In CXF 3.2.0 we have also introduced a server-side NIO extension which is based on the very first JAX-RS API prototype done by Santiago Pericas-Geertsen. The client NIO API prototype was not ready but the server one had some promising start. It was immediately implemented in CXF once a long-awaited 1st 2.1 API jar got published to Maven.
However, once the JAX-RS 2.1 group finally resumed its work
and started working on finalizing NIO API, the early NIO API was unfortunately dropped (IMHO it could've stayed as an entry point, 'easy' NIO API),
while the new NIO API did not materialize primarily due to the time constraints of the JCP process.
The spec leads did all they could but it was too tight for them to make it right. As sad as it was, they did the right decision, rather then do something in a hurry, better do it right at some later stage...
It was easily the major omission from the final 2.1 API. How long JAX-RS users will wait till the new JAX-RS version will get finalized with the new NIO API becoming available to them given that it takes years for major Java EE umbrella of various specs be done ?
In meantime the engineering minds in SpringBoot and RxJava and other teams will come up with some new brilliant ways of doing it. There will be not 1 but several steps ahead.
Which brings me to this point: if I were to offer a single piece of advice to Java EE process designers, I'd recommend them to make sure that the new features can be easily added after the EE release date with the minor EE releases embracing these new features to follow soon, without waiting for N years. If it were an option then we could've seen a JAX-RS 2.2 NIO in say 6 months - just a dream at the moment, I know. The current mechanism where EE users wait for several years for some new features is out of sync with the competitive reality of the software industry and only works because of the great teams around doing EE, the EE users loyalty and the power of the term 'standard'.
Anyway, throwing away our own implementation of that NIO API prototype now gone from 2.1 API just because it immediately became the code supporting a non-standard feature was not a good idea.
It offers an easy link to the Servlet 3.1 NIO extensions from the JAX-RS code and offers the real value. Thus the code stayed and is now available for the CXF users to experiment with.
It's not very shiny but it will deliver. Seriously, if you need to have a massive InputStream copied to/from the HTTP connection with NIO and asynchronous callbacks involved, what else do you need but a simple and easy way to do it from the code ? Well, nothing can be simpler than this option for sure.
Worried a bit it is not a standard feature ? No, it is fine, doing it the CXF way is a standard :-)