Archiv der Kategorie: GWT

Non invasive GWT and Spring integration

If you are working with GWT 1.6, you probably would like to have a look at this here:

Obviously I am not the only one looking for a way to integrate my Spring backend into some GWiT application. After searching for a while I didn’t find any suiting solution. There are some interesting approaches (like GWT Widget Library SL and using the maven plugin), but being new to GWiT I did not want to give up the GWiT Development Shell neither the embeded Tomcat. I wanted the integration to be less invasive as possible.

So here is what I did…

First problem I had to solve was how to get my configuration elements into the tomcat configuration. I needed to add some listeners to the web.xml (e.g. to start the Spring container and Acegi security):

Here is a snippet from my web.xml showing how to use the listeners provided by the Spring Framework to start an application context and to configure log4j properly:

GWiT does not provide any extension point for custom configuration, so I had to add the configuration elements directly to the provided Tomcat web.xml. I cannot make changes to the web.xml in the ROOT web application, since this web.xml is re-generated by GWiT every time the Development Shell is started. Fortunately, the default web.xml stored in the “conf” directory of the embedded Tomcat is generated only one. So I found some place to add my configuration elements. Unfortunately the configuration there is not reusable, so I am having double configuration here: configuration for the development and configuration for the deployment.

Next, I wanted to easily have access to my Spring beans. JSF developers have a variable resolver and have access to their Spring beans for free. I wanted that too. Again, being new to GWiT, I didn’t want to loose the features provided by the IDE (Eclipse + GWT-Designer in this case). It is nice to simply say “add new remote service” in the IDE and get everything wired out of the box. For different reasons I do want to expose my Spring beans automatically to the GWiT application. JSF developers have the services for free at the SERVER SIDE: the services do not get exposed via RPC automatically. I wanted something similar: my GWiT remote service implementation should have access to the Spring backend for free without exposing anything automatically.

I wanted some sort of dependency injection. Since the servlets are managed by the servlet container and the servlet container does not know anything about dependeny injection of Spring beans into servlets I had to do it myself.

First, I needed some sort of markup to identify what to inject. I was using JDK 5 syntax on none client sources, so I used an annotation. I could also have used some marker interface for my services, but I found this to be to invasive, I did not touch my backend files.

All I wanted to do is to add this annotation to my setter methods in my GWiT remote service implementations. So I extended the RemoteServiceServlet provided by GWiT and implemented the auto injection.

In short: in the initialization of the servlet I pickup the Spring application context and do the injection for all setter methods annotated.

In the following GWiT remote service implementation I use this injection to get access to my Spring login service (that itself uses acegi, the user DAO and a session scoped bean):

While this leads to a lot of delegation code, I am still happy with the layer separation and the easy usage.

When playing with GWiT 1.5 (build from the trunk) a few weeks ago, I noticed that GWiT started overriding the web.xml in the embeded Tomcat configuration directory. I just thought “oh no!”, but next thing I noticed was that it stopped overriding the application web.xml in the ROOT webapps folder. Let’s see if it stays this way…

GWT deferred binding tutorial

I just found a very good tutorial for the undocumented deferred binding featureof GWT. In this example the deferred binding is used to create a databinding framework for GWT.

While introducing deferred binding, it shows a very interesting data binding idea for GWT. Instead of solving the binding to the UI elements at runtime, this approach solves the problem during the compilation. This saves ressources at the client and even simplifies the development.

Using Acegi to secure a GWT application

I am on the train to Munique on my way to the OOP 2008. Today I spent some time refactoring a GWT application (we are moving to GWT 1.5 built from the svn trunk) and thought I could use the travel time to post about GWT and Acegi.

GWT changes the way we develop internet applications. The web application is not only rich, but it also runs standalone in the client browser. Shortly spoken, a GWT application is a statically loaded set of html, css and javascript files. No serverside web technology gets touched in this process. The webserver delivering the files does not count here… Once the application starts running on the browser, it will start loading data from the server using http requests. This data will further on be displayed in the browser by the GWT application. An application delivered in Javascript is a security nightmare, since the code on the client side is readable and can be easily manipulated.

There are mainly two aspects in a GWT application scenario that might be secured:

– the GWT application
– the RPC services offered by the server

Securing the GWT application is an impossible task. It is possibe to only give access to the Javascript application after the user has identified itself, but once the Javascript code gets transfered to the client there is nothing that can be done to avoid manipulation.

Securing the RPC services offerend by the server is by far much more easier. Assuming the server is written in Java, there are many ways to secure the services offered: services in the GWT are RPC are implemented by Servlets. Restricting access to the servlets would do the job quite efficiently.

The approach described here does not try to hide the GWT application from the world. The application cannot be secured in the browser and, in many cases, it is the data and the serverside actions that must be secured. A GWT application that cannot connect to its RPC services because of bad credentials is useless. If the application algorithm is worth securing, then hiding the application is the best option available. Unfortunately, this cannot be done with GWT, it would be like a dog trying to bite it’s own tail. After all, the user must identify itself before, and then the application can be dowloaded to the browser. GWT does not known anything about lazy loading parts of an application at runtime. A simple way to implement hiding the application is to host the application in a JSP page ahd have this page secured. Not only the JSP must be secured, but also all static files composing the GWT application. Only after authentication succeeds, the JSP page can be accessed, bootstraping the Javascript application. The JSP should create a server session with the provided credentials, the GWT application would use RPC to gain access to this information and to know who has logged in. This is a server side approach, and here is where I tend to use Acegi.

Acegi is a very interesting Java security framework. It is often used with Spring and with web applications, but it also can be used standalone (without Spring) and to secure any kind of Java application. With AOP techniques its usage can be non invasive to the application code.

In our project we did not hide the GWT application. First, it is an intranet application. Second, the application is useless without the RPC services and the provided data.

The idea on our approach is to download the GWT application to the browser and let the application display a modal login dialog box. The information provided by the user will then be sent to a unsecured login service on the server. The server performs the authentication through acegi, creates a server side session with the appropriate credentials.

Een if some user hacks the Javascript to bypass the login dialog, nothing will happen. The navigation tree is populated by RPC: each user has a different navigation tree, depending on user preferences, stored searches and security roles.

The login RPC call returns the login information required by the GWT application to know that a successful or unsuccessful login has occured. On a successful login, the GWT application starts loading the data needed through the secured services in the server. The server allows this services to be called because the client has created a valid server side session with apropriate credentials. Acegi will secures the services based on this information.

Finally, our GWT application checks regularly if the session has not been invalidated (or if the server is respoding at all) and performs a client-side logout if needed.

Feedback is welcome!

No, I did not post it, but talking about provocation…

The posting under URL

is not from me. I do not know who posted it there and put my name below it. The title, the content and the original author are completely wrong! What happened? Hello?

The original post to the cited content comes from Fabrizio Giudici’s Blog, you can read it here:

And, BTW, the whole web application era we are in is a provocation to developers. We witnessed a browser war, where the competitors glanced by adding text flashing capabilities to their proprietary html. We have now 2007 and our web browser still don’t know the difference between a text, number and date field. So rich clients are IMHO the only technical sane solution for rich internet applications. But the insanity comes from potentially million users out there, sitting in front of browsers (already installed and configured to bypass the evil firewall), ready to use, ready to surf, just ready.

If Google Gears is the tool helping me to use my web application instead of rich clients, I’ll go for it. The browsers had their chance. I am sorry they didn’t take it.