Thursday, February 08, 2007

Mixing Iron and Clay: Implementing CMS and Portal as one product

For the last few weeks I've had the joy of working with a portal solution that also provides excellent CMS features. More often than not, portal meets CMS by being a Portal with some junky editor functionality inside, but there are also some CMS'es that feature bits'n'pieces of functionality like portlets, widgets and plugins.

The portal I've been working seems to strike in the middle. It has the richest CMS interface I've come across in a portal product (content structure, versioning, editing all neatly packed into a comfy filthy fat applet client that actually works), and every bit of content is actually an instance of a portlet by its own right (how they've managed to keep the performance so high is quite impressive, and a mystery to me).

This means that not only can I choose each piece of content from a large variety of portlet packages, but I can also transcend content between different states, which is a very handy portlet feature.

However, there is a flip-side to the coin:

Portlets are tiny web-applications. They are powerful, and great once you have them configured. But this means you have to configure them , states and all. You have to compile them, build them, deploy them into the portal, create instances of them and run them before you get to see results.

Additionally, the only content-level storage supported by the portlet spec is portlet preferences, a simple little rag of string data that is used to reference the content of the portlet instance. The spec is a portal-spec, not a CMS spec.

This annoys me as a developer to some degree because I can't utilize the functionality of the underlying content! I can't iterate over the current collection of content. I can't store data in the portlets other than in the portlet preferences, and I can't retreive data other than using the predefined content-portlets that came with the product.

So when I extend the functionality, customizing the CMS if you will, I have to create entire webapps; verticals: database, control layer, web layer and all, in addition to configuring the portlets, states, and then having the deployment hassle of it all.

And at that point I wonder what is the point of having a portal after all....

The answer lies somewhere in the component architecture of the portal. A portal can use many different verticals throughout the application. A webapplication normally uses only one. Oh allright, the webapp can use several as well, but the configuration is all stuck in one place. A portal can combine many different packages of portlets, each having their own web-app configuration, say for instance using their own MVC framework.

This is great, because by the time people get tired of the portlets using Oracle DB with Struts on top, you can replace it with a new DerbyDB/WebWork vertical, exchanging one component that hopefully does not disturb the others. This great is because.. (dramatic pause for today's word of wisdom)...

A framework or technology has a lifespan of max five years. If you throw out and exchange the old components piece by piece, you have a product that will live forever.

I wish there was some third option, getting nice and clean modular webapps without the heavy portal framework. I know there is a solution out there, we just need somebody to sit down and figure it out, then share the knowledge with the rest of us.