reroi

Zero-dependency library for creating reactive systems with maximum control.

Content

Overview
Implementation
Transactions
Other Functions
- peek
- destroy
Examples
- React
  - React connector
  - Shopping Cart

Overview

You can read the overview article here: [https://blog.pungy.me/articles/reroi]

Every reactive system has defining characteristics that determine its behavior. Here is a list of them, along with how they are implemented in reroi:

Execution flow: Synchronous
Change propagation: Push-Based
Update process: Dataflow
Dependency graph: Explicitly defined by the programmer (data flow differentiation)
Cycle dependencies: Not handled automatically
Transactions: Fully supported
Evaluation:
- derivations: Lazy
- listeners: Proactive
Determinism: Deterministic in practice, but might be non-deterministic due to caching and laziness.

Implementation

The key features of reroi:

Reactive entities are Type Constructors.
No side-effect subscription - You only subscribe to entities that you explicitly list as dependencies.
Control of execution order - you can manipulate when your reaction will be recalculated.
Full-featured Transactions - Group multiple state changes into atomic operations that can be rolled back if any part fails.
High-Order Reactive Entities - reactive entities can contain other reactive entities, enabling complex and dynamic dataflow patterns.

Here is a basic example of reroi:

import * as R from 'reroi'

const _name_    = R.val("Michal")
const _surname_ = R.val("Smith")

const _surnameUpper_ = R.derive(_surname_, str => str.toUpperCase())

const _fullName_ = R.deriveAll(
    [_name_, _surnameUpper_],
    (name, surname) => name + " " + surname
)

console.log(R.read(_fullName_)) // Michal Smith

R.listen(
    _fullName_,
    fullName => console.log("Hello, " + fullName)
)

R.write(_name_, "George")
// log: Hello, George SMITH

console.log(R.read(_name_)) // George
console.log(R.read(_fullName_)) // George SMITH

The _name_ and _surname_ are ReactiveValue<string>. The _fullName_ is ReactiveDerivation<string>. These can be generalized as: type Reactive<A> = ReactiveValue<A> | ReactiveDerivation<A>.

listen provides a way to proactively react to changes in any Reactive entity.

Typically, variable names for reactive entities are wrapped with _ around them. So, they kind of float on the water :))

In the following text, the term reactive entity refers to Reactive<A>. Otherwise, clarification will specify whether it is a derive or a value.

Reactive type

Any reactive object is a type constructor. This means it cannot be used as a plain value, but rather as a container of something.

Similar to how you treat a Promise, you cannot read the value directly; you need to unwrap it first. For example, you can unwrap a promise with await, like so:

import * as R from 'reroi'

const reactive_a: Reactive<number> = R.val(10)
const promise_a: Promise<number> = Promise.resolve(10)

console.log(await promise_a) // 10
console.log(R.read(reactive_a)) // 10

There are two types of reactive objects:

R.val (Read-Write): Can be used with R.read and R.write.
R.derive (Read-only): Can be used only with R.read.

NOTE: ReactiveValue and ReactiveDerivation do not have any internal properties or methods. Every operation on them should use functions that accept them as parameters.

val: ReactiveValue

ReactiveValue or val is an independent container with some value inside it. To read it, pass it to read (which can also consume ReactiveDerivation). To modify it, set the new value with write.

function val<V>(value: V): ReactiveValue<V>;

derive: ReactiveDerivation

The ReactiveDerivation, created with derive, is a way to create a new computed value derived from an existing Reactive entity.

function derive<V, V2>(
    _source_: Reactive<V>,
    computation: (value: V) => V2,
    props?: { priority?: Priority },
): ReactiveDerivation<V2>;

_source_: A single reactive dependency we deriving from.
computation: A function that takes a single value, based on the source. The return value becomes the state of the derivation.
props:
- priority: See priorities. Default is priorities.base.

The result of the derivation is cached between calls and recomputed only after dependency updates.

import * as R from 'reroi'

const _cost_ = R.val(200)
const _discounted_ = R.derive(_cost_, cost => cost * 0.85) // 15% discount

console.log(R.read(_discounted_)) // 170, computed
console.log(R.read(_discounted_)) // 170, cached

R.write(_cost_, 100)
R.write(_cost_, 500)

// 85 was ignored since not read
console.log(R.read(_discounted_)) // 425, computed

NOTE: Derivation update is a passive listener, meaning the computation is not called immediately after a dependency update, only upon direct reading. For an active listener, use listen.

IMPORTANT EXCEPTION: If a derive is in the dependency list for listen/listenAll, it will be recomputed when the listener is about to execute.

deriveAll: ReactiveDerivation of multiple sources

The ReactiveDerivation, created with derive, is a way to create a new computed value derived from an existing Reactive entity.

function deriveAll<Vs extends Array<unknown>, V2>(
    _sources_: { [K in keyof Vs]: Reactive<Vs[K]> },
    computation: (value: Vs) => V2,
    props?: { priority?: Priority },
): ReactiveDerivation<V2>;

sources: A list of reactive dependencies we deriving from.
computation: A function that takes a list of values from the dependencies, and returns a state of the derivation.
props:
- priority: See priorities. Default is priorities.base.

const _a_ = val("a")
const _b_ = val("c")
const _d_ = val("d")
const _e_ = val("e")

const deps = [_a_, _b_, _c_, _d_, _e_]
const _word_ = deriveAll(deps, sources =>
    sources.reduce((str, x) => str + x, ""))

read(_word_) // abcde

const _f_ = val(10)

const _compound_ = deriveAll([_e_, _f_], ([str, num]) =>
    str.toUpperCase() + ": " + num.toFixed(5))

read(_compound_) // E: 10.00000

Reading and Writing

write

function write<A>(
    _value_: ReactiveValue<A>,
    newValue: A | ((value: A) => A),
    props?: { literateFn?: boolean },
): ReactiveValue<A>;

value: Reactive value to write to.
newValue: Value producer or a plain value.
props:
- literateFn: Treat a function passed as newValue as a value, not as a value producer.

import { val, read, write } from 'reroi'

const _x_ = val(10)

write(_x_, 20)
expect(read(_x_)).toBe(20)

write(_x_, x => x * 2)
expect(read(_x_)).toBe(40)

// LiterateFn
const _lazyX_ = val(() => 10)
const x20 = () => 20
write(_lazyX_, x20, { literateFn: true })

expect(read(_lazyX_)).toBe(x20)

write has no memoization, and even if you write the same value repeatedly, it will always propagate changes to dependencies:

import { val, write, listen } from 'reroi'

const _x_ = val(5)
listen(_x_, x => console.log(`I'm on ${x}`))

write(_x_, 10) // I'm on 10
write(_x_, 10) // I'm on 10
write(_x_, 10) // I'm on 10

read

function read<V>(_reactive_: Reactive<V>): V;

If it's a ReactiveValue, returns the associated value.
If it's a ReactiveDerivation, computes the value if not cached.

Listening

listen

Active listener with side effects on single dependency:

type Unsub = () => void;

function listen<V>(
    _source_: Reactive<V>,
    sideEffect: (value: V) => void,
    props?: { priority?: Priority, immediate?: boolean },
): Unsub;

_source_: A single reactive dependency we listen and fire effect on update.
sideEffect: An effect called on update of the source. sideEffect is called on every dependency update.
props:
- priority: See priorities. Default is priorities.base.
- immediate: Call the listen effect upon declaration. Default is false.

listenAll

Active listener with side effects:

type Unsub = () => void;

function listenAll<Vs extends Array<unknown>, V2>(
    _sources_: { [K in keyof Vs]: Reactive<Vs[K]> },
    sideEffect: (values: Vs) => void,
    props?: { priority?: Priority, immediate?: boolean },
): Unsub;

_sources_: A list of dependencies.
sideEffect: An effect to be called on update of dependencies.
props:
- priority: See priorities. Default is priorities.base.
- immediate: Call the listen effect upon declaration. Default is false.

Priorities

For details on controlling evaluation order with prioritization, read here: controlling evaluation order

Basically, reroi DOES NOT automatically resolve issues with evaluation order. If your derivation or listener subscribes to a source more than once (e.g., implicitly through a chain), it will update multiple times.

import * as R from 'reroi'

const _price_ = R.val(0)

const _tax_ = R.derive(
  _price_,
  price => price * 0.08, // 8% tax
)
const _shipping_ = R.derive(
  _price_,
  price => price > 50 ? 0 : 5.00, // free shipping over $50
)

R.listenAll(
    [_price_, _tax_, _shipping_],
    ([price, tax, shipping]) => {
        const total = price + tax + shipping
        console.log(`Final price: $${total.toFixed(2)} (incl. tax: $${tax.toFixed(2)}, shipping: $${shipping.toFixed(2)})`)
    }
)

R.write(_price_, 20.00)
// Final price: $26.60 (incl. tax: $1.60, shipping: $5.00)
// Would be logged THREE TIMES

This occurs because the listener subscribes to _price_ three times: explicitly in the dependency list, and implicitly through _tax_ and _shipping_ (_price_ updates _tax_, which updates the listener).

This behavior is intended. To correct it, use priorities, which explicitly define the position of a listener or derivation in the dependency list of its sources.

The fixed solution looks like this:

import * as R from 'reroi'

R.listen(
    _price_, // We only need to subscribe to the root dependency.
    (price) => {
        const total = price + R.read(_tax_) + R.read(_shipping_)
        console.log(`Final price: $${total.toFixed(2)} (incl. tax: $${tax.toFixed(2)}, shipping: $${shipping.toFixed(2)})`)
    },
    { priority: R.priorities.after(R.priorities.base) },
)

Now, the internal dependency graph would look like this:

price
 |
 +---> tax
 +---> shipping
 |
 +---> listener

The listener executes after shipping and tax.

Levels

A priority level is essentially a number. There are three default levels:

priorities.highest: Maximum priority level, executed before all others.
priorities.base: Default priority for dependencies, equivalent to 0.
priorities.lowest: Opposite of highest - executed after all others.

import { val, listen } from 'reroi'

const _msg_ = val("")
const log = console.log

listen(_msg_, (msg) => log("3: " + msg), { priority: 3 })
listen(_msg_, (msg) => log("2: " + msg), { priority: 2 })
listen(_msg_, (msg) => log("4: " + msg), { priority: 4 })
listen(_msg_, (msg) => log("1: " + msg), { priority: 1 })

write(_msg_, "Hi?")

// 4: Hi?
// 3: Hi?
// 2: Hi?
// 1: Hi?

No priority should be higher than highest, neither lower than lowest.

highest: 1000.
lowest: -1000.

2000 of levels should be enough for all ;)

priorities.before

function before(p0: ReactiveDerivation<unknown> | Priority): Priority;

Helper function that provides a priority higher than the one passed as an argument. You can also pass a derivation, meaning your priority will be higher than that of the derivation.

priorities.after

function after(p0: ReactiveDerivation<unknown> | Priority): Priority;

after is the opposite of before. It decreases the priority relative to the one passed.

Prioritization usage

import * as R from 'reroi'

const base = R.priorities.base

expect(R.priorities.before(base)).toBe(1)
expect(R.priorities.after(base)).toBe(-1)

const _a_ = R.val(0)
const _der1_ = R.derive(_a_, a => a + 1, { priority: 15 })

expect(R.priorities.before(_der1_)).toBe(16)
expect(R.priorities.after(_der1_)).toBe(14)

Rather than memorizing numerical directions, use the helpers :)

Transactions

For an overview of transactions, read here: transactions.

Transactional write

transaction.write

The function interface for transaction.write is complex due to compositional transactions:

type TransactionState<R, E> = TransactionSuccess<R> | TransactionError<E>
type TransactionFN = <R, E, C>(aVal: R, context: C) => TransactionState<R, E>

function writeT<R, E, C = {}, ID extends string = string>(
  _value_: ReactiveValue<R>,
  newValue: TransactionFN<R, E, C> | R,
  id?: ID
): ReactiveTransaction<R, E, C, ID>;

Generics:

R: Result of the transaction to be written to ReactiveValue.
E: Type of error with which the transaction can be rejected.
C: Context of the transaction. Used for composition.
ID: ID of the transaction for further use in context.

Parameters:

value: Reactive value to write the transaction result to.
newValue: Either a function that accepts the current value and transaction context, or a plain value to resolve the transaction with.
id: Optional ID of the transaction.

import { val, transaction, read } from 'reroi'

const _a_ = val('a')
const tr = transaction.write(_a_, 'A')

expect(read(_a_)).toBe('a')

ReactiveTransaction

The transaction object returned by transactional write provides an interface for the created transaction. The primary method of interest is run, which executes the transaction.

export interface ReactiveTransaction<
  R = unknown, // Might be success with
  E = unknown, // Might be error with
> {
  run(): TransactionState<R, E>;
}

import * as R from 'reroi'

const _a_ = R.val('a')
const tr = R.transaction.write(_a_, 'A')

tr.run();
expect(R.read(_a_)).toBe('A')

Helper functions

transaction.success

Use this inside a TransactionFN to produce a success value to be written to the ReactiveValue.

type TransactionSuccess<R> = { value: R }
const success = <R>(value: R): TransactionSuccess<R> => ({

import * as R from 'reroi'

const _a_ = R.val('a')
const tr = R.transaction.write(_a_, () => R.transaction.success('A'))

tr.run();
expect(R.read(_a_)).toBe('A')

transaction.error

Use this inside a TransactionFN to reject the transaction execution.

type TransactionError<E> = { error: E }
function error<E>(error: E): TransactionError<E>

import * as R from 'reroi'

const _a_ = R.val('a')
const tr = R.transaction.write(_a_, () => R.transaction.error('A'))

tr.run();
expect(R.read(_a_)).toBe('a')

transaction.isSuccess

Checks whether the transaction result was success.

function isSuccess<R, E>(transaction: TransactionState<R, E>): transaction is TransactionSuccess<R>

import * as R from 'reroi'

const _a_ = R.val('a')
const tr = R.transaction.write(_a_, () => R.transaction.success('A'))

const state = tr.run();
expect(R.transaction.isSuccess(state)).toBe(true)

if (R.transaction.isSuccess(state)) {
    console.log(state.value) // 'A'
}

transaction.isError

Checks whether the transaction result was error.

function isError<R, E>(transaction: TransactionState<R, E>): transaction is TransactionError<E>

import * as R from 'reroi'

const _a_ = R.val('a')
const tr = R.transaction.write(_a_, () => R.transaction.error('A'))

const state = tr.run();
expect(R.transaction.isError(state)).toBe(true)

if (R.transaction.isError(state)) {
    console.log(state.error) // 'A'
}

transaction.mapS

The following utilities are optional and provide a more functional programming flavor to reroi :)

Transaction is essentially an IO (Either E R) ADT with Functor and Foldable type classes.

Maps TransactionSuccess<R> to TransactionSuccess<R2>. Can be used to peek into a success transaction's value.

import * as R from 'reroi'

const _a_ = R.val('a')
const tr = R.transaction.write(_a_, () => R.transaction.success('A'))

const beautify = R.transaction.mapS((value: string) => `I was success with: "${value}"`)

const state = beautify(tr.run())

if (R.transaction.isSuccess(state)) {
    console.log(state.value) // I was success with: "A"
}

transaction.mapE

Maps TransactionError<E> to TransactionError<E2>.

import * as R from 'reroi'

const _a_ = R.val('a')
const tr = R.transaction.write(_a_, () => R.transaction.error('A'))

const beautify = R.transaction.mapE(error => `I was rejected with: "${error}"`)

const state = beautify(tr.run())

if (R.transaction.isError(state)) {
    console.log(state.error) // I was error with: "A"
}

transaction.fold

Folds or reduces the transaction result into a single value.

import * as R from 'reroi'

const _a_ = R.val('a')
const trS = R.transaction.write(_a_, () => R.transaction.success('A'))
const trE = R.transaction.write(_a_, () => R.transaction.error('A'))

const toBoolean = R.transaction.fold(
    () => false, // on error 
    () => true,  // on success
)

console.log(toBoolean(trS.run())) // true
console.log(toBoolean(trE.run())) // false

Composing transactions

The key strength of transactions is composition. You can combine multiple transactions into a single one, preserving atomicity and other transactional properties.

import * as R from 'reroi'

const _name_ = R.val("George")
const _surname_ = R.val("Kowalski")

R.listenAll([_name_, _surname_], ([name, surname]) => {
    console.log(`Hello, ${name} ${surname}!`)
})

const tr = R.transaction.compose(
               R.transaction.write(_name_, "Grzegosz"),
               R.transaction.write(_surname_, "Smith"),
           )

tr.run()
// Only once:
// Hello, Grzegosz Smith!

No changes applied until entire transaction is completed

Even if one part of the transaction completes, no values are written until the entire transaction finishes:

import * as R from 'reroi'

const _a_ = R.val("a")
const _b_ = R.val("b")
const _c_ = R.val("c")

const _A_ = R.derive(_a_, (a) => a.toUpperCase())

const tr = R.transaction.compose(
    R.transaction.write(_a_, "F"),
    R.transaction.write(_b_, () => {
        console.log(R.read(_a_)) // a
        console.log(R.read(_A_)) // A
        return R.transaction.success("B")
    }),
    R.transaction.write(_c_, () => {
        console.log(R.read(_b_)) // b
        return R.transaction.success("C")
    }),
)

No changes applied if any transaction is error

import * as R from 'reroi'

const _a_ = R.val("a")
const _b_ = R.val("b")
const _c_ = R.val("c")

R.listen(_a_, console.log)
R.listen(_b_, console.log)
R.listen(_c_, console.log)

const tr = R.transaction.compose(
    R.transaction.write(_a_, "A"),
    R.transaction.write(_b_, () => R.transaction.error("error")),
    R.transaction.write(_c_, "C"),
)

tr.run()

console.log(R.read(_a_)) // logs: 'a' // Wasn't changed
// No console logs from `listen` reactions

This allows rejecting the entire transaction if any part fails, but it raises questions:

How to access new values from previously success actions?
How to compute a derivation's new state?

The answers involve transaction context and peek.

Transaction context and peek

During execution, the ctx parameter (second argument in the handler for transaction.write) provides access to context. Assign an id to actions to identify values in the context.

peek allows previewing how a derivation's value would look based on provided dependency values. It calls the inner function of derive without affecting the derivation.

The API is more complex, but it enables true transactional behavior.

import * as R from 'reroi'

const _name_ = R.val("George")
const _surname_ = R.val("Kowalski")
const _fullName_ = R.deriveAll(
    [_name_, _surname_],
    ([name, surname]) => name + " " + surname
)

const _messagePool_ = R.val<Array<string>>([])

const addPerson = (name, surname) => (
    R.transaction.compose(
        //                  r_val   val   id
        R.transaction.write(_name_, name, "name"),
        R.transaction.write(_surname_, surname, "surname"),
        R.transaction.write(_messagePool_, (pool, ctx) => {
            const fullName = R.peek(_fullName_, [ctx.name, ctx.surname])

            pool.push(`The user "${fullName}" has been added!`)
            return pool
        })
    )
)

addPerson("Oda", "Nobunaga").run()

console.log(
    R.read(_messagePool_).at(-1) // The user "Oda Nobunaga" has been added!
)

Other Functions

peek

Reads a derive with dependencies provided as a list in the second parameter. Completely pure and does not affect the derivation in any way.

function peek<R extends ReactiveDerivation>(_derive_: R, dependencies: R['dependencies']): R['value'];

destroy

You may need to destroy a derivation so it no longer reacts to changes (and no one reacts to its changes).

To ensure all links are cleared and avoid garbage accumulation, use destroy.

function destroy(
    _derivation_: ReactiveDerivation,
): void;

Once destroyed, it notifies all listeners, which stop listening. If it was the last dependency, dependents are cascadedly destroyed.

Examples

List of complete examples of reroi usage.

React

To connect reroi with React, create a custom hook: useReactive (this could become a separate package).

React connector

The useReactive hook listens to updates from _reactive_, stores the value in a ref, and forces a state update (since useState memoizes identical values, which reroi does not).

import { useEffect, useReducer, useRef } from "react";
import { Reactive, listen } from "reroi";

export function useReactive<V>(_reactive_: Reactive<V>): V {
  const [, forceUpdate] = useReducer(() => [], [])
  const listener = useRef<ReturnType<typeof listen> | null>(null)
  const value = useRef(R.read(_reactive_))

  useEffect(
    () => {
      if (listener.current !== null) {
        // unsub from old reactive
        listener.current()
      }

      listener.current = listen(
        _reactive_,
        v => {
          value.current = v
          forceUpdate()
        }
      )
      return listener.current
    },
    [_reactive_]
  )

  return value.current
}

Based on this hook, reroi serves as an effective state manager. Here is an example.

Shopping Cart

Try on codesandbox

An app with numerous reactions, featuring a shopping cart with discounts: add items, observe price increases, apply discount rates based on total price, and add/remove discount rates. Note the clean state usage: no prop drilling, no complex structures.

The entire state is in model.ts, connected to components via the useReactive hook.

Name		Name	Last commit message	Last commit date
Latest commit History 33 Commits
src		src
.eslintignore		.eslintignore
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
TODO.md		TODO.md
dts-bundle-generator.config.json		dts-bundle-generator.config.json
eslint.config.js		eslint.config.js
package.json		package.json
pnpm-lock.yaml		pnpm-lock.yaml
tsconfig.json		tsconfig.json
vite.config.ts		vite.config.ts

License

PunGy/reroi

Folders and files

Latest commit

History

Repository files navigation