@benev/slate

🪨 @benev/slate

🚧 prerelease wip under constructon subject to change

• frontend ui framework, built on lit
• wonderful web components
• versatile views
• hipster hooks syntax
• satisfying state management
• useful utilities
• top-tier typescript typings

most devs misunderstand how to leverage web components.

you're not supposed to make your whole app out of web components.
they're too cumbersome — web components are html-native, not typescript-native — so they don't take typesafe props, they're referred to by html tag names with poor ide support, and juggling their dom registrations is a pain.

this is why views are important, and are a central feature of slate — views are almost the same as components (they can use shadow-dom), except that views are ergonomically handled via javascript, they accept props, they don't need registration (they're simply imported) — and views enjoy full typescript typings.

you want to think of web components as the tip of your iceberg — they are the entrypoints to your ui — they are the universal control surfaces to help html authors interact with your systems — but below the surface, most of your internals can be made of easily-composable views.

👷 quick start

1. install slate

   npm i @benev/slate
   

2. prepare your app's frontend and context

   import {setup, Context} from "@benev/slate"
   
   export const slate = setup(
     new class extends Context {
       theme = css`
         * {
           margin: 0;
           padding: 0;
           box-sizing: border-box;
         }
       `
     }
   )
   

3. import html and css template functions

   import {html, css} from "@benev/slate"
   

⚙️ components

you can create custom html elements that work in plain html or any web framework.

slate.shadow_component — "carbon"

const styles = css`span {color: yellow}`

export const MyCarbon = slate.shadow_component({styles}, use => {
  const count = use.signal(0)
  const increment = () => count.value++

  return html`
    <span>${count}</span>
    <button @click=${increment}>increment</button>
  `
})

slate.light_component — "oxygen"

export const MyOxygen = slate.light_component(use => {
  const count = use.signal(0)
  const increment = () => count.value++

  return html`
    <span>${count}</span>
    <button @click=${increment}>increment</button>
  `
})

deploying your components

• register components to the dom

  import {register_to_dom} from "@benev/slate"
  
  register_to_dom({
    MyCarbon,
    MyOxygen,
  })
  

• now use your components via html

  <section>
    <my-carbon></my-carbon>
    <my-oxygen></my-oxygen>
  </section>
  

🖼️ views

views are just like components, but are not registered to the dom as custom html elements.
instead, they are used via javascript.
you import them, and inject them into your lit-html templates.
they accept js parameters called props, and are fully typescript-typed.

slate.shadow_view — "obsidian"

const styles = css`span {color: yellow}`

export const MyObsidian = slate.shadow_view({styles}, use => (start: number) => {
  const count = use.signal(start)
  const increment = () => count.value++

  return html`
    <span>${count}</span>
    <button @click=${increment}>increment</button>
  `
})

• auto_exportparts is enabled by default.
  • auto exportparts is an experimental obsidian feature that makes it bearable to use the shadow dom extensively.
  • if auto_exportparts is enabled, and you provide the view a part attribute, then it will automatically re-export all internal parts, using the part as a prefix.
  • thus, parts can bubble up: each auto_exportparts shadow boundary adds a new hyphenated prefix, so you can do css like ::part(search-input-icon).

slate.light_view — "quartz"

export const MyQuartz = slate.light_view(use => (start: number) => {
  const count = use.signal(start)
  const increment = () => count.value++

  return html`
    <span>${count}</span>
    <button @click=${increment}>increment</button>
  `
})

deploying your views

• use a quartz view

  html`<div>${MyQuartz(123)}</div>`
  

  • quartz views are beautifully simple
  • they just take props as arguments
  • without any shadow-dom, they have no stylesheet, and without a wrapping element, they have no attributes
• use an obsidian view

  html`<div>${MyObsidian([123])}</div>`
  

  • obsidian views need their props wrapped in an array
  • when rendered, obsidian views are wrapped in a <obsidian-view> component, which is where the shadow root is attached
  • obsidian views will accept a settings object

    html`
      <div>
        ${MyObsidian([123], {
          content: html`<p>slotted content</p>`,
          auto_exportparts: true,
          attrs: {part: "cool", "data-whatever": true},
        })}
      </div>
    `
    

🪝 use hooks — for views and components

core hooks

• use.state
  works like react useState hook

  const [count, setCount] = use.state(0)
  const increment = () => setCount(count + 1)
  

• use.setup
  perform setup/cleanup on dom connected/disconnected

  use.setup(() => {
    const interval = setInterval(increment, 1000)
    return () => clearInterval(interval)
  })
  

• use.prepare
  initialize a value once

  const random_number = use.prepare(() => Math.random())
  

• use.init
  perform a setup/cleanup, but also return a value

  const scene = use.init(() => {
  
    // called whenever dom is connected
    const scene = setup_3d_scene_for_example()
  
    return [
      canvas, // value returned
      () => scene.cleanup(), // cleanup called on dom disconnect
    ]
  })
  

signal hooks

• use.signal
  create a reactive container for a value (inspired by preact signals)

  const count = use.signal(0)
  const increment = () => count.value++
  

  you can directly inject the whole signal into html

  html`<span>${count}</span>`
  

• use.computed create a signal that is derived from other signals

  const count = use.signal(2)
  const tripled = use.computed(() => count.value * 3)
  console.log(tripled.value) //> 6
  

• use.op
  create an OpSignal in a loading/error/ready state, and it can hold a result value

  const count = use.op()
  count.run(async() => fetchCount("/count"))
  

flatstate hooks

• use.flatstate
  create a reactive object (inspired by mobx and snapstate)

  const state = use.flatstate({count: 0})
  const increment = () => state.count++
  

useful accessors

• use.context
  access to your app's context, for whatever reason

  // wait for all flatstate reactions to complete
  await use.context.flat.wait
  
  // wait for all signal reactons to complete
  await use.context.signals.wait
  
  // access your own custom data you put on the context
  use.context.my_own_custom_data
  

• use.element ~ carbon, oxygen, obsidian
  access the underlying html element

  use.element.querySelector("p")
  

• use.shadow ~ carbon, obsidian
  access to the shadow root

  use.shadow.querySelector("slot")
  

• use.attrs ~ carbon, oxygen
  declare accessors for html attributes

  const attrs = use.attrs({
    start: Number,
    label: String,
    ["data-active"]: Boolean,
  })
  

  set them like normal js properties

  attrs.start = 123
  attrs.label = "hello"
  attrs["data-active"] = true
  

  get them like normal js properties

  console.log(attrs.start) // 123
  console.log(attrs.label) // "hello"
  console.log(attrs["data-active"]) // true
  

  components rerender when any attributes change from outside

🥇 plain elements — gold and silver

gold and silver are "plain" elements, which are alternatives to LitElement.
they're used as primitives underlying our carbon and oxygen components.
for most cases you probably want to stick with carbon/oxygen, and only use gold/silver when you're doing some funky sorcery, or you yearn to go back to a simpler time, without hooks.

consider these imports for the following examples:

import {GoldElement, SilverElement, attributes} from "@benev/slate"

gold element — shadow-dom element

export class MyGold extends GoldElement {
  static styles = css`span {color: blue}`

  #attrs = attributes(this as GoldElement, {
    label: String
  })

  #state = slate.context.flat.state({
    count: 0,
  })

  render() {
    return html`
      <span>${this.#state.count}</span>
      <button @click=${() => this.#state.count++}>
        ${this.#attrs.label}
      </button>
    `
  }
}

silver element — light-dom element

export class MySilver extends SilverElement {

  #attrs = attributes(this as SilverElement, {
    label: String
  })

  #state = slate.context.flat.state({
    count: 0,
  })

  render() {
    return html`
      <span>${this.#state.count}</span>
      <button @click=${() => this.#state.count++}>
        ${this.#attrs.label}
      </button>
    `
  }
}

deploying plain elements

if you want plain elements to have reactivity or have the context's css theme applied, you'll want to run them through slate.components before you register them:

register_to_dom({
  ...slate.components({
    MyGold,
    MySilver,
  }),
})

🔮 deferred context

you can extend the context with anything you'd like to make easily available:

export const slate = setup(new class extends Context {
  my_cool_thing = {my_awesome_data: 123}
})

but since your component modules have to import slate, you might not want to be instancing your context at import-time — so you can defer the creation of your context until later at run-time:

export class MyContext extends Context {
  my_cool_thing = {my_awesome_data: 123}
}

export slate = setup<MyContext>()

//
// ... later, in a different module ...
//

// assign your deferred context at runtime
slate.context = new MyContext()

// just be sure to set context before you register your components
register_to_dom(myComponents)

🛠️ standalone utilities

if you're using slate's frontend components and views, you'll probably be using these utilities via the use hooks, which will provide a better developer experience.

however, the following utilities are little libraries in their own right, and can be used in a standalone capacity.

🛎️ signals

signals are a simple form of state management.

this implementation is inspired by preact signals.

• signal tower

  import {SignalTower} from "@benev/slate"
  
  const signals = new SignalTower()
  

  • signal towers are completely separated from one another
  • you probably only want one in your app (unless you're running isolated tests)
  • you could export your single signal tower from a module, and then import it all over your app
• signals — they hold values

  const count = signals.signal(0)
  const greeting = signals.signal("hello")
  
  count.value++
  greeting.value = "bonjour"
  
  console.log(count.value) //> 1
  console.log(greeting.value) //> "bonjour"
  

• track — react when signals change

  signals.track(() => console.log("doubled", count.value * 2))
  //> doubled 2
  
  count.value = 2
  //> doubled 4
  

• html templating — you can omit .value

  html`<p>count is ${count}</p>`
  

• op signal — to represent async operations

  const json = signals.op<MyJson>()
  
  console.log(json.loading) //> true
  
  await json.run(async() => {
    const data = await fetch_remote_data()
    return JSON.parse(data)
  })
  
  console.log(json.ready) //> true
  console.log(json.payload) //> {"your": "json data"}
  

• computed — signal derived from other signals

  count.value = 1
  
  const tripled = signals.computed(() => count.value * 3)
  
  console.log(tripled.value) //> 3
  

• wait — for debounced tracking

  const tripled = signals.computed(() => count.value * 3)
  console.log(tripled.value) //> 3
  
  count.value = 10
  console.log(tripled.value) //> 3 (too soon!)
  
  await signals.wait
  console.log(tripled.value) //> 30 (there we go)
  

🥞 flatstate

flatstate help you create state objects and react when properties change.

flatstate is inspired by mobx and snapstate, but designed to be super simple: flatstate only works on flat state objects, only the direct properties of state objects are tracked for reactivity.

flatstate basics

• create a flatstate tracking context

  import {Flat} from "@benev/slate"
  
  const flat = new Flat()
    // what happens in this flat, stays in this flat.
    // you probably only want one for your whole app.
  

• make a flat state object

  const state = flat.state({count: 0})
  

• setup a reaction

  flat.reaction(() => console.log(state.count))
    //> 0
  
  state.count++
    //> 1
  

  • flatstate records which state properties your reaction reads
  • flatstate calls your reaction whenever those specific properties change
  • your reaction can listen to more than one state object

flatstate details

• reactions are debounced -- so you may have to wait to see state changes

  const flat = new Flat()
  const state = flat.state({amount: 100})
  
  state.amount = 101
  console.log(state.amount) //> 100 (old value)
  
  await flat.wait
  console.log(state.amount) //> 101 (now it's ready)
  

• you can stop a reaction

  const stop = flat.reaction(() => console.log(state.count))
  
  stop() // end this particular reaction
  

• clear all reactions on a flatstate instance

  // clear all reactions on this flat instance
  flat.clear()
  

flatstate reactions

• so first, there's a simple one-function reaction:

  flat.reaction(() => console.log(state.count))
  

  • flatstate immediately runs the function, and records which properties it reads
  • then, anytime one of those properties changes, it runs your function again
• you can also do a two-function reaction:

  flat.reaction(
    () => ({count: state.count}),
    ({count}) => console.log(count),
  )
  

  • now there's a separation between your "collector" and your "responder"
  • the collector "passes" relevant data to the responder function
  • flatstate calls the responder whenever that data changes
• there's also something called "deepReaction"

  flat.deepReaction(() => console.log(state.count))
  

  • it's the same as "reaction", but it has "discovery" enabled
  • discovery means the collector is checked again for every responder call
  • it's less efficient, but allows you to respond to deeply nested recursive structures
• there's also .auto and .manual reactions
  • these allow you to set options like discovery and debounce (you can turn off the debouncer)
  • but that's bigbrain stuff that you'll have to read the sourcecode about

flatstate advanced

• multiple flatstate instances are totally isolated from each other

  const flat1 = new Flat()
  const flat2 = new Flat()
  

• create readonly access to a state object

  const state = flat.state({count: 0})
  const rstate = Flat.readonly(state)
  
  state.count = 1
  await flat.wait
  console.log(rstate.count) //> 1
  
  rstate.count = 2 // !! ReadonlyError !!
  

  • btw, you can use readonly on anything, not just flatstate

flatstate integration with frontend elements

• let your components rerender on flat state changes

  import {apply} from "@benev/slate"
  
  const MyElement2 = mixin.flat(flat)(MyElement)
    // can also be a class decorator
  
  const elements2 = apply.flat(flat)(elements)
  

  • this works on any BaseElement, which includes LitElement, GoldElement, SilverElement, carbon, and oxygen

💫 op

utility for ui loading/error/ready states.

useful for implementing async operations that involve loading indicators.

you get a better dev-experience if you use ops via signals, but here is the documentation for plain ops on their own, without signals.

• create some ops

  import {Op} from "@benev/slate"
  
  Op.loading()
    //= {mode: "loading"}
  
  Op.error("a fail occurred")
    //= {mode: "error", reason: "a fail occurred"}
  
  Op.ready(123)
    //= {mode: "ready", payload: 123}
  

• you can run an async operation and keep things synchronized

  let my_op = Op.loading()
  
  await Op.run(op => my_op = op, async() => {
    await nap(1000)
    return 123
  })
  

• you can create op signals that have op functionality built in

  const count = use.op()
  
  count.run(async() => {
    await sleep(1000)
    return 123
  })
  

• functions to interrogate an op

    //        type for op in any mode
    //                 v
  function example(op: Op.Any<number>) {
  
    // branching based on the op's mode
    Op.select(op, {
      loading: () => console.log("op is loading"),
      error: reason => console.log("op is error", reason),
      ready: payload => console.log("op is ready", payload)
    })
  
    const payload = Op.payload(op)
      // if the mode=ready, return the payload
      // otherwise, return undefined
  }
  

🪈 pipe

• pipe data through a series of functions
• maybe you've done silly nesting like this:

  // bad
  register_to_dom(
    apply.signals(signals)(
      apply.flat(flat)(
        apply.css(theme)(
          requirement.provide(context)(elements)
        )
      )
    )
  )
  

• now you can do this instead:

  import {Pipe} from "@benev/slate"
  
  // good
  Pipe.with(elements)
    .to(requirement.provide(context))
    .to(apply.css(theme))
    .to(apply.flat(flat))
    .to(apply.signals(signals))
    .to(register_to_dom)
  

• call .done() when you want to return the result

🧐 more useful utils

no time to document these fully, but they're there

• debounce — is a pretty good debouncer
• deep_equal — compare json
• deep_freeze — make stuff immutable
• explode_promise — get a promise flipped inside-out
• generate_id — generate a crypto-random hexadecimal id string
• pub — easy pub/sub tool
• requirement — pass required data to a group of things