Web Components hypothesis

[tucnak]

I chose to use Svelte for this project.

Much can be said about the unreasonable efficiency of dramatic writing as a UI/UX paradigm.

In the recent years, many chat interfaces have been created to accommodate LLM's, RAG, and various agentic architectures. I believe they may have rushed to implement primitives that are not really good, and fundamentally limiting. Do we really need to be stuck in chat threads circa 2000's? A play is able to order events, speech and actions, and each primitive reads the same regardless of subject: not just for the actors that would take active part by being in the room, but also for the reader that would passively observe the unfolding events.

README

Now, it's true that the project is a mess, currently, I'm very happy about my choice. However, the more I think about this project and what I need it for specifically, the more I'm willing to recognise that its primitives are really simple. That is to say, it would be a shame, if it couldn't be consumed simply, too. Let's investigate whether we could target Web Components, how hard would that be, and most importantly—say, would React people appreciate it?

First big question: our dependencies, think Carta but also the transient ones, could they somehow impede our progress? I don't understand Web Components very well, or Svelte, for that matter—what are the constraints here? Are there inherent incompatibilities we have to be aware of? Svelte compiles, and it's supposedly very efficient! I'm curious how well it translates to Web Components. Hypothetically speaking, Papir's lightweight footprint could make it attractive for embedded.

Think e-ink displays!

The goal here is to make an informed decision about whether Web Components is a viable target for us.

[tucnak]

The jury is out!

That is to say, Web Components is absolutely the direction we're taking on. The original "clever paper" demo was written in Svelte 4, and later when it became Papir, we had contemplated a lot whether we should support Svelte 5. They changed a lot, and my initial impression was that it went in the wrong direction. Most importantly, they removed slots as such, replacing them with snippets; the consensus so far is such that it would be a massacre to support both 4 and 5, especially considering we were not able to get a 5 build in the first place.

However, courtesy of Web Components, this is no longer a problem for us! You can see how it's done practically in the v4 documentation on Custom elements. There are some limitations! Roughly speaking, you can't have any two-way bindings, everything is passed through attributes (nothing new here) and interaction between any two Web Components happens through HTML events.

I'll be making a few changes myself shortly.