XPages: The UI Toolkit and the App Framework

Jun 17, 2020, 9:19 PM

Tags: java xpages
  1. XPages: The UI Toolkit and the App Framework
  2. The RuntimeEnvironment Idiom
  3. NSF ODP Tooling 3.1.0: Dynamically Including Web Resources

Lately, one of my client projects has been picking up the pace on the years-long effort of taking a giant XPages app, making the business logic portable, and incrementally cutting down on the "XPage-iness" of it all. I expect that this will be a recurring source of blog posts, and this one is about distinguishing between "XPages the UI toolkit" and "XPages the web app framework".

XPages in an NSF

Coming from a Domino perspective, there is no distinction between the two, and that's largely because of the path we took to get here. Other than existing inside the same NSF, the distinction between "classic" Notes apps (web or client) and XPages couldn't be more stark. Legacy design elements are only shared in ways that are completely divorced from their original UI presentation (for the better), and the runtimes - as much as legacy elements can be said to have a "runtime" - are entirely distinct.

An XPages app can be thought of as conceptually a "normal" Java WAR-based webapp housed inside an NSF, and it has a lot of the trappings: classes in WEB-INF/classes, libraries in WEB-INF/lib, and an OSGi-style WebContent folder for miscellaneous files. It's not technically a normal webapp - there's no "web.xml" and the XPages outer "LCD" runtime is actually more like one giant webapp that acts like many - but it's close.

Critically, though, the Domino HTTP router only routes requests for ".xsp" files or "/xsp/" folders to your app's XPages environment, and this is the biggest technical and conceptual impediment. You can't (within an NSF) intercept just any incoming request and process it as you would in a normal webapp. You can kind of shim your way into it with servletFactory, but it's a fiddly process and limited to "/xsp/..." URLs.

Additionally, a running XPages app only exists in a very constrained way between requests. While the JSF-level "Application" and the Servlet-level session exist, you don't work with a per-app ServletContext the way you do in a Servlet webapp. You can hook in with ApplicationListeners and similar constructs, but they're still based on the lifecycle of the XPages app, which comes into existence only on the first request and dies (usually) half an hour after the last.

These, plus the specifics of Domino data access, combine to make the "XAgent" - an abomination of a concept - the catchall replacement for specialized rendering, batch processing, and even scheduled tasks.

XPages the View Engine

These are all accidents of history, though. They stem from the firm requirement that existing Domino HTTP behavior remain intact even with its brain transplant, as well as the "soft" requirement that XPages in an NSF pretend to be "forms with repeats and partial refresh".

At its core, XPages is "just" a web view engine: its only job is to accept a request from an HTTP client and return some HTML. The concepts it uses to accomplish this - components, renderers, managed beans, themes - are all incidental to the main task. This is the "V" part of MVC. Admittedly, even without the NSF compromises, XPages bleeds beyond its assigned third of the triad, and it inherited this from JSF. JSF is also billed as MVC, but it completely subsumes the "Controller" part and partially eats the "Model" part with its bean management.

Still, though, even a domineering framework like JSF slots in as just one component of a normal webapp, rather than being the whole thing as XPages is in an NSF. For example, take the app behind this blog, which partially looks like this:

Java and JSP resources in the blog

It uses JSP as its view template engine, but is it a "JSP app"? Not really. The fact that it uses MVC 1.0 is more important to understanding it, but that's really an extension to JAX-RS. You could make a strong case that it's a "JAX-RS app", especially when you expand the "Services" section in Eclipse:

REST services in the blog

That covers more of it, but still leaves parts out. It has application-wide beans by way of CDI, entirely-UI-free scheduled tasks kicked off from a ServletContextListener, and core business logic and model objects that are kept in a module that doesn't even know about the Servlet API.

It's layered, but the layers are explicable and the distinctions create a tremendous amount of flexibility. I could, if I wanted, change to ThymeLeaf for the front end with essentially no friction, JSF or Vaadin with only mildly more, or to a client JS REST UI by chopping off the top two layers outright.

Okay, So?

This description isn't a call to action - there's nothing inherently wrong about an XPages app in an NSF, especially a small-to-medium one - but this will be an important part of the conceptual groundwork in the months to come. To figure out what to do with all these piles of XSP markup and framework-specific business logic we have, we'll have to do a lot of deconstruction.

New Comment