What Makes XPages "Not Modern"?

Aug 23, 2019 1:53 PM

Tags: xpages
  1. What Makes XPages "Not Modern"?
  2. Medium-Term Ways to Improve XPages

A big part of figuring out how to move past XPages is discussing what makes it no longer modern in the first place. Some of these reasons will be gimmes - out-of-date or outright-missing features - while some will be less about what XPages is strictly capable of and instead more about how XPages development and deployment works. I'm going to leave out some of the ones that are just from recent lack of development attention, such as Bootstrap 4 support remaining in alpha state.

Admittedly, I suspect that this post will come across as an enumeration of complaints, but I don't really mean it that way. XPages served us well for a while, but it's one thing to say "ah, XPages, it's a bit old and creaky" and another to think about just how pervasive that is.

Technical Limitations

The most straightforward reasons are the Java technologies that are included in the XPages stack, but are significantly out of date. This includes big-ticket core standards like Servlet 2.4 (faux 2.5), the forked JSF 1.x, JavaMail 1.3, pre-Unified EL, and JAX-RS 1.1; lesser-known standards like Activation 1.1; and even third-party libraries that show up in the fixed XPages classpath like Apache BeanUtils 1.6, Apache Collections 2.1, Apache Logging 1.0, and Guava 19 (a recent addition).

Some of these can be worked around relatively simply (including a newer JAR in your app/plugin or adding an OSGi bundle), but others are either more pernicious (like Logging and Guava, causing runtime ClassLoader trouble) or are outright showstoppers. I've done a lot of work to bring newer and missing Jakarta EE standards to XPages, but the ancient Servlet version is a hard limit on a lot of things. In addition to its own limitations like the lack of Web Sockets, third-party libraries and implementations often expect Servlet 3 as a bare minimum, and nowadays make use of Servlet 4.

Self-Imposed Hurdles

Beyond the technical limitations that largely come from outdated implementations, XPages is afflicted by what I'll deem "self-imposed hurdles": difficulties that spring from choices that Lotus/IBM/HCL actively made or chose to not make that resulted in things being a bit more difficult than in other environments. These decisions were often even good decisions in the aggregate, but they have tradeoffs.

The first big one is the use of the NSF container. Overall, this is a positive: not only did it create a more unified development experience with classic Notes elements, but the NSF itself brings attributes like external-to-the-webapp ACLs and seamless replication, which are extra addons at best with other frameworks. However, because the NSF is essentially a proprietary binary blob, it means that it can't readily participate with other tooling. You can't crack one open with a ZIP file tool, inside-JVM reflection/classpath tools can stumble over it, and it can't inherently participate in source control or automated builds. As above, some of that can be worked around with immense effort, but even then not nearly as smoothly as elsewhere.

Along the lines of the NSF, due to Designer's lineage and (presumably) the state of technology at the time that XPages was integrated, the normal XPages development process has no participation in the larger unified Java development world that Maven largely ushered in. Even besides not fitting into the build system, this introduces just a bit more friction in normal development. Yes, an XPages app can use many third-party Java libraries by just adding them to the Code/Jars section, but they don't benefit at all from version management or automated fetching of source and Javadoc. It may sound like a small thing, but it makes a world of difference once you're used to it.

That plays in to a general Domino/Designer issue that stretches back to the introduction of Java to the platform: the libraries include no parameter info or Javadoc. Designer has its help sidebar for Notes.jar classes and the Javadoc exists for a subset of the XPages stack as of 9.0.1, but they're not integrated in the IDE, and large swaths of the platform don't have any Javadoc at all. This leads to stumbling blocks where you'll seek out documentation for a class that nominally exists in XPages, but you have to be wary of the "Since" note for every method (and even then you may be thrown for a loop, like UIViewRoot#getViewMap() that's "since" 2.0 but does exist in XPages).

Then there's a big sticking point that's created a weird cargo cult within the community: the default Java policy. While the file itself does indeed come from stock Java (including the adorable "properies" typo), the way it interacts with the SecurityManager and ClassLoader setup in XPages has an outsized negative impact. XPages apps butt heads with this policy constantly, in particular when using reflection - which is something that third-party libraries do often as a matter of course. It's had the effect of making many people shy away from using third-party libraries or code that uses normal Java features like this out of a partially-justified desire to not tweak the settings.

The last one in this category is the use of OSGi generally and Equinox specifically. I can't knock the original choice too hard: OSGi is a logical pick for making a large, pluggable system and IBM institutionally had tremendous experience with it by way of Eclipse and WebSphere. However, while it's a solid choice for system architecture, it's often very awkward for app architecture. And, hypothetically, XPages is designed to avoid this. Like the various Java EE servers that are implemented with OSGi but run plain-old WAR-based apps, a basic XPages app isn't meant to care about the OSGi layer at all. However, a combination of the above security-policy troubles, Designer's limitations as an IDE, explicit blocks on what classes an NSF can access, and the severely-limited EE services provided by the XPages runtime means that OSGi plug-in development came immediately to the fore. And developing plug-ins for XPages is hard. It involves all sorts of ceremony and incantations just to get started, and maintaining even a well-structured plug-in project is much more difficult than a normal Maven or Gradle project. Some of this is OSGi, some of it is Equinox specifically, and some of it is unique to XPages, but it all combines to make it all a daunting process even for people who aren't put off by Java-the-language. So, much like with the security policy, you have a large subset of the XPages community that for either developer or admin reasons don't develop XPages libraries and often don't bring in any outside of that HCL ships in the box.

Development Model

And, finally, I'd like to cover the development model created by XPages even when everything is firing on all cylinders.

The first one is the lack of MVC (or similar) structure. XPages was introduced as bringing MVC to Domino development, in large part thanks to the underlying JSF nature, but this was something of a dirty trick. While it's theoretically possible to make a cleanly-structured XPages app, the framework provides no useful model layer, and Designer is almost actively hostile to this kind of separation. And, because the framework doesn't give any guidance and few of us in the community came in as seasoned web developers, the discussion about XPages and MVC is largely about each person's individual version of "MVC-style" development, and Lord knows I'm as guilty as anyone. This, like the lack of dependency management, is something that you don't fully appreciate until you work with a system with real structure and clear guidance on how to do it.

The second point in this section is one that I actually debated putting on this list, and that's XPages's use of server-side state. Depending on who you talk to, this alone is something that makes XPages dead in the water, an ancient relic of a forgotten era. Personally, I'm not so universally down on the notion: I feel that the worries about scalability are usually based on hypotheticals, and being able to make complicated dynamic apps maintained by the server cuts down on the work of writing and securing REST APIs for absolutely everything. Still, it's pretty clear that the server-state model has fallen out of favor, and that's worth at least considering when determining "not modern".

So What About Workarounds?

A running theme through all of this has been "X is behind the times or weird, but can be overcome with a bunch of work". And this shows up a ton in practice: a lot of XPages developers have gone to Herculean lengths to bring in other technologies or carve out a way to do structured programming, and it sometimes works. Others eschew XPages UI components entirely and build JavaScript apps backed by REST services written with the ExtLib components, JAX-RS, or other mechanisms. Especially since the Java 8 switch, you can do a lot to drag XPages forward.

But that doesn't make it "modern" really. This is all work fighting against the platform, work that doesn't need to be done with other tooling. In other environments, if you want to use the latest version of a spec, you make sure your container is up-to-date and then you just use it. If you want to bring in a dependency, you declare it in your "pom.xml" and it shows up, source and all. And then there are the things that other environments provide for you, like OpenAPI documentation, and concepts that don't even exist in XPages, like fault tolerance. Maybe some of those things could be brought to XPages too, but, again, it'd be a struggle every step of the way. There'd always be some weird thing to do with the Servlet version, or Tycho, or ClassLoaders, or loggers, or some of the many other little asterisks that accompany the platform.

The Work To Move Forward

My plan for future entries in this series is to discuss some of the specific steps that I've been taking with my largest active client project to prepare it for the future. That future is as-yet-undefined - it may stay within the Domino web container, it may not - but the good part is that a lot of the work is in common regardless of where we take it, and I think that it will prove useful to others facing similar situations.

New Comment