Earlier today, I published version 2.5.0 of the XPages Jakarta EE Support project. It's mostly a consolidation and bug-fix release, but there are few interesting features and notes about the implementation. Plus, as teased in the post title up there, there's a looming problem for the project.
There are two main new features in this version.
First, I added some configurable CORS support for REST services. Fortunately for me, RestEasy comes with a CORS filter by default, and it just needs to be enabled. I wired it up using MicroProfile Config to read some values out of xsp.properties:
I also added support for using the long-standing @WebServlet annotation. Though REST services will generally do what you want, sometimes it's handy to use the lower-level Servlet capability, and now you can do so inline:
There were a couple specs where I had previously either copied the source into the repository (CDI, Mail) or had maintained a local branch fork (NoSQL). Those were always uncomfortable concessions to reality, but I decided to look further into ways to handle that.
For NoSQL, part of it was what I talked about in my last post: using Eclipse Transformer to make use of
javax.* compiled binaries and source converted to
jakarta.* automatically. But beyond that, it had the same problem that I had forked Mail for. Namely, it hits the same trouble that lots of non-OSGi code does in an OSGi context, where it uses
ServiceLoader in a non-extensible way. Though I have an open PR to make use of the pseudo-standard "HK2" ServiceLoader provider, waiting for that would mean continuing the local-build trouble.
Instead, for all of these cases I made use of OSGi's Weaving capability to re-write those parts of the class files on the fly. While this is a bit unfortunate, it works well in practice. The only real down side for now is having to be a bit more careful when bumping the versions in the future, but this type of code changes very rarely.
The Looming Wall
While this has been going swimmingly, I've started to hit some real impediments with Domino's Java version. The next release of Jakarta EE, version 10, requires Java 11 as a minimum. This is similar to the move Equinox (Domino's OSGi framework of choice) made just under two years ago, and which has itself bitten me with a blockage to upgrading Tycho to version 2.0 and above. Java 11 is about four years old now, and is no longer even the latest LTS release, so this all makes sense.
I've known this was coming for a while, but incompatible versions of JEE specs and implementations started to trickle in over the past year, leading to me leaving notes for myself about maximum versions. JEE 10 itself is fairly imminent now, so I'll be capped at the ones released with JEE 9 a while ago.
So I've been pondering my options here.
In one sense, I solved this problem years ago. The Domino Open Liberty Runtime project has had the ability to download any version of open-source Java that you want, and I expanded it last year to let you pick from several common flavors. Liberty maintains a breathless pace of advancement, adding official support for Java 18 the month after it came out. If one wants to run JEE apps on Domino, that's the most complete way. However, though it does its job technologically well, it's not exactly a natural fit for Domino developers in its current state.
But I've been considering anew a notion I had years ago, which is to write an extension for Liberty so that it reads class files and resources out of an NSF directly. In some early investigation a bit ago, this started to appear quite doable. In theory, I could write an adapter that would take an incoming request for "foo.nsf" and then read files out of the NSF in the same way XPages does, but instead feeding them to Liberty's runtime. Doing this would essentially implement all remaining JEE and MicroProfile specs in one fell swoop on top of the "any Java version" support, but would add the fault-prone attribute of running a separate process and proxying requests to it. In practice, that setup has proven itself good, but it's certainly more complicated than the "single process on port 80" deal that Domino's HTTP is now.
That route also wouldn't inherently support XPages, which would be something of an impediment to the XPages JEE project's original remit. That's something I've also pondered, and in theory I could make an auto-vivifying version of the XPages Runtime project that grabs all the pertinent XPages bundles from the current server and patches them into the Liberty server as an extension feature, similar to how all the built-in Liberty features work. This could be done, but I'll admit that I balk a bit at the prospect. Though I run XPages outside Domino constantly, it's with full knowledge of the tradeoffs and special considerations. Getting a normal NSF-based XPages app to run in this way would take some additional work.
Anyway, those options could work, but none of them are great. The true fix would naturally be for HCL to move to a newer Java version in Domino's HTTP stack, but I don't control that, so I'll content myself with considering what to do in the mean time. Admittedly, pondering this sort of thing is enjoyable in its own right. Also fortunately, even without tackling this, there's still plenty of stuff in the pile for me to tackle as the fancy strikes me.