Hi, everybody! I just wanted to clarify the curre...

# webassembly

[bashor]

Hi, everybody!

I just wanted to clarify the current situation of supporting WebAssembly in Kotlin.

We are working on a dedicated Kotlin to WebAssembly compiler backend.
So all works on wasm support in Kotlin/Native (through LLVM) were suspended.

For the new compiler backend, we have set the following goals:
• produce small binaries -- to achieve it we decided don’t use own memory manager and use builtin one instead (see GC proposal);
• fast compilation while developing -- for that we are going to be able to generate binaries directly;
• better integration with hosts, especially with their GC to avoid leaks.
Besides GC proposal we also depend on some other proposals that aren’t finalized yet. It’s hard to predict when all of them will be finalized and implemented.
We have joined WebAssembly community group to work on proposals along with other stakeholders. Also, we are collaborating with some of Wasm VM vendors to get things done as early as possible and with our interests in mind.

As well as other kotlin compiler backends it’s developed inside a public repository, so actual sources could be found here.

We’ll try to keep you up to date, stay tuned!

[Robert Jaros]

Any very approximate ETA on new backend?

[Svyatoslav Kuzmich [JB]]

We don’t have an ETA. As Zalim mentioned, it depends on quite a lot of external factors, and those are hard to estimate.

[Robert Jaros]

But perhaps months, years or decades ? :-)

[Big Chungus]

@Robert Jaros expect decades and be happily surprised when it comes any earlier 😄

[sdeleuze]

@bashor That’s great news thanks for sharing !

[bashor]

@Robert Jaros someday, after GC proposal is ready 😉

[sdeleuze]

What is the current status of the GC proposal ? Any stakeholders in addition to you?

[Svyatoslav Kuzmich [JB]]

There are 2.5 competing GC proposals. Browsers started to implement one of them, but discussion is still ongoing. First V8 prototype is almost ready, but it will be under a flag until things are sorted out.

[sdeleuze]

Ok thanks for your feedback it seems good progresses even if still frustrating to not see the GC support moving faster.

[Zach Klippenstein (he/him) [MOD]]

Is this backend being built on top of the new kotlin IR?

[bashor]

@Zach Klippenstein (he/him) [MOD] yes

[Jilles van Gurp]

Awsome. I'm curious about the long term strategy for this. Particularly interesting would be the integration into the js eco system & of course kotlin js in terms of tools, multi platform builds, etc. Ultimately, there should be little advantage in using the js transpiler over the wasm compiler.

The comments regarding GC are interesting, I think that was a reason that wasm is so far mainly used with languages that don't need that. C# blazor seems to be an exception but comes with relatively large binary sizes because they probably bundle one.

[corneil]

I believe this is going to be a game changer! I can see it in desktop, mobile, embedded and small server environments. It will be Kotlin vs Rust in this space.

[sdeleuze]

Yeah but different target since Rust targets non GC variant and is more a low level language.

[sdeleuze]

But they could be complementary

[corneil]

Exactly. It will depend on how quickly you want to develop or what your constraints are.

[darkmoon_uk]

@Jilles van Gurp You mis-spelled elimination of the JS ecosystem 👌

[darkmoon_uk]

(Honestly, it can't come quickly enough)

[Jilles van Gurp]

@darkmoon_uk, I need type safe languages to save me from typos.

[napperley]

Would be curious to know what type of GC Kotlin WASM will be using.

[elizarov]

It will be using WASM GC. The same GC that is already implemented by the corresponding execution engine (e. g. Orinoco in a modern V8).

[Big Chungus]

@bashor, from your point of view (being much more involved with wasm proposals), when would you expect to have gc ready? This year, next year, next decade? Looking for a very rough ETA

[bashor]

I hope during the next few months we will get some prototype under a flag (turned off by default) inside V8 (Chrome's engine), and I hope we will get the same inside SpiderMonlkey (Firefox's engine) at least next year. Since they would be early prototypes I can't even be sure that they will be available in release versions of browsers, even under a flag.

It's really hard to say right now when it will be production-ready and importantly when it will be widespread. I think it will take years, but (hope) not a decade.

[Robert Jaros]

It took ~3 years to create wasm itself. "Just" adding gc should not be more difficult 😉

[Jorrit]

Don't get me wrong, I'm excited about the development. But if I read this correctly, using Kotlin for WASM is pretty much a no-go right now with the LLVM version deprecated and the current work several years out. Unfortunate. I have no idea about which other proposals you're waiting for, but is there any chance for a transitional memory manager to hold us over until the host GC is widespread?

[Jilles van Gurp]

@Jorrit I think both v8 (chrome & node.js) and the firefox implementation are moving pretty rapidly. Probably if Jetbrains were doing this a few years ago (like the blazor guys) it would have made sense to invest in custom memory management & GC. But as things are evolving building on what ships with standard implementations is the best use of time. It will in any case take time for tools and libraries to come together as well. I'm guessing Jetbrains is targeting the 1.5 release with a release version of this.

[Robert Jaros]

Anyone has any experience with alternative solutions for Kotlin and Wasm? E.g. like java bytecode compilers (JWebAssembly, TeaVM)?

[bashor]

@Jorrit could you please share more about your usecase? Why it’s critical to use wasm for it?

[bashor]

IMO, the current state of wasm is not suitable yet for developing real/production web application, since calls to and from wasm are expensive. The main use cases for wasm today are reuse code written in C/C++ and number crunching.

[Big Chungus]

I can second that. I've written VDOM KMP project that compiles from common sourceSet to both JS and WASM. And WASM performs SIGNIFICANTLY slower

[Big Chungus]

Here's a compiled bundle of all three versions (js, wasm debug and wasm release) if anyone wants to see for themselves

bundle.zip

[bashor]

If you want to write a web application using Kotlin today we recommend you Kotlin/JS in the future we are going doing our best to make migration/interop between Kotlin/JS and Kotlin/Wasm smooth.

[Robert Jaros]

But products like Blazor WebAssembly gain a lot of attention right now. They will be years ahead of Kotlin if we will just sit and wait for next years 😿

[bashor]

AFAIK Blazor is far from production-ready in terms of size and speed.

[Big Chungus]

Guys, let's not kid ourselves. WASM will not be a valid alternative to JS for web UIs until direct access to DOM apis is implemented. Until then it's just a utility tool to import c libt to web and speed up heavy load processing operations.

[Jorrit]

Performance for number crunching. But it's not critical at this point, just looking around for what the end-product could look like one day (if ever) and what the right tool for the job is. Kotlin is nice to target JS now and when the time comes 'upgrade' to WASM, but if that's 5 years out there's no sense even worrying about it. Somewhat off-topic, for Kotlin/JS I wonder if this - https://discuss.kotlinlang.org/t/kotlin-js-performance/10643 - is still a relevant issue, though?

[Robert Jaros]

It's not about using it in production. It's about learning, playing and developing new solutions. And it's hard to do it, when the only tools available are deprecated and new will be available in a few years.

[bashor]

We don’t wait, we are working on Kotlin for Web, today it’s Kotlin/JVM on server-side and Kotlin/JS on client-side. Also, we are working on Kotlin/Wasm, collaborate with Wasm VM vendors to bring the future closer.

[bashor]

The migration to other Kotlin targets will be as smooth as much of code will be written in pure Kotlin. And for that, as The Community, we need cover more and more cases in pure Kotlin.

[bashor]

Please note, we are not waiting until the technology (Wasm, GC, fast interop, etc.) will be production-ready. We are working with VM vendors on prototypes on both sides, and anyone will be able to play with them much earlier, not in a few years.

[Big Chungus]

To further emphasise how unperformant WASM to JS calls are, I've ran performance test on the same application (but WASM and JS variants) compiled from the same kotlin file for the same ammount of time and here's how they compare in terms of time spent scripting:
• JS variant: 2901 ms
• WASM variant: 4316 ms

[napperley]

Some people have mentioned using WASM for Serverless, and Edge IoT, however that is best left to Kotlin/Native where those software development areas are already covered through the Linux targets, and the Linux integration that is required is already there.

[Jilles van Gurp]

@napperley for most edge computing setups, wasm is the only thing that is going to be able to run there. Kotlin native is pretty much where wasm support will be as it is basically is going to leverage llvm.

[Big Chungus]

To follow up with my previous WASM vs JS findings, it's nice to notice that WASM release bundle is MUUUCH smaller than JS:
• JS: 2.49 MB
• WASM: 424 KB
Even WASM debug bundle is just 2.85 MB, not that much larger than JS release bundle

[bashor]

@Big Chungus could you please provide more info about your measurements? What kind of operations do they perform? How did you measure the size? Is it size on disk? If so, maybe you took into account unnecessary files like .map? Have you tried the new JS compiler backend? What libraries you have as dependencies? Maybe you could share sources?

[Big Chungus]

Sure, will compile all that a bit later today

[Big Chungus]

Ok, looks like previous size measurements were a bit off, here's the update (and bundles attached):
• JS-Legacy: 226KB
• JS-IR: 122KB
• WASM Debug: 3MB
• WASM Release: 436KB
The sample project uses kvdom and forked kotlinx-html libraries and renders a simple vDOM-based app. All source code is in common code, that's compiled to both, JS and WASM. For performance test mentioned earlier I just ran "Surprise me!" button (that starts rendering random count between 100 & 1000 of elements each render cycle) for 10s on chrome dev-tools on linux machine.

Full source code is available here .

Hope that helps.

bundle.zip

[bashor]

@Big Chungus thanks for the update