Solid 2.0 RC: The Big <Reveal>
solidjs.com
solidjs.com
Obviously having an LLM write these kinds of things is the norm now, but what doesn't have to be the norm is the writing style. The way Claude (and other popular LLMs) write is so obnoxious and verbose. I think a simple pass of "make this more terse and to the point" would do wonders here.
But you aren't wrong. The marketting heavy parts are where the tone shows. And honestly those are the parts I write worst. Exactly why I lean on AI there. When I ran before/after comparisons past reviewers, the polished versions kept winning. Maybe that's the norm creeping into how all of us write now. I notice it in my own sentences too. The words I choose. But given the choice between slightly synthetic and authentically clumsy, I tested both and shipped the one that read better.
I'm at least one example of someone who prefers authentically clumsy. Users like me will be less likely to give Solid a chance if they cannot see that there are real humans with a vision behind it.
And there we have it. "AI"s prime use case. Best instead of good enough.
:)
The AI tone in the announcement is light enough to be barely noticeable, but it’s there. Note that posting AI-generated comments to HN is explicitly forbidden.
Anyone who has spoken to me knows I talk in run-on sentences. I never complete a thought without interrupting myself. And I write the way I talk. I used to heavily edit that out of my writing, but now AI uses the proper punctuation and fills in where I miss words. It's like what I hear in my head but better. Because it is grammatically sound.
After:
Anyone who's spoken to me knows I talk in run-on sentences. I never complete a thought without interrupting myself. And I write the way I talk. I used to spend whole editing passes beating that out of my drafts. Now AI handles the punctuation and fills in the words I skip. It's what I hear in my head, but grammatically sound.
The result is almost always terser, more clear, and to the point. And usually all it takes is one more pass over to adjust anything I wouldn't say. In this case I kept the after verbatim just to illustrate my point.
What's an "editing pass"?
That's really the heart of the problem - in an effort to sound exciting, many parts of the article are just wasting the reader's time rather than simply explaining how things work clearly and directly.
- meander freely and use the character limit to help you truncate
- split into one tweet per point, with one code snippet screenshot + quick explainer
- either bullets like this or 2-3 sentences each
Then your article becomes a slight expansion of the thread, where you can adjust language to make it flow better instead of starting with fluff and trying to carve it into something solid
(I think of it like mobile-first design applied to writing)
If the reason you don't like meandering or bad grammar is because you associate it with low quality/low effort writing, then the same can be said about standard LLM output. You may have gone through many iterations to land at this specific AI generated article, which stands up to your level of scrutiny over the content, but to anyone who doesn't scrutinize it just comes off like any other AI generated article.
Everybody using AI to write slop do so because they think they can't write better but we all prefer a bad written article to the "correct" slop we are getting nowadays.
E.g. "The problem that never went away was async. Every framework, including ours, treated it as a condition that happened to it."
Async treated as a condition?
And a condition happens?
And on grammar, the second it is ... the framework??
I rather suspect this is second-order output. Writings of a brain rotted by "AI".
Shame, because enough of the message gets through the fog to tell me the whole message would be worth receiving. I wish the project every success.
I had to rewrite the drafts very heavily in my own voice in order for people to enjoy reading them —- not that I’m a great writer, but I am human and people tend to appreciate that.
https://hackmd.io/@0u1u3zEAQAO0iYWVAStEvw/rJM9ws3Kbg https://hackmd.io/@0u1u3zEAQAO0iYWVAStEvw/H1Q8XTMSbe
The published post reminded my of where I got to after several rounds of review and cleanup.
We have frameworks essentially living in a completely synchronous world--like a 2D plane--and they can't tell that there is another dimension out there except for when something inserts itself into its view out of nowhere. The protocol that we arrive at to deal with async is convoluted, because it is a state that is basically inflicted on the framework. That's what we sought to remedy.
What’s also interesting to me is that it seems like it’s colorless. Your async state is used just like any other synchronous signal state.
Really excited to start playing with it!
If anyone hasn't tried it I highly recommend it. Simply put: Solid makes the "dumb and naive code" the optimal code. It takes all the pain of React away.
I agree with your sentiment that we've pushed web frontend performance as far as we should bother. I think optimizing calls is the next frontier. On that end, I like the approach for async, but maybe I have a suggestion?
Solid needs a tightly coupled request library that emits signals/resources directly. That "server-side" data binding layer would be the missing glue. I think server actions sort of get at that, but I've never really been satisfied with them. Just my 2c.
Step one of all my projects is always wrap a bunch of fetch calls into whichever correct incantation of solid primitives. Plus, the solid compiler should be able to early load those requests via static analysis and improve real life FCP / LCP times without the developer needing to do gymnastics.
I picked up on the async headlines and the infrastructure pieces. We talked a lot more about the new reactivity in the beta release. The plan from here as indicated in the article is a series of followups explaining parts of Solid 2.0 in more detail.
The key takeaways are: 1. We've solved declarative async, race conditions/complexity are reduced by the structure of your code.
2. We've collapsed the metaframework layer. Every Solid project out of the box is as capable as a metaframework based on the features you want to use. You can just start a simple client app and go from there.
The result is we are setting the stage that both Humans and AI can do things a single way and fall into the "Pit of Success" by the shape of the solution and the guarantees of the model.
Integrating async as first class citizen into the core is indeed interesting. I wonder how it impacts the ergonomics and state reasoning in the end.
Need to test it out.
(I see a lot of people complaining about AI use in writing the announcement. What they forget is, when you spend time actually writing and debugging code, these kind things that can be outsourced to LLMs should be. The important stuff is elsewhere and that's what matters. Not a blog post. Its purpose is to tell the users what to expect and the post does that well.)
the way you do something is the way you do everything. AI slop in one place of your work makes people think there's AI slop in other places too.
We can throw up a JS Framework Benchmark.. I obviously tested it a bunch locally, but mostly as a regression test. I also used Octane's new benchmark suite. Same sort of conclusions though. Everyone basically doing the same thing, some with compilers and some with more manual wiring (I haven't gone in and made PRs to update those to add that to Solid implementations in theirs), to the point this isn't a contention point. It was when everyone was as slow as React. But now everyone is more or less as fast as Solid.
Bundle size there is a regression though. Its natural cost of this feature. Async everywhere, means async doesn't tree shake. Hello World went from 4.7kb to 9.8kb simply because we can't tree shake it. I'm ok with that given what we deliver. Not everyone will be I'm sure. Svelte has also grown in size similarly over the same time. Basically leaving only minimalist solutions in that ~5kb range.. ie Preact + Signals etc..
But it also isn't really an announcement headline. So it is where we are at. I guess the takeaway. Still fast, still incredibly small considering what it does.
I've always appreciated Solid's DX/mental model (I _never_ have to think "when does this re-render?") so v2.0's continued wins on DX are remarkable. I never thought a release could be about how much can we _cut_. Primitives at work!
https://gist.github.com/aleclarson/0f4266d63fd83c7a5ea551244...
I just hoped that announcement would be another blog post.
SolidStart 2.0 has a very important role of moving current projects into the future. Better tooling was key and it was that rewrite work that allowed me to hoist relevant parts into the core. Now every Solid project benefits from no FOUC in dev, preloading tag emission during SSR streaming etc...
I think this is inevitable collapse with AI. We want to look for our tools in a single place. It doesn't have to differentiate between React and Next when React is the one actually providing the features.
Aside, what sort of blog post content are you looking for?
urgh