> I hope in the process of doing it we will find new ways of doing things.
HTML & CSS themselves have become a major bottleneck to quality and creativity. The arcane layout model, the baggage of backwards compatibility, the cognitive dissonance — played out over decades of design-by-committee — between "this is a document engine" and "this is an app engine."
The rendering-thread-is-the-main-thread architecture of JS plus the JS garbage collector cause jitters & frame drops that most people don't consciously recognize, but everyone subconsciously recognizes. These little bits of jank are why consumers can recognize webview-wrapped apps vs. native apps.
Don't get me wrong — HTML and CSS and JS have brought us far, and the zero-trust execution environment that is the browser is an amazing feat of humanity.
I hope the "new ways of doing things" you describe include a major innovation on HTML and CSS and JS. WebAssembly makes this possible today — and I have dedicated the last few years of my life to proving this concept[0], and I hope others explore similar avenues. We deserve a better substrate, and this can be done without reinventing the browser.
I disagree. There has never been a better markup language. HTML allows precise control of atomic elements that make up a component. It has been the best thing for accessibility we ever came up with.
I can imagine no better way for readability and accessibility than a text-first representation of content, which is what HTML is. I will even argue that having the modern web emerge from a text-content first approach is the best thing that could have happened to us.
All other content representation infrastructures that weren't text first, like java applets or flash died because they were no match for HTML.
Agreed! Markup-as-content-backbone is one of many things HTML/CSS do extremely well, and I agree not building around something like this was a major weakness of Flash.
Silverlight sought to do this with XAML, but it had other problems (required a plugin right as plugins were on their way out; CLR was a huge dependency; too locked down and proprietary under Ballmer leadership)
Can you imagine a text-first content representation, which is visually editable with the UX of a vector design tool? This is the rabbit hole that drives our work on Pax.
Care to point us to a better combination than HTML/CSS when it comes to layouting flexible interfaces?
Don't get me wrong, I also could imagine better markup languages for that purpose. But everything I have seen in the wild was worse in multiple, show-stopping ways.
Maybe instead of coming up with a new thing, we just need a clear way to solve the pain points and put the solutions into CSS4 and HTML6
Both Figma autolayout and HStack / VStack are careful subsets of flexbox; not quite the same thing. Consider whether a marble statue is "just a subset" of the block that contained it.
None of these is a perfect technology, but I believe the most practical answer to the question driving this thread[0] is to study/understand the landscape, then build towards a better future.[1]
[0] > Care to point us to a better combination than HTML/CSS when it comes to layouting flexible interfaces?
> Alternate layout engines for the web might be a fun experiment, PhD thesis, or talent retention program, but it's not practical.
Flexbox was once an "alternate layout engine for the web," as was Flash player, as is Figma. Framer, Retool, and Squarespace all offer alternate layout engines tailored for visual building. All of these seem practical to me.
1. not Figma or Flash: it's not practical or performant to manipulate HTML/CSS to achieve the creative freedom of a vector design tool[0]
2. the rest of these that build on HTML: not a single one of them exposes that code for manual editing, so they're not developer tools and their "alternate layouts" are proprietary + locked away.
The root issue: HTML was not designed to be a substrate for design.
[0] I don't claim this casually; I spent several years seeking to do exactly this with github.com/famous/famous and https://www.haikuanimator.com/
`Waiting for fonts.googleapis...`, pax website was blank for a couple of minutes, might want to reconsider making third-party fonts non blocking on your website so it doesn't appear broken/empty when a third-party CDN edge is slow.
It's one of the core things that has enabled the web to be as useful as it is. One of the things that draws people in, and keeps them using it.
Yes, there are problems with what we have. But if you break compatibility, you'll either not be adopted, or part of the crowd that the audience yells at for taking away their favourite things. You'll kill efforts and bury knowledge bases.
I don't think the world should abandon HTML, nor break backwards compatibility across the HTML spec. The first cars drove on roads designed for horses, and horses are still around. At no time did we gather a committee and decree that horses were deprecated.
Taken to an extreme, "don't break backwards compatibility" has an insidious failure mode, which is "don't innovate." The car could not have come about without a willingness to break backwards compatibility with horse drawn carriages, plows, mills, hitching posts, etc.
The adoption of a radically new technology like this is voluntary, collaborative, and progressive. Provided it offers enough value to exceed the switching costs, there's no need to kill efforts or bury knowledge bases.
Doesn't the rendered page run in a different thread to its JS? If the website does use some SPA framework to lazy-render the components, load the new view, or execute some logic, maybe then, but when I say run `(() => { const s = Date.now(); while (Date.now() - s < 10000); })();` and try to scroll the page, it doesn't freeze.
Interesting project. Have you tested any non-XML markup languages for the user interface declaration so that metadata doesn't take as much space as data in
<Text>Hi {firstName} </Text>
and also requiring double the number brackets
There are modern cleaner markup alternatives like kdl
We chose an XML-like syntax because it's clear & explicit & established (HTML) — you can know where you are in a hierarchy clearly at any point thanks to closing tags. The major downside is verbosity, both for reading and for typing. Pax's closing tags compile away so they aren't transmitted across the network like they are for HTML.
We predict 95%+ of Pax will be written by machines, especially visual design tools and LLMs, so verbosity becomes less important (LLMs may even benefit from explicit closing tags.) We're innovating on multiple axes, including language, so we chose to make the syntax itself as boring and uninnovative as possible.
But clarify is exactly what's lacking because it's verbose, it obscures content at no benefit to simple matching bracket highlights
Also, will it be read 95% of the time by machines during design? That's not what the demo shows with the side-by-side xml and output, which I expect is a much more common workflow than 5%
The innovating part is what puzzled me and prompted this question - why cling to the old garbage when you're doing new design?
Point taken! And yes, especially for developers (our core target demographic,) you are right that hand-writing should account for >5% of code volume.
Syntax is fairly contentious. "One man's trash is another man's treasure," on your note of "old garbage." It's hard to please everyone, however:
We could offer syntax "skins," since the data storage mechanism is a layer separated from the AST (thus different ASTs/syntaxes could de/serialize from the same persisted data.) So folks who want closing tags can have them; folks who want a YAML-like format could have it; folks who want a KDL-like language could have it.
At the language level, Pax's distinguishing characteristic is that it's the union of an expression language and a markup language; this is the reason we couldn't use an off the shelf markup language / parser. But again, KDL or YAML could be extended with PAXEL to make pax-kdl and pax-yaml flavors alongside pax-xml.
We had to start somewhere, and starting as close as possible to the markup lingua franca (HTML) made sense to us. It'd be a dream for Pax to be loved / adopted enough that we or anyone else cares to make a syntax skin.
I won't claim Pax is better than any of these, but that we seek to solve different goals.
Unlike Flash: driven by a markup language, fully open source, no plugin required, solves a11y, compiles to native apps incl. mobile
Unlike Silverlight: fully open source, no plugin required, no heavy VM
Unlike JavaFX: no plugin required, no heavy VM, visual builder is a vector design tool
Unlike Flutter: designed for web (small footprint, a11y out of the box); first-party and foundationally integrated visual builder, visual builder is a vector design tool
Thanks for explanations. However, instead of making yet another framework, wouldn't it be better to contribute to Flutter (e.g. optimize its wasm target, add a vector visual builder), so there is one strong open cross-platform alternative to HTML/CSS/JS?
Pax's primary goal is "designability" — to enable a vector design tool to read & write the same content & behavior you may write by hand. (.pax files)
Our solution requires language constraints — a hermetic separation of concerns between the declarative description language (.pax) and a Turing-complete programming language (starting with Rust.)
These language constraints and rendering requirements[0] are so core to our solution that building anywhere other than the systems level was not tenable.
[0] Specifically, the runtime must render in "design tool coordinates," to enable a vector design tool authoring UX. This authoring experience must also be extremely fast, like Figma's. The rendering engine must be designed around this requirement and neither HTML/CSS's or Flutter's were.
Swift is pretty next level compared to Rust. Rust does seem like a fairly constant level of slow and linearly slows down as the project gets larger. Swift tends to have huge degradations on seemingly innocent changes that make the type inference go off the rails. Although at least if you spend the time to track these down you can fix the worst offenders.
Pardon me for not getting your context, but are compile times a big issue in software development? I have never programmed professionally and all my experiences with code is from a couple of classes taken in college a decade ago.
When doing UI things, a common workflow is to have code and the app side-by-side, make a small tweak in the code, recompile/hot-reload, look at he result, repeat. A long compile time makes workflow a pain.
In general, you're right. But there are at least 2 times where they're absolutely vital- anytime you're dealing with a UI and data exploration in data science (since you make a lot of frequent, small changes in the goal of fine tuning something.) Everything else, best practices has good testing and partial compilations to make it moot. There's probably some other contexts that make it valuable, but I've never had to deal with those.
There's been a lot of work on this in the last few years with Windows support, preliminary Android support is also being worked on and should appear at some point in Swift 6.
There probably won't be a cross-platform UI layer and many Apple frameworks won't work, although apparently the Foundation Swift rewrite already works. For a lot of companies though, simply being able to share business logic will be a big plus. At my current company we have a lot of talented Swift people and existing code written in the language so being able to share anything would be a huge advantage.
I'm biased but I personnaly find the language really productive to work with, runs fast enough for my needs and let's me target more and more platforms.
Actually, it sounds like they only just decided to use it in the future once Swift 6 is released. I would guess much of it will still be in C++ for a while.
For everything. Swift is—at least in theory—a general purpose language, it's not exclusively for Apple technologies.
> First off, Swift has both memory & data race safety (as of v6). It's also a modern language with solid ergonomics.
> Something that matters to us a lot is OO. Web specs & browser internals tend to be highly object-oriented, and life is easier when you can model specs closely in your code. Swift has first-class OO support, in many ways even nicer than C++.
> The Swift team is also investing heavily in C++ interop, which means there's a real path to incremental adoption, not just gigantic rewrites.
> > First off, Swift has both memory & data race safety (as of v6)
> But v6 is not released yet, right
As of Swift 5 there is zero data race protection and the process model (DispatchQueues if memory serves) is woefully. No advantage over fork and much more convoluted
I am well clear of the Apple development world now (thank goodness) but the tools were of very poor quality, albeit very nice looking, as of this time last year
For the sake of my friends still in the world I hope Swift 6 is better, but I fear the dumpster fire that is Xcode is too far gone to rescue.
The comparison with Rust demonstrates the utility of "design by committee ". Rust is far from perfect, but feels like the future where Swift feels like warmed up Objective C
> As of Swift 5 there is zero data race protection and the process model (DispatchQueues if memory serves) is woefully. No advantage over fork and much more convoluted
Swift Concurrency is the replacement for Dispatch and has been around since Swift 5.5 in (IIRC) 2021. It’s a completely different system, uses lightweight tasks (a la tokio in rust, or goroutines in go, etc), has a concept of “Sendable” for thread-safety a la rust’s Send, async/await, and a native `actor` type, among other things.
Swift 5.5 didn’t get all the way towards rust-style data race safety due to a few things they had to make warnings instead of errors (to avoid breaking existing code), and introducing keywords like `@preconcurrency` when importing frameworks that predate Swift Concurrency, to facilitate incremental adoption. They’ve also been adding more checks in each minor release to tighten things up.
IIUC Swift 6 is mainly going to turn all the warnings into proper errors and tweak some defaults so that you get proper data race protection on a default compile.
Point is, it’s totally inaccurate to say that Dispatch Queues is all that exists in Swift 5. You’ve had much better stuff for a while now (although SC still has a ton of issues worth discussing.)
> Swift 5.5 didn’t get all the way towards rust-style data race
When I experimented with it it was trivial for one thread to interfere with another. So Swift got nowhere towards data race safety. Still stuck in the 1990s
I know not what you mean "Swift Concurrency". When I was doing it all we had was DispatchQueue which was an obfuscation of `fork`. Quite shameful really.
I think the main point is that Swift is a failure.
"although SC still has a ton of issues worth discussing" once I would have cared, but this year (decade, century) I am just very glad putting meat in my fridge no longer depends on those rouges from Apple who treated me so badly when I was sweating so hard making software for their platforms (not to mention paying them so much money). In 2024, for a company like Apple, for their flagship developer offering, why would anyone still have "a ton of issues" with it?
Apple is now an example of why popularity is a terrible metric to estimate quality of technical offerings. What a shame. How far the mighty have fallen
> Okay, so you're saying you don't know what Swift Concurrency
I just looked it up. It is Swift's version of async/await. That is a different thing from threads. I know what that is, used it a lot, because using threads was such a nightmare in Swift.
> language with many compile time guarantees of thread safety
From two separate threads you can access the same memory. No trouble (apart from crashes memory corruption....) at all.
Async/await is always a bad idea, and without a garbage collector it is a nightmare (look at the mess Rust has gotten into). Whatever, async/await is no replacement for parallel programming with threads. It is a different beast.
> I just looked it up. It is Swift's version of async/await.
Since you “just looked it up”, maybe don’t make blind assertions about something you clearly don’t know very much about?
It’s a lot more than async/await. It is a way to offer compile time guarantees about thread safety (through Sendable, which is part of SC), it’s an actor model (allowing shared mutable state to be isolated, leveraging async/await for coordination so that callers can pause if the actor is currently servicing another message) and a bunch more stuff.
I explained all this in my post you replied to, maybe read the whole thing before making wrong claims about stuff you spent 1 minute looking up?
> Whatever, async/await is no replacement for parallel programming with threads.
Is it not for the vast majority of use-cases?
Sure, you can use async/await without parallelism, via a single-threaded runtime to just get single-threaded concurrency, but with a multi-threaded worker-pool async/await-like tasks or fibers I think mostly cover the use-cases you'd have for parallelism?
You have to make sure that you e.g. don't starve other tasks via having no yield points in a task that does a lot of computation (if you're doing cooperative tasks which Swift is doing iirc), but that's not a big one, and can mostly be solved by the runtime too (e.g. Go had cooperative fibers for a long time, until they chose to introduce preemption).
Async/await may or may not be a replacement for highly concurrent and parallel programming, depending on what is the execution model of the async runtime.
If Swift's model is anything like .NET's lightweight Tasks + async/await or Rust's async/await Futures and Tasks as implemented by Tokio or async-std, then it is such replacement.
> Whatever, async/await is no replacement for parallel programming with threads. It is a different beast.
Have you missed Tasks and Task Groups as well? And Actors? For now, they are an abstraction over threads, and IMO a good one. It’s actors + structured concurrency, borrowing from Kotlin‘s Coroutines and sprinkling some Erlang on top. Additionally, in Swift there is AsyncSequence + AsyncStream, a (woefully incomplete) Kotlin Flow alternative.
Xcode 16 rewrote much of the autocomplete behaviors and the new engine is blazing fast in terms of UI blockage compared with 15, and stable. Everything to do with tagging etc not just completion itself.
I don't know if this is irony or you've been in Apple ecosystem way too long.
I had to do a very simple macOS app for my personal consumption recently and XCode dx is nothing to write home about. The only reason I finished the project there was that I couldn't set up vs code quickly for Swift/Cocoa project. I had to endure the slow compilation time, slow reaction time of the IDE UI. You make a change you keep seeing squiggly lines for a while, it's as though the UI is booting up each time. It was a horrible experience coming from IntelliJ and VS Code daily experience.
Computer is 32Gb Apple M1 Pro. Imagine what will happen on some 8Gb i5 macbook.
Swift is useless outside of macOS/iOS dev… the thing doesn’t even have namespacing. I don’t know for what reason someone would use it outside for Apple environment.
You can disambiguate two types with the same name from different libraries, e.g. `Factotvm.URL` and `Foundation.URL`. Do you mean something more full-featured? You are not prefixing types with three letters, if that's what you think has to be done.
I don't know if it's still the case, but there was an annoyance where you couldn't have a type with the same name as the package. But that is hardly a lack of namespaces.
Objective-C had some minimal adoption outside of Apple (probably due to NextStep nostalgia), so if Objective-C managed to get some traction, Swift will do it to, probably.
However, Apple's history is very much stacked against Swift becoming a mainstream language outside of Apple's platform.
HN majority doesn't like hearing that ladybird et al might just be wandering around, even if the goal is catnip for the bleachers, and we should be skeptical this is the year of multiplatform Swift, because it wasn't last year if you actually tried it. Or the year before last. Or the year before that. Or the year before that year. Or the year before that one.
I’m slightly more ambivalent than you about it. Swift is a nice language and has better ergonomics than C++ and I imagine a Swift codebase might find more contributors than a C++ one (maybe I’m wrong about that!)
I also think it’s separate from the dream of “multiplatform Swift”. For that you need a healthy package ecosystem that all works cross platform, Swift doesn’t have that. But a lot of Ladybird is written at a low enough level that it won’t matter so much.
Problem is Swift engineer supply is low, there's not a viable business case to learn Swift because it's not actually viable cross-platform for development unless you have $X00 million to throw at your macOS/iOS team to build it from scratch, platform by platform (to wit, sibling comment re: Arc Browser)
So best case we're looking at: Swift isn't ready yet, the next major version will be, and we can't build UI with it, so we'll put in the effort in to bootstrap a cross-platform ecosystem and UI frameworks. Or maybe we'll just do our business logic in it? It's a confusing mess that is irrational. Even with great blessings of resources. ex. $X00M that Arc has obtained one incremental platform after a year. And "all" they had to do was Swift bindings for WinRT and connect it to the existing C++ engine.
All of this is easy to justify if we treat it as an opportunity to shoot for how we wish software could work in theory, instead of practice. I hope I'm wrong but after being right the last few years, I'm willing to say it's naive wishcasting out loud, even though its boorish. I see it as unfortunately necessary, a younger me would be greatly misled by the conversations about it on HN. "We've decided to write the browser in Swift!" approaches parody levels of irresponsible resource management and is a case study in several solo engineer delusions that I also fall victim to.
It's genuinely impossible for me to imagine anyone in my social circle of Apple devs, going back to 2007, who would think writing a browser engine in Swift is a good idea. I love Swift, used it since pre-1.0, immediately started shipping it after release, and that was the right decision. However, even given infinite resources and time, it is a poor fit for a browser engine, and an odd masochistic choice for cross-platform UI.
> Problem is Swift engineer supply is low, there's not a viable business case to learn Swift because it's not actually viable cross-platform for development
The Swift business case is that in many situations native is strongly preferable than cross-platform. Excluding some startups that wants to go to market super fast and consulting companies that have to sell the cheapest software possible, usually the benefits of native outweighs the ones of cross platform.
For this reason now there are plenty of companies of all sizes (faangs included) that build and maintain native apps with separate iOS/Android teams. There are very good business reasons to learn Swift or Kotlin in my opinion.
Right -- no one read my comment and thought I meant Swift was unnecessary or businesses don't use it. Contextually, we're discussing Swift for cross-platform dev.
Well I was trying to be kinder than just leaving you downvoted and confused why. I guess I shouldn't have bothered, my apologies. Hope your week gets better!
I wonder what convinced Andreas Kling to abandon his own language Jakt [1] in favour of Swift.
In the long run, it would be good to have high-level languages other than Java that have garbage collection (at least optionally) and classes, and that are still capable of doing cross-platform system development. I don't know if Swift fits that bill, besides cross-platform ecosystem (a la Java), submitting the language for ISO standardization (not just open sourcing one implementation) would be a good indication of being serious about language support.
> In the long run, it would be good to have high-level languages other than Java that have garbage collection (at least optionally) and classes, and that are still capable of doing cross-platform system development.
One of the major differences between ladybird as part of serenity and ladybird the separate project is using 3rd party libraries. When what you are building are is for fun and you build everything yourself it makes sense to also build a language.
Ladybird as a separate project has the goal though of something usable in the shorter term. So similarly with switching to 3rd libraries for things I don't think it makes sense to spend potentially years first building the language before building the browser.
You might be powerfully dry here, imparting a Zen lesson. If not, or if you, dear reader, doesn't see it: it is worth meditating on that there was a language, an OS, and a web browser. Then, on common characteristics in decision-making that would lead to that. Then, consider the yaks that have to be shaved, the ISO standardization / more than one implementation hints at this.
It’s not really a question of fairness. The existing codebase is C++, the new stuff is Swift. Hence the comparison.
I’ve written both Rust and Swift while being an expert in neither. I wouldn’t say Swift has no pluses in comparison, reference counting is often a lot easier to reckon with than lifetimes, for one. I’m actually curious what a large multithreaded Swift codebase looks like with recent concurrency improvements. Rust’s async story isn’t actually that great.
I agree. People's perspectives differ. It abhor `async/await` in Rust. It has poisoned the well IMO for asynchronous Rust programming. (I adore asynchronous programming - I do not need to pretend my code is synchronous)
But that is taste, not a comment on poor engineering!
The lack of the borrow checker in Swift is what makes it approachable for newcomers, as opposed to Rust which is a harsh mistress.
But reference counting is such a silly idea. Swift really should have a garbage collector with a mechanism or subset to do that very small part of programming that cannot be done with a garbage collector. That would have been design!
I fear that Swift is going to get a borrow checker bolted on - and have the worst of both worlds....
Reference counted objects have deterministic lifetimes, like Rust or C++. Garbage collected languages don't have that. Essentially a reference counter automates some* of the ideas of the borrow checker, with a runtime cost as opposed to a compile time one.
The great thing about automatic reference counting is you can elide the incrementing and decrementing a lot of the time. Stronger guarantees, such as saying "this object is single-threaded" lead to even more optimizations.
I don’t think Apple would want to sacrifice the determinism that refcounting provides for garbage collection. iOS apps are still more buttery smooth than Android in many cases.
I mean this year we did have the porting of Arc browser to windows. I use it my gaming PC and it is starting to feel like it has a similar level of polish to the MacOS version.
Ladybird exists and is funded because the main author is an extremely skilled software developer with a background in building web browsers.
He also built a totally from scratch operating system, and built a very productive and skilled community around it, which branched into building their own web browser as SerenityOS has an 'everything from scratch' policy. How is Serenity not an interesting feat of engineering?
You seem to call it a 'toy' because of where the development is at - this is grossly unfair criticism. By your measure, is all software that isn't at a 100% finished state a toy?
Why is it weird that a company that built their entire business on the web wouldn't want there to be a web browser monopoly controlled by an advertising company? Shouldn't you fund projects that will be beneficial for your company in the future?
I think they were referring to the version of Ladybird taking advantage of Swift v6. I agree in part that it is only nice idea until a production ready binary gets released; right now it is just experimental.
Is there a good software engineering resource about the complexity of a web engine? I mean, we all know that is complex but what are the critical areas. Performance is one, compatibility another.
This is a good post; as far as Ladybird is concerned while they may have started things as fun, they seem to have taken a turn towards seriousness recently.
The problem with these smaller webbrowsers is that you have nobody to sue if the browser turns out to leak your personal information or credentials etc. Therefore it is better to stick with browsers made by big corporations.
I'm curious what exactly you think suing Google or Apple in such a case would accomplish?
On the (long) odds that you get all the way to a successful class-action suit, the lawyers get rich, and the class members eventually end up with a free year of credit monitoring
(we have seen this over and over again in settlements for high-profile data breaches)
This is true for any software though, so if we push that reasoning to its extreme, no one should ever use any free and open source software, which I find a little bonkers.
Some people just get warm fuzzies at the idea of theoretically being able to sue someone if they need to. Usually these are people who've never actually looked into what it would take to even start a lawsuit.
I hope in the process of doing it we will find new ways of doing things.