Simple and tiny reactive state library.
Statin is heavily inspired by MobX. It was created as an attempt to get a MobX like reactivity in as little code as possible.
Features:
- Tiny, <4KB min, <2KB gz.
- Fully typed.
- View source friendly, it's just a single medium sized file.
- Clean, straight forward, no-magic API.
- No proxies, decorators, or other abstractions that introduce non-standard behavior to wrapped objects.
- Error handling and recovery.
- Circular reaction detection.
Requirements:
WeakMap
: 96.48% browser support as of January 2022 (IE 11+).Symbol
: 95.57% browser support as of January 2022 (Edge+).
Bindings:
Examples:
- Quick interactive example.
- TodoMVC app example using preact, statin, and poutr.
npm install statin
import {Signal, signal, computed, reaction} from 'statin';
// Create a signal
const number = signal(4);
// Get value
console.log(number()); // 4
// Set value
number(5);
// Create a signal that holds an array of other signals
const numbers = signal<Signal<number>[]>([number]);
// Create a computed signal that updates only
// when one of its dependencies changes.
const sum = computed(() => numbers().reduce((sum, value) => sum + value(), 0));
// Create a reaction
const dispose = reaction(() => {
console.log(`Sum of all numbers is ${sum()}.`);
});
// console.log: Sum of all numbers is 5.
// Update observed signal
number(6);
// console.log: Sum of all numbers is 6.
// Add a new number signal to numbers by mutating it
numbers.edit((array) => array.push(signal(2)));
// console.log: Sum of all numbers is 8.
// Cancel reaction
dispose();
Statin was created as a state keeping library for Drovp, which is a non-trivial app with a lot of state requirements. Drovp is as of time of writing in a stable beta, and has no known state/statin related issues.
I consider statin to be stable and feature complete. That's not to say there aren't any issues or improvements to be made. If you've noticed something I've missed, feel free to create an issue.
Everything exported by the module:
Interface returned by signal()
creator below.
interface Signal<T extends unknown = unknown> {
(): T;
(value: T): void;
value: T;
changed: () => void;
edit: (editor: (value: T) => void) => void;
// Aliases
get: () => T;
r: () => T;
set: (value: T) => void;
w: (value: T) => void;
}
function signal<T extends unknown = unknown>(initial: T): Signal<T>;
Creates a signal that will trigger its observers any time a new value is assigned to it. Signal is a function that returns a value when called with no arguments, or sets a value when called with 1 argument. Example:
const name = signal('John');
// Read value
console.log(name()); // 'John'
// Set value
name('Mike');
console.log(name()); // 'Mike'
NOTE: Observers are only notified when value being assigned is different than the current one. Assigning same value will not cause any reactions to run.
Signal also has these properties and methods:
This is a reference to the current signal value. You can use it to read/write the value in non-reactive way. Only touch this if you really know what you're doing, as misuse will lead to stale and/or incorrect state.
() => void;
A method that allows manually sending changed signal to all of the signal's observers.
(editor: (value: T) => void) => void;
A convenience method to edit mutable signal values such as arrays, maps, sets,... Example:
const set = signal(new Set<string>());
// Mutates the set and sends changed signal afterwards
set.edit((set) => set.add('foo'));
// This is essentially just a shorthand for
set.value.add('foo');
set.changed();
Value returned by editor function is ignored as signal.edit()
is meant for mutating. If you want to instead swap the value for a new one, just do signal(newValue)
.
Interface returned by computed()
creator below.
interface Computed<T extends unknown> {
(): T;
}
function computed<T extends unknown>(compute: (prev: T | undefined) => T): Computed<T>;
Creates a cached computed value that re-computes and sends changed signal to its observers when any of the signals it depends on changes. compute()
function receives the previous value as its 1st argument.
IMPORTANT: Writing into signals inside computed(() => { ... })
is not allowed, and will throw an error.
Example:
const n = signal(3);
const nFactorial = computed(() => factorial(n()));
// 1st computation of `factorial(3)`
console.log(nFactorial()); // 6
// returns cached result from 1st
console.log(nFactorial());
n(4);
// 2nd computation of `factorial(4)`
console.log(nFactorial()); // 24
When computed throws, the error is stored, and thrown to each consumer until the computed state is invalidated by any of the signals it depends on (signals read before the error happened). This way computed values can recover.
function action<T extends unknown>(fn: () => T): T;
Calls fn()
in an action context, which queues and de-duplicates all signal changes, and sends them to their observers only after it completed. Example:
const value = signal('init');
reaction(() => console.log(value())); // console.log: 'init'
value('foo'); // console.log: 'foo'
value('bar'); // console.log: 'bar'
action(() => {
value('foo');
value('bar');
});
// console.log: 'bar'
NOTE: Signals read inside an action are not tracked! This means if you execute an action inside a reaction, the signals read inside an action will not be added as dependencies to the parent reaction.
function createAction<T extends (...args: unknown[]) => unknown)>(run: T): T;
A convenience function to wrap common methods in an action. Example:
const foo = signal(5);
const addToFoo = createAction((amount: number) => foo(foo() + amount));
addToFoo(2);
console.log(foo()); // 7
// This is essentially just a shorthand for:
const addToFoo = (amount: number) => action(() => foo(foo() + amount));
Wrapped methods inherit argument and return types.
Why do you need this if it's pretty much the the same amount of characters? When used on a class method, it prettifies better imo :)
type Disposer = () => void;
interface ReactionAction<T extends unknown> {
(dispose: Disposer): T;
}
export interface ReactionEffect<T extends unknown> {
(value: T, dispose: Disposer): void;
}
interface ReactionOptions {
immediate?: boolean; // default: false
onError?: (error: Error, dispose: Disposer) => void;
}
function reaction<T extends unknown>(action: ReactionAction<T>, options?: ReactionOptions): Disposer;
function reaction<T extends unknown>(
action: ReactionAction<T>,
effect: ReactionEffect<T>,
options?: ReactionOptions
): Disposer;
A reaction creator that returns a disposer. It has two supported call signatures:
reaction(action, options?)
const foo = signal(5);
const dispose = reaction((dispose) => {
// This will run when reaction is created,
// and every time `foo` signal changes.
console.log(foo());
// Can also dispose from inside
dispose();
});
reaction(action, effect, options?)
const foo = signal(5);
const dispose = reaction(
// This runs on init and every time `foo` changes
(dispose) => foo(),
// This runs only when `foo` changes,
// or also on init when immediate is enabled.
// Signals used in effect are not tracked.
(foo, dispose) => console.log(`foo changed to ${foo}`),
// Tell reaction that effect should be run on init as well
{immediate: true}
);
Type: boolean
optional
Default: false
In reaction(action, effect)
, enabling this will cause effect to also be called for the initial action call.
Type: (error: Error, dispose: Disposer) => void
optional
By default, errors in reactions are caught and just logged to the console. This is so that reactions can recover from errors and don't break execution of other reactions in queue.
Specifying onError()
listener will pass the error to the handler instead of logging it to the console.
If onError()
handler itself throws, the error is caught and logged to the console.
interface OnceOptions {
onError?: (error: Error, dispose: Disposer) => void;
}
function once(action: (dispose: Disposer) => void, effect: () => void, options?: OnceOptions): Disposer;
Sets up a single time reaction that runs action to subscribe to all signals inside it, and then triggers effect, once, when any of them changes.
This is a lower level single time reaction that needs to be recreated every time the effect is triggered. It's used to implement reaction
above, as well as bindings for UI libraries.
Type: (error: Error, dispose: Disposer) => void
optional
By default, errors in once are caught and just logged to the console. This is so that reactions can recover from errors and don't break execution of other reactions in queue.
Specifying onError()
listener will pass the error to the handler instead of logging it to the console.
If onError()
handler itself throws, the error is caught and logged to the console.
function toJS(value: any): any;
Converts all signals in value
, deeply, into pure serialized JavaScript primitives. value
itself can also be a signal.
The purpose of this is to serialize data that is currently held inside signals. Example:
class Human {
name: Signal<string>;
age: Signal<number>;
constructor(name: string, age: number) {
this.name = signal(name);
this.age = signal(age);
}
}
const human = new Human('John Doe', 20);
Now if we throw the human
object into:
JSON.stringify(human);
We get this string:
{"name":"John Doe","age":20}
And if we throw it into:
toJS(human);
We get this object:
{
name: 'John Doe',
age: 20
}
In 99.9% cases serializations need to run only occasionally, or are at least throttled, therefore the implementation trades speed for size, and essentially all this does is JSON.parse(JSON.stringify(value))
. This is still quite fast, but in a case where speed is of utmost importance, a separate implementation would be necessary.
The other side effect of this implementation is that any value that is not serializable into JSON (such as functions) will be dropped.
If any error happens inside a reaction, action, effect, or computed getter, by default you get a generic error message like Error in Reaction22: ....
.
To help statin identify and name the source of the error, simply name your functions like so:
// Naming normal functions
reaction(
function getLoggedUsersCount() {
return users().fulter((user) => user.isLoggedIn()).length;
},
function logLoggedUsersCount(count) {
console.log(`There is ${count} logged in users.`);
}
);
// Naming arrow functions
const getLoggedUsersCount = () => users().fulter((user) => user.isLoggedIn()).length;
const logLoggedUsersCount = (count) => console.log(`There is ${count} logged in users.`);
reaction(getLoggedUsersCount, logLoggedUsersCount);
If action above throws, you'll get Error in getLoggedUsersCount: users(...).fulter is not a function
.
Naming functions this way also has a benefit that these names will be removed during minification (when building for production), which automatically makes your bundle smaller with no extra effort.
Alternatively, you can use the nameFn()
utility, which simply attaches displayName
property to the function, and statin will pick it up instead of the default name:
import {nameFn, reaction} from 'statin';
reaction(nameFn('MyReaction', () => {}));
Everything in statin is synchronous. There is no setTimeout
or setImmediate
shenanigans going on.
For example, when triggering a signal change:
foo('bar');
all side effects happen before it returns, UNLESS called inside an action:
action(() => foo('bar'));
in which case they happen just before the action returns.
Signals read inside an action are not tracked. This means that in this example:
reaction(() => {
foo();
action(() => bar());
});
reaction will not track the bar()
signal.
Signals changed inside a reaction will not re-trigger the current reaction. For example, in this snippet:
reaction(() => {
foo(); // adds foo() as a reaction dependency
foo('bar'); // changes foo()
});
changing foo()
DOESN'T cause the reaction to run again after it's finished.
On the other hand, changing the signal in an effect WILL run the reaction again:
reaction(
() => foo(),
() => foo(foo() + 1)
);
The example above will trigger circular reaction error after 100 cycles.
No matter if error happens in reaction, action, or effect, the state will continue functioning and be reactive. Reaction that threw the error won't cancel current effects queue, and will continue to be reactive until disposed manually, providing error recovery.