as commands:

Command cmd = new RecordingCommand("Create Library") {
        protected void doExecute() {
            Library library = LibraryFactory.eINSTANCE.createLibrary();

            // these modifications require a write transaction in this editing domain
            res.getContents().add(library);
            library.setName("Main Branch");
        }
    };

try {
    ((TransactionalCommandStack) domain.getCommandStack()).execute(cmd, null); // default options
} catch (RollbackException rbe) {
    ErrorDialog.openError(
        getWorkbench().getShell(), "Transaction Rollback",
        "Transaction rolled back due to validation errors.", rbe.getStatus());
}

Listening for Model Changes

Listeners can be added to the editing domain to find out about changes that occur (directly and indirectly) in the resource set and its contents. There are two call-backs supported by the editing domain: transactionAboutToCommit (pre-commit) calls notify the listener of what changes a command has performed before it completes (commits) them, and resourceSetChanged (post-commit) indicates what changes a command has committed after it has completed normally (or what changes occurred during reading the resource set).

ResourceSetListener listener = new ResourceSetListenerImpl() {
        protected void resourceSetChanged(ResourceSetChangeEvent event) {
            // analyze the complete list of notifications that occurred during
            //    execution of the command to figure out how the model has changed
            analyzeChanges(event.getNotifications());
        }
    });
    
// attach the listener to the editing domain
domain.addResourceSetListener(listener);

The resource set listener can make changes to the resource set, but these are chained onto the current command and are not executed until after all other notifications have been processed:

// notify only on additions to the contents of a resource
NotificationFilter filter = NotificationFilter.createFeatureFilter(
        EcorePackage.eINSTANCE.getEResource(),
        Resource.RESOURCE__CONTENTS).and(
    NotifierFilter.createEventTypeFilter(
            Notification.ADD, Notification.ADD_MANY)
));

// create a new listener with this filter
ResourceSetListener listener = new ResourceSetListenerImpl(filter) {
        protected Command aboutToCompleteCommand(ResourceSetChangeEvent event) {
            // create a command to follow-up the changes that we observed
            Command cmd = analyzeChanges(event.getNotifications());
            
            // the command will be executed by the command stack
            return cmd;
        }
    });
    
// attach the listener to the editing domain
domain.addResourceSetListener(listener);

Notice two things about the previous example: first, the listener's pre-commit call-back method returns a command for future execution. The stack executes this command after completing the round of call-backs. It cannot simply "piggy-back" on the current command's transaction because it is closing. Secondly, it declares a filter to restrict the notifications that it will receive. The filter is a composite of two filters, stipulating that only ADD or ADD_MANY notifications are wanted, and these only from the contents lists of resources. If when a command completes, no notifications match the filter, then this listener is not invoked at all. Otherwise, only the notifications that do match are provided to the listener.

Listeners can also force rollback of the currently executing command if it must not be permitted to commit its changes. This is accomplished by throwing a RollbackException:

ResourceSetListener listener = new ResourceSetListenerImpl() {
        protected Command transactionAboutToCommit(ResourceSetChangeEvent) throws RollbackException {
            // validate the changes and force rollback if not acceptable
            if (!analyzeChanges(getNotifications())) {
                throw new RollbackException("Didn't like the changes.");
            }
            
            return null;  // no need to perform further changes
        }
    });

ResourceSetListeners can also be declared on an extension point, to be loaded on creation of an editing domain. This approach has the advantage of automatically attaching the listener to an editing domain before the client has been activated. The extension can specify one or more unique identifiers of editing domains to listen to; these are mapped to editing domain instances in the shared TransactionalEditingDomain.Registry. The listener class implementing this extension should initialize itself with an appropriate filter.

<extension point="org.eclipse.emf.transaction.listeners">
    <listener class="com.example.MyResourceSetListener">
        <editingDomain id="com.example.MyEditingDomain"/>
    </listener>
</extension>    

Undoing Model Changes

Transactions (commands) are undone via the command stack, as usual:

Command cmd = SomeCommand();

try {
    ((TransactionalCommandStack) domain.getCommandStack()).execute(cmd, null); // default options
} catch (RollbackException rbe) {
    ErrorDialog.openError(
        getWorkbench().getShell(), "Transaction Rollback",
        "Transaction rolled back due to validation errors.", rbe.getStatus());
}

// undo the last executed transaction
domain.getCommandStack().undo();

// redo it
domain.getCommandStack().redo();
@see org.eclipse.emf.transaction.TransactionalEditingDomain @see org.eclipse.emf.transaction.TransactionalCommandStack @see org.eclipse.emf.transaction.RecordingCommand @see org.eclipse.emf.transaction.RunnableWithResult @see org.eclipse.emf.transaction.ResourceSetListener