Skip to content

Latest commit

 

History

History
418 lines (298 loc) · 11.5 KB

README.md

File metadata and controls

418 lines (298 loc) · 11.5 KB
Raw

proxy-memoize

CI npm size discord

Intuitive magical memoization library with Proxy and WeakMap

Project status

It's stable and production ready. As the technique behind it is a bit tricky with Proxy, there might still be some bugs especially with nested memoized selectors.

Note: Nesting memoized selectors is theoretically less performant.

Our docs and examples are not very comprehensive, and contributions are welcome.

Introduction

Immutability is pivotal in more than a few frameworks, like React and Redux. It enables simple-yet-efficient change detection in large nested data structures.

JavaScript is a mutable language by default. Libraries like immer simplify updating immutable data strucutres.

This library helps deriving data from immutable structures (AKA, selectors), efficiantly caching results for faster performance.

This library utilizes Proxy and WeakMap, and provides memoization. The memoized function will re-evaluate the original function only if the used part of argument (object) is changed. It's intuitive in a sense and magical in another sense.

How it works

When it (re-)evaluates a function, it will wrap an input object with proxies (recursively, on demand) and invoke the function. When it's finished it will check what is "affected". The "affected" is a list of paths of the input object that are accessed during the function invocation.

Next time when it receives a new input object, it will check if values in "affected" paths are changed. If so, it will re-evaluate the function. Otherwise, it will return a cached result.

The cache size is 1 by default, but configurable.

We have 2-tier cache mechanism. What is described so far is the second tier cache.

The first tier cache is with WeakMap. It's a WeakMap of the input object and the result of function invocation. There's no notion of cache size.

In summary, there are two types of cache:

  • tier-1: WeakMap based cache (size=infinity)
  • tier-2: Proxy based cache (size=1, configurable)

Install

npm install proxy-memoize

How it behaves

import { memoize } from 'proxy-memoize';

const fn = memoize(x => ({ sum: x.a + x.b, diff: x.a - x.b }));

fn({ a: 2, b: 1, c: 1 }); // ---> { sum: 3, diff: 1 }
fn({ a: 3, b: 1, c: 1 }); // ---> { sum: 4, diff: 2 }
fn({ a: 3, b: 1, c: 9 }); // ---> { sum: 4, diff: 2 } (returning a cached value)
fn({ a: 4, b: 1, c: 9 }); // ---> { sum: 5, diff: 3 }

fn({ a: 1, b: 2 }) === fn({ a: 1, b: 2 }); // ---> true

Usage with React Context

Instead of bare useMemo.

const Component = (props) => {
  const [state, dispatch] = useContext(MyContext);
  const render = useCallback(memoize(([props, state]) => (
    <div>
      {/* render with props and state */}
    </div>
  )), [dispatch]);
  return render([props, state]);
};

const App = ({ children }) => (
  <MyContext.Provider value={useReducer(reducer, initialState)}>
    {children}
  </MyContext.Provider>
);

Usage with React Redux & Reselect

Instead of reselect.

import { useSelector } from 'react-redux';

const getScore = memoize(state => ({
  score: heavyComputation(state.a + state.b),
  createdAt: Date.now(),
}));

const Component = ({ id }) => {
  const { score, title } = useSelector(useCallback(memoize(state => ({
    score: getScore(state),
    title: state.titles[id],
  })), [id]));
  return <div>{score.score} {score.createdAt} {title}</div>;
};

Using size option

The example above might seem tricky to create memoized selector in component. Alternatively, we can use size option.

import { useSelector } from 'react-redux';

const getScore = memoize(state => ({
  score: heavyComputation(state.a + state.b),
  createdAt: Date.now(),
}));

const selector = memoize(([state, id]) => ({
  score: getScore(state),
  title: state.titles[id],
}), {
  size: 500,
});

const Component = ({ id }) => {
  const { score, title } = useSelector(state => selector([state, id]));
  return <div>{score.score} {score.createdAt} {title}</div>;
};

The drawback of this approach is we need a good estimate of size in advance.

Usage with Zustand

For derived values.

import { create } from 'zustand';

const useStore = create(set => ({
  valueA,
  valueB,
  // ...
}));

const getDerivedValueA = memoize(state => heavyComputation(state.valueA))
const getDerivedValueB = memoize(state => heavyComputation(state.valueB))
const getTotal = state => getDerivedValueA(state) + getDerivedValueB(state)

const Component = () => {
  const total = useStore(getTotal)
  return <div>{total}</div>;
};

Usage with immer

Disabling auto freeze is recommended. JavaScript does not support nested proxies of frozen objects.

import { setAutoFreeze } from 'immer';
setAutoFreeze(false);

API

getUntracked

This is to unwrap a proxy object and return an original object. It returns null if not relevant.

[Notes] This function is for debugging purpose. It's not supposed to be used in production and it's subject to change.

Examples

import { memoize, getUntracked } from 'proxy-memoize';

const fn = memoize(obj => {
  console.log(getUntracked(obj));
  return { sum: obj.a + obj.b, diff: obj.a - obj.b };
});

replaceNewProxy

This is to replace newProxy function in upstream library, proxy-compare. Use it at your own risk.

[Notes] See related discussoin: dai-shi/proxy-compare#40

memoize

Create a memoized function

Parameters

  • fn function (obj: Obj): Result

  • options {size: number?, noWeakMap: boolean?}?

    • options.size (default: 1)
    • options.noWeakMap disable tier-1 cache (default: false)

Examples

import { memoize } from 'proxy-memoize';

const fn = memoize(obj => ({ sum: obj.a + obj.b, diff: obj.a - obj.b }));

Returns function (obj: Obj): Result

memoizeWithArgs

Create a memoized function with args

Parameters

  • fnWithArgs function (...args: Args): Result

  • options Options?

    • options.size (default: 1)

Examples

import { memoizeWithArgs } from 'proxy-memoize';

const fn = memoizeWithArgs((a, b) => ({ sum: a.v + b.v, diff: a.v - b.v }));

Limitations and workarounds

Inside the function, objects are wrapped with proxies and touching a property will record it.

const fn = memoize(obj => {
  console.log(obj.c); // this will mark ".c" as used
  return { sum: obj.a + obj.b, diff: obj.a - obj.b };
});

A workaround is to unwrap a proxy.

const fn = memoize(obj => {
  console.log(getUntracked(obj).c);
  return { sum: obj.a + obj.b, diff: obj.a - obj.b };
});

Memoized function will unwrap proxies in the return value only if it consists of plain objects/arrays.

const fn = memoize(obj => {
  return { x: obj.a, y: { z: [obj.b, obj.c] } }; // plain objects
});

In this case above, the return value is clean, however, see the following.

const fn = memoize(obj => {
  return { x: new Set([obj.a]), y: new Map([['z', obj.b]]) }; // not plain
});

We can't unwrap Set/Map or other non-plain objects. The problem is when obj.a is an object (which will be wrapped with a proxy) and touching its property will record the usage, which leads unexpected behavior. If obj.a is a primitive value, there's no problem.

There's no workaround. Please be advised to use only plain objects/arrays. Nested objects/arrays are OK.

Input object must not be mutated

const fn = memoize(obj => {
  return { sum: obj.a + obj.b, diff: obj.a - obj.b };
});

const state = { a: 1, b: 2 };
const result1 = fn(state);
state.a += 1; // Don't do this, the state object must be immutable
const result2 = fn(state); // Ends up unexpected result

The input obj or the state must be immutable. The whole concept is built around the immutability. It's faily common in Redux and React, but be careful if you are not familiar with the concept.

There's no workaround.

Input can just be one object

const fn = memoize(obj => {
  return { sum: obj.a + obj.b, diff: obj.a - obj.b };
});

The input obj is the only argument that a function can receive.

const fn = memoize((arg1, arg2) => {
  // arg2 can't be used
  // ...
});

A workaround is to use memoizeWithArgs util.

Note: this disables the tier-1 cache with WeakMap.

Comparison

Reselect

At a basic level, memoize can be substituted in for createSelector. Doing so will return a selector function with proxy-memoize's built-in tracking of your state object.

// reselect
// selecting values from the state object requires composing multiple functions
const mySelector = createSelector(
  state => state.values.value1,
  state => state.values.value2,
  (value1, value2) => value1 + value2,
);

// ----------------------------------------------------------------------

// proxy-memoize
// the same selector can now be written as a single memoized function
const mySelector = memoize(
  state => state.values.value1 + state.values.value2,
);

With complex state objects, the ability to track individual properties within state means that proxy-memoize will only calculate a new value if and only if the tracked property changes.

const state = {
  todos: [{ text: 'foo', completed: false }]
};

// reselect
// If the .completed property changes inside state, the selector must be recalculated
// even through none of the properties we care about changed. In react-redux, this
// selector will result in additional UI re-renders or the developer to implement
// selectorOptions.memoizeOptions.resultEqualityCheck
createSelector(
  state => state.todos,
  todos => todos.map(todo => todo.text)
);

// ----------------------------------------------------------------------

// proxy-memozie
// If the .completed property changes inside state, the selector does NOT change
// this is because we track only the accessed property (todos.text) and can ignore
// the unrelated change
const todoTextsSelector = memoize(state => state.todos.map(todo => todo.text));

Related projects

proxy-memoize depends on an internal library proxy-compare. react-tracked and valtio are libraries that depend on the same library.

memoize-state provides a similar API for the same goal.