Purview

What if your React components ran on the server-side? The server renders components to HTML and sends it to the client. The client renders HTML and notifies the server of DOM events. The server executes event handlers and lifecycle events, and it maintains the state of each component.

With this architecture, your components can directly make database queries, contact external services, etc, as they're running exclusively on the server. There's no more REST or GraphQL; the client-server interface is abstracted away, and all you deal with are standard components, event handlers, and lifecycle events.

Below is a snippet of an example; see full example code here.

import Purview, { css } from "purview"
import { Sequelize, QueryTypes } from "sequelize"

const db = new Sequelize("sqlite:purview.db")

class Counter extends Purview.Component<{}, { count: number }> {
  async getInitialState(): Promise<{ count: number }> {
    // Query the current count from the database.
    const rows = await db.query<{ count: number }>(
      "SELECT count FROM counters LIMIT 1",
      { type: QueryTypes.SELECT },
    )
    return { count: rows[0].count }
  }

  increment = async () => {
    await db.query("UPDATE counters SET count = count + 1")
    await this.setState(await this.getInitialState())
  }

  render(): JSX.Element {
    return (
      // Atomic CSS-in-JS support is built-in.
      <div css={css({ border: "1px solid #333", padding: "10px" })}>
        <p>The count is {this.state.count}</p>
        <button onClick={this.increment}>Click to increment</button>
      </div>
    )
  }
}

Benefits

• Make database queries, contact external services, etc. directly within your components, with no need for REST or GraphQL.
• Extensive type-checking: Comprehensive JSX typings ensure that your HTML tags/attributes, event handlers, component props, etc. are all statically type-checked, courtesy of TypeScript.
• Server-side rendering is the default, so you get fast time to first meaningful paint.
• Persistent two-way WebSocket connections allow the server to trigger updates at any time. You can push realtime changes from your database or external services directly to the client with a simple call to this.setState().
• Client-side virtual DOM diffing for efficient updates.
• Your front-end and back-end are both encapsulated into reusable components. It's easy to see and modify the functionality of any part of your page.

Caveats

• Every event and re-render incurs a network round-trip cost. Applications that require minimal latency (e.g. animations, games) are not well suited for Purview. That being said, many applications are primarily CRUD-based, and hence work well under Purview's architecture.
• Not React compatible due to the differences listed below, so you can't use existing React components/libraries with Purview.
• You can't directly access the DOM within your components. For example, if you need to attach listeners to window, that's currently unsupported.

Installation

1. Install with npm:

   $ npm install purview

2. Set your JSX transform to be Purview.createElem. For TypeScript, in your tsconfig.json, you can do this like so:

   {
     "compilerOptions": {
       "jsx": "react",
       "jsxFactory": "Purview.createElem"
     }
   }

   For other compilers/transpilers, you can use the JSX comment pragma: /* @jsx Purview.createElem */.

   You can also reference our full tsconfig.json, which enables various strict TypeScript features that we'd recommend.

Usage

1. Write components by extending Purview.Component.
2. Send down (a) the server-rendered HTML of your component and (b) a script tag pointing to Purview's client-side JS file. If you'd like to enable Purview's CSS-in-JS support, also send (c) the server-rendered CSS (a style tag) in the head of your document.
   • For (a), call Purview.render(<Component />, req), where Component is your root component, and req is the standard request object, of type http.IncomingMessage, from express or http.createServer. This returns a promise with HTML.
   • For (b), either serve the JavaScript in Purview.scriptPath directly (see example below) or, in an existing client-side codebase, import "purview/dist/browser".
   • For (c), call Purview.renderCSS(req), where req is the standard request object, of type http.IncomingMessage, from express or http.createServer. This returns a promise with HTML containing a style tag that should be inserted in the head of the document. Make sure to call this after Purview.render() in step (a), as the CSS depends on the components that are being rendered.
3. Handle WebSocket connections by calling Purview.handleWebSocket(server, options), where server is an http.Server object. If you're using Express, call http.createServer(app) to a create a server from your app object. Then call server.listen() instead of app.listen() to bind your server to a port.
   • options should be an object with one key: origin, whose value is a string.
   • origin should be the protocol and hostname (along with the port if it's non-standard) of the server (e.g. https://example.com). This is used to perform WebSocket origin validation, ensuring requests originate from your server. You can set origin to null to skip origin validation, but this is not recommended.
   • Note that, if you incorrectly specify origin, the page will keep refreshing in an attempt to re-connect the WebSocket.

Below is a full working example:

import Purview, { css } from "purview"
import { Sequelize, QueryTypes, DataTypes } from "sequelize"
import * as http from "http"
import * as express from "express"

const db = new Sequelize("sqlite:purview.db")

// (1) Write components by extending Purview.Component. The two type parameters
// are the types of the props and state, respectively.
class Counter extends Purview.Component<{}, { count: number }> {
  async getInitialState(): Promise<{ count: number }> {
    // Query the current count from the database.
    const rows = await db.query<{ count: number }>(
      "SELECT count FROM counters LIMIT 1",
      { type: QueryTypes.SELECT },
    )
    return { count: rows[0].count }
  }

  increment = async () => {
    await db.query("UPDATE counters SET count = count + 1")
    await this.setState(await this.getInitialState())
  }

  render(): JSX.Element {
    return (
      // Atomic CSS-in-JS support is built-in.
      <div css={css({ border: "1px solid #333", padding: "10px" })}>
        <p>The count is {this.state.count}</p>
        <button onClick={this.increment}>Click to increment</button>
      </div>
    )
  }
}

async function startServer(): Promise<void> {
  // (2) Send down server-rendered HTML and CSS alongside a script tag with
  // Purview's client-side JavaScript.
  const app = express()
  app.get("/", async (req, res) => {
    const counterHTML = await Purview.render(<Counter />, req)
    // Make sure to call Purview.renderCSS() after Purview.render(), as the CSS
    // is determined by the components rendered on the page.
    const styleHTML = await Purview.renderCSS(req)
    res.send(`
      <html>
        <head>
          ${styleHTML}
        </head>
        <body>
          ${counterHTML}
          <script src="/script.js"></script>
        </body>
      </html>
    `)
  })
  app.get("/script.js", (_, res) => res.sendFile(Purview.scriptPath))

  // (3) Handle WebSocket connections.
  const server = http.createServer(app)
  const port = 8000
  Purview.handleWebSocket(server, {
    origin: `http://localhost:${port}`,
  })

  // Reset database and insert our initial counter.
  db.define("counter", { count: DataTypes.INTEGER }, { timestamps: false })
  await db.sync({ force: true })
  await db.query("INSERT INTO counters (count) VALUES (0)")

  server.listen(port, () => console.log(`Listening on localhost:${port}`))
}

startServer()

Differences from React

Purview mimics React in many ways, but differs significantly when it comes to event handlers, controlled form inputs, and getInitialState().

Event handlers

Because your components run on the server-side, your event handlers are not passed standard DOM event objects. Instead, Purview determines relevant information associated with certain events and creates its own event objects. Here's a description of the event object that Purview passes to your handler for various event types:

• onInput: The event object is of type InputEvent<T> = { name: string, value: T }, where name is the name of the input and value is its value. T is boolean for checkboxes, number for <input type="number">, and string for all other inputs.

• onChange: The event object is of type ChangeEvent<T> = { name: string, value: T }, where name is the name of the input and value is its value. T is boolean for checkboxes, number for <input type="number">, string[] for <select multiple> and string for all other inputs.

• onKeyDown, onKeyPress, and onKeyUp: The event object is of type KeyEvent = { name: string, key: string }, where name is the name of the input and key is the key that was pressed.

• onSubmit: The event object is of type SubmitEvent = { fields: Record<string, unknown> }. fields is a mapping of form field names to values. It is your responsibility to perform validation on fields for both the types and values, just as you would do if you were writing a server-side route handler.

  When you add an onSubmit handler, the default action of the submit event is automatically prevented (i.e. via event.preventDefault()). This stops the browser from navigating to a different page.

All other event handlers are passed no arguments.

Controlled Form Inputs

If you specify a value attribute on a text input or textarea, a checked attribute on a radio/checkbox input, or a selected attribute on an option, the form input will be controlled. Upon each re-render, the value will be forcibly set to the value you specify.

Unlike React, a controlled form input's value can be modified, but it'll be reset to the specified value when re-rendered. To prevent modification, use the standard readonly or disabled HTML attributes.

Purview does not let you specify a value attribute for select tags like React does. Instead, you must use the selected attribute on option tags, just like you would in regular HTML. Purview controls the select given at least one option has a selected attribute.

If you want to set an initial, uncontrolled value, use the attribute defaultValue for text inputs and textareas, defaultChecked for radio/checkbox inputs, and defaultSelected for options.

Do note that events require a round-trip to the server, so controlling form inputs is more expensive than in React. That being said, it's quite fast given a reasonable Internet connection, and this expense can often be ignored.

getInitialState()

Components can define a getInitialState() function that returns a promise with the initial state of the component. This can be used to e.g. fetch information from a database or service prior to the component rendering.

The call to Purview.render() returns a promise that resolves once all initial state has been fetched and components have been rendered. This prevents the user from seeing a flash of empty content before your components load their state.

Do not assign/modify instance variables on your components within getInitialState(). On page load, when the WebSocket connection is established, Purview will re-initialize all components with their saved state from the last render, and it won't call getInitialState(). Hence, any instance variables you assign/modify within this function may not be reflected.

Other differences

In addition to the above, Purview also differs from React in the following ways:

• The only supported lifecycle methods are componentDidMount(), componentWillReceiveProps(), and componentWillUnmount().
• Context, refs, fragments, error boundaries, portals, and hooks are unsupported.

Atomic CSS-in-JS support

Purview comes with built-in atomic CSS-in-JS support. To enable CSS-in-JS, make sure to send the rendered CSS to the client per step 2 of the usage instructions.

To use CSS-in-JS, first call Purview's css function to generate a set of CSS properties. Note that the object passed to the css function is typechecked thanks to CSSType, and it expects camelCased keys representing CSS properties or pseudo classes with nested CSS properties (see example below). Then, pass the return value to the css={...} attribute of any standard JSX element.

const buttonStyles = css({
  padding: "6px 8px",
  fontSize: "1.4rem",
  backgroundColor: "#ccc",
  ":hover": {
    textDecoration: "underline",
  },
})

class Button extends Purview.Component<{}, {}> {
  render(): JSX.Element {
    return <button css={buttonStyles}>Button</button>
  }
}

Note that you can compose CSS rules by passing multiple objects to the css function:

const blueStyles = css({ backgroundColor: "blue" })
// Combines all styles in buttonStyles and in blueStyles, returning a new object
// representing the joint properties. If there are conflicts, styles in
// blueStyles will take precedence over styles in buttonStyles.
const blueButtonStyles = css(buttonStyles, blueStyles)

class BlueButton extends Purview.Component<{}, {}> {
  render(): JSX.Element {
    return <button css={blueButtonStyles}>Blue Button</button>
  }
}

The above code can be expressed more succinctly by using Purview's styledTag helper function, which will accept a tag name and CSS, and return a new component that renders that tag with the given CSS:

const BlueButton = styledTag("button", buttonStyles, blueStyles)
// BlueButton can now be used just like any other component. If the following
// JSX is rendered, it'll appear the same as in the example above:
<BlueButton>Blue Button</BlueButton>

Styles do not need to be static--you can generate and change them dynamically (even based on props/state, but be careful about user input), and Purview will update the DOM appropriately. Purview checks that all CSS rules are valid and will let you know if there are issues.

When Purview renders the button in the example above, the markup will look like the following:

<!-- in the head -->
<style>
  .p-a { padding-top: 6px }
  .p-b { padding-right: 8px }
  .p-c { padding-bottom: 6px }
  .p-d { padding-left: 8px }
  .p-e { font-size: 1.4rem }
  .p-f { background-color: blue }
  .p-g:hover { text-decoration: underline }
</style>

<!-- in the body -->
<button class="p-a p-b p-c p-d p-e p-f p-g">Blue Button</button>

In the markup above, notice how blueButtonStyles has been split up into multiple CSS classes, each of which has one declaration. This is called atomic CSS-in-JS, and it's described in detail by Sébastian Lorber.

In short, atomic CSS-in-JS solves the following problems (among others, all identified by Christopher Chedeau in his talk):

• Global namespace: CSS has one global namespace. If two selectors happen to use the same class, tag, or ID, the rules could unintentionally affect one another. Purview's atomic CSS-in-JS is locally scoped because it generates unique class names for each unique property and is directly integrated into JSX.
• Dead code elimination: It's hard to identify CSS selectors that are no longer used without auditing a website in full. With atomic CSS-in-JS, we get dead code elimination for free: TypeScript and ESLint/TSLint can identify unused variables, and only the CSS that is included in a rendered component via the css={...} attribute is actually sent to the client.
• Non-deterministic resolution: If an element has two different classes that define conflicting CSS properties, the properties that take precedence are based on the specificity of the CSS selectors, or if the specificity is the same, the ordering of the selectors in the stylesheet. This leads to unexpected behavior. Atomic CSS-in-JS has deterministic resolution--the last property passed to the css function is the one that takes precedence, as in the BlueButton example above.
• Minification: With classic CSS, each new component requires new CSS rules to be written, so CSS grows linearly. With Atomic CSS-in-JS, if an individual rule has been used in the past by any component, it is reused thanks to the single-purpose classes. This allows CSS to grow much more slowly; large sites like Facebook have experienced significant savings per Sébastian's article linked above (413KB before to 74KB after).

Note that traditional CSS-in-JS frameworks, like Styled Components and Emotion, don't work with Purview, since Purview components do not run in the browser.

Error handling

Normally, when an error occurs in the context of a Purview component, it will appear as top-level uncaughtException or unhandledRejection events. Purview provides the option to observe or modify these errors with a custom error handler before they appear within these top-level events.

In order to add a custom error handler, pass an onError function as an option to Purview.render(). When provided, the handler is invoked for these classes of Purview errors:

• Errors within event callbacks.
• Errors in the component render path.
• Errors when setting initial component state, after WebSocket connection is established.

If your handler does not itself throw an error, Purview will re-throw the error in order to prevent execution of further code.

app.get("/", async (req, res) => {
  function onError(error: unknown): void {
    // Instrument the error here, such as to add context:
    error.userID = req.userID
    // Or send it to a monitoring service:
    logError(error)
  }

  const appHTML = Purview.render(<App />, req, { onError })
  // ...
})

Using Purview with load balancing / multiple processes

When a user makes a request to a web server using Purview, Purview will initialize and render the root component (and any sub-components) on the server, and it'll return the rendered HTML to the client. By default, Purview will retain the component state in memory on the server.

Purview comes with client-side JS that sets up a persistent WebSocket connection. This connection is used by the client to notify the server of events that occur, and for the server to send re-rendered HTML to the client.

When the client makes a request to initiate the WebSocket connection, Purview re-uses the component state that it had saved in memory. This abstraction breaks down if there is load balancing where multiple processes (or servers) are handling requests--the initial web request may go to a different process/server than the WebSocket request, so storing the component state in memory is not sufficient.

In order to handle this case, you can tell Purview how to store and load component state by overriding the functions in Purview.reloadOptions. It's generally advised to serialize and persist component state in a central database that any process/server can access--then, even if the WebSocket request goes to a different process/server, that process/server can reload the component state.

Component state is represented by the StateTree type, which contains the state of all components in a hierarchical, tree-like structure. If you put types that aren't easily serializable in state, you may need to recursively traverse the StateTree and convert any such types accordingly. For this reason, it's generally recommended to use JSON-serializable values in component state.

The reload functions are as follows:

// Stores the state tree and associates it with the provided ID. This should
// act like an update or insert (upsert) operation, since the same ID may be
// used to update a state tree.
Purview.reloadOptions.saveStateTree(id: string, tree: StateTree): Promise<void>

// Returns the previously stored state tree with the given ID, or null if no
// such tree exists.
Purview.reloadOptions.getStateTree(id: string): Promise<StateTree | null>

// Deletes the state tree with the given ID. It's recommended to also delete
// sufficiently old trees (i.e. trees that haven't been updated in a few days)
// within this function.
Purview.reloadOptions.deleteStateTree(id: string): Promise<void>

If you're using atomic CSS-in-JS, you'll also need to tell Purview how to save and reload CSS state. CSS state is represented by the CSSState type, which is JSON serializable.

// Stores the CSS state and associates it with the provided ID. This should
// act like an update or insert (upsert) operation, since the same ID may be
// used to update CSS state.
Purview.reloadOptions.saveCSSState(id: string, cssState: CSSState): Promise<void>

// Returns the previously stored CSS state with the given ID, or null if no
// such state exists.
Purview.reloadOptions.getCSSState(id: string): Promise<CSSState | null>

// Deletes the CSS state with the given ID. It's recommended to also delete
// sufficiently old state (i.e. state that hasn't been updated in a few days)
// within this function.
Purview.reloadOptions.deleteCSSState(id: string): Promise<void>

Note that these reload functions are also used when a WebSocket connection is closed/re-established--in this way, they help account for disconnects and flaky connections.

Unique component names

In order to reload component state as described in the section above, Purview needs to be able to associate state with a component, which means it has to be able to identify a component with some unique, serializable data type. Purview uses the name of the component class by default as the unique identifier. If two component classes have the same name, Purview will throw an error at runtime and inform you of the conflict accordingly. Note that you may alternatively define a static getUniqueName function to return a unique name for your component, which would then let you use the same class name for multiple components.

Because the unique name is used for reloading the state of the component for the initial web socket connection, any disconnects, and code restarts, it should be deterministic--i.e. even if the app is restarted or a different process handles a request, the unique name for a class should remain the same. If the name changes, Purview won't be able to reload the state correctly and the page may not work as expected until it is refreshed.

Inspiration

Phoenix Live View -- https://www.youtube.com/watch?v=Z2DU0qLfPIY

Contributors

Karthik Viswanathan -- [email protected]

If you're interested in contributing to Purview, get in touch with me via the email above. I'd love to have you help out and potentially join the core development team.

License

Purview is MIT licensed.