> It is not a goal to support cross-compilation. Code compiled in a training run must run on the same CPU architecture, with the same set of features, in subsequent production runs.
Your link is about a completely separate feature, called the AOT code cache (targeted for JDK28, not released yet). The post is about the AOT cache, which forms the foundation for this feature, but is separate and already available since JDK24. The JEP for the feature the post is about is https://openjdk.org/jeps/483
For the past several years, Java features are released piecemeal, with each piece released as soon as it can be useful to some significant portion of users or users who can offer useful feedback (and when it's not expected to restrict future pieces). The AOT cache is already used in production despite not being as user-friendly as it could be, because the companies using it find it valuable.
The first use case is the very common one (for Java) where developers control deployment (internal deployment). The last basic building block, caching machine code, will be released next year in JDK 28, and then the feature is likely to evolve to better support other use cases, such as external deployment.
Yes. And this is the right tradeoff vs the ossification you saw previously in Java (and we're starting to see now in Python, see the recent ejection of the CPython JIT), especially now when you can say "GPT, source dive the JDK and figure out which cutting edge is causing the bleeding."
> What would be needed is some sort of compatibility switch that disables all optional CPU features. Sure, you will get some reduced performance of the compiled code, but in exchange, a faster startup time.
I think other AOT-compiled languages don't opt in for the lowest set of CPU features -- they just detect them at runtime and use whenever present, all in compiled machine code.
It's normally an opt-in thing I think, since it's not an unconditional win: you have to figure out which code paths are worth it, and then where to trade off binary bloat vs the runtime cost of dynamic dispatch. In rust for example:
What are the most common ways to improve warmup times, actually? Looked a bunch into this, primarily native image, aot and crac, all of them super messy to get to work with a normal JVM webapp. I understand the warmup concept is inherent in JIT compilation, but how are big companies like Netflix handling this to quickly scale microservices horizontally etc?
Native images shouldn't be too messy anymore, Oracle put a lot of effort into DX improvements and all major web frameworks are now aware it's an important and growing use case.
Struggling to build native images should be a sign you're reliant on libraries that are dangerously outdated and abandoned.
So it is an antipattern nowadays to manipulate the classpath after building? Iirc native image require static classpath, but it is quite common to inject e.g. environment specific data via the classpath. Should one just read from the filesystem instead, or what’s the strat there?
They can definitely be pretty messy in some ways; even the standard library doesn't work out of the box perfectly. I have certainly struggled to get image manipulation with java.awt (e.g. thumbnail generation for a fairly standard web use case) to work in a native image.
Yes (or something along those lines), it's just that this isn't the first use case addressed by this capability, which is being delivered gradually (as we normally do in the JDK), and not even all the basic building blocks are out yet.
https://openjdk.org/jeps/544
> Non-Goals
> It is not a goal to support cross-compilation. Code compiled in a training run must run on the same CPU architecture, with the same set of features, in subsequent production runs.
IMHO, that would be much better than running into crashes and having to figure out the cause yourself.
"This is what I would like AOT cache to mean."
The first use case is the very common one (for Java) where developers control deployment (internal deployment). The last basic building block, caching machine code, will be released next year in JDK 28, and then the feature is likely to evolve to better support other use cases, such as external deployment.
I think other AOT-compiled languages don't opt in for the lowest set of CPU features -- they just detect them at runtime and use whenever present, all in compiled machine code.
https://doc.rust-lang.org/beta/core/arch/#dynamic-cpu-featur...
There are plenty of other AOT/JIT cache approaches in the Java ecosystem already, with more mature deployment experience of several years, namely:
OpenJ9, Azul, PTC, Aicas, the Android cousin since Android 7
Technology based on process snapshots CRaC for Amazon OpenJDK distro (Correcto), and OpenJ9, is also another approach available for some time.
https://youtu.be/U2j4bKgKr-s?t=3577
Struggling to build native images should be a sign you're reliant on libraries that are dangerously outdated and abandoned.
Instead, this AOT feature is not very ergonomic to use at all, with too many pitfalls to be of broad use IMO.