[xwiki-devs] [XEclipse] Design improvements discussion

Dear all, I am starting to take a deeper look at XEclipse and I think there are some improvements at the design level that, IMHO, are worth to be taken into account. These improvements leverage Eclipse platform's features that, to some extent, have been re-implemented in XEclipse. 1) Deferred tree content managers. I am a big fan of this feature and I tend to use it whenever I can. Basically a deferred tree content manager allows you to "fetch" tree children in a deferred way. One example is the "SVN Repository" or "CVS Repository" views in your Eclipse IDE. When you expand a node it displays a "Pending..." element and, when the fetching job is over, the actual elements (i.e., files and directories available on the repository) are displayed. The same mechanism could be used in XEclipse to fetch Spaces and Pages. The good thing about this approach is that the manager takes care of everything: * A tree node that is collapsed before the fetching job is terminated deletes the corresponding job; * Changing the input element of the tree invalidates all the active jobs * You can expand different tree nodes without waiting that the previous ones terminate their fetching job: in this case you will have several "Pending..." indicators in the tree. You can play a bit with the "SVN Repository" view and see what I mean. The best news is that all of this already implemented and available in the eclipse platform through the DeferredTreeContentManager[1] class. Using the deferred tree content fetching mechanism eliminates the need of having additional logic in XWikiSpace for understanding when pages are ready (i.e., all the code that manages isPagesReady in XWikiSpace.java). You will simply have a method getPages that could take possibly forever and that directly fetches the pages from the XMLRPC source (or from the offline cache). I think that this will also have impact on the Decorators that are used to wrap XWiki* elements in order to provide a visual feedback on what's going on (i.e., all the GuiUtils.runOperationWithProgress will be unnecessary since the deferred tree content manager will take care of everything) 2) Adapters I noticed that there is a TreeAdapter interface that basically is used for providing labels, children and so on. All classes that can be "displayed" in a tree implements this interface (i.e., XWikiConnection, XWikiPage, XWikiSpace). I think that this "pollutes" the model because, for example, getPages and getTreeChildren basically return the same thing and, therefore, are duplicated methods. Moreover in this way you put "GUI"-oriented stuff in the model and this is not good. The Eclipse platform has a mechanism for providing adapters that are able to give information about model elements to GUI components[2]. Not surprisingly they are called Adapters and are handled via extension points. Basically you register an AdapterFactory that provides IWorkbenchAdapters. An IWorkbenchAdapter[3] actually contains what was declared in XEclipse TreeAdapter. When you need structural information about an object, you can then query the platform for an adapter that matches a given interface and is suitable for a given object. In our case I can register an adapter for XWikiSpace that implements the getChildren method and that will simply call the getPages method of the XWikiSpace model object. In this way I can clearly separate the "GUI" code from the "model" code and leave Eclipse handle the rest. It is worth to notice that many GUI content providers directly support IWorkbenchAdapters out of the box, so if you carefully write your model adapters you will never have to write a content/label provider; but even when you write your own content provider you can always put a lookup to the adapter and then talk to it for getting the information to display (this is actually a standard practice). In the attachment you can find a sample implementation of the "deferred tree" way of displaying the XWiki navigator that makes use of adapters. Notice the getChildren methods in XWikiSpaceAdapter (declared in AdapterFactory.java). It calls a method myGetPages that I've added for testing purposes to the IXWikiSpace interface and that basically does the following straight simple thing: public Collection<IXWikiPage> myGetPages() throws ConfluenceException, SwizzleConfluenceException { Confluence rpc = getConnection().getRpcProxy(); String spaceKey = getKey(); List<Object> pages = rpc.getPages(spaceKey); List<IXWikiPage> result = new ArrayList<IXWikiPage>(); for (int i = 0; i < pages.size(); i++) { PageSummary pageSummary = (PageSummary) pages.get(i); XWikiPage xWikiPage = new XWikiPage(this, pageSummary); result.add(xWikiPage); } return result; } Of course this is a proposal to discuss about. But I think that we should spend some time on it because we could end up with a lighter design and a more maintainable product. WDYT? Cheers, Fabio [1] http://help.eclipse.org/help33/index.jsp?topic=/ org.eclipse.platform.doc.isv/reference/api/org/eclipse/ui/progress/ DeferredTreeContentManager.html [2] Of course the adapters mechanism is not limited to GUI components but can be used to adapt an object to whatever interface we need to. [3] http://help.eclipse.org/help33/index.jsp?topic=/ org.eclipse.platform.doc.isv/reference/api/org/eclipse/ui/model/ IWorkbenchAdapter.html These are the XML lines in plugin.xml that defines the org.eclipse.core.runtime.adapters extension point for registering the AdapterFactory and types: <extension point="org.eclipse.core.runtime.adapters"> <factory adaptableType="org.xwiki.plugins.eclipse.model.IXWikiConnection" class="fm.AdapterFactory"> <adapter type="org.eclipse.ui.progress.IDeferredWorkbenchAdapter"></adapter> </factory> <factory adaptableType="org.xwiki.plugins.eclipse.model.IXWikiSpace" class="fm.AdapterFactory"> <adapter type="org.eclipse.ui.progress.IDeferredWorkbenchAdapter"></adapter> </factory> </extension>