innerself

A tiny view + state management solution using innerHTML.

innerHTML is fast. It's not fast enough if you're a Fortune 500 company or even if your app has more than just a handful of views. But it might be just fast enough for you if you care about code size.

I wrote innerself because I needed to make sense of the UI for a game I wrote for the js13kGames jam. The whole game had to fit into 13KB. I needed something extremely small which would not make me lose sanity. innerself clocks in at under 50 lines of code. That's around 600 bytes minified, ~350 gzipped.

innerself is inspired by React and Redux. It offers the following familiar concepts:

• composable components,
• a single store,
• a dispatch function,
• reducers,
• and even an optional logging middleware for debugging!

It does all of this by serializing your component tree to a string and assigning it to innerHTML of a root element. It even imitates Virtual DOM diffing by comparing last known output of components with the new one :) I know this sounds like I'm crazy but it actually works quite nice for small and simple UIs.

If you don't care about size constraints, innerself might not be for you. Real frameworks like React have much more to offer, don’t sacrifice safety, accessibility, nor performance, and you probably won’t notice their size footprint.

innerself was a fun weekend project for me. Let me know what you think!

You need to know a few things before you jump right in. innerself is a less-than-serious pet project and I don't recommend using it in production.

It's a poor choice for form-heavy UIs. It tries to avoid unnecessary re-renders, but they still happen if the DOM needs even a tiniest update. Your form elements will keep losing focus because every re-render is essentially a new assignment to the root element's innerHTML.

When dealing with user input in serious scenarios, any use of innerHTML requires sanitization. innerself doesn't do anything to protect you or your users from XSS attacks. If you allow keyboard input or display data fetched from a database, please take special care to secure your app. The innerself/sanitize module provides a rudimentary sanitization function.

Perhaps the best use-case for innerself are simple mouse-only UIs with no keyboard input at all :)

• A moment lost in time. - a first-person exploration puzzle game by @michalbe and myself. I originally wrote innerself for this.
• Innerself Hacker News Clone - a Hacker News single page app by @bsouthga with innerself as the only dependency. Also serves as an example of a TypeScript innerself app.
• Reach/Steal Draft Tracker - a fantasy football draft tracker by @bcruddy that tests the rendering performance with 300+ table rows backed by an expressjs server.
• TodoMVC - a TodoMVC app based on innerself by @Cweili.

$ npm install innerself

For a more structured approach @bsouthga created innerself-app. Use it to bootstrap new innerself apps from a predefined template.

innerself expects you to build a serialized version of your DOM which will then be assigned to innerHTML of a root element. The html helper allows you to easily interpolate Arrays.

import html from "innerself";
import ActiveTask from "./ActiveTask";

export default function ActiveList(tasks) {
    return html`
        <h2>My Active Tasks</h2>
        <ul>
            ${tasks.map(ActiveTask)}
        </ul>
    `;
}

The state of your app lives in a store, which you create by passing the reducer function to createStore:

const { attach, connect, dispatch } = createStore(reducer);
window.dispatch = dispatch;
export { attach, connect };

You need to make dispatch available globally in one way or another. You can rename it, namespace it or put it on a DOM Element. The reason why it needs to be global is that the entire structure of your app must be serializable to string at all times. This includes event handlers, too.

import html from "innerself";

export default function ActiveTask(text, index) {
    return html`
        <li>
            ${text} ${index}
            <button
                onclick="dispatch('COMPLETE_TASK', ${index})">
                Mark As Done</button>
        </li>
    `;
}

You can put any JavaScript into the on<event> attributes. The browser will wrap it in a function which takes the event as the first argument (in most cases) and in which this refers to the DOM Element on which the event has been registered.

The dispatch function takes an action name and a variable number of arguments. They are passed to the reducer which should return a new version of the state.

const init = {
    tasks: [],
    archive: []
};

export default function reducer(state = init, action, args) {
    switch (action) {
        case "ADD_TASK": {
            const {tasks} = state;
            const [value] = args;
            return Object.assign({}, state, {
                tasks: [...tasks, value],
            });
        }
        case "COMPLETE_TASK": {
            const {tasks, archive} = state;
            const [index] = args;
            const task = tasks[index];
            return Object.assign({}, state, {
                tasks: [
                    ...tasks.slice(0, index),
                    ...tasks.slice(index + 1)
                ],
                archive: [...archive, task]
            });
        }
        default:
            return state;
    }
}

If you need side-effects, you have three choices:

• Put them right in the on<event> attributes.
• Expose global action creators.
• Put them in the reducer. (This is considered a bad practice in Redux because it makes the reducer unpredictable and harder to test.)

The dispatch function will also re-render the entire top-level component if the state changes require it. In order to be able to do so, it needs to know where in the DOM to put the innerHTML the top-level component generated. This is what attach returned by createStore is for:

import { attach } from "./store";
import App from "./App";

attach(App, document.querySelector("#root"));

createStore also returns a connect function. Use it to avoid passing data from top-level components down to its children where it makes sense. In the first snippet above, ActiveList receives a tasks argument which must be passed by the top-level component.

Instead you can do this:

import html from "innerself";
import { connect } from "./store";
import ActiveTask from "./ActiveTask";
import TaskInput from "./TaskInput";

function ActiveList(state) {
    const { tasks } = state;
    return html`
        <h2>My Active Tasks</h2>
        <ul>
            ${tasks.map(ActiveTask)}
            <li>
                ${TaskInput()}
            </li>
        </ul>
    `;
}

export default connect(ActiveList);

You can then avoid passing the state explicitly in the top-level component:

import html from "innerself";
import { connect } from "./store";

import ActiveList from "./ActiveList";
import ArchivedList from "./ArchivedList";

export default function App(tasks) {
    return html`
        ${ActiveList()}
        ${ArchivedList()}
    `;
}

Connected components always receive the current state as their first argument, and then any other arguments passed explicitly by the parent.

innerself comes with an optional helper middleware which prints state changes to the console. To use it, simply decorate your reducer with the default export of the innerself/logger module:

import { createStore } from "innerself";
import withLogger from "innerself/logger";
import reducer from "./reducer"

const { attach, connect, dispatch } =
    createStore(withLogger(reducer));

I know, I know. But it works! Check out the examples:

• example01 - an obligatory Todo App.
• example02 by @flynnham.
• example03 illustrates limitations of innerself when dealing with text inputs and how to work around them.

The update cycle starts with the dispatch function which passes the action to the reducer and updates the state.

When the state changes, the store compares the entire string output of top-level components (the ones attached to a root element in the DOM) with the output they produced last. This means that most of the time, even a slightest change in output will re-render the entire root.

It's possible to dispatch actions which change the state and don't trigger re-renders. For instance in example01 the text input dispatches CHANGE_INPUT actions on keyup events. The current value of the input is then saved in the store. Crucially, this value is not used by the TaskInput component to populate the input element. The whole thing relies on the fact that the native HTML input element stores its own state when the user is typing into it.

This limitation was fine for my use-case but it's worth pointing out that it badly hurts accessibility. Any change to the state which causes a re-render will make the currently focused element lose focus.

React is of course much smarter: the Virtual DOM is a lightweight representation of the render tree and updates to components produce an actual diff. React maps the items in the Virtual DOM to the elements in the real DOM and is able to only update what has really changed, regardless of its position in the tree.

Here's an interesting piece of trivia that I learned about while working on this project. React only re-renders components when their local state changes, as signaled by this.setState(). The fact that it also looks like components re-render when their props change derives from that as well. Something needs to pass those props in, after all, and this something is the parent component which first needs to decide to re-render itself.

When you think about how you can connect components with react-redux to avoid passing state to them from parents it becomes clear why behind the scenes it calls this.setState(dummyState) (which is an empty object) to trigger a re-render of the connected component :) It does this only when the sub-state as described by the selector (mapStateToProps) changes, which is easy to compute (and fast) if the reducers use immutability right. In the best case scenario it only needs to compare the identity of the sub-state to know that it's changed.