Should JavaScript be split into two languages?
devclass.com
devclass.com
That is, the Hotspot VM was such a phenomenal engine that lots of other languages sprung up to take advantage of that: Closure, Scala, Kotlin, etc.: https://en.m.wikipedia.org/wiki/List_of_JVM_languages . Even with the Java language itself, syntactic changes happen much more frequently than VM-level bytecode changes.
With an interpreted language like JavaScript, the dividing line is a little grayer, because the shippable code isn't bytecode, it's still just text. But it still seems to make sense to me to target a a "core", directly interpretable language, and then let all the syntactic sugar be precompiled down to that (especially since most JS devs have a compilation step now anyway). Heck, we basically already did this with asm.js, the precursor to WebAssembly.
asm.js came about because it was a very optimizable subset of JavaScript, then it was superseded by WebAssembly, then the proposal in TFA is basically asking for asm.js back, but perhaps the better answer is to make WebAssembly fully support all of what JS could originally do.
This is perhaps why as I get older I sometimes feel like I want to get out of software development and become a goose farmer like that dude on LinkedIn - lots of times feels more like spiraling in circles than actually advancing.
Look at the stronghold grip of C/C++ and how long it’s taken Rust to gain a meaningful foothold in those realms for example.
Google wanted to flat replace JS once already, that was the entire origin of Dart. They only pivoted to the cross platform mobile framework as its primary target after it failed to gain traction as a standard
Dartium was cancelled and AdWords team, having just migrated from GWT to AngularDart, saved the Dart team.
Eventually many left the team.
Somewhere at Google, Flutter started and when they decided to replace JavaScript on their original design, they decided to reach out to the Dart team.
So Dart got a new purpose in life, being Flutter's language.
In this process it was rebooted from dynamic language, into a static type one, having JIT and AOT toolchains.
How long Flutter, and by association Dart, remain relevant remains to be seen.
It was actually really fortunate in that for a long time it didn’t have a big community behind it and they just put a lot of very smart language designers on the team where they had ten years to try various approaches and learn from the mistakes of not only themselves but others without a lot of outside noise.
But there’s no other language I would prefer to write applications in. It’s just a really nice mix of ergonomic, expressive and powerful.
Additionally we already have LLVM all over the place, alongside JVM and CLR, it is the most deployed compiler infrastructure with contributions at the same level as the Linux kernel.
Javagator / Jazilla!
https://www.cnet.com/tech/tech-industry/javagator-down-not-o...
I feel like the WASM fervor has more to do with the fact people don’t enjoy using Frontend tools or JavaScript etc. vs looking at the actual utility tradeoffs
>> WASM isn’t the silver bullet everyone seems to cling to.
Copy from JavaScript to WebAssembly:
Use TextEncoder to convert a JS String to Uint8Array
Copy the bytes from the Uint8Array to WemAssembly.Memory
Copy from WebAssembly to JavaScript:
Copy the bytes from WebAssembly.Memory into a Uint8Array
Use TextDecoder to convert from Uint8Array to JS String
https://github.com/WebAssembly/js-string-builtins/blob/main/...
"the overhead of importing glue code is prohibitive for primitives such as String, ArrayBuffer, RegExp, Map, and BigInt where the desired overhead of operations is a tight sequence of inline instructions, not an indirect function call"
I guess the more elegant and universal stringref proposal is DEAD now !?
https://github.com/WebAssembly/stringref/blob/main/proposals...
I don't really mind, as it keeps the wasm bytecode cleaner.
I agree about keeping wasm bytecode cleaner. The core plus simd stuff is such a great generalization of the ARM and X86 CPUs we mostly use. The idea of gunking it all up with DOM related stuff is distasteful.
We don’t yet have consensus on this proposal in the Wasm standardization group, and we may never reach there, although I think it’s still possible. As I understand them, the objections are two-fold:
WebAssembly is an instruction set, like AArch64 or x86. Strings are too high-level, and should be built on top, for example with (array i8).
The requirement to support fast WTF-16 code unit access will mean that we are effectively standardizing JavaScript strings.
And dealing with strings isn't fun in many other languages or runtimes or OSes.
e.g.1. C# "Strings in .NET are stored using UTF-16 encoding. UTF-8 is the standard for Web protocols and other important libraries. Beginning in C# 11, you can add the u8 suffix to a string literal to specify UTF-8 encoding. UTF-8 literals are stored as ReadOnlySpan<byte> objects" - https://learn.microsoft.com/en-us/dotnet/csharp/language-ref...
e.g.2. Erlang/BEAM/Elixir: "The Erlang string type is implemented as a single-linked-list of unicode code points. That is, if we write “Hello” in the language, this is represented as [$H, $e, $l, $l, $o]". The overhead of this representation is massive. Each Cons-cell use 8 bytes for the code point and 8 bytes for the pointer to the next value. This means that the 5-byte ASCII-representation of “Hello” is 5*16 = 80 bytes in the Erlang representation." - https://medium.com/@jlouis666/erlang-string-handling-7588daa...
This refers just to Erlang's string() type, not BEAM strings in general; it's just a bad default. If you're not using binaries, you're doing it wrong, and that's exactly why Elixir's strings are UTF-8 binaries.
It supports nearly arbitrary imports of anything you want. How much more flexibility do you need? You could provide an `eval` function to run arbitrary code with a small amount of effort.
Is the problem that Emscripten and/or Rust haven't laid it all out on a platter?
Anything you can do in JavaScript, including access to the DOM, can be put into a JavaScript function. You can import that function into a WebAssembly Module, and you can use WebAssembly Memory to transfer large or complicated data as an efficient side channel. It all works.
This is what StackOverflow tells me (2020):
> Unfortunately, the DOM can only be accessed within the browser's main JavaScript thread. Service Workers, Web Workers, and Web Assembly modules would not have DOM access. The closest manipulation you'll get from WASM is to manipulate state objects that are passed to and rendered by the main thread with state-based UI components like Preact/React.
> JSON serialization is most often used to pass state with postMessage() or Broadcast Channels. Bitpacking or binary objects could be used with Transferrable ArrayBuffers for more performant messages that avoid the JSON serialization/deserialization overhead.
This feels like "we can have DOM access at home" meme.
Web Workers can't directly access the DOM in JavaScript either. This is not a WebAssembly problem. If you want a Web Worker to manipulate your document, you're going to post events back and forth to the main thread, and Web Assembly could call imported functions to do that too.
I don't even know what he's on about with Preact/React...
Save the following as "ergonomic.html" and you'll see that WebAssembly is manipulating the DOM.
<!doctype html><title>Not that hard</title>
<script type="module">
document.addEventListener('DOMContentLoaded', () => {
/* Compile this module with wat2wasm to make the binary below:
(module
(import "env" "easy" (func $easy (param i32)))
(func $run (param) (result)
(call $easy (i32.const 123))
(call $easy (i32.const 456))
)
(memory $mem 1)
(export "run" (func $run))
(export "mem" (memory $mem))
)
*/
const binary = new Uint8Array([
0, 97, 115, 109, 1, 0, 0, 0,
1, 8, 2, 96, 1, 127, 0, 96,
0, 0, 2, 12, 1, 3, 101, 110,
118, 4, 101, 97, 115, 121, 0, 0,
3, 2, 1, 1, 5, 3, 1, 0,
1, 7, 13, 2, 3, 114, 117, 110,
0, 1, 3, 109, 101, 109, 2, 0,
10, 14, 1, 12, 0, 65, 251, 0,
16, 0, 65, 200, 3, 16, 0, 11,
]);
const imports = {
easy(arg) {
const div = document.createElement("div");
div.textContent = "DOM this: " + String(arg);
document.body.appendChild(div);
}
};
const module = new WebAssembly.Module(binary);
const instance = new WebAssembly.Instance(module, { env: imports });
instance.exports.run();
});
</script>
I'd like to believe a simple standalone example like this would be enough to get people to shutup about the DOM thing, but I know better. It'll be the same people who think you need to link with all of SDL in an Emscripten project in order to draw a line on a canvas.
> This feels like "we can have DOM access at home" meme.
And I'm sure somebody (maybe you) will try to move the goal posts and claim some other meme applies.
Thanks for confirming that WebAssembly still cannot manipulate DOM in 2024.
It can only call custom javascript functions that manipulate DOM AND I need to write some arcane function signature language for every DOM manipulating function I want to call.
I'll give another 4 years and see if they fixed this.
You really don't know that you can create WebAssembly in other languages?!? I used WAT to keep the example short, but that's clearly lost on you.
> I'll give another 4 years and see if they fixed this.
In that time, there are lot of things you could be learning. Embracing ignorance and belligerence isn't like to serve you well in the long term.
Around 10 years ago, I was having lunch in a food court and overheard "Luckily I don't have to use javascript, just jquery".
Around 5 years ago, a co-worker admitted he still had issues distinguishing what functionality was python and what came from Django (web framework), despite having used them both daily for years. He thought it was because he learned both at the same time.
I wouldn't be surprised if this was more of the same, and just getting worse as we pile more and more abstractions on top.
but what bothers me a bit is that this example still uses custom javascript code.
i tried to find and answer but essentially what appears to be missing is the ability to access js objects from wasm. to access the document object it looks like i need a wrapper function in js:
jsdocument(prop, arg){
document[prop](arg)
}
maybe more elaborate:
callDOMobj(prop, arg, prop2, arg2){
document[prop](arg)[prop2](arg2)
}
getDOMobj(prop, arg){
var len = objlist.push(document[prop](arg))
return len-1;
}
callDOMobj(pos, prop, arg){
objlist[pos]["prop"](arg)
}
solving this problem is what i mean when i ask for direct access to the DOM. i believe such an interface should be written only once so that everyone can use it without having to reinvent it like it appears to be necessary at the moment.
It's nice of you to say so. Thank you.
> can you see what i am getting at here?
I mostly can, but I'm not sure we're clear what we're talking about yet.
I see a lot of people who repeat something about "WebAssembly isn't usable because it can't manipulate the DOM". Ok, so I show an example of WebAssembly manipulating the DOM. That should put that to rest, right? If not, I'm curious what they meant.
> building up some kind of API that allows me to access and manipulate any DOM object via a set of functions that i can import into WASM to work around the fact that i can't access document and other objects directly,
This is a shortcoming in the language implementation, or the library for the language. The machinery is already there at the WebAssembly level. If your language is low level (Rust, C, or C++), and doesn't have what you want, you could roll your own. If your language is high level (Python or Lua), you're at the mercy of the person who built your version of Python.
The core of WebAssembly is a lot like a CPU. It's analogous to AMD64 or AArch64. It'd be weird to say you need changes to your CPU just to use a function called `getElementByName()` or `setAttribute()`. Some WebAssembly extensions have added features to make that "CPU" more like a Java style virtual machine. There are (or will be) garbage collected references to strings, arrays, and structs. This might make it better for implementing Go and Java style languages, and it could help with a fresh implementation of Python or Pike too. And maybe some of those changes will give controlled access to JavaScript style objects.
There's a last discussion to be had about performance. Maybe the bridge between WebAssembly imports and exports is too slow for intensive use. That's a debate that should be backed up with benchmarks of creative solutions. Maybe accessing JavaScript strings is so common, so important, and so slow that it really does require an enhancement to the standard.
The current active proposal for it is the Component Model: https://component-model.bytecodealliance.org/design/why-comp....
https://blog.stenmans.org/theBeamBook/#_compiler_pass_core_e...
But what if it was?
What I would like to see is:
- a bytecode "language" that roughly corresponds to Javascript semantics, and that is what the engines interpret (and JIT compile)
- browsers still include a compiler to compile JS sourcecode to the bytecode. Possibly wasm could work, although it would need a lot more functionality, like native support for GC, and DOM access, etc.
- browsers include a disassmbler/decompiler to improve debugging the bytecode
Then simple sites, and development can use plain JS source, but for higher performance you can just deploy the pre-compiled bytecode.
Java is a product of the JVM, which was the innovation, not the reverse. A successful language moving post-success to a new byte code format would be as far as I know unprecedented.
The idea that JavaScript is an interpreted language is also fairly shaky. It’s JIT compiled as soon as it arrives to your browser. Honestly, a modern JS engine is not different from any other VM.
The question as you rightfully pointed really is what do you send to the browser and under it lies the fundamental question of what is a browser actually. Is it a way to browse hypertext content or a standardised execution environment?
Splitting the language might make sense from an engineering perspective but what about all the extra energy and bandwidth that will be needed?
1) JavaScript, the original assembly language of the internet, does not need new language features.
2) JavaScript, the front-end web development language is a fractal of infinitely many sub-languages that transpile back to ES5.
The proposal, as I read it, is: Let's stop adding front-end web features to the assembly language; it doesn't get easier, better or faster if we change the underlying, slowly adopting and hard-to-optimize foundation.
When you want a new language feature, add it to the fractal part that transpiles back to the part well-supported and highly optimized in existing runtimes. The only loss is that you need to transpile, so your build pipeline becomes non-trivial. But it's either "code assembly" or "get a modern build tool".
There are still some new language features that need to be transpiled, but most projects do not need to worry about transpiling cost/let/arrow functions/etc.
I mean even newer features like nullish coalescing and optional chaining are at 93-94% support.
At the end of the day, I would say tools like babel for transpiling are less and less important. Yes, you still use a bundler because the web has a lot of runtime constraints other native applications don’t have (gotta ship a tiny bundle so the page loads fast), but it’s better for the language features to be implemented in the VM and not just faked with more JS.
I did assess the ES6 coverage of ~97% a month ago.
I just evaluated that while it sounds high, 3% of people is a lot of people to cut off if your JavaScript is essential.
E.g. Firefox sits at ~2.7% browser market share. (Not incidentally the part that doesn't support ES6, but it's a demography the size of my own.)
These are probably the 3% that won’t affect your business much. They’re more likely to be on older hardware and also have less discretionary income. Or browsing on really weird hardware that is also unlikely to lead to a sale.
JavaScript for some scripting, any other language for bigger applications.
But there's still far to go. Large parts of the browser API are still not directly available in WASM.
I very much look forward to WASM reaching stability. It's very enjoyable to run Rust code in the browser.
WebAssembly can call arbitrary JavaScript through imports. You could literally provide an `eval` function if you were motivated to.
Providing access to an already proven DOM would be the better solution.
A) Get your hands dirty and write what you want. Once.
B) Chant along with the mob who doesn't even understand what they're asking for.
C) Wait several years for some super complicated solution to be designed by committee.
I wouldn't even want direct access to the DOM if we had it today. The DOM as an API is atrocious.
Instead, I want a set of nice functions that do things like put a graphical chart on the page - all in one call. Or one call to pass a bunch of 3D triangles or splats to visualize in a WebGL canvas. Or one call to play some audio samples. Or a function to poll for recording audio. And so on...
I choose option A.
wait for a framework that implements option A.
if option A works, why aren't there any frameworks yet that implement it?
maybe all the framework devs are waiting for C?
but why?
you could be right about A but at present the majority view seems to be that C is the right option. which is what pushes me into going with B because i have no interest in developing my own framework.
if a framework appears that implements option A i'll gladly consider it. (just as long as it isn't tightly coupled with a backend)
The download isn't much different to a typical website. That Flutter demo in wasm is 2 megabytes.
Avalonia UI's WebAssembly uses canvas in C#: https://avaloniaui.net/
Uno Platform's WebAssembly implementation uses the DOM rather than drawing to canvas: https://platform.uno/
Uno's philosophy is to use platform native controls. The benefit is that you get platform native characteristics, the cost is it will never be exactly the same in each browser and platform.
A 2mb base floor before any code or assets is not acceptable for most use cases.
A random drawn rectangle is not a UI, it’s not accessible, not inspectable, not part of the de facto OS native toolkit.
If all we wanted is a random cross-platform canvas element to draw onto from a vm, it could be solved in a weekend. There are million examples of that.
Of course it is. All screen based user interfaces are blinking lights.
> it’s not accessible
It's best to read the documentation first. It's a low effort thing to do:
https://docs.flutter.dev/ui/accessibility-and-internationali...
https://medium.com/flutter/accessibility-in-flutter-on-the-w...
> The Flutter team would like to eventually turn the semantics on by default in Flutter Web. However, at the moment, this would lead to noticeable performance costs in a significant number of cases, and requires some optimization before the default can be changed
But good. That's a kind of progress.
did you mean there are snappy webapplications running in WASM? if you have any examples, i'd be curiuos to learn more.
> [...] go through the javascript host, which is slow
And now you admit:
> this is going beyond my level of experience [...]
> how much slower, i don't know
I guess people just repeat what they hear without questioning or understanding it, and then it becomes dogma.
> did you mean there are snappy webapplications running in WASM?
No. I meant that all existing web apps go through the "javascript host", using JavaScript. So if any of them are fast enough, and some certainly are, the problem isn't the "javascript host".
i am only talking about webapps running inside WASM. are there any WASM based webapps that are as fast as pure js webapps?
Lol, your question asked me what I meant. I told you what I meant.
> are there any WASM based webapps that are as fast as pure js webapps?
You can browse links from Google for examples and benchmarks. Maybe one of these will scratch your itch, but I won't vouch for any of them:
https://madewithwebassembly.com/
But really, JavaScript and WebAssembly are both very fast. I don't think speed is the reason to choose one or the other.
For me, I like WebAssembly because it lets me program in languages other than JavaScript. JavaScript makes me want to scratch my eyes out.
That doesn't have to be true.
Eventually WASM will get direct access to the full browser API, without going through JavaScript.
The browser exposes a browser API to the JavaScript VM it hosts, so things like the DOM are available.
Those things aren't available in other JavaScript VMs, like Node. (There's no DOM to interact with.)
And they're not yet available in the WASM VM in the browser, either.
The reason is that the WASM APIs/ABIs have not stabilised. It takes time to make right, but there is progress.
Eventually WASM will get direct access to the full browser API, without going through JavaScript.
well, that is what i am waiting for. my point is that it's not the case yet, while the gp seemed to suggest that it's not needed because access through the host is available
I also have to wonder if people are excited about replacing Javascript, why they would want to have HTML/CSS/DOM on top of WASM. A different front-end UI tech could be much better than slow, old DOM.
Keep the single threaded event loop approach but kill the JS semantics.
https://developer.chrome.com/blog/wasmgc/
https://v8.dev/blog/wasm-gc-porting
But languages like C# want more features in WasmGC:
https://github.com/dotnet/runtime/issues/94420
No direct DOM access yet. You still have to use JavaScript glue code to get at the DOM.
I want fixed-size buffer-backed structs for JS. Basically a DataView as a C struct. This would massively benefit interop and solve some shortcomings of DataView.
There was a proposal for a binary AST for JS several years ago [1]. Why not just use that as JS0? It's separate and can offer new possibilities as well.
How would this be useful for a JS0?
And Scala.js has shipped it. [1] Although technically experimental, it has no known bugs and it has full support of things like manipulating DOM objects from Scala.js-on-Wasm code.
[1] https://www.scala-js.org/news/2024/09/28/announcing-scalajs-...
Also... do you really think its wise to rewrite v8 to target WebAssembly?
Here's a demo of Dart and Flutter compiled to WebAssembly:
https://flutterweb-wasm.web.app/
WebAssembly enables you to use any language. And when you can use any language, why would you use JavaScript?
Google has started migrating parts of Google Sheets to WebAssembly. They're compiling Java to WebAssembly and seeing a 100% performance increase:
https://web.dev/case-studies/google-sheets-wasmgc
Amazon has been migrating its Prime Video app from JavaScript to WebAssembly. They're compiling Rust to WebAssembly and they've seen increased performance and lower memory usage:
https://www.amazon.science/blog/how-prime-video-updates-its-...
On a serious note I don’t see the point in turning browsers into an OS on top of the OS. I know it’s some kind of Google wet dream so they can suck up even more data than they already do but still. If you want to ship applications, just do that. The sandboxing should be done at the OS level where it belongs.
I'd rather we just move to native cross platform applications and stop using a document browser to build interactive applications.
What's more likely is that all of this will probably be eclipsed by LLM and virtual assistants - which could be controlled by native apps with a dynamically generated GUI or voice.
I think APIs exposing data and executing functions will fundamentally change what we think the web is.
Here you go. Do both native and wasm:
Flutter example:
It’s unfortunate there isn’t a more native “app like” UI toolkit. Especially on mobile, web apps generally are bad and a lot of the reason is trying to shoehorn an app experience onto the dom.
If you're using Safari it's true that Safari's WebAssembly implementation is behind the other browsers. But that's a Safari problem more than a WebAssembly problem.
Go back to garbage "cross platform" UI toolkits and having to help users manage software dependencies on their machine? No thanks.
Cool stuff, let's kill web for good.
You can choose to just work with the core and maybe a minimal sugar library. Which will probably be faster and don't include "all the language features that were forced upon you".
- at work they expect me to write code with the latest features
- my colleagues write code with the latest features and I have to review/extend/build upon using the same style
- the community, books, copilots, LLMs, libraries, tooling are "forcing" all that new stuff upon me
- etc.
Transparent polyfills (or anything that really needs to be loaded "first") also don't work, since async means you can't specify the order to load things. This means that every single module you write has to explicitly mention which exact polyfill it's going to end up using (hope you don't accidentally skip one or specify a conflicting polyfill) ... or you just abandon modules and load your polyfills with a normal script (which means that now your source is a bastard mixture).
Lazy imports are technically possible via dynamic imports, but unnecessarily annoying and break in all sorts of places. Granted, the standardization of the "leaky browser abstraction" makes this pretty awful regardless.
Yet every language has either that or BigDecimal. Even if Google's frontend devs haven't found a use, there also exist JS devs outside of Google who certainly have found uses (though possibly more of them on the backend).
Similarly, not every developer has a compilation step in their JS work. And there are places where you can't have one, e.g. in the browser console. Develop the language instead of tons of incompatible tools.
in pike, bigint and int are integrated in such a way that the transition is automatic. there is no overflow but as long as the values fit in an int, that is used internally to keep the code fast.
However, that works for int and bigint, but Number (double precision) can represent numbers that BigInt can not, and BigInt can represent numbers which Number can not. There isn't a graceful way to automatically promote/degrade one to the other in all cases, and a silent conversion will do the wrong thing in many cases.
This is where you say something about "exact" vs "inexact" as though that will hand wave it away.
btw, i just checked, typeof() no longer shows the difference between int and bigint. it did in the past if i remember correctly
Saying, "That's nice." is a cliche condescension. You're free to disagree, but I think his intent was clear.
I'm not familiar with this debate, but how is that a hand wave? The article describes a reasonable-sounding way to extend the tower with a second dimension of precision. Following those rules, you would never just convert between bigint and float, but an expression involving both would output a float.
float(0.5) +
bigint(9007199254740993)
== float(9007199254740992)
Basically, ULP-level inaccuracy is a problem inherent to having float at all, even without bignum interactions. They would be a menace even if you had a pure tower from 32 bit int to double to complex to more.
The real point is that you can get some non-intuitive answers from letting that numeric tower make conversion decisions for you. It's just a rule, and it's not an amazing rule.
Then it's even less of issue. Yes if you convert to a float you get rounding, what did you expect when you introduced a float?
It's somewhat unintuitive but that's the nature of floating point.
> The real point is that you can get some non-intuitive answers from letting that numeric tower make conversion decisions for you. It's just a rule, and it's not an amazing rule.
But again, you can have the same kind of issue without bignums. It's not a tower problem it's a float problem.
And that says nothing about whether implicit conversions are a good idea or not.
But more importantly, I'm saying that the problematic rounding can occur even if your tower does not have both bigint and float. It can happen even if every layer can completely represent every value of the layer above it. Do you have any complaints that are unique to a tower that has both bigint and float, and don't apply to towers that only have float?
To elaborate on that, an implicit cast directly from a single bigint to a single float won't happen with the rules in the wikipedia article. You'd have to do something like bigint+float, which can have horrible rounding errors, but those horrible rounding errors are also present in float+float.
And you can even have these problems without a tower. So I don't see how the bigint and float scenario is an argument against towers.
N < Z < Q < R < C < H
ℕ ⊂ ℤ ⊂ ℚ ⊂ ℝ ⊂ ℂ ⊂ ℍ
But no programming language can have "Reals" (they aren't computable), so floats are a common/useful approximation. And in actuality `bigint?` doesn't imply `floating?`, and `floating?` doesn't imply `bigint?`. Neither is a strict subset of the other, and because of this you can easily find examples where implicit conversion does something "questionable". You've made it about rounding errors, but I'm trying to criticize something about pretending they are subtypes/subsets. Claiming it's a tower and hand waving about exact/inexact doesn't make it a tower, and so I think implicit conversion for these is a poor choice.
You can have little subset relations for implicit conversions:
float32? implies float64?
float64? implies complex64?
float32? implies complex32?
complex32? implies complex64?
fixint? implies bigint?
fixint? implies rational?
bigint? implies rational?
fixint? implies float64?
> I'm trying to criticize something about pretending they are subtypes/subsets. Claiming it's a tower and hand waving about exact/inexact doesn't make it a tower
I thought we established right away that it's not a single tower. The description in the wikipedia page is two towers with links between them. (Or at least it's two if you don't waste effort on things like having both float64 and complex32.)
But I don't see any hand waving. The relationships and conversions are very clear. That's why I interpreted your complaint as being more about the specific operation. So with your correction, I need you to explain where you see hand-waving.
If you just don't like the name "Tower" for an implementation that has both bignums and floats then okay I agree I guess?
Where did we say that? The first picture on the Wikipedia page shows the tower as a linear stack of items from set theory. The Scheme predicates are named similarly. This is the appealing myth.
> The description in the wikipedia page is two towers with links between them.
Not on the page I'm seeing. Are you reading the English page? At the bottom, I see a tree of abstract types (sets).
This shows that you can traverse (Integer to Rational to Real) and (Float to Real) to find the common abstract type Real. But there isn't actually a Real type you can do operations with. You've got concrete BigInt and Float64, and even if Real is implemented as a C-style tagged-union of the two types, you still need to pick one or the other for doing operations like addition. Then the Scheme standard says stuff like, "try to be exact when you can, but inexact is ok sometimes". So all the set theory justification is out the window, and it's really just an ad hoc rule.
It's just not as elegant as it seems, and it gives an unsound justification to making implicit conversions.
> If you just don't like the name "Tower" then okay I agree I guess?
Please don't do that. I've tried to clarify details in response to your questions, but if you're just going to dismiss it with some snarky crap like that then you can go fuck yourself.
Reply if you want, but I'm guessing we're done here.
In the section where the wikipedia page talks about exact and inexact, the specific thing you were calling out, it says "Another common variation is to support both exact and inexact versions of the tower or parts of it; R7RS Scheme recommends but does not strictly require this of implementations. In this case, similar semantics are used to determine the permissibility of implicit coercion: inexactness is a contagious property of numbers,[6] and any numerical operation involving both exact and inexact values must yield inexact return values of at least the same precision as the most precise inexact number appearing in the expression, unless the precision is practically infinite (e.g. containing a detectable repetend), or unless it can be proven that the precision of the result of the operation is independent of the inexactness of any of its operands (for example, a series of multiplications where at least one multiplicand is 0).".
I want to especially highlight the phrase "exact and inexact versions of the tower or parts of it" Which I then reacted to by saying "The article describes a reasonable-sounding way to extend the tower with a second dimension of precision." Once you have two dimensions it's no longer a single tower. I thought that was the common ground that we were talking on, that if you use that method it's not a true tower anymore.
> Please don't do that. I've tried to clarify details in response to your questions, but if you're just going to dismiss it with some snarky crap like that then you can go fuck yourself.
That wasn't snark. I am really trying to understand your argument, because it looks like we've been talking about different things the entire time.
I had thought we established from the very start that the description on the wiki page wasn't actually a single tower. If you are still trying to convince me it's a more complicated graph, then I agree with you, and I don't understand how we got so far without that being clear. Sorry for sounding reductionist about it.
So please, honest question for clarification, do you object to the graph of number types described by that paragraph, do you object to using the word "tower" to talk about it, or do you object to both? Please don't get mad at me for asking, or think I'm trying to dismiss you.
And if someone builds a pure tower that goes int32, double, complex, quaternion, do you think that's inherently self-defeating because it can't live up to the promises of a tower? It doesn't have the issue of floats versus bignums; it's strict subsets all the way down.
> do you object to using the word "tower" to talk about it
No, I don't really care about the terminology, except when it helps to communicate.
> do you object to the graph of number types described by that paragraph
I think the problem boils down to using a flawed analogy to arrive at a conclusion and then pretending the conclusion is sound and elegant. There are really two things going on:
First, we've got a tower, or tree, or DAG of "abstract" types. These are mathematical constructs or Platonic ideals. So you can build a tower that says "All Integers are Rationals" and "All Rationals are Reals". And it's supported by Set Theory! So you conclude that you can use an Integer anywhere that a Rational or Real is allowed. Then, knowing that we're going to apply this to a programming language, you add "All Floats are Reals". Fine, we've got abstract Floats, and it looks lovely.
Second we've got actual "concrete" data types. These are things like Float64, Int32, or BigInt. Importantly, you can't have an implementation of Real anything. In general, Real numbers can't be processed on a Turing machine. You can have a tagged union of Computable things, but that's not really the same as "Real" in bold quotes.
Ok, so the mistake comes when you try to combine those first and second sets of things. We say concrete BigInt is like the abstract Integers, and concrete Float64 is like the abstract Floats. So far so good. Then we look at the abstract tower, we decide that Integers and Floats need to become Real, so we say BigInt and Float64 need to use Reals to get a common type. But there is no common type. We said the concrete types are analogous to the abstract types and made an unsound conclusion.
Finally, we write the compiler, and reality hits us. So we go back to the standard and add some bits about "Some things really should be Exact. Conforming implementations should try to avoid Inexact when they can." It's not a separate tower - it's a bandaid for flawed logic.
Anyways, this is all a bit too philosophical. I'm not actually passionate about it, but our discussion kept going, and you kept asking, so I kept trying to explain. Most people like implicit conversions in their programming languages, and so you've got to make up some rules. I just don't like pretending the rules are not ad hoc, and it's nothing a smug lisp weenie should really be smug about.
Assuming the obvious implementation of complex and quaternion built on two or four doubles, it's fine. Each type represents a set that is a proper subset of the next type in the list.
Annoyingly, it'll all go to crap if you have int64 though.
The numeric tower in Scheme describes general number types with above of in the tower graphic (in the Wikipedia article) meaning subtype of. double precision floats and arbitrary precision integers are representations of numbers. Both would also be Real numbers.
If performance and complexity really are the primary concerns then the language must stop pandering to children. We already know what high performance looks like. I wrote about it here: https://github.com/prettydiff/wisdom/blob/master/performance...
If the goal really is higher performance and lower complexity the most desirable solution is to create a new language with forced simplicity directly in the image of JavaScript, where simple means less and not easy, and transition to that new language slowly over time. Yes, I understand Google went down this road in the past with Dart, but Dart solved for all the wrong problems and thus was dead on arrival. Instead the idea is to take something that works well and shave off all the bullshit reducing it down to the smallest possible thing.
Forced simplicity means absolutely ignoring all vanity/stylistic concerns and instead only focusing on fewer ways of doing things. As an example consider a language that requires strong typing like TypeScript and thus thereby eliminates type coercion. Another example is operators that have a single purpose (not overloaded), single character, and no redundant operators.
Will there be a lot of crying about vanity bullshit... yes, absolutely. Just ignore it because, you cannot reasonably expect to drive above 120mph on a child's tricycle. If people wish to offer their own stylistic considerations they should include performance metrics and explanations how their suggestions reduce quantity of operations without unnecessary abstraction.
Language Evolution: Problems, and What Can We Do About It? - https://news.ycombinator.com/item?id=41795190 - Oct 2024 (1 comment)
Proposal of JavaScript becoming a compiled language: JS0 and JSSugar - https://news.ycombinator.com/item?id=41764825 - Oct 2024 (2 comments)
And btw, the TypeScript tooling scene is far from being able to get standardized. TypeScript is basically a Microsoft thing, and we don't see a single non-official TypeScript tool can do type-checking. And there's no plan to port the official tools to a faster language like Rust. And the tsc is not designed for doing tranditional compiler optimizations. The TypeScript team made it clear that the goal of tsc is to only produce idiomatic JavaScript.
I agree that the tooling/UI around this could be better, but by focusing on this approach, things like Typescript get better as well.
Are those being supplied with every website you use?
If the browser starts treating JS as assembly, then there would probably be a greater onus for features like this.
For example: TypeScript's sourceMap [1], Elm's time-travelling debugger [2], Vue.js DevTools [3], just to name a few I've tried. Especially well-typed languages tend to behave well at run-time once they pass type-checking. Or rather, I have not made enough front-end code to discover transpiler bugs.
[1]: https://www.typescriptlang.org/tsconfig/#sourceMap [2]: https://elm-lang.org/news/time-travel-made-easy (2014 [3]: https://devtools.vuejs.org/
So if I agree with GP then I just haven't found the right tooling yet?
As easy, certainly. But how are they easier?
Elm's debugger lets you step forwards and backwards in the application's state.
TypeScript's type system lets you catch bugs before you run the code.
Vue.js's DevTools extend the browser's with a component-based overview, so you can interactively see what's going on at a high level of abstraction. (I'm sure something similar exists for most frameworks similar to Vue.js, and possibly even frameworks made in vanilla ES5, I'm just picking one I've tried.)
With vanilla ES5 you get interactive debugging.
Seriously frontend is already the most fragmented and fast changing area of web there is. Don’t split the language.
Who cares? If backwards compatability is maintained then this fails to have any impact on my experience as a developer. It sounds like the VM maintainers are busy making their own lives hell. Not my problem.
I do. Maybe if someone programs in one language it's okay for them to keep up with language changes, but if you have to constantly juggle multiple languages it becomes a real chore to stay up to date with every one of them.
Thankfully.. both maintain reasonable backwards compatability where security is not otherwise implicated.
You still need to be aware of them when you encounter unfamiliar syntax.
A proposal rooted in attempts to improve the language would be one thing. This appears to be about Google struggling with technical competencies and not having the budget to do the right thing.
“A Google engineer presented … JavaScript VMs (virtual machines), they say, are “already very complex because of pressure to be fast,” which compromises security, and “feels especially bad” when new features do not get adoption.“
“The foundational technology of JavaScript should be simple, according to the proposal, because security flaws and the “complexity cost” of the runtimes affects billions of uses, whereas the benefits are restricted to developers and applications that actually use that complexity to advantage.”
Maybe try something besides c++ for V8 if you are having security issues?
The apathy towards proper tail calls in V8 leads me to distrust Google’s language proposals. But now that abdication appears to possibly have been a canary? Perhaps even prior to the pandemic they couldn’t keep up with maintaining the V8 C++ codebase, and that’s why PTCs got skipped?
By the way, what happened to Golang and Dart?
I think if it was that simple, it would be done that way already (maybe it is, for some features). Two big arguments for doing the "desugaring" offline are the (1) speed, and (2) security of the browser. Those two things also conflict somewhat if addressed on the client, since faster but more complex compiler code increases the surface area for potential exploits.
But if you do this compile step offline, you don't need to worry about compromising the performance or security of the browser.
The problem is distributing the runtime(s). By having developers transpile to a small core, anyone can freely invent new language features without waiting for the rest of the internet to download support for them.
If you read the original slides from the proposers, they're presenting a framing where there is an inherent tension between "serving the user" and "helping the developer". They argue that there is too much of the latter, and that a formalized splitting should push more to the former.
From an end-user perspective, it definitely makes more sense that the js doesn't have to be transformed locally before it can interpreted. I think your suggestion is not compatible with the motivations of the proposal.
JS0 should be a subset of current JS
JS1 should be current JS
JsSugar should be current JS plus future features
BigInt failing to materialize I think has more to do with ergonomics around it, they’re a bit unwieldy and there aren’t able to be used with the built in Math object functions.
They also have zero JSON support out of the box which is a huge miss.
Honestly it should have been roadmapped to replace the built in Number type
But the idea is that it should have been proposed with a roadmap of what it would look like to have it eventually supplant Number
I like that JavaScript now has modules/imports, destructuring, Proxies, async/await, etc. These were all new features at one point, But yeah, why did Symbol.species get in? Seems like it’s to enable some odd subclassing pattern? I’m an anti-OOP zealot, so my hot take would be that maybe OOP subclassing is unnecessarily complex already, so stuff like that shouldn’t make it in. We got the OOP syntactic sugar, which is enough. Stop there.
How much of the extra complexity is from stuff like that that is rarely used? Maybe we just need to be a lot more conservative about what makes it in, but stopping changes and forcing everything into more tooling complexity is not the direction I’d like to go in. We need to reduce tooling, not increase it.
It is now becoming rare when I see any serious project that still uses JavaScript, everyone I know is using TypeScript and I don't recall any job posts not requiring TypeScript. What are the standards bodies doing? They are still implementing hacks upon JavaScript instead of seeing the writing on the wall.
Maintaining JS engines is difficult because of the old stuff that few developers actually use, it would make a lot more sense to start deprecating those features and adding the new ones developers actually want.
I started using alternatives to NodeJS because I don't feel like I should subject myself to a compilation step if I don't have to.
And it was used by some browsers, there was just no consensus between different vendors due to politics. The problem largely solved itself by.. only one vendor remaining, chromium.
If anything, I expect those existing VMs to slowly be replaced by WebAssembly due to how crucial and complicated that very specific sandbox requirement is - and how useful that is once you have it working reliably.
Personally I never want to run untrusted code on any of my computers outside of a robust sandbox. I look forward to a future where every application I might install runs in a sandbox that I can trust.
The more important thing to consider, however, is the fact that CLR, JVM, etc. provide internal memory safety whereas Wasm runtimes don't.
e.g. a C program that goes sufficiently out of bounds on an array is guaranteed to segfault in the C runtime, but that runtime error does not necessarily occur on a wasm target. That is to say, the program in the sandbox can have totally strange runtime behavior -- still, defined behavior according to wasm -- although the program has undefined behavior in the source language. In the case of JVM languages, this can't really happen.
https://news.ycombinator.com/item?id=9732827
The Web is an evolving system too large and long-lived for any single company, stable consortium, or standards body capable of doing the deed to do it, so none of Java, Flash (AVM), .NET/CLR, NaCl/PNaCl, Dart, and others I have forgotten about ever had a chance to take over.
JS got out first and evolved through several jumps into https://asmjs.org/, a typed (as in static types) subset suitable with AOT+JIT techniques of hosting near-native-speed code such as Unreal Engine 3. https://brendaneich.com/2013/03/the-web-is-the-game-platform...
Java was mismanaged as a plugin (and only ever a plugin -- no deep or even shallow browser integration worth talking about) by Sun, who tried getting it into Windows after Microsoft was killing Netscape (Microsoft then killed Java in Windows, pulled trigger on .NET; Oracle later bought Sun).
Flash had its day but fell to HTML5 and fast JS, Adobe threw in the towel well before Wasm announcement, even salted the earth re: good Flash tools instead of retargeting them at the Web.
Google was a house divided all along but had absolutely no plan for getting PNaCl supported by Apple, never mind Mozilla or Microsoft. I told them so, and still get blame and delicious tears to drink as I sit on my Throne of Skulls, having caused all of this by Giant-Fivehead mind control (testimony from one of my favorite minions at https://news.ycombinator.com/item?id=9555028).
https://news.ycombinator.com/item?id=1905155
https://kripken.github.io/talks/2020/universal.html#/ (from Alon Zakai in 2020)
I'm not a huge fan of WASM but it's easy to see that the authors would clearly not want to leave control in the hands of Microsoft or Oracle (and as a result all of us are hostages to Google instead because of evil that is Chromium).
https://ecma-international.org/publications-and-standards/st...
> As told in JavaScript: The First Twenty Years, Brenden Eich joined Netscape in April 1995.
> [..]
> However, Eich didn’t think he’d have to write a new language from scratch. There were existing options available — such as the research language, Scheme, or a Unix-based language like Perl or Python. So when he joined, Eich “was expecting to implement Scheme in the browser.” But the increasingly fractious politics of the software companies of the day (it was, basically, everyone against Microsoft) soon saw the project take a more creative turn.
> On 23 May 1995, Sun Microsystems launched a new programming language into the world: Java. As part of the launch, Netscape announced that it would license Java for use in the browser. This was all well and good, but Java didn’t really fit the bill for the web. Java is a general-purpose programming language that promised Write Once, Run Anywhere (WORA) functionality, but it was too complicated for web designers and other non-programmers to use. So Netscape decided it needed a scripting language, which was a trendy term at the time for a smaller, easier to learn programming language.
There's a whole lot more interesting stuff but I think that part directly answers most of what you're wondering.
The JVM and CLR are poor compilation targets for C and C++, because those languages weren't designed to target those runtimes and those runtimes weren't designed to run those languages. (C++/CLI isn't C++.) It's possible to get something working, and a few people have tried, but you run into a lot of impedance mismatches and compatibility issues. I think you would see people run into a lot more problems trying to get their code running on the JVM or CLR than they in fact run into trying to get it running on WebAssembly. (Though I think the CLR is less bad about this than the JVM.)
As for the idea of using LLVM bitcode as an interchange format, we don't have to guess how that would have gone, because it was actually tried! Google implemented this in Chrome and called it PNaCl, and some sites and extensions relied on it for a while. They ultimately withdrew it in favor of WebAssembly. I don't understand all the reasons why it failed, but I think part of the problem is that it ran into a bunch of "the spec is whatever LLVM happens to do" type issues that were real problems for would-be toolchain authors and made the other browser vendors (including Apple, LLVM's de facto primary corporate sponsor) reluctant to support it. WebAssembly has a relatively short and simple standard that you can actually read; writing a WebAssembly interpreter is an undergraduate exercise, though of course writing a highly performant one is much more work.
Also, as far as I can tell, LLVM hasn't at all been optimized to death for the use case of runtime code generation, where the speed of the compiler is about as important as that of the generated code. The biggest dynamic language I know that uses LLVM is Julia, which is a decently big deal, but the overwhelming majority of LLVM usage is for ahead-of-time compilation of languages like C, C++, Swift, and Rust.
On a bigger-picture note, I'm not sure I at all understand why adopting an existing bytecode language would have made things easier. Yes, it would have been much easier to reuse existing Java code if the JVM had been adopted, or to reuse existing C# code if the CLR had been adopted, but those options are mutually exclusive; the goal was something that would work at least okay for all the languages. Python doesn't have a stable bytecode format, and Rust and Haskell compile to LLVM bitcode (which LLVM has no problem lowering to WebAssembly since WebAssembly was designed to make that straightforward), so I don't see how those languages are in any way disadvantaged by the choice of WebAssembly as the target bytecode language instead of some alternative.
Or are your concerns about I/O? That's a bigger can of worms, and you'd need to explain how you imagine this would work, but the short version is that reusing the interfaces that existing OSes provided would not have worked well, because the browser has a different (and in many ways better) security model.
Look at the Java bytecode, and you'll see it features such things as a goto with an arbitrary offset: https://en.m.wikipedia.org/wiki/List_of_Java_bytecode_instru...
They had to build a verifier that attempts to ensure the bytecode isn't doing anything bad. That proved to be fairly difficult, and comes at a considerable cost.
Also, "the proposed solution is not to backtrack on existing features" makes very little sense. If you're going to split something into core and "compiles down to core", then a LOT of features can be moved out of core because they're just (definitely worth keeping, but not necessary in core if that split were made) convenience APIs.
Incorrect: https://ecma-international.org/about-ecma/history/
It is a disaster the moment you try to run same code on different platforms.
While it is admittedly confusing to have all these different flavors of JS, I don’t think this proposal is actually as radical as it seems.
https://www.amazon.science/blog/how-prime-video-updates-its-...
So now a huge swaths of use cases are going to be killed by this change. E.g. AdBlock, NewPipe and yt-dlp - how is that better? All of them (expect maybe AdBlock) rely on parsing incoming JS from YouTube which will be rendered obsolete by WebAssembly blob.
There is already too much exhaustion around switching frameworks and paradigms in the JS world, but I guess everyone likes getting jerked around by corpos and evangelists these days.
On the Backend there are very few issues, outside of FFI only being in unstable for Deno I suppose, but you could frankly be running the same old Express API you did a decade ago and be perfectly fine.
If you’re burnt out on changes and keeping up with things I think the issue is mostly a “you” issue. You don’t have to chase down the latest hypes or fads. In fact I think you almost never should.
This is an incredibly disingenuous response. You maybe like the world this way. It doesn’t mean there isn’t room for change or improvement away from Javascript.
What's wrong with VanillaJS?
It’s not just the flavor of the week frameworks, it’s libraries and best practices. Want to work with dates? Do you use moment? Nope that’s deprecated, what do you use? Which moment successor? How do you write react? Classes or functions? You can’t use hooks with classes, so you better update to functions. On and on you run into a decision tree because of the shifting target of javascript. It causes a lot of churn to be migrating and updating to new systems, especially when the new hire can’t help because they don’t understand prototypal inheritance.
I can tell you such stories about any language, it’s not unique to JS. Welcome to working with people.
I had always imagined that if the DoJ took any action it would be to cleave the ad business away from Google. Although if they went so far as to take action against GCP I bet Amazon, Amazon Marketplace, and AWS would start to get sweaty palms
That said I also see the rationale for that. JS is already getting heavy and we should limit feature creep.
https://github.com/search?q=BigInt+language%3ATypeScript&typ...
https://github.com/search?q=BigInt+language%3AJavaScript&typ...
Is it though? Sounds more like a single team's preference, and not necessarily a team of FE engineers
Why is wasm so damn limited?
Hot take
No, it won't be faster, if you only optimize a lesser language. If you have higher level code running your optimizer can do more than if it only has a low level version.
No, it won't be more secure. Js0 might be more secure, but if we sites all run any of dozens of different tools those tools are going to be creating the vulnerabilities. It's shifting where security issues occur, and creating more of them.
I'm terrified this could happen. JS has gotten so much better over time. We are so close to being able to not need transpilers. This sounds like such an absurd cop out for browsers to say, meh, we just don't want to do the work to implement. Being so close & then saying, sorry, you must use big toolchains to develop for the web is a monstrously bad future.
I don't see how a browser running JS0 can be any less secure than a browser running JS
Browsers operate at a scale never seen before. Imagine all the extra energy and bandwidth needed if core functionality is moved to the application code when you have billions of users.
What we need is more native functionality (implemented in the JS engine with C++ or Rust) to have as less user land code as possible.
As an example, imagine how much energy, bytes, and CPU would be saved if browsers and JS engines included reactivity and JSX. Or if browsers included an API similar to jQuery.
Quick math to grasp the scale: 100kb * 1 billion users = 100TB of data that needs to be transferred and parsed many times, every single day. It's absurd.