Intuitive magical memoization library with Proxy and WeakMap
The API is complete. Unless we find some major issues, it will be fixed. Before reaching v1, we would like to collect more bug reports and best practices. There are no obvious/known issues at the moment, but there are some limitations and workarounds.
In frontend framework like React, object immutability is important. JavaScript itself doesn't support forcing immutability. Several libraries help encouraging immutable coding style, like immer. While immer helps updating an object, this library helps creating a derived value from an object, a.k.a. selector.
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.
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)
npm install proxy-memoize
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
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>
);
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>;
};
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.
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>;
};
Create a memoized function
fn
function (obj: Obj): Resultoptions
{size: number?}?
import memoize from 'proxy-memoize';
const fn = memoize(obj => ({ sum: obj.a + obj.b, diff: obj.a - obj.b }));
Returns function (obj: Obj): Result
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.
import memoize, { getUntrackedObject } from 'proxy-memoize';
const fn = memoize(obj => {
console.log(getUntrackedObject(obj));
return { sum: obj.a + obj.b, diff: obj.a - obj.b };
});
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(getUntrackedObject(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 by 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.
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.
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 create a wrapper.
const memoizeWithArgs = (fnWithArgs, options) => {
const fn = memoize(args => fWithArgs(...args), options);
return (...args) => fn (args);
};
Note: this will essentially bypass the tier-1 cache with WeakMap.
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));