GWT - MVP

I think the overall idea of activities is that they are short-lived instances, so they're effectively "lazily created" (actually, a new instance is created each time one is needed) and they don't need to be "awoken" because if they're not currently in use they're already "dead" and garbage collected. The ActivityManager (actually its associated ActivityMapper) will decide whether a particular activity is needed (and then instantiate it, possibly going through a GWT.runAsync for code splitting); the activity will listen to events its interested in *during its lifetime* (e.g. whether some object has changed or has been added or deleted, so it can update its view); but when it's done (stopped or cancelled), it's simply thrown away (the event bus passed to the start() method is a ResettableEventBus so all handlers have been automatically unregistered for you, which as a side effect allows the activity to be garbage collected). This is the (AIUI) intended use, but nothing forces you to write such short-lived instances: you can very well use singletons, but then you'll have the additional task of maintaining state between "runs" (start/stop or start/cancel), in which case your activity can listen to events from the event bus after being stopped/cancelled (just use the "real" event bus instead of the ResettableEventBus passed to the start() method); but it won't "ask to be revealed": navigation is handled at another layer, triggered on the PlaceController and handled by ActivityManagers.

I think the overall idea of activities is that they are short-lived instances, so they're effectively "lazily created" (actually, a new instance is created each time one is needed) and they don't need to be "awoken" because if they're not currently in use they're already "dead" and garbage collected. The ActivityManager (actually its associated ActivityMapper) will decide whether a particular activity is needed (and then instantiate it, possibly going through a GWT.runAsync for code splitting); the activity will listen to events its interested in *during its lifetime* (e.g. whether some object has changed or has been added or deleted, so it can update its view); but when it's done (stopped or cancelled), it's simply thrown away (the event bus passed to the start() method is a ResettableEventBus so all handlers have been automatically unregistered for you, which as a side effect allows the activity to be garbage collected). This is the (AIUI) intended use, but nothing forces you to write such short-lived instances: you can very well use singletons, but then you'll have the additional task of maintaining state between "runs" (start/stop or start/cancel), in which case your activity can listen to events from the event bus after being stopped/cancelled (just use the "real" event bus instead of the ResettableEventBus passed to the start() method); but it won't "ask to be revealed": navigation is handled at another layer, triggered on the PlaceController and handled by ActivityManagers.

> is so called REST like URLs possible ? /employees/1 This is no more or less "REST like" than the above (I assure you!); and yes it's possible (the issues then are to identify objects that are not yet persisted to the server; and of course mapping the "employees" to an EntityProxy class, representing the class+id couple in the Place, and having dedicated "find" methods to retrieve the object from the server, as without an EntityProxyId you won't be able to use RequestFactory.find()) > does it mean we have to write custom getPlace, getToken, to convert > tokens to places ? Yes. That or writing your own PlaceHistoryMapper that won't use PlaceTokenizer's at all (i.e. implement PlaceHistoryMapper in a concrete class and not use the GWT.create() magic to generate the implementation).

there's absolutely nothing in GWT proper related to MVP actually; I refuse to call Activities an "MVP framework", it has nothing to do with MVP in my opinion

try to disconnect activities from MVP: Activities don't force you to do MVP, and you'll use MVP beyond Activities.

you can separate your project into two modules: A module containing model and persistence code that is created by Spring Roo A GWT web application that you create by hand. Make the first (Roo) module a dependency of the second (GWT) module. Basically you're using Roo to create a JAR library that's used by your web application. As long as you don't run the controller command the Roo won't add any web application code to your module.

The MVP Framework is an app framework that makes it easy for you to connect Data Presentation Widgets with backend data. Using this framework you create views that are focused on displaying data, Activities and an ActivityManager which are the "presenters", responsible for handling self-contained actions, and RequestFactories that fetch and propagate model changes throughout your app. To make developing apps of this style easier, the 1.1 M1 release of Spring Roo, can generate and maintain the boilerplate code associated with connecting your app's components with GWT's MVP Framework.

've started prototyping with Activities and can now give some more feedback on how it compares to my previous own implementation of a PlaceManager and how I used it with MVP.

your ChatBoxPresenter isn't related to a "place", i.e. it's "awoken" based on a "business event", not a navigation event; and in other words, it's not an Activity: case made. Activities (ActivityManager) is limited in scope to reacting to place changes and navigation (not that you couldn't use the Activity "contract" in other scenarios, such as your "chat box", but the ActivityManager wouldn't be the right tool for the job, you'd have to find/write another "activity manager" for your different use case). In other words: GWT 2.1 Activities won't replace GWTP as a whole (and I believe, as I already said in the past, that it's not its goal either).

Start moving the code of addStock to the view. Try not to use widget code on the presenter. Then create an interface and implement in the presenter: class SomeViewHandlers {     void deleteStock(Stock code); } You already have the deleteStock method... so just add "implements SomeViewHandlers" to it. Then you need to give to the display that interface (generally in the constructor): display.setViewHandlers(this); Then in the presenter lease the method as: private void addStock(JsArray<Stock> stocks) { display.addStock(stocks); } And in the view copyPaste the code of the method and change this:  removeButton.addClickHandler(new ClickHandler() {        @Override        public void onClick(ClickEvent event) {          *viewHandlers.*deleteStock(code);        }      });

I believe the reason we've not created View and Presenter interfaces to date is because there are two different styles of MVP widely in use, only one of which allows the view to call the presenter as in your example. Which leaves us in the funny position that the new MVP framework is missing *formal* definitions of View and Presenter. Personally, I think it's a good thing that GWT Activities and Places are independent of views and presenters so as not to force you into one model. I think View and Presenter as you've described would fall in the category of things that are not quite core code, but are nevertheless useful abstractions that probably need a home in the GWT source somewhere for reference. We'll chew on this a bit for 2.1.1...

Clients of ValueStore subscribe to particular Properties of particular Ids, and values and validation errors are pushed into them asynchronously as they become available. Typically the code to set up these subscriptions is generated around UiBinder templates.

In a nutshell: RequestFactory is a scheme for referring to server side entities via Id and Property objects. It relies upon ValueStore, which is a more general system for dealing with data binding and client side validation in general.

The way we're minimizing code right now, is by creating some useful abstract tokenizers that handle the common use cases (no parameter tokenizer, key-value pair tokenizer). With those there are corresponding abstract tokenizers that handle creating the hashCode and equals methods. Aside from the 1:1 between Place and Activity, which bothers me in principle but may not be a practical concern just yet, it's coming along pretty nicely.

MVC implies that the server handles much of the view - with a GWT application the server is reduced to a glorified DAO. All you really need is integrate services from the backend into the frontend. What we really missing is something like RMI for the browser - GWT's RPC is as good as it gests for this purpose and and libraries like the GWT-SL allow you to directly publish Spring managed POJOs as services over RPC into the client.

Yeah, the one problem with UiBinder and MVP is this pattern collision.... UiBinder says, "view are directly attached to views". MVP says, "view's are attached to presenters, if you want to chain views the presenters control this". Consequently, you can't get UiBinder to create your @UiFields (i.e. empty constructor) and you can't get Gin to @Inject them into the view either... because they are in the presenter. Unless of course you use @Named+Singleton bindings in Gin and then both presenter and view will be injected with the same object. This is the one bit of boilerplate we're still writing.

I looked into that, and (unless I'm wrong), I think that @UiField(provided=true) will cause the UiBinder to look in the .ui.xml file for argument to satisfy Foo. In my case I am trying to "inject" an EventBus into the widget, not a visual element.

use the history and pass the information as parameters in the history token.

has a number of advantages provided the parameter is serializable (and small enough). For example, it could let the user bookmark the detail page of "Lady_Gaga". A bit trickier when the presenter is a dialog box, but totally doable.

It however clarify things about how Google sees MVP in GWT. They're even adding some MVP "framework" to GWT 2.1 (IsWidget and Activities, where an Activity is more or less your presenter, from what I understood).

RequestFactory on the other hand is a new thing for efficient CRUD operations on entity objects, which plugs more-or-less directly into the new Data widgets. Those widgets do blur the line a bit, but they're not really about presentation "logic", and they're somehow MVP- based themselves.

if your concern is testing your presenter, then how about changing your MVP from the "Passive View" pattern (the one showed by Ray Ryan at I/0 2009) to the "Supervising Controller" pattern (the one used by Cell widgets internally and by applications generated by Spring Roo): http://martinfowler.com/eaaDev/uiArchs.html Sure you test a bit less things as you leave more of the logic into the view than with a "Passive View", but in my experience it also makes tests much more easier to write and more readable, as you need far less mocks and stubs (no MockButton, MockTextBox, etc. only a MockView).

What is the best way to implement that ? 1) only one couple of presenter/view for object1 and this couple manages the display of object2 in each tab 2) one presenter/view object for object1 and one presenter/view for each instance of object2 ? in other words, do I have one couple of presenter/view by object model ?