@GergelyOrosz I find turning off thinking and steering more has been a good middle ground.
Also I plan with no thinking on SOTA models, and build with cheaper models (notice once a plan is in place the implementation tends to be 90% correct)
@piliponful@yacineMTB From memory it was a front end role for Rokt https://t.co/uDMuByyr9n
not sure why I got rekt like that tho T-T was hoping for gnarly react or DOM questions at least
@goodbyesahil@yacineMTB I got this https://t.co/uDMuByyr9n
Tried to do it first by reducing the regex to be simplier and iterated on the * complexity but never did it
@afraidoftime_@yacineMTB Yeah, there’s lots of parallels for sure. I love chess puzzles and feel like you can derive from first principles a bit more than LC but there is patterns that you can pick up
@GergelyOrosz My counter to this is I’ve refactored so much using agent and largely because when something works you’re half way there cause now all iterative refactor is purely for non sloppifying and it’s nice cause I can read and comment faster than I can type
@realamlug Wonder if you'll take bun's scope as inspiration. A postgres driver, s3 driver etc...
i feel like kv + sql + blob is most of what I do anyway so it's a good scope
@netadror@shadcn shadcn is great, https://t.co/udEZTEgdH9 i feel is the OG token pattern (was the CSS library for radix!).
shadcn tokenised colors with tailwind scale. Would've loved if they also standard font size and padding size too.
@devinjameson The moment complexity hits React you either spray and pray your state update works or you move everything to store (Redux/Zustand/Xstate) which really is just Foldkit but less guarantee of correctness
Problem:
I was building an chat app with @qavernX and for the most part it works.
All the over-engineering delights like local-sync, virtualization, actual markdown support, inline image support, etc...
https://t.co/jF3SV1iQkJ
Under the hood, we were using PlateUI which under it's hood uses prose-mirror. The issue with that is React will fight prose-mirror on rendering and became really hard to wire in complex logic and feel confident things are working.
https://t.co/lr6i7q3Hjg
I had a few choices
1. Continue building on top of React + Prosemirror and force it to work
2. Wait for NYT to release their solution and use it
3. Build my own one
Me and @qavernX also recently attended a hackathon and felt the power of expressing intent and things just working when we used Effect v4 + effect-atom.
I then stumbled on https://t.co/kN2LHaeVN5 and had the idea of owning the whole flow from service/repo to the rendering engine with foldkit.
Currently, building this rich text editor from scratch has been me applying Gall's principle and it's been fun! Things are working, tests are catching stuff, extending it gives me confidence.
Once the rich text editor is in a good state, I'll tackle the local-sync and see if this is doable as well.
Interestingly, the failure modes doesn’t come from composing messages. It comes from the DOM handlers triggering messages.
Keyboard events, highlighting…
The other thing is because I’m iterating this an adding more complexity. I haven’t found the need yet to lock down the API with tests like story and scene.
Can push up the repo and am open to advise on best practices
Building a Rich Text Editor from scratch!
Taking inspiration from prosemirror with nodes
What's nice about foldkit is state + message forces correctness that I can test like below and it makes adding complexity simple.