Lessons from Mixing Rust and Java: Fast, Safe, and Practical | Hacker News Reader

Lessons from Mixing Rust and Java: Fast, Safe, and Practical

medium.com

115 pointskillme200842 comments

Do not write the bindings manually. Just use the amazing uniffi-rs library from Mozilla.

You can generate bindings for multiple languages. It supports error handling on both sides and latest versions also support native async integration.

I've used it to reuse the same Rust engine in iOS and Android apps and write native UI.

https://github.com/koofr/vault

The bindings are inefficient doing excessive cloning. But if performance is not a concern then its fine.

Couldn’t you manually edit after generating so you could thereby optimize the code as needed?

Thanks! Anywhere you hit issues?

Not really. But I didn't use async (was not supported yet when I started using it).

Bindings were easy, everything else (building, linking,...) was a bit pain to setup because there no good examples.

1. The article seems kind-of shallow. I didn't see any concrete (qualitative or quantitative) remarks about the "fast" part. I don't doubt you have reasons to do this - but I expected some information on what component are you writing using Rust + JNI, and how it helped? Or is it just a demo?

At some point, repeated calls into the JNI are counter-productive to performance, since the JIT can not optimize them. Pinning affecting garbage collection is another potential drawback if any of your rust calls are long lived. If we don't measure and just conclude "we are fast because rust is faster than Java, and we took average of both speeds", it's a disservice.

2. Also, I see unsafe for each call? I'd rather isolate this into a class / different file, since in JNI only few types of calls are possible. (method returning one of the primtive types, an object or `void`). This is the approach I took in dart jnigen. (Though there, the call is Dart -> Java, not Java -> Native language).

    unsafe {
      env.call_method_unchecked(
                java_logger,
                logger_method,
                ReturnType::Primitive(Primitive::Void),
                &[JValue::from(format_msg(record)).as_jni()]
      );
    }

3. I believe some details are missing here. What's native_add_one mapped to? And how is tokio futures awaited from Java? I believe that's the important part you should be presenting.

    public CompletableFuture<Integer> add_one(int x) {
        long futureId = native_add_one(x); // Call Rust
        return AsyncRegistry.take(futureId); // Get CompletableFuture
    }

4. Also please don't use ChatGPT for writing anything. It totally derails the reader by mentioning irrelevant details and long winded corporate conclusion at the end of every sentence.

This article summarizes our experience from a commercial project that runs on an in-vehicle Android system. In this project, we needed to invoke Rust code(DB) from Java(App), so we couldn't directly use the project’s source code for demonstration. Instead, we created a demo project: https://github.com/GreptimeTeam/rust-java-demo

1. I agree that using Rust doesn't necessarily mean faster performance; it simply gives you the opportunity to implement some compute-intensive modules in Rust, which is a possible approach.

2. This is a great suggestion, and we organized our project in the same way. You don’t need to use unsafe for every call. However, if you want to call JNI APIs from Rust, unsafe is required.

3. Sorry, some details were missing here. We use AsyncRegistry(Java) as an intermediary. Before initiating an async operation in Rust, we need to call Java code in advance to register a future and obtain a unique future ID. After the async execution completes, we retrieve the registered future by its ID, and then complete it or complete it exceptionally depending on the async result. You can refer to this code: https://github.com/GreptimeTeam/rust-java-demo/blob/90ffa0ba... and https://github.com/GreptimeTeam/rust-java-demo/blob/90ffa0ba...

4. This article was not generated by AI; it’s just that our official blog has a fixed template at the end. Sorry for the inconvenience.

Thanks for the clarifications. Good if you mention the background in the medium post. Otherwise it reads like a PoC demo.

5. How did you handle java local and global ref lifetimes in rust callee? Was it assumed that java caller owns all the refs and freed after the rust computation returns? Or did your calls mostly involve byte buffers and primitive types? That latter is a sweet spot but not always feasible.

Why didn't OP use Panama or the modern Java native APIs which are far better than JNI?

Android?

Proving the point about Android being Google's J++, and Kotlin their C#.

Does OP target Android?

I agree that Java + native is the way to go.

But does rust really give you an edge over C/C++?

Here is how you do JNI with C++: http://move.rupy.se/file/jvm.txt

So simple it's ridiculous!

Then you can use RegisterNatives to give C++ API to the Java side instead of the stub (Java calls C++ .dll/.so) thing...

Having written a few JNI libraries in Rust (jni-rs specifically) myself, I'd argue that it catches & prevents a lot of potential mistakes when using JNI, especially regarding error handling/refs/types. I would pick using jni-rs any day over raw JNI. The simplicity of the C JNI API is also the cause of much pain when your project expands in scope.

Your comment is exactly the reason why while I find Rust a cool language, I would be using C++ instead.

That is the systems language most JVM implementations make use of, alongside Java, and what is directly supported by JNI tooling, including on Java IDEs mixed language debugging.

And in what concerns Java, native is anyway the synonym for unsafe.

However to each their own.

C++ is a perfectly good programming language if you never make mistakes. So the problem is that of course you'll make mistakes, and the people who wrote your compiler make mistakes, and the C++ committee makes mistakes, and the language's inventor makes mistakes, if you work in a team the other team members and the tooling both also have mistakes. Other than these issues, sound choice.

You can bash C++ as much as you feel like, if it makes you happy.

I am well aware of the issues with C++'s flaws, sometimes it is easier to deal with such flaws, than adding a new layer into the C++ sandwich of debugging tools, build toolchain and IDE tooling for a given language runtime, and dragging a whole team for the ride as well.

What about removing LLVM dependency from Rust, switch to Cranelift, so that rustc isn't hindered by C++ mistakes on LLVM?

Sure, or perhaps since LLVM devs like niches so much they might decide to RIIR.

For me the big advantage of Cranelift isn't that it's written in Rust it's that they seem to have invested more into coherent semantics. I do not want to write code which is correct but is miscompiled because the compiler internal semantics are nonsense and that's a small but noticeable problem in LLVM.

You can write contorted (but safe) Rust which LLVM just plainly miscompiles, the rust layer is like "Make local variable A, and local variable B" Ok says LLVM, "and now is the address of A the same as the address of B?" "No", says LLVM those are different, not the same variable so different addresses. OK says the Rust. Now, subtract the smaller from the larger and tell me the number you got. "Zero" triumphantly proclaims LLVM having deduced that we don't need A or B so there's no need to store them anywhere, but forgotten that it promised their addresses aren't the same... Oops.

That example is silly, but it's hard to be sure how many large codebases might tickle equivalent LLVM miscompilation, which is not good.

Might be, but then don't expect any language runtime, or compiler toolchain, that is written in C++ to be taking pull requests written in Rust anytime soon, including Rust own compilers, the GCC ongoing support effort even more so, and that was my whole point.

Regardless of how easier it is to make mistakes in C++ versus Rust.

Because those mistakes get fixed on LLVM devs' time, not on parent's.

Yeah, but parent clearly enjoys the outcome of all universities and compiler vendors, that despite their mistakes, pump into optimizations passes, and advanced compiler algorithms into LLVM, even though they happen to be written in C++.

I thought JNI was being deprecated in favor of the new FFM interface.

https://openjdk.org/jeps/472

FFM is the recommended alternative, but JNI is not being deprecated. There are some things that JNI can do that FFM can't, in particular - initiating calls from native code to arbitrary Java methods. FFM only supports upcalls to Java in the form of Java callbacks that are passed to the native code. However, this is enough for the vast majority of Java-native interactions, so FFM should be preferred in new code (it's much easier to use than JNI).

Notice the delivery version (24). That was literally just sent out in March. The FFM interface was stabilized in 22. 21 is the latest LTS version which is the furthest most orgs will go.

That doesn't mean you should happily go out and build a new cathedral of technical debt.

This says:

> It is not a goal to deprecate JNI or to remove JNI from the Java Platform.

It says they want to put a safety barrier in front of JNI, so you'll have to explicitly indicate that you want a module to use JNI.

I know that's the goal, but I was doing deeper reading and didn't understand the nuances here. It felt deprecated, and I remember reading that you should prefer FFM.

That's one of the reasons I posted it. A lot of knowledgeable people here can chime in more details. And a sibling comment did!

Others should consider Chicory, enabling JVM apps to run wasm code (compiled from Rust or many other languages) at high speed, securely in-process.

JNI introduces a lot of other challenges and sacrifice in terms of portability, memory and retaining observability and control within the JVM.

Chicory even supports AOT compilation to translate wasm bytecode to JVM bytecode for native performance.

https://github.com/dylibso/chicory

WHich part of the Java app was slow for you? And did you check on FFI (foreign function interface) ?

Another useful library in this space that allows you to avoid manually writing bindings is https://github.com/duchess-rs/duchess

Should be possible with Zig as well. Is this one up to date? https://github.com/zig-java/jui

I found mixing Bun and Rust works pretty well. Bun has gotten many cool new things recently which feel great to use and it has a nice FFI API. So I have Next.js apps running in Bun runtime and anything CPU bound is written in Rust and called via FFI.

I would have expected Zig, given Bun, or using Deno instead, if Rust as extension language.

> I would have expected Zig

You can use Zig instead if it meets your needs, but last time I checked, the Zig ecosystem was lacking compared to Rust.

> or using Deno instead, if Rust as extension language

Deno? There was a thread the other day that Deno is dead. I wouldn't even compare Deno and Bun because Bun is just waaay better. Even before Bun already felt very polished and nice to use, and now it even has native postgres and s3 clients. And it doesn't try to sell you a KV database or some shit.

The point I was making was using the same language the VM makes use of, instead of adding extra layers.

I mean, I guess there would be some point if you, e.g., forked bun and wrote new APIs in Zig. But if you are using FFI it doesn't really matter because the interface is C ABI anyway; there's no extra layers in using a language different from the one your runtime uses.

JS developer use tooling older than 3 years challenge. Level: impossible.

It's not only JS. The tooling was just trash everywhere before Rust started gaining traction. E.g. for Python there is uv now, and I am definitely not going back to freaking pip.

Me neither. But only because uv is compatible with pip. I am not going to use bun / dinosaur whatever instead of node at this point.

> But only because uv is compatible with pip.

Take this: Bun is compatible with Node.

It is cozy enough in JVM/CLR land, the problem is when we need to get out of the druid mountain.

java and practical don't belong in the same sentence

Reply on news.ycombinator.com