Raible Designs

[TSE] The Holy Grails of Web Frameworks with Guillaume LaForge

Under the hood, Grails uses Spring MVC. It has support for "flash scope" between requests.

I find it funny that flash scope is so popular these days, we've had this in AppFuse for four years. However, web frameworks didn't add native support for it until it had a name (provided by Rails). To be fair to Struts Classic, they had support for it before Rails was even invented.

Rather than JSPs, Grails uses Grails Server Pages, which look much like JSPs. Grails uses SiteMesh by default and allows you to easily change the layout used with a meta tag.

<meta name="layout" content="main"/>

Most of the dynamic attributes in a GSP are rendered using the various "g" tags. There's dynamic taglibs for logic (if, else, elseif), iterating, linking, ajax (remoteFunction, remoteLink, formRemote, submitToRemote), form (select, currencySelect, localeSelect, datePicker, checkBox), rendering (render*, layout*, paginate), validation (eachError, hasError, message) and UI (i.e. richtexteditor). [Read More]

RE: Five things I hate about AppFuse

Karsten Voges has written a nice critique of AppFuse titled Five things I hate about AppFuse. I started a new JSF+Hibernate project with AppFuse yesterday, so I definitely feel some of his pain. Let's examine his points one-by-one:

1. It's nice to choose between the usage of JSP 2.0 or before, but making the changes (to all jsps and the web.xml) every time I build my app sucks.

I absolutely agree. I spent a good hour modifying build.xml, web.xml, etc. to switch my app from JSP 1.2 to JSP 2.0 yesterday. You can modify your build.xml to permanently switch to JSP 2.0, but it doesn't get everything. I'll create a separate script for 1.9.3 and the upcoming 1.9.4 to make this cleaner.

2. Seperating the classes in web, services and dao is good, but I hate the building of jar-Files for the different layers. Just take the classes under dao and service and copy them into the war or move them over to the webapps folder as it is done with the web classes!

Yeah, I've proposed doing this a couple of times on the user mailing list. It always gets shot down by existing users. I'm sure it'd be possible to write a script to do this, but it'd be no fun to write.

3. Eclipse crashed with OutOfMemory errors. Always when trying to open the build file. The build file is really, really long, with lots of stuff in there. IMO 50% of it could be deleted.

Hmmm, I rarely have OOM errors with Eclipse, but I also used Bruce's tip for increasing Eclipse's available memory. If you're using Windows, here's how to bump up Eclipse's memory.

4. Generation of Hibernate-Mapping files. I really hate it to look within a jarfile how the Hibernate mapping file looks like. It is nice to get it generated, but I prefer to be able make adjustments to it by hand to try out things quickly. And it is quite hard to enter special SQL statements in an Hibernate file, if it gets overwritten all the time.

You can create/modify the Hibernate mapping files by hand if you prefer. From the FAQ:

If you have an @hibernate.class tag on a POJO - hibernatedoclet will generate the mapping file into build/dao/gen. If you have a mapping file (*.hbm.xml) file for your POJO in the src/dao/**/model/* directory, it will overwrite the generated version. If you don't want to worry about the two conflicting - just remove the @ sign from @hibernate.class in your POJO and put your hbm.xml file in the model directory.

No build.xml modification are need for this to work. The "package-dao" target will include these mapping files. If you want to get rid of the hibernatedoclet process, you can do that- but make sure and run it first - and then copy all of the generated hbm.xml files into your model directory.

5. I don't like to get my struts.xml merged from many sources. I like to have one struts-config file holding all my struts configuration.

I agree this is kinda painful, but so is developing with Struts. You should be able to move the generated struts-config.xml into web/WEB-INF and remove the <strutsconfigxml> from build.xml to get the behavior you're looking for.

As far as throwing out the build.xml, I'm actually planning on doing that with my current project. I will keep it in place for the development phase, but I hope to move the application into a large build system once I'm done. Since it's all Java code and XML in the end, this shouldn't be hard to do. I did it when migrating to Maven 2, so I know it's possible. As far as Karsten's opinion of Maven 2, he may be right - but I hope to make a strong effort to make it very useable when using AppFuse. In fact, I hope to make it possible for users to use their IDE their entire time, with no need to run any Maven commands. Of course, that could be a pipe dream - only time will tell.

As far as sounding like the BileBlog, the more you rag on AppFuse, the better. Remember, screaming users are a good thing.

Seam 1.0

I've posted my thoughts on Seam 1.0 to my Virtuas blog. What are your thoughts?

It's great to see the release of Seam 1.0. Seam is similar to many full-stack frameworks like Rails, Rife and AppFuse in that it gives you all the pieces you'll need to build a kick-ass web application.

I've blogged my thoughts on Seam before, so there's no need to do that again. I like the idea, especially the lack of interfaces and the 3-files-for-each page idea. However, I don't know that this concept will fly with Java developers. I agree there's a need to simplify, but many of us are mesmerized by the de-coupling that Spring gives us. So now we're programming to interfaces, and every-so-often swapping implementations. I don't know that we can switch to this simpler model. And then there's the "EJB" thing. I think there will be a fair amount of developers that don't use EJB3 simply because it has the "EJB" name. The best thing the EJB Expert Group could have done for EJB3 would be to give it a new name.

The other thing I worry about with Seam is that it wasn't developed from an existing application. AFAIK, it didn't get extracted from a real-world application that had all the problems that Seam solves. I know that Gavin is a smart guy, and he's probably seen these problems in the real world, but there's nothing like developing a real-world application with a technology - and then extracting the framework from that.

In reality, I'm probably jealous. Seam has some really cool features, JBoss has done a great job of marketing it, and it seems to be a really cool way to develop applications. If I'm going to make AppFuse a direct competitor to Seam, it's gonna be quite the uphill battle.

How do you determine a good MaxPermSize?

I know I'll probably get beat up for not knowing my JVM Turning parameters. I admit that I should know them better than I do. Hopefully this post will help us all understand them a bit better.

Ever since I upgraded appfuse.org to AppFuse 1.9.1, it's been experiencing OOM issues. They've been so bad that the site is lucky if it stays up for more than an hour. I've done a fair amount of performance testing on a single AppFuse application (and gotten very good numbers), so I was pretty puzzled by the whole situation.

To reproduce the problem, I downloaded all 5 demos to my machine and began profiling with JProfiler. Nothing stood out, but I was able to reproduce the problem by clicking through all the different applications. While testing, I had my JAVA_OPTS set to -Xms256M -Xmx384M.

After staring at JProfiler for hours, I gave up and sent my findings to the AppFuse mailing list. After going back and forth with several ideas, Sanjiv came up with the winner.

Did you try increasing the max perm size (-XX:MaxPermSize=256m)? Max Perm size is running out of memory and not necessarily the main memory. Class metadata stuff is placed in the perm memory (google for more details) and since we're using Spring, Hibernate and Tapestry which all use a lot of reflection, proxying etc, it's not surprising that max perm size is running out of memory.

Based on his advice, I added -XX:MaxPermSize=256m to my JAVA_OPTS, fired up JProfiler/Tomcat and began hammering my local instance with WAPT. 15 minutes later, with 20 simultaneous users, the heap and memory were humming along nicely with no issues. I made the change on appfuse.org and it's been up every since.

This experience has motivated me to start adding "-XX:MaxPermSize=256m" to all my JAVA_OPTS. Is this a good idea? If so, is 256m a good value to use? If not, what's the best way to determine (or guess) the proper value for this setting?

Jetty 6 Maven Plugin now works with SiteMesh (and Equinox)

The 1.6 version of Equinox contains commented-out settings for Maven 2 Jetty Plugin. The reason these are commented-out is because this plugin didn't work with SiteMesh at the time. I checked again today, and it looks like they got it fixed. See Brett's post titled Developing with Jetty: Where Have You Been All My Life? to see why this plugin is so cool.

Using this plugin (or the JettyLauncher in Eclipse) makes it pretty damn easy to do develop Java webapps. There's no longer a deploy cycle, just save and refresh your browser. IMO, it's almost as good as using a scripting language or developing with HTML/CSS/JavaScript.

I'd love to see someone develop a TomcatLauncher, a WinstoneLauncher and Maven 2 Plugins for both. AppFuse works with Winstone 0.8.1 (a wicked fast servlet container with a good story behind its name).

In other Jetty news, Jan Bartel posted a nice tutorial today titled How To Use JOTM as the XA Transaction Manager in Jetty6.

[ANN] Equinox 1.6 Released

For a good story of how Equinox helps, see Wayland Chan's Equinox to the rescue blog post.

This release's major new features are Tapestry 4.0 and WebWork 2.2.1 upgrades. In addition, I changed to use Maven's Standard Directory Layout. It makes IDE and using Maven plugins much easier, so it's a natural progression.

This release does not contain Maven support for running the integration tests with Cargo. This is because Cargo still seems a lot more complicated with Maven than with Ant. Hopefully I'll be able to figure out an easy way to get test-all functionality with Maven and Cargo in the next release.

All of the frameworks used in Equinox, as well as its build/test system is explained in Spring Live. A summary of the changes are below (detailed release notes can be found in JIRA):

• Added custom exception page for Tapestry, as well as tapestry-flash.
• Changed birthday date input to use WebWork's DatePicker component.
• Added support for pre-compiling JSPs when building with Maven (on by default).
• Added createDatabaseIfNotExist=true to jdbc.properties.mysql to auto-create the database when using MySQL.
• Changed classes that extend *SpringContextTests to use AUTOWIRE_BY_NAME so more than one instance of an interface is supported.
• Dependent packages upgraded:
  • Cargo 0.7
  • DisplayTag 1.1
  • Hibernate 3.1.2
  • Scriptaculous 1.5.2
  • Tapestry 4.0
  • WebTest build 1168
  • WebWork 2.2.1

Download. For more information about installing the various options, see the README.txt file.

Demos:

• Spring
• JSF
• Struts
• Tapestry
• WebWork

Known Issues: The Tapestry-Flash JAR was built with JDK 1.5 - so you'll need JDK 5 to run the Tapestry version. Howard Lewis Ship said he'd fix this tonight or tomorrow. Also, if you're on Unix, you'll need to run "ant fixcrlf" before you install anything. Finally, downloading dependencies might not work the first time. Running the "ant" or "mvn" command multiple times usually solves the problem.

See the roadmap for what's coming in the next release.

Large sites powered by Java web frameworks and Tiles + WebWork

Yesterday, I delivered a Comparing Web Frameworks seminar that included Struts, Spring MVC, WebWork, JSF and Tapestry. This was for a client that's in the process of re-working an extremely high traffic site (50+ servers currently) from Servlets + JSPs to a web framework. They love the idea of Tiles (and know how to use it) as well as plan on integrating many Ajax features.

We quickly eliminated Struts because of ActionForms since they're planning on moving to persisted POJOs. Spring MVC and JSF had a notch up because they work with Tiles. However, JSF has reportedly had scalability issues. Furthermore, it's the most-complained about framework out there. One attendee noted how she was impressed with the low number of complaints about WebWork.

WebWork doesn't integrate with Tiles (but probably will soon) and they were concerned about SiteMesh performance with large pages (1MB + of text). While I believe SiteMesh can do almost everything that Tiles can do, I also agree that Tiles is a good technology. Furthermore, the "advanced features" of SiteMesh to be largely undocumented, which can be a barrier for adopting it as a "development standard".

Spring MVC was dinged because it doesn't have built-in Ajax support like WebWork and Tapestry (via Tacos). However, it's support for Tiles might just make it the one they choose - especially since they plan on using Spring in the middle-tier/backend. While they loved the idea of Tapestry, they didn't think they could afford the learning curve and I don't know enough about the @Border component to verify if it has all of Tile's functionality.

One interesting thing that came up was the list of high-volume sites using these various web frameworks. Tapestry seems to come out on top when you look at the list of well-known sites. However, I'm sure there are plenty I don't know about. If you know of high-volume sites using any of these five frameworks, please let me know. I'm looking for major sites with millions of hits per day. Here's my current list (extra points for fancy templating with SiteMesh/Tiles + Ajax widgets):

• Struts: None that I know of off the top of my head, but I'm sure there are plenty.
• Spring MVC: None that I know of.
• WebWork: JavaBlogs (don't know if this exactly qualifies as high-volume, there aren't that many Java developers). WebWork also has a few products based on it (i.e. Jive, JIRA, Confluence), but these companies also employ WebWork committers.
• JSF: None that I know of.
• Tapestry: NHL.com, TheServerSide.com (similar comments to JavaBlogs) and Zillow.com.

Thanks!

Related: How To use Tiles like SiteMesh and SourceLab's Web application technologies comparison (with performance numbers!).

Update: FWIW, I figured out How to use Tiles with WebWork and wrote a short howto for doing dependency injection with SiteMesh.

Does JPOX suck?

There's an ongoing effort in Roller to migrate from Hibernate to JDO. Mostly, this is due to Apache's silly rule about no L/GPL dependencies - even if they're downloaded separately. I think this is a valiant effort, especially if JDO performs as well as Hibernate.

However, it was interesting to see the following message on the mailing list this morning:

i have experience using jdo, and jpox in particular, with a commercial product. first, you probably already know this, but jdo is dead (from a spec perspective anyway). it will be phased out in favor of ejb3 persistence. maybe that transition will be graceful, maybe not. i see jpox has ejb3 on their roadmap, but not sure what that means.

second, jpox has really, very atrocious performance issues. the jpox folks admit that performance is a low priority, as they are an ri. if someone wants the details on this, i can dig them up.

Interestingly enough, this message is from a Sun employee. It's interesting to hear someone from Sun say that "jdo is dead". What are you thoughts? Should Roller change their persistence backend just to satisfy Apache?

Of course, now you'll tell me your favorite Apache-licensed persistence framework and why it's worked so well for you. The real question is - are you willing to re-write Roller's backend using it?

Edit Java webapps Redux: Jetty Launcher and Equinox

A few weeks ago, I realized I was developing webapps the hard way, by re-deploying everytime I made a change. It's important to have a build process that can create a WAR and deploy, but it's also important to have a system where you can edit/save/view changes w/o ever deploying.

I spent some time this weekend working with Jetty Launcher, Eclipse 3.1 and the latest version of Equinox. Below are instructions on how to make the two work together. Once you've completed the steps below, you should be able to launch Jetty and edit/save Java files or JSP files in Equinox - and the changes will show up immediately in your webapp. No deployment or Ant/Maven interaction necessary.

I've tested this setup on Eclipse 3.1, using both OS X and Windows XP.

Step 1: Install Jetty and Launcher

Download and install Jetty. I tested 5.1.4 and 5.1.6 and both seem to work (6.0.0 Beta does not). In Eclipse, go to Help » Software Updates » Find and Install. Select Search for new features to install and click Next. Click on the New Remote Site button and enter "Jetty Launcher" for the name and "http://jettylauncher.sourceforge.net/updates" for the URL. Click OK, continue to download and restart your workspace.

Step 2: Install Equinox and create Eclipse project files

Download Equinox 1.5 beta 1 and extract to your workspace (I generally use c:\Source on Windows and /Users/${username}/dev on OS X). Download Maven 2.0, install it, and add $M2_HOME/bin to your $PATH. From the command line, cd into the equinox directory and type "mvn eclipse:eclipse". Get a cup of coffee or soda while you wait for Maven to download all the dependencies.

Once the project files have been created, open Eclipse and go to File » New » Project » Java Project. Click Next and type "equinox" in the Project name box. Click Finish to begin importing the project. You'll probably get an error like the following. Click OK to continue.

Click Next to continue (I had to click it a few times before it took me to the next screen). On the next screen (Define the Java build settings), select the web directory and click Configure inclusion and exclusion filters. Click the Add button for Exclusion patterns and enter "WEB-INF/classes/" (the trailing slash is important).

You're not done yet, now you need to define the M2_REPO variable that points to all the downloaded dependencies. Click the Libraries tab and then the "Add Variable" button as seen below.

Click the Configure Variables button and add a new one with a name of M2_REPO and Path of to your local Maven repository (/Users/${username}/.m2/repository on OS X and C:\Documents and Settings\${username}\.m2\repository on Windows). Click OK, Cancel and then Finish.

Step 3: Configure Jetty Launcher for Equinox

In Eclipse, go to Run » Debug and select Jetty Web in the Configurations panel. Click the New button. In the "Use Options" section, use "web" for the webapp root and (optionally) "/equinox" for the context name.

Jetty has issues running applications that use commons-logging, so you'll need to turn off all the INFO logging from the projects that Equinox uses. To do this, click on the Arguments tab and add the following to the list of VM arguments:

-DDEBUG -DDEBUG_PATTERNS=org.appfuse -DDEBUG_VERBOSE=-1

This can be placed on the line below the -Djetty.home argument. For more information on logging in Jetty, see the Jetty logging and debugging tutorial.

Step 4: Start Jetty in debug mode and modify to your hearts content

Click Apply and then Debug, and watch Jetty startup. If you go to http://localhost:8080/equinox/users.html in your browser, you should see a log message like the following:

11:44:25.855 DEBUG  [SocketListener0-1] org.appfuse.web.UserController.handleRequest
(UserController.java:24) >29> entering 'handleRequest' method...

You should be able to click on "UserController.java:24" to navigate to the UserController.java class. In this class, modify the log.debug(...) message, save the file and hit refresh on your browser. The console should spit out your updated log message. If it didn't, go to Window » Preferences » Workspace and make sure Build automatically is checked.

As far as I can tell, edit/save/refresh will work for .java and .jsp files, but not for .xml files. For that, the Jetty Launcher adds a Stop/Restart Jetty icon to your Eclipse toolbar. This setup seems to work great, except for the fact that you can't see when Jetty is done starting up in the console.

NOTE: I tried to get a similar setup working with the Tomcat Eclipse Plugin (v3.1 beta) and the Eclipse Web Tools Project (v0.7.1). Neither worked as smoothly, and the WTP wouldn't even deploy to Tomcat.

Editing Java webapps instead of edit/deploy/reload

For the last few years, I've always done Java webapp development the hard way. Yeah, I'm the guy that makes Dion cringe (although I'm pretty sure he's not referring directly to me). I edit a class/jsp/xml file and run "ant deploy reload". Then I wait a few seconds for my context to reload in Tomcat. Luckily, I do mostly test-first development, so it's rare that I have to open my browser to test stuff. However, with the power of CSS and Ajax, manual testing in a browser is becoming more and more useful (although Selenium may solve that).

I've long resisted the power of the IDE, b/c I've always trusted Ant and felt confortable with the command line. However, I'm ready for a change. I'm ready to start developing Equinox and AppFuse-based applications using the edit/save/auto-reload cycle. So how do I get started? Where's the instructions for setting up my IDEs to work this way?

I prefer to use Eclipse and IDEA for development - so I'll likely try to get this working in both. If I get it working, I'll make sure and provide good documentation so others can do the same. I'm also willing to make any changes in project structure to make this happen; modifying build.xml (or pom.xml) to accomodate shouldn't be too difficult.