Migrating from Java 8 to Java 17: A Comprehensive Guide to New Features | Hacker News Reader

Migrating from Java 8 to Java 17: A Comprehensive Guide to New Features

igorstechnoclub.com

66 pointsIgor_Wiwi67 comments

I love Java, I love the ecosystem. It's been a 25 year journey.

I will be the first to admit that I have not adopted "modern Java", mostly with all of the functional stuff (notably all of the collection streams and what not).

I certainly enjoy easy lambdas, but I don't necessarily code stuff up that way.

I'm not big on locked down classes, as a general rule. I won't say I never use "private", but it wouldn't bother me if that didn't exist. I can't say I'd "never" use records, but I'm so used to the boiler plate, I don't really see it any more.

My one issue, though, is simply how heavy handed the Java 9 module system is. I have a desktop app, and I use the module system to lighten up the JRE that it bundles. But that's an edge case (to me).

I'm just not a fan of it, I find it gets in the way more than anything, and it just seems more forced than optional.

Outside of building JREs, does anyone use the module system for anything?

My understanding is that the module system is primarily for library developers and not application developers. You can use the module system to strongly encapsulate your internals, which allows you to change those internal implementations without having to worry that some client is depending on it.

It’s for both really. The module system comes into play when application developers what to ship a jlinked runtime. You can make the size of your JVM executable down to like 35MB or smaller.

GraalVM uses it to build its native executables.

You can build GraalVM native images from modularized code, but you can also build them from non-modularized code.

Wouldn't the native executable be smaller using modularized code?

I don't see why it would be. The native image static analysis decides on the level of individual classes, methods, and even class fields what to include in the image and what not. It doesn't compile entire packages or modules into the image, only parts it considers reachable. And the module system shouldn't influence this reachability analysis I think.

Weird example with the "Java 8 exception handling":

  catch (Exception e) {
    if (e instanceof IOException) {
    } else if (e instanceof SQLException) {
    } else if (e instanceof NumberFormatException) {
    } else {
    }
  }

Older Javas already allowed choosing which Exceptions (and subclasses) to catch directly.

Definitely a bad example.

catch (Exception | IOException e) { … }

Is there a reason you choose to go to 17 but not to 21? 21 is also a LTS version.

The latest version of The Well Grounded Java Developer was released in 2022, when 17 was the latest LTS version. The high quality of that book and being able to prescribe it is one reason why someone might prefer it.

I think OpenJDK 17 might also the highest LTS installable one on Debian 12 out of the box right now, but that's a lot less certain.

Indeed, OpenJDK 21 isn't available in Bookworm's repo, just Sid's: https://tracker.debian.org/pkg/openjdk-21

As far as I'm aware, the only officially supported way of shipping Java applications is by building custom runtimes with jlink. So if you want to play by the rules, it doesn't matter what the distributions ship.

Ubuntu ships multiple runtimes for every supported distribution and does backports FWIW.

`jlink` was released for Java 9, I believe, so it's unlikely people who are upgrading from Java 8 already have the infrastructure in place to seamlessly transition to that instead of however they were managing their JAR files before.

"But this should be simple! It's almost like a Go binary!" I've learned not to underestimate how long it can take to change infrastructure. DevOps isn't seen as a profit center in most places, and they might well have much bigger things to worry about than transitioning from a kind of working JARs-on-VMs approach.

Jlink/jpackage UX isnt the best either. They mostly produce installers not a single binary that you can just run. Project Leyden has some interesting work around this called Hermetic Java which plans to ship one binary with the VM + jar together. It’s coming out of some folks at Google. I’m really looking forward to it.

Really now? Now that's interesting. I thought GraalVM might be a good option in that space https://www.graalvm.org/ last I checked, but something literally called Hermetic Java has my interest piqued.

GraalVM creates native AOT compiled binaries. The goal of hermetic is to ship the normal JVM. This is where is started a year ago: https://cr.openjdk.org/~jiangli/hermetic_java.pdf

I’ve been mildly watching the mailing list and they’ve had small progress.

If you're using AWS ElasticBeanstalk Tomcat distribution, 21 isn't supported yet. I realize this isn't a huge segment of the market, but it's probably not the only example.

https://docs.aws.amazon.com/elasticbeanstalk/latest/platform...

21 is supported for Amazon Linux 2023 and the JavaSE ElasticBeanstalk offering, though.

Some large application servers may not support Java 21 yet.

New features… ? We are on Java 22

Some businesses are migrating to Java 11 as we speak :)

Unhealthy ones maybe /s

I meant new features of Java 17

Java is a good, fun language that gets the job done.

Good for both everything and nothing at the same time.

It's really good at causing pain

interesting how all languages evolve a „record“ construct.

The payment processing case is nostalgic if a bit ironic, the team I worked at used this exact pattern 6 years ago or so in C#, and the expressiveness has only improved since then.

Yes, C# is better by every measure and has had everything we can only dream of for decades. You can stop coming to every discussion of Java with the same old story.

The problems with C# are many.

It's tied to .NET at the hip - there are no 3rd party implementations (Mono is effectively dead, correct me if I'm wrong), it's development is effectively solely driven by Microsoft, and every few years you have to rewrite all your things or get stuck on EOL versions of frameworks or abandoned tooling. The community is also vastly different in that it does not (yet) value open source and many different implementations of ideas (ie. there's an official Microsoft way to do X and that's what everyone does).

C# might be a fine language, but it's coupling to .NET and Microsoft are kind of a liability.

Java has a ton of warts, and it's not the language I'd choose today. However, it's ecosystem is FOSS-first and is immense in size and scope, has numerous options for any type of library/framework (take a loot at HTTP clients, or logging frameworks, or web frameworks, etc), and while Oracle primarily drives development they don't get the de facto final say anymore. There's also many JVM implementations and many class library implementations - all with different pro's and cons.

Perhaps .NET will get there one day. It is, after all, relatively new at FOSS/XPlat.

At least the company driving .NET releases it under MIT and does not have a trap SDK that should you make a mistake of using makes your company liable to legal action by Oracle.

Surely you're not gonna say that OpenJDK and GraalVM (both owned by Oracle) is somehow an issue, would you? And Mono lives well in dotnet/runtime, serving targets like Browser WASM and iOS (until NativeAOT replaces it in full at that target, at some point).

But you're not using .NET, so no need to worry. No one is gonna pry Java from your hands, but we'll get to laugh at the choice - it is both slower in performance-oriented code and more verbose in LOB one. Pretty much the only driving reason behind choosing it is if the company is an Apache/Java shop or when a vendor offers an SDK in Java/Python/JS only due to their popularity which is a common issue in the industry.

There's dozens of JDK & Class Library implementations. If you use the Oracle one - well, that's on you.

Most people don't use the Oracle JDK anymore.

And OpenJDK is GPL licensed - and has no strings attached. What you are referencing are people using Oracle's First Party JDK (which has a base of OpenJDK + Secret Sauce). Oracle wants money for that.

> No one is gonna pry Java from your hands, but we'll get to laugh at the choice

I don't think anyone in Javaland are concerned. .NET usage is a tiny blip on the radar compared to Java/JVM ecosystem.

So you have to pay for performance features too rather than getting to contribute to them and using them in the open? Unfortunate.

You are exposing your JVM ignorance I'm afraid.

No, you don't need to pay for performance. The only people paying Oracle are those who choose to.

There are many JVM vendors - each with their own secret sauce, pros/cons and add-ons. Some charge to use, some charge for support, some just give it away. Almost all offer a for-free OpenJDK version as well.

Vendors include some of the biggest names in tech, such as Amazon and IBM, and many others you might only be aware of if you're in the ecosystem such as Eclipse Foundation, Azul Systems, Bellsoft, and more.

I would say the Oracle JDK/JVM is likely in the minority of JDK/JVM deployments today.

Oracle deliberately positioned themselves to be just another JDK/JVM vendor a few years back. All of the vendors can offer support etc. Oracle has no special control over OpenJDK these days, even if they pay the most full-time staffers to work directly on it.

> Oracle has no special control over OpenJDK these days, even if they pay the most full-time staffers to work directly on it.

The contribution graph would suggest otherwise: https://image.slidesharecdn.com/jcconf2021openjdkcontributio...

> No, you don't need to pay for performance.

Then what is the selling point of that, if you say crucial features are not closed-sourced and gated behind costly licensing?

I feel like this is among many other reasons that people get PTSD which causes them to throw the baby (C# of du jour bundled in public perception with Java) out with the bathwater (Java ecosystem) and move to, inferior in many ways, Golang.

> The contribution graph would suggest otherwise

The contribution graph says precisely what I claimed.

> Then what is the selling point of that, if you say crucial features are not closed-sourced and gated behind costly licensing

Each vendor can do what they want. As I have already said twice, some vendors give away their inhouse-built JVM/JDK and only charge for support, some charge to use it at all, and some just give it away entirely for free. Motivations for each vary by company.

The point is, you have a ton of choice within the JVM ecosystem. You have 1 choice for .NET... the ecosystem and community are not the same. The JVM ecosystem and community are vastly better by any measurement.

C# developer's only valid criticisms are levied at the Java language - failing to realize the slow development pace is deliberate and prevents people from having to rewrite everything every couple years...

Despite that, there are many languages that run on the JVM - so you don't even need to use Java to experience the greatness of the JVM ecosystem and community.

I'd rather not experience "the greatness of the JVM ecosystem and community" nor I have to rewrite code every two years, the only big break was ~8 years ago with .NET Framework -> FOSS .NET ordeal, all subsequent changes are minor and incremental and pretty much in the libraries only. In general, this view is unfortunate but not unexpected, and is common with Java or, for some reason, Go.

There's a world of different projects, beyond back-end applications, written in C# or even F# like games, scientific instrumentation and analysis, desktop and mobile applications (multi-platform or otherwise), trading systems, malware and anti-malware, and more. It's a really versatile and powerful tool.

Let's talk again when the project Valhalla is done and Panama vectors are actually usable for writing implementations that can compete with C++.

GraalVM is released under a free license. GraalVM CE is open source.

Why the large jump from Java 8 to 17?

Many organizations are stuck on Java 8 because Java 9 made breaking changes. (Java 11 also made breaking changes.) Many of these changes also broke some popular libraries at the time that hacked into the JDK internals, so it meant a significant refactor for software that might be in maintenance mode already.

To clarify Java made changes to things they never promised would be a part of backward compatibility. Libraries were using things from JDK internals/implementation details, things that were never guaranteed to be there.

Is this analogous to the Python 2 to 3 jump?

It isn't that severe of a break. Most of the friction was with the libraries with transitive dependencies, like ASM, that depend on a particular bytecode version, using internal classes that were removed (e.g. sun.misc.BASE64Encoder), or the new module system encapsulating APIs that were intended to be internal-only (usually fixable with JVM runtime flags.)

A lot of organizations that use Java are very conservative with their tech changes, and didn't see a compelling reason to upgrade. This was especially true since Android Java still used Java 8, so many popular libraries continued to support Java 8 for their Android userbase.

No. Java 22 will happily execute code built for Java 6.

Java's version history crawled along at a snail's pace until Java 10, when they switched from a feature-based to a time-based release schedule. The release cycle has been pretty quick after that, since that happened only after 2017 or so.

That, combined with some pretty large changes in the underlying security model, mean that a lot of places have been stuck on Java 8 for a real long time. It's much harder to move from Java 8 to Java 9/10/11 in a lot of places than it is to then go from 9/10/11 to 17, for multiple overlapping tech debt cleanup reasons.

next LTS is 11, which I believe is already out of support, so 17 would be the next logical LTS.

the company I worked in, just recently moved to Java 17. So it's a bit personal

Reply on news.ycombinator.com