[ReleaseCandidat]

It doesn't have access to _anything_ (in the browser, in other runtimes there is WASI with POSIX functions). Everything has to be imported from or exported to JS.

And currently using anything but C, C++ or Rust isn't feasible, as the runtime needed for a GC is way too big. A Haskell "Hello World" for example is about 1MB (even after running `wasm-opt` on the generated WASM binary).

[progmetaldev]

I do agree things like the GC are very large if we're comparing against standard JavaScript, so I don't believe the technology is ready for standard customer facing websites. For things like internal applications/extranet type applications, I think that the runtime download cost is minimal compared to the functionality that you are given with a proper framework (I can only speak of Microsoft Blazor, but that's just language ignorance, and I know there are other that fit the model as well). As a web developer that also writes utilities to run on the desktop, for things such as ETL or fixing bad data, being able to not switch to another language or even really framework is a huge boon for my time. I know JavaScript, but being able to rarely have to deal with it keeps my head in what I'm solving, rather than having to context switch between interface and server.

[yebyen]

> isn't feasible

This is kind of subjective, and in the context of "is this a JS killer" which is what you're answering, I'd agree, it makes sending the Wasm to the browser a bit of a non-starter that it requires a large bundle most of which is simply boilerplate for whatever runtime, without some type of local storage and persistent cache it's difficult to imagine using Ruby in a Wasm for example. If you're deploying to a container, where you're able to use a cache warmer to ensure the wasm is ready before it's called, then a 1mb binary might not be such a big issue.

(I mean, it is still a big issue because the point was fast cold starts, and big assemblies mean no fast cold starts, but again, subjective value judgments... 1mb isn't too big in many cases and I'd wager most of the cases that I'd really care about. But in a browser...)

But if you're not trying to run the Wasm in a browser then it's still potentially quite useful. You might not be running all of your Wasm in a browser and it might still be a JS killer, and all of those might not be in conflict.

[ReleaseCandidat]

Well, the main reason why I'm taking a deeper look at WASM is because I'm creating a language and compiler that is optimized to compile to WASM. While I don't think that my compiler will be the JS killer, I do think that WASM languages using a (the WASM) GC have a future.

[patmorgan23]

I mean Microsofts Blazor framework can use WASM to run C#(and its runtime) in the browser.

[titzer]

Virgil compiles to Wasm and brings its own GC as well, and the runtime is on the order of about 4KB. The GC is a simple semi-space copying collector and the compiler prunes away dead code from libraries, so binaries are pretty small. So overall I don't think this is a Wasm issue as it is mostly a language runtime size issue.

[plopz]

In the browser, can it integrate with SharedArrayBuffer, worker's postMessage or transferrable offscreen canvas?

[ReleaseCandidat]

Depends what you mean by "integrate". It always has to import or export JS functions or memory. WASM (without WASI) has no direct connection to "the outside World", everything has to be routed via JS FFI. So you can import or export a SharedArrayBuffer to communicate with JS. https://developer.mozilla.org/en-US/docs/WebAssembly/JavaScr...

But you need the WASM thread extension (for atomic access to the shared memory) to be able to use shared memory.

[connicpu]

WASM essentially can call javascript functions that were imported, and I believe javascript is able to read WASM's memory (a big help for transferring strings). If you're using something like Rust, all the glue code to call any JS APIs can be automatically generated for you with wasm-bindgen, so it really isn't a huge problem usability wise. It's just not great for performance.

[dragonwriter]

> And currently using anything but C, C++ or Rust isn't feasible

Someone should tell Anaconda that they can't do this, then: https://pyscript.net/

[tyingq]

That uses Pyodide. From Pyodide's own pages:

"At present a first load of Pyodide requires a 6.4 MB download, and the environment initialization takes 4 to 5 seconds."

So, yes, it works. But there's plenty of situations where a big download followed by a 5 second stall is a non-starter.