Dominic Gannaway

There's custom elements, and there is the web component specs. I'm conflating them as they typically need to, to get a deserved output. Also what are native components? What Svelte, Vue and Solid do is not the same as generating native Swift code from your abstraction.

Well I mean React has done that with React Native and its many other renderer targets. I mean, it's arguably still the major draw to React today. I'm not saying React is the best choice to use, but that is a killer feature – and I speak from seeing it implemented in practice at scale.

I don't think web components are a good enough abstraction. If you're leveraging AI, you definitely want to be working with a higher abstraction so that it can make your application portable. The DOM should be purely seen as one of many render targets IMO.

I'm not convinced on vibe coding. I think the bubble around it is going to pop, and it will end up relegated as a prototyping workflow, a bit like how building a prototype of Figma is.

I'd love to be wrong, but I've just seen too much from the perspective of the infra/framework/tooling author-side.

Evergreen won’t get you far though. Without libraries you’ll just create no better than jQuery fluff. CSS standards have changed dramatically over the last year too

Standards change as regularly as frameworks tbh lol. Also what really is vanilla JS these days? You need some kind of structure and framework, even if it’s not React etc, otherwise scaling things becomes difficult - security, testing, accessibility, A/B testing, performance and i18y can be unwieldy

For context, they work really well outside of writing code outside of framework "template" logic. Like they're super powerful at fixing regexes and mundane tasks that would take me 20mins, but now takes me a few mins.

I should probably play around with Vue and Solid with Cursor/Claude Code more to see how they fair. So far it's been a really frustrating experience to say the least, I'm just having to correct everything that they suggest changing each time.

Svelte 5 fairs better, but then AI gets confused around template modules vs non template modules when it comes to runes. This might get better if Svelte opens up declaring state and other bits of logic in control-flow, but it's still not great.

The amount of times I've had to correct AI output because it doesn't quite understand control-flow around JSX is astonishing. I still feel like JSX in its current state isn't the ideal thing we should be using in 2025.

I still think there's a missed opportunity to create a frontend framework that works really well with AI tooling.

Yep precisely. Such a shame as control flow becomes far easier to reason with when I converted some complex components.

There are a few ways, one is to use a standard for loop, or to use Object.entries on the collection and capture the key.

Thank you so much :)

I think some folks from the Lexical community might be interested in this (even though you're using PM). I'll ask around :)

I recently switched from Svelte to work on some other things internally. I wanted to try out some new things!

I left the Inferno project back in 2017. Just haven't had the bandwidth to maintain it whilst working full-time on another project. I tried, but I burned out trying.

AI is great when you provide a ceiling in context, but when you ask it to do abstract things where you don't know what you want – then you'll see wildly different results.

There's more to it than that, but that is one slice of feeling I have about it. The other is the cost of development to make a new framework in 2025. It's just so much higher than it was back in 2015 with Inferno. It's not just financial, it's a time cost too.

I don’t think it’s that big a deal in reality. You’ll likely be using some i18n library anyway rather than raw strings.

No, the function won't ever rerun. It basically breaks down to something signal-like, but not quite signal driven. The $double value above is derived above. If you didn't want it to be derived, you'd need to untrack the value you assign it (if it's reactive).

Interested in Signals? Enjoy learning by reading code? Check out github.com/jkup/learn-s...

Yeah for the DOM sorry. If that’s your render target.

Yeah I tried this many moons ago with Inferno. I avoided using the prototype though as it was slow setting this in some VMs. Ultimately it didn’t change much as people tend to use spread often in real world cases and you lose this optimisation quickly. It was better to just use template cloning.

Your modules would stay as .inferno modules when published. Just like .svelte modules are today

Yep, except statically it would be the case. There’s no runtime of doing it the same way as components do not re-render.

Yep

Firstly, it’s JS first. Which makes it more approachable to React developers. I also personally prefer it. By being JS focused it also made it easier for AI to manipulate too in my research.

Secondly the reactivity is clearly defined by the $ prefix. This also applies to object properties.

Here's what it looked like:

bsky.app/profile/true...

Shows the code: ```js export component App() { let $show = true; <div class="container"> <h1>{"Inferno X"}</h1> let $count = 0; let $double = $count * 2; <button onClick={() => { $count += 1; }}>{'Count: ' + $count}</button> <span>{'Double: ' + $double}</span> if ($show) { <div>{"Showing"}</div> } <button onClick={() => { $show = !$show; }}>{$show ? 'Hide' : 'Show'}</button> </div> } ```