A tree data structure that emits events on updates, even if the modification is triggered by one of the leaves, making it easier to think in a reactive way.
JavaScript HTML
Clone or download
Latest commit 1698024 Feb 19, 2018



A tree data structure that emits events on updates, even if the modification is emited by one of the leaves, making it easier to think in a reactive way.

Build Status npm version

Are you looking for an immutable.js alternative? Freezer is made with React.js in mind and it uses real immutable structures. It is the perfect store for you application.

Are you looking for a Redux alternative? Using Freezer you don't even need a flux framework, just listen to its update events to refresh your UI. Goodbye boilerplate code!.

What makes Freezer special is:

  • Immutable trees to make fast comparison among nodes.
  • Eventful nodes to notify updates to other parts of the app.
  • No dependencies.
  • Lightweight: ~9KB minified (much less if gzipped).
  • Packaged as UMD module that can be loaded everywhere.
  • Uses common JS array and objects to store the data, so you can use it with your favourite libs like lodash, underscore or ramda

Do you want to know more?



Freezer is available as a npm package.

npm install freezer-js

It is possible to download a file to use it directly in the browser. Grab the full version (~20KB) or minified one (~9KB).


You can play with this example in JSBin.

// Browserify/Node style of loading
var Freezer = require('freezer-js');

// Let's create a freezer object
var freezer = new Freezer({
    a: {x: 1, y: 2, z: [0, 1, 2] },
    b: [ 5, 6, 7 , { m: 1, n: 2 } ],
    c: 'Hola',
    d: null // It is possible to store whatever

// Let's get the frozen data stored
var state = freezer.get();

// Listen to changes in the state
freezer.on('update', function( currentState, prevState ){
    // currentState will have the new state for your app
    // prevState contains the old state, in case you need
    // to do some comparisons
    console.log( 'I was updated' );

// The data is read as usual
console.log( state.c ); // logs 'Hola'

// And used as usual
state.a.z.forEach( function( item ){
    console.log( item );
}); // logs 0, 1 and 2

// But it is immutable, so...
state.d = 3; console.log( state.d ); // logs null
state.e = 4; console.log( state.e ); // logs undefined

// to update, use methods like set that returns new frozen data
var updated = state.set( 'e', 4 ); // On next tick it will log 'I was updated'

console.log( state.e ); // Still logs undefined
console.log( updated.e ); // logs 4

// freezer's data has changed!
freezer.get() !== state; // true
freezer.get() === updated; // true

// The nodes that weren't updated are reused
state.a === updated.a; // true
state.b === updated.b; // true

// Updates can be chained because the new immutable
// data node is always returned
var updatedB = updated.b
    .push( 50 )
    .push( 100 )
    .set(0, 'Updated')
; // It will log 'I was updated' on next tick, just once

// updatedB is the current b property
freezer.get().b === updatedB; // true

// And it is different from the one that started
updated !== freezer.get(); // true
updated.b !== updatedB; // true
console.log( updated.b[0] ); // updated did't/can't change: logs 5
console.log( updatedB[0] ); // logs 'Updated'
console.log( updatedB[4] ); // logs 100
updatedB.length === 5; // true: We added 2 elements and removed 1

// Untouched nodes are still the same
state.a === freezer.get().a; // still true
updated.a === freezer.get().a; // still true

// Reverting to a previous state is as easy as
// set the data again (Undo/redo made easy)
freezer.set( state ); // It will log 'I was updated' on next tick

freezer.get() === state; // true

Why another state holder?

Freezer is inspired by other tree cursor libraries, specifically Cortex, that try to solve an inconvenience of the Flux architecture:

  • If you have a store with deep nested data and you need to update some value from a child component that reflects that data, you need to dispatch an action that needs to look for the bit of data again to update it. That may involve a lot of extra code to propagate the change and it is more painful when you consider that the component already knew what data to update.

On the other hand, data changes always flowing in the same direction is what makes the Flux architecure so easy to reason about. If we let every component update the data independently, we are building a mess again.

So Freezer, instead of letting the child component update the data directly, gives every node the tools to make the change. The updates are always made by the root of the store and the data can keep flowing in just one direction.

Imagine that we have the following tree structure as our app state:

Initial tree

And we have a component responsible for handling the state.c.f ( the yellow node ) part of the data. Its scope is just that part of the tree, so the component receives it as a prop:

// The component receives a part of the freezer data
this.props.branch = { h: 4, i: 5};

Eventually the component is used to update state.c.f.h = 8. You can dispatch an action with the frozen node as the payload ( making it easier for your actions to know what to update ), or even use the node itself to make the change:

this.props.branch.set( {h: 8} );

Then, Freezer will create a new immutable data structure ( a new state for your app ) starting from the top of the tree, and our component will receive a new branch to render. The state ends up like this: Updated tree

Since the whole tree is updated, we can have the main app state in one single object and make the top level components re-render in a reactive way to changes that are made deep in the store hierarchy.

Freezer is strongly influenced by the way that Facebook's Immutable.js handles immutabilty. It creates a new tree every time a modification is required, referencing the non modified nodes from the previous tree. Sharing node references among frozen objects saves memory and boosts the performance of creating new frozens.

Using immutability with React is great, because you don't need to make deep comparisons in order to know when to update a component:

shouldComponentUpdate: function( nextProps ){

    // The comparison is fast, and we won't render the component if
    // it does not need it. This is a huge gain in performance.
    return this.props.prop != nextProps.prop;

Instead of learning the set of methods needed to use Immutable, Freezer's API is much simpler; it uses common JS objects and arrays to store the data, so you can start using it right now. It also makes Freezer much more lightweight (Minified, Immutable is ~56KB and Freezer ~9KB).


Create a freezer object using the constructor:

var freezer = new Freezer({a: 'hola', b:[1,2, [3,4,5]], c: false });

Freezer can be initialized with an object or an array:

var arrayStore = new Freezer( [1, 2, {foo: 'bar'}] );

A freezer object can accept options on initialization:

var freezer = new Freezer({hi: 'hello'}, {mutable: true, live:true});
Name Type Default Description
mutable boolean false Once you get used to freezer, you can see that immutability is not necessary if you learn that you shouldn't update the data directly. In that case, disable immutability in the case that you need a small performance boost.
live boolean false With live mode on, freezer emits the update events just when the changes happen, instead of batching all the changes and emiting the event on the next tick. This is useful if you want freezer to store input field values.
freezeInstances boolean false It's possible to store class instances in freezer. They are handled like strings or numbers, added to the state like non-frozen leaves. Keep in mind that if their internal state changes, freezer won't emit any update event. If you want freezer to handle them as freezer nodes, set 'freezerInstances: true'. They will be frozen and you will be able to update their attributes using freezer methods, but remember that any instance method that update its internal state may fail (the instance is frozen) and wouldn't emit any update event.
singleParent boolean false Freezer allows to add the same node to different parts of the state tree. Updating that node will update all its references that the current state contains. This is a nice feature but it's not ideal if you don't want circular dependencies in your state, in that case set it to true.

And then, Freezer's API is really simple and only has 2 methods: get and set. A freezer object also implements the listener API.


Returns a frozen object with the freezer data.

// Logs: {a: 'hola', b:[1,2, [3,4,5]], c: false }
console.log( freezer.get() );

The data returned is actually formed by arrays and objects, but they are sealed to prevent their mutation and they have some methods in them to update the store. Everytime an update is performed, get will return a new frozen object.

set( data ) & set( key, value )

Replace the current frozen data with new one.

// An example on how to undo an update would be like this...
var freezer = new Freezer({a: 'hola', b:[1,2, [3,4,5]], c: false }),
    state = freezer.get()

var updated = state.set({c: true}); // You can also state.set(c, true)
console.log( updated.c ); // true

// Restore the inital state
freezer.set( state );
console.log( freezer.get().c ); // false


Every time the data is updated, an update event is emited on the freezer object. In order to use those events, Freezer implements the listener API, and on, once, off and emit methods are available on them.

If you need to use the events but you don't want to give access to the complete store, you can use the getEventHub function:

var f = new Freezer({my: 'data'}),
  hub = f.getEventHub()

// Now you can use freezer's event with hub
hub.on('do:action', function(){ console.log('Do it!') });
hub.emit('do:action'); // logs Do it!

// But you don't have access to the store data with it
hub.get(); // undefined

Update methods

Freezer data has three different types of nodes: Objects, Arrays and leaf nodes. A leaf node can't be updated by itself and needs to be updated using its parent node. Every updating method returns a new immutable object with the new node result of the update:

var freezer = new Freezer({obj: {a:'hola', b:'adios'}, arr: [1,2]});

var updatedObj = freezer.get().obj.set('a', 'hello');
console.log( updatedObj ); // {a:'hello', b:'adios'}

var updatedArr = freezer.get().arr.unshift( 0 );
console.log( updatedArr ); // [0,1,2]

// {obj: {a:'hello', b:'adios'}, arr: [0,1,2]}
console.log( freezer.get() );

Both Array and Object nodes have a set method to update or add elements to the node and a reset method to replace the node with other data.

set( keyOrObject, value )

Arrays and hashes can update their children using the set method. It accepts a hash with the keys and values to update or two arguments: the key and the value.

var freezer = new Freezer({obj: {a:'hola', b:'adios'}, arr: [1,2]});

// Updating using a hash
freezer.get().obj.set( {b:'bye', c:'ciao'} );

// Updating using key and value
freezer.get().arr.set( 0, 0 );

// {obj: {a:'hola', b:'bye', c:'ciao'}, arr: [0,2]}
console.log( freezer.get() )

reset( newData )

Reset/replaces the node with new data. Listeners are preserved if the new data is an array or object, so it is possible to listen to reset calls.

var freezer = new Freezer({ foobar: {a: 'a', b: 'b', c: [0, 1, 2] } });

var newfoobar = { foo: 'bar', bar: 'foo' };

var reset = data.foobar.reset(newfoobar);

console.log( reset ); //{ foo: 'bar', bar: 'foo' }

Util methods


Freezer nodes are immutable. toJS transforms Freezer nodes to plain mutable JS objects in case you need them.

// Require node.js assert
var assert = require('assert');

var data = {obj: {a:'hola', b:'adios'}, arr: [1,2]},
    freezer = new Freezer( data )

assert.deepEqual( data, freezer.get().toJS ); // Ok


When pivot is called in a node, all the changes requested in the descendant nodes will return the updated pivoted parent. The pivot will be removed on the next tick.

var freezer = new Freezer({
    people: {
        John: {age: 23},
        Alice: {age: 40}

// If we don't pivot, the updated node is returned
update = freezer.get().people.John.set({age: 18});
console.log( update ); // {age: 18}

// If we want to update two people at
// a time we need to pivot
var update = freezer.get().people.pivot()
    .John.set({age: 30})
    .Alice.set({age: 30})
console.log( update );
// {people:{ John: {age: 30}, Alice: {age: 30} }

The pivot method is really handy because when you have access to a node and update its children, it is the only way of getting the node updated to modify other children.

The pivot is removed on the next tick. This way it won't interfere with other parts of the app.


Using now in a node emits the update method immediately.

var freezer = new Freezer({ test: 'hola' });

freezer.on('update', function( currentState, prevState ){

freezer.get().set({test: 'change'});
// logs 'changed' and then 'event' on the next tick

freezer.get().set({test: 'adios'}).now();
// logs 'event' first and 'changed' after

Use it in cases that you need immediate updates. For example, if you are using React and you want to store an input value outside its component, you'll need to use now because the user can type more than one character before the update method is emited, losing data.

Using Freezer's live option is like using now on every update.

Object methods

remove( keyOrKeys )

Removes elements from a hash node. It accepts a string or an array with the names of the strings to remove.

var freezer = new Freezer({a:'hola', b:'adios', c:'hello', d:'bye'});

var updated = freezer.get()
    .remove('d') // Removing an element
    .remove(['b', 'c']) // Removing two elements

console.log( updated ); //{a: 'hola'}

Array methods

Array nodes have modified versions of the push, pop, unshift, shift, splice and sort methods that update the cursor and return the new node, instead of updating the immutable array node ( that would be impossible ).

var freezer = new Freezer({ arr: [4,3,2,1,0] });

    .sort( (a,b) => a < b ? -1 : 1 ) // [0,1,2,3,4]
    .push( 5 ) // [0,1,2,3,4,5]
    .pop() // [0,1,2,3,4]
    .unshift( 'a' ) // ['a',0,1,2,3,4]
    .shift() // [0,1,2,3,4]
    .splice( 1, 1, 'a', 'b') // [ 0, 'a', 'b', 2, 3, 4]

Array nodes also have the append and prepend methods to batch insert elements at the begining or the end of the array.

var freezer = new Freezer({ arr: [2] });

    .prepend([0,1]) // [0,1,2]
    .append([3,4]) // [0,1,2,3,4]


Freezer objects emit update events whenever their data changes. It is also possible to listen to update events in an intermediate node by creating a listener on it using the getListener method.


Returns a listener that emits an update event when the node is updated. The listener implements the listener API.

var freezer = new Freezer({ arr: [2] }),
    state = freezer.get(),
    listener = state.arr.getListener()

listener.on('update', function( state, prevState ){
    console.log( 'Updated!' );
    console.log( state, prevState );

state.arr.push( 3 ); //logs 'Updated!' [2,3] [2]

Listener API

Freezer instances and listeners implement an API influenced by the way Backbone handles events. The main event that Freezer emits is update, and it is emitted on every node update.

on( eventName, callback )

Register a function to be called when an event occurs.

once( eventName, callback )

Register a function to be called once when an event occurs. After being called the callback is unregistered.

off( eventName, callback )

Can unregister all callbacks from a listener if the eventName parameter is omitted, or all the callbacks for a eventName if the callback parameter is omitted.

emit( eventName [, param, param, ...] )

Trigger an event on the listener. All the extra parameters will be passed to the registered callbacks. emit returns the return value of the latest callback that doesn't return undefined.

  .on('whatever', function(){
    return 'ok';
  .on('whatever', function(){
    // This doesn't return anything

console.log(freezer.emit('whatever')); // logs 'ok'

Use emit with promises to return the value of an asynchronous function when it has completed. For instance:

  .on('whatever', function(data){
    return Ajax.post('/path', data); // This async operation will return a promise

freezer.emit('whatever', data)
  .then((ajax_response) => {
    // Now you can work with ajax_response, the result of the async operation

Event hooks

Freezer objects and nodes also emit beforeAll and afterAll events before and after any other event. Listeners bound to these events also receive the name of the emited event in the arguments.

var Store = new Freezer({a: 1});
Store.on('beforeAll', function( eventName, arg1, arg2 ){
    console.log( event, arg1, arg2 );

Store.get().set({a: 2}); // Will log 'update', {a:2}, {a:1}
Store.emit('add', 4, 5); // Will log 'add', 4, 5

This is a nice way of binding reactions to more than one type of event. It is possible to add some changes to the state inside the beforeAll event, so they will be available for the update event handlers.

var Store = new Freezer({a: 1});
Store.on('beforeAll', function( eventName, arg1, arg2 ){
  if( eventName === 'update' && arg1.a === 2 ){
    arg1.set({message: 'You changed a to be 2!'});

Store.on('update', function( state ){
  console.log( state.message );

Store.get().set({a: 2}); // Logs 'You changed a to be 2!'

Batch updates

At some point you will find yourself wanting to apply multiple changes at a time. The full tree is re-generated on each change, but the only tree you probably need is the final result of all those changes.

Freezer nodes offer a transact method to make local modifications to them without generating intermediate frozen trees, and a run method to commit all the changes at once. This way your app can have really good performance.

var freezer = new Freezer({list:[]}),
    state = freezer.get()

// transact returns a mutable object
// to make all the local changes
var trans = state.list.transact();

// trans is a common array
for( var i = 0; i < 1000; i++ )

// use it as a normal array
trans[0] = 1000; // [1000, 1, 2, ..., 999]

// the store does not know about the yet
freezer.get().list.length == 0; // true

// to commit the changes use the run method in the node

// all the changes are made at once
freezer.get().list; // [1000, 1, 2, ..., 999]

Transactions are designed to always commit the changes, so if you start a transaction but you forget to call run, it will be called automatically on the next tick.

It is possible to update the child nodes of a node that is making a transaction, but it is not really recommended. Those updates will not update the store until the transaction in the parent node is commited, and that may lead to confusion if you use child nodes as common freezer nodes. Updating child nodes doesn't improve the performance much because they have a transacting parent, so it is recommended to make the changes in the transaction node and run it as soon as you have finished with the modifications to prevent undesired behavior.

Usage with React

Creating data-driven React applications using Freezer as your only app state holder is really simple:

1 Wrap your top React component in order to pass the app state as a prop. 2 Re-render on any state change.

That's it, you have your reactive app running:

// My only store
var freezer = new Freezer({/* My initial state */});

var AppContainer = React.createClass({
    render: function(){
        // 1. Your app receives the state
        var state = freezer.get();
        return <App state={ state } />;
    componentDidMount: function(){
        var me = this;
        // 2. Your app get re-rendered on any state change
        freezer.on('update', function(){ me.forceUpdate() });

You can use freezer's update methods in your components, or use it in a Flux-like way without any framework.

Instead of calling actions we can emit custom events, thanks to the open event system built in Freezer. Those events accept any number of parameters.

// State is the Freezer object
freezer.emit('products:addToCart', product, cart);

A dispatcher is not needed either, you can listen to those events directly in the Freezer object.

freezer.on('products:addToCart', function (product, cart) {
    // Update the app state here...

Listener methods that update the state are called reactions, ( we are building reactive applications, are't we? ). It is nice to organize them in files by their domain, as though they were flux stores, but with the difference that all the domains store the data in the same Freezer object.

If you need to coordinate state updates, you can emit new events when a reaction finishes, or listen to specific nodes, without the need for waitFor.

This is all it takes to understand Flux apps with Freezer. No complex concepts like observers, reducers, payloads or action creators. Just events and almost no boilerplate code. It's the simplest way of using React.

You can check this approach working in the TodoMVC sample app, or in the flux comparison project.




MIT licensed