Oracle JDK 24 debuted on Tuesday with 24 JDK Enhancement Proposals, or JEPs as they're known in the Java programming community.
The synchronicity of version number and JEP count "was not intentional, just kind of turned out that way, but it's nice," Georges Saab, Oracle SVP of development for the Java Platform, told The Register.
The last time version numbers and JEPs came close to aligning was JDK 16 with 17 JEPS – an off-by-one coincidence.
"However, these are not the only changes that are in the release," Saab added. "There's on the order of over 3,000 smaller RFEs or bug fixes, issues that have been fixed that are in the release."
Java 24's arrival coincides with the advent of the JavaOne 2025 conference in Redwood Shores, California, the site of Oracle's headquarters prior to its 2020 decampment to Austin, Texas. Incidentally, Big Red has announced plans to move its headquarters to Nashville, Tennessee.
Java 24 is a feature release, meaning that Oracle will support it for six months. Its successor, Java 25, is due in September and will be supported for eight years as an LTS (long-term support) release.
Feature releases primarily interest Java developers, while corporate users tend to prefer LTS releases for their long-term stability.
"We're getting a lot of feedback from developers that people really appreciate the six-month cadence of releases," said Saab. "It's giving developers more features more quickly, while at the same time we have the opportunity to sort of balance that with what enterprises want, which is just rock-solid stability."
The 24 JEPs in the JDK 24 release describe projects undertaken to improve the Java programming language, and as usual these are in various stages of completion.
"A couple I'd call out here in particular," said Saab. "The Class-File API is going to be great for people who are working at that level. It effectively provides a standardized way of doing the kinds of things that in the past you might have used [bytecode and manipulation framework] ASM for.
"Stream Gatherers is also a really interesting one. I would also say take a look at the things that have come in from Project Leyden over here on performance and runtime, like Ahead-of-Time Class Loading & Linking."
Donald Smith, VP of product management, said that those developing Java have been working on features to better support AI workloads. He pointed to Project Panama, which has been complete for several releases, and its Foreign Function & Memory API.
Noting that historically Java has been difficult to connect to native libraries, he said: "The JNI [Java Native Interface] was very brittle, but the Foreign Function & Memory API made it almost trivial to connect into a native library, which is very relevant to AI developers or developers trying to integrate business logic with AI, especially inference engines, which are often native libraries. Now you can easily wrap those and connect to them instead of using JNI, using the Foreign Function & Memory API."
Among the current crop of JEPs, further work has been done on the Vector API, designed to make vector computations reliably compile at runtime in a way that optimizes vector instructions for different CPU architectures. The API isn't complete yet – it's in its ninth incubation.
"It's basically waiting for Valhalla to start to come on board," said Smith, who suggested it would help vector computation for AI inference on certain RISC and ARM-based processors.
Java is also being outfitted for Post-Quantum Crypto, with foundational changes that will enable future support for cryptography algorithms designed to withstand the anticipated but not yet realized data processing power of quantum computers.
"We're kind of at a state now where there's consensus that it's not a matter of if, but a matter of when we need to be ready for the deprecation of traditional cryptographic schemes," said Smith.
Smith said the relevant JEPs are not really intended for developers at the moment.
"These are APIs that researchers and cryptographic providers are going to be using, including ourselves," he explained. "So these are still very much in the building blocks stage. JDK 24 introduces key derivation for quantum resistant computing, key encapsulation, and then the actual signing of whatever it is you happen to be signing."
The intent is to get feedback from the technical community and once there's more confidence that the implementations are correct, to then build out TLS support, code signing support, and all the niceties that developers of higher-level applications expect.
In conjunction with these rather complicated concerns, the Java team is also focused on basic renewal by ensuring that those learning how to program have access to Java resources. Over the past decade, computer science courses have shifted toward languages such as Python and JavaScript to teach new students about programming.
That hasn't been lost on Oracle's Java crew, which has been working on various ways to make Java less verbose and evangelize its benefits. The latest gambit is the launch of Learn.java.
"If you're brand new to the language and the platform, some of the stuff here can be a bit overwhelming," Smith explained. "And that's why we're announcing Learn.java. This is going to be a new portal launching next week, at JavaOne, dedicated to teachers and students. And the goal here is people will graduate from here to Dev.java and Inside.java."
Despite Oracle's interest in improving Java as a platform for AI workloads, it's noteworthy that Java insiders remain a bit hesitant to embrace AI for code development.
"That's definitely something that we've been looking at," said Saab. "As it is, we're obviously trying to balance the productivity and advances you can see there with some of the challenges of general-purpose tools around IP and stuff like that."
Saab pointed to Oracle's Code Assist and said it's shaping up well.
"That team, although they obviously have a broader remit than just generative AI for Java, consider Java as one of their initial and interesting targets," said Saab.
Beyond concerns about AI models ingesting valuable IP, Saab explained that AI works well with abundant training data but struggles with novel code generation.
"We're making new code and the possibility of writing code in new ways that there just aren't lots of examples of yet," said Saab. "We hope that in a few years there might be many, many examples of people using Java 24 or Java 25. But those don't exist yet."
Asked about the extent to which the government's effort to steer developers away from languages that lack memory safety guarantees like C and C++, Smith said the Java team at Oracle is proud of memory safety provided by the Java Virtual Machine (JVM) – written mostly in C++ – and is not looking to rewrite the JVM in Rust.
"We have documented our expectations in this area in the JEP draft 'Integrity By Default,'" said Smith. "Historically there have been issues when crossing the JVM boundaries such as with native interfaces. That is why the Foreign Function and Memory API very carefully considered the memory safety differences between Java and any underlying native library it is accessing.
"Java developers interact with off-heap memory through an API that allows for broad control over the handling and control of memory segments. It still allows programs to perform unsafe operations with native code and data, but must be done explicitly and the JDK will warn developers about such operations by default. We have no plans to reimplement the JVM in other languages or platforms." ®