A minimal declarative component set designed for LLM generation #5
Conversation
| - `<hstack>` - flexbox with row flex direction. | ||
| - Attributes | ||
| - `gap` - creates a gap between items. Defaults to a standard gap size. | ||
| - `<vstack>` - flexbox with column flex direction. | ||
| - Attributes | ||
| - `gap` - creates a gap between items. Defaults to a standard gap size. | ||
| - `<spacer>` - an empty element with aggressive flex grow and flex shrink to allow it to fill additional available space. |
There was a problem hiding this comment.
No <grid>? You can jerryrig one using stacks I suppose.
There was a problem hiding this comment.
I think we want to avoid fine-grained visd primitives for now, in favor of use-case specific declarative/semantic layout components.
I think flexbox is a minimal layout module that should work for everything else. Notable to me that React Native and SwiftUI have both converged around it. And hstack/vstack/spacer offer the minimal markup surface.
| - `<button>` | ||
| - `<textfield>` | ||
| - `<rich-textfield>` - can wait | ||
| - `<input type="text">` |
There was a problem hiding this comment.
<input type="number> is another good candidate imo
There was a problem hiding this comment.
Idle thought, if we integrate models like Whisper we could even have an input type to accept voice/audio.
|
Take the following with a grain of "chris wouldn't have put in the effort to write a whole RFC for an experiment so maybe he should keep his opinions to himself." Not stated in this RFC (but I, for one, would enjoy reading) is a rationale for why an off-the-shelf component set and/or state management solution is not considered. At the surface level, it seems like the shortest path to answering a question like, "can an LLM organize basic declarative components into something more useful," is one where the components and/or state management does not need to be implemented from scratch. (Thanks for being bothered to write this BTW) |
|
@cdata do you have a component library in mind? Would be a great exercise to take it and figure out the delta between it and our requirements |
|
No, I didn't have a library in mind. But, there is so much unmentioned prior art for the web as to make me wonder about the rationale for something brand new. Authoring good/robust components from scratch is a time consuming and thankless exercise, after all. For some token cases of component libraries, I would suggest picking apart https://ui.shadcn.com/ or https://shoelace.style/. |
|
TY! I'll dig into these. |
|
Closing out stale PRs, please reopen if this is relevant |
Rendered