Develop Reference

React fast global redux-like variable

I'm making a front-end application using react/webgl and I need to be vary of performance improvements since almost everything must be rendered real-time and dynamically.
I need to render something on a canvas and need to use some variable's globally across many different components, but they need to be updated fast. Technically redux is what I need, however accessing dispatched variables takes time and causes crucial performance issues.
So instead I opted in to use useRef() which solves the “slow” issue but now I cannot update it’s value across different components. Using useRef() solves my issue but since it's not globally accessible it causes problems on other parts of the application.
Declaration of the variable looks like this:
import { useRef } from 'react';
const WebGLStarter = (props) => {
...
const myValue = useRef();
myValue.current = someCalculation();
function render(myValue.current){
...
requestAnimationFrame(function () {
render(myValue.current);
});
}
...
}
Currently someCalculation() is on the same component as it's declaration. I want to use someCalculation() on a different file but I can't do it beacuse useRef() won't allow me to. And again, I can't use redux because it's slow.
TL;DR : I need something similar to redux but it needs to be fast enough to not cause performance issues on an infinite loop.

Create a context with the ref. Wrap your root with the provider, and use the hook to get access to the ref when you need to use/update it:
import { createContext, useRef, useContext } from 'react';
const defaultValue = /** default value **/;
const MyValueContext = createContext(defaultValue);
const MyValueContextProvider = ({ children }) => {
const myValueRef = useRef(defaultValue);
return (
<MyValueContext.Provider value={myValueRef}>
{children}
</MyValueContext.Provider>
);
};
const useMyValue = () => useContext(MyValueContext);
To use in components call the useMyValue hook. This would give you direct access to the ref, and since you don't update any state (just change the ref's current property) it won't cause re-renders:
const WebGLStarter = (props) => {
const myValue = useMyValue();
myValue.current = someCalculation();
...
};

Children useCallback dependency hell

From what I understand you use useCallback to prevent rerendering so I've been using it in every function and my spider senses are telling me it already sounds bad.
But the story doesn't ends there, since I've been using it everywhere I'm now passing dependencies to all my child components that they shouldn't need to worry about like in the following example :
EDIT // SANDBOX: https://codesandbox.io/s/bold-noether-0wdnp?file=/src/App.js
Parent component (needs colorButtons and currentColor)
const ColorPicker = ({onChange}) => {
const [currentColor, setCurrentColor] = useState({r: 255, g:0, b: 0})
const [colorButtons, setColorButtons] = useState({0: null})
const handleColorButtons = useCallback((isToggled, id) => {
/* code that uses colorButtons and currentColor */
}, [colorButtons, currentColor])
return <div className="color-picker">
<RgbColorPicker color={currentColor} onChange={setCurrentColor} />
<div className="color-buttons">
{
Object.entries(colorButtons).map(button => <ColorButton
//...
currentColor={currentColor}
onClick={handleColorButtons}
colorButtons={colorButtons}
/>)
}
</div>
</div>
}
1st child (needs style and currentColor but gets colorButtons for free from its parent)
const ColorButton = ({currentColor, onClick, id, colorButtons}) => {
const [style, setStyle] = useState({})
const handleClick = useCallback((isToggled) => {
/* code that uses setStyle and currentColor */
}, [style, currentColor, colorButtons])
return <ToggleButton
//...
onClick={handleClick}
style={style}
dependency1={style}
dependency2={currentColor}
dependency3={colorButtons}
>
</ToggleButton>
}
2nd child (only needs its own variables but gets the whole package)
const ToggleButton = ({children, className, onClick, style, data, id, onRef, ...dependencies}) => {
const [isToggled, setIsToggled] = useState(false)
const [buttonStyle, setButtonStyle] = useState(style)
const handleClick = useCallback(() => {
/* code that uses isToggled, data, id and setButtonStyle */
}, [isToggled, data, id, ...Object.values(dependencies)])
return <button
className={className || "toggle-button"}
onClick={handleClick}
style={buttonStyle || {}}
ref={onRef}
>
{children}
</button>
}
Am I doing an anti-pattern and if so, what is it and how to fix it ? Thanks for helping !

React hook useCallback
useCallback is a hook that can be used in functional React components. A functional component is a function that returns a React component and that runs on every render, which means that everything defined in its body get new referential identities every time. An exception to this can be accomplished with React hooks which may be used inside functional components to interconnect different renders and maintain state. This means that if you save a reference to a regular function defined in a functional component using a ref, and then compare it to the same function in a later render, they will not be the same (the function changes referential identity between renderings):
// Render 1
...
const fnInBody = () => {}
const fn = useRef(null)
console.log(fn.current === fnInBody) // false since fn.current is null
fn.current = fnInBody
...
// Render 2
...
const fnInBody = () => {}
const fn = useRef(null)
console.log(fn.current === fnInBody) // false due to different identity
fn.current = fnInBody
...
As per the docs, useCallback returns "a memoized version of the callback that only changes if one of the dependencies has changed" which is useful "when passing callbacks to optimized child components that rely on reference equality to prevent unnecessary renders".
To sum up, useCallback will return a function that maintains its referential identity (e.g. is memoized) as long as the dependencies don't change. The returned function contains a closure with the used dependencies and must thus be updated once the dependencies change.
This results in this updated version of the previous example
// Render 1
...
const fnInBody = useCallback(() => {}, [])
const fn = useRef(null)
console.log(fn.current === fnInBody) // false since fn.current is null
fn.current = fnInBody
...
// Render 2
...
const fnInBody = useCallback(() => {}, [])
const fn = useRef(null)
console.log(fn.current === fnInBody) // true
fn.current = fnInBody
...
Your use case
Keeping the above description in mind, let's have a look at your use of useCallback.
Case 1: ColorPicker
const handleColorButtons = useCallback((isToggled, id) => {
/* code that uses colorButtons and currentColor */
}, [colorButtons, currentColor])
This function will get a new identity every time colorButtons or currentColor changes. ColorPicker itself rerenders either when one of these two are set or when its prop onChange changes. Both handleColorButtons and the children should be updated when currentColor or colorButtons change. The only time the children benefit from the use of useCallback is when only onChange changes. Given that ColorButton is a lightweight component, and that ColorPicker rerenders mostly due to changes to currentColor and colorButtons, the use of useCallback here seems redundant.
Case 2: ColorButton
const handleClick = useCallback((isToggled) => {
/* code that uses setStyle and currentColor */
}, [style, currentColor, colorButtons])
This is a situation similar to the first case. ColorButton rerenders when currentColor, onClick, id or colorButtons change and the children rerender when handleClick, style, colorButtons or currentColor change. With useCallback in place, the props id and onClick may change without rerendering the children (according to the above visible code at least), all other rerenders of ColorButton will lead to its children rerendering. Again, the child ToggleButton is lightweight and id or onClick are not likely to change more often than any other prop so the use of useCallback seems redundant here as well.
Case 3: ToggleButton
const handleClick = useCallback(() => {
/* code that uses isToggled, data, id and setButtonStyle */
}, [isToggled, data, id, ...Object.values(dependencies)])
This case is elaborate with a lot of dependencies but from what I see, one way or the other, most of the component props will lead to a "new version" of handleClick and with the children being lightweight components, the argument to use useCallback seems weak.
So when should I use useCallback?
As the docs say, use it in the very specific cases when you need a function to have referential equality between renders ...
You have a component with a subset of children that are expensive to rerender and that should rerender much less often than the parent component but rerender due to a function prop changing identity whenever the parent rerenders. To me, this use case also signals bad design and I would attempt to divide the parent component into smaller components but what do I know, maybe this is not always possible.
You have a function in the body of the functional component which is used in another hook (listed as a dependency) which is triggered every time due to the function changing identity whenever the component rerenders. Typically, you can omit such a function from the dependency array by ignoring the lint rule even if this is not by the book. Other suggestions are to place such a function outside the body of the component or inside the hook that uses it, but there might be scenarios where none of this works out as intended.
Good to know connected to this is ...
A function living outside a functional component will always have referential equality between renders.
The setters returned by useState will always have referential equality between renders.
I said in the comments that you can use useCallback when there is function doing expensive calculations in a component that rerenders often but I was a bit off there. Let's say you have a function that does heavy calculations based on some prop that changes less often than a component rerenders. Then you COULD use useCallback and run a function inside it that returns a function with a closure with some computed value
const fn = useCallback(
(
() => {
const a = ... // heavy calculation based on prop c
const b = ... // heavy calculation based on prop c
return () => { console.log(a + b) }
}
)()
, [c])
...
/* fn is used for something, either as a prop OR for something else */
This would effectively avoid calculating a and b every time the component rerenders without c changing, but the more straightforward way to do this would be to instead
const a = useMemo(() => /* calculate and return a */, [c])
const b = useMemo(() => /* calculate and return b */, [c])
const fn = () => console.log(a + b)
so here the use of useCallback just complicates things in a bad way.
Conclusion
It's good to understand more complicated concepts in programming and to be able to use them, but part of the virtue is also to know when to use them. Adding code, and especially code that involves complicated concepts, comes at the price of reduced readability and code that is harder to debug with a lot of different mechanisms that interplay. Therefore, make sure you understand the hooks, but always try to not use them if you can. Especially useCallback, useMemo and React.memo (not a hook but a similar optimization) should, in my opinion, only be introduced when they are absolutely needed. useRef has its very own use cases but should also not be introduced when you can solve your problem without it.
Good work on the sandbox! Always makes it easier to reason about code. I took the liberty of forking your sandbox and refactoring it a bit: sandbox link. You can study the changes yourself if you want. Here is a summary:
It's good that you know and use useRef and useCallback but I was able to remove all the uses, making the code much easier to understand (not only by removing these uses but by also removing the contexts where they are used).
Try to work with React to simplify things. I know this is not a hands-on suggestion but the more you get into React, the more you will realize that you can do things co-operating with React, or you can do things your own way. Both will work but the latter will result in more headache for you and everybody else.
Try to isolate the scope of a component; only delegate data that is necessary to child components and constantly question where you keep your state. Earlier you had click handlers in all three components and the flow was so complicated I didn't even bother to fully understand it. In my version, there is just one click handler in ColorPicker that is being delegated down. The buttons don't have to know what happens when you click them as long as the click handler takes care of that. Closures and the ability to pass functions as arguments are strong advantages of React and Javascript.
Keys are important in React and it's good to see that you use them. Typically, the key should correspond to something that uniquely identifies a specific item. Good to use here would be ${r}_${g}_${b} but then we would only be able to have one sample of each color in the button array. This is a natural limitation but if we don't want it, the only way to assign keys is to assign a unique identifier, which you did. I prefer using Date.now() but some would probably advise against it for some reason. You could use a global variable outside the functional component too if you don't want to use a ref.
Try to do things the functional (immutable) way, and not the "old" Javascript way. For example, when adding to an array, use [...oldArray, newValue] and when assigning to an object, use {...oldObject, newKey: newValue }.
There are more things to say but I think it's better for you to study the refactored version and you can let me know if you wonder about anything.

How can we implement componentWillUnmount using react hooks?

The method componentWillUnmount() is invoked immediately before a component is unmounted and destroyed. If we use useEffect with an empty array ([]) as the second argument and put our function in return statement it will be executed after the component is unmounted and even after another component will be mounted. This is done for performance reasons as far as I understand. In order not to delay rendering.
So the question is - how can we call some function using hooks before a component gets unmounted?
What I am trying to do is an application which saves user's input as he types (without submitting form). I use setInterval to save updated text every N seconds. And I need to force save updates before the component will unmount. I don't want to use prompt by react router before navigating. This is an electron application. I appreciate any thoughts or advice on how to implement such functionality.
Update
Unfortunately, Effects with Cleanup run after letting the browser paint. More details can be found here: So What About Cleanup?. It basically means that cleanup is run after a component is unmounted and it is not the same as executing code in componentWillUnmount(). I can clearly see the sequence of calls if I put console.log statements in the cleanup code and in another component. The question is whether we can execute some code before a component is unmounted using hooks.
Update2
As I can see I should better describe my use case. Let's imagine a theoretical app which holds its data in a Redux store. And we have two components with some forms. For simplicity, we don't have any backend or any async logic. We use only Redux store as data storage.
We don't want to update Redux store on every keystroke. So we keep actual values in the local component's state which we initialize with values from the store when a component mounts. We also create an effect which sets up a setInterval for 1s.
We have the following process. A User types something. Updates are stored in the local component state until our setInterval callback is called. The callback just puts data in the store (dispatches action). We put our callback in the useEffect return statement to force save to store when the component gets unmounted because we want to save data to store in this case as soon as possible.
The problem comes when a user types something in the first component and immediately goes to the second component (faster than 1s). Since the cleanup in our first component will be called after re-rendering, our store won't be updated before the second component gets mounted. And because of that, the second component will get outdated values to its local state.
If we put our callback in componentWillUnmount() it will be called before unmounting and the store will be updated before the next component mounts. So can we implement this using hooks?

componentWillUnmount can be simulated by returning a function inside the useEffect hook. The returned function will be called just before every rerendering of the component. Strictly speaking, this is the same thing but you should be able to simulate any behaviour you want using this.
useEffect(() => {
const unsubscribe = api.createSubscription()
return () => unsubscribe()
})
Update
The above will run every time there is a rerender. However, to simulate the behaviour only on mounting and unmounting (i.e. componentDidMount and componentWillUnmount). useEffect takes a second argument which needs to be an empty array.
useEffect(() => {
const unsubscribe = api.createSubscription()
return () => unsubscribe()
}, [])
See a more detailed explanation of the same question here.

Since the introduction of the useLayoutEffect hook, you can now do
useLayoutEffect(() => () => {
// Your code here.
}, [])
to simulate componentWillUnmount. This runs during unmount, but before the element has actually left the page.

The question here is how do you run code with hooks BEFORE unmount? The return function with hooks runs AFTER unmount and whilst that doesn’t make a difference for most use cases, their are some where it is a critical difference.
Having done a bit of investigation on this, I have come to the conclusion that currently hooks simply does not provide a direct alternative to componentWillUnmount. So if you have a use case that needs it, which is mainly for me at least, the integration of non-React libs, you just have to do it the old way and use a component.
Update: see the answer below about UseLayoutEffect() which looks like it may solve this issue.

I agree with Frank, but the code needs to look like this otherwise it will run only on the first render:
useLayoutEffect(() => {
return () => {
// Your code here.
}
}, [])
This is equivalent to ComponentWillUnmount

Similar to #pritam's answer, but with an abstracted code example. The whole idea of useRef is to allow you to keep track of the changes to the callback and not have a stale closure at the time of execution. Hence, the useEffect at the bottom can have an empty dependency array to ensure it only runs when the component unmounts. See the code demo.
Reusable hook:
type Noop = () => void;
const useComponentWillUnmount = (callback: Noop) => {
const mem = useRef<Noop>();
useEffect(() => {
mem.current = callback;
}, [callback]);
useEffect(() => {
return () => {
const func = mem.current as Noop;
func();
};
}, []);
};

After a bit of research, found that - you could still accomplish this. Bit tricky but should work.
You can make use of useRef and store the props to be used within a closure such as render useEffect return callback method
function Home(props) {
const val = React.useRef();
React.useEffect(
() => {
val.current = props;
},
[props]
);
React.useEffect(() => {
return () => {
console.log(props, val.current);
};
}, []);
return <div>Home</div>;
}
DEMO
However a better way is to pass on the second argument to useEffect so that the cleanup and initialisation happens on any change of desired props
React.useEffect(() => {
return () => {
console.log(props.current);
};
}, [props.current]);

I got in a unique situation where the useEffect(() => () => { ... }, []); answers did not work for me. This is because my component never got rendered — I was throwing an exception before I could register the useEffect hook.
function Component() {
useEffect(() => () => { console.log("Cleanup!"); }, []);
if (promise) throw promise;
if (error) throw error;
return <h1>Got value: {value}</h1>;
}
In the above example, by throwing a Promise<T> that tells react to suspend until the promise is resolved. However, once the promise is resolved, an error is thrown. Since the component never gets rendered and goes straight to an ErrorBoundary, the useEffect() hook is never registered!
If you're in a similar situation as myself, this little code may help:
To solve this, I modified my ErrorBoundary code to run a list of teardowns once it was recovered
export default class ErrorBoundary extends Component {
// ...
recover() {
runTeardowns();
// ...
}
// ...
}
Then, I created a useTeardown hook which would add teardowns that needed to be ran, or make use of useEffect if possible. You'll most likely need to modify it if you have nesting of error boundaries, but for my simple usecase, it worked wonderfully.
import React, { useEffect, useMemo } from "react";
const isDebugMode = import.meta.env.NODE_ENV === "development";
const teardowns: (() => void)[] = [];
export function runTeardowns() {
const wiped = teardowns.splice(0, teardowns.length);
for (const teardown of wiped) {
teardown();
}
}
type Teardown = { registered?: boolean; called?: boolean; pushed?: boolean } & (() => unknown);
/**
* Guarantees a function to run on teardown, even when errors occur.
*
* This is necessary because `useEffect` only runs when the component doesn't throw an error.
* If the component throws an error before anything renders, then `useEffect` won't register a
* cleanup handler to run. This hook **guarantees** that a function is called when the component ends.
*
* This works by telling `ErrorBoundary` that we have a function we would like to call on teardown.
* However, if we register a `useEffect` hook, then we don't tell `ErrorBoundary` that.
*/
export default function useTeardown(onTeardown: () => Teardown, deps: React.DependencyList) {
// We have state we need to maintain about our teardown that we need to persist
// to other layers of the application. To do that, we store state on the callback
// itself - but to do that, we need to guarantee that the callback is stable. We
// achieve this by memoizing the teardown function.
const teardown = useMemo(onTeardown, deps);
// Here, we register a `useEffect` hook to run. This will be the "happy path" for
// our teardown function, as if the component renders, we can let React guarantee
// us for the cleanup function to be ran.
useEffect(() => {
// If the effect gets called, that means we can rely on React to run our cleanup
// handler.
teardown.registered = true;
return () => {
if (isDebugMode) {
// We want to ensure that this impossible state is never reached. When the
// `runTeardowns` function is called, it should only be ran for teardowns
// that have not been able to be hook into `useEffect`.
if (teardown.called) throw new Error("teardown already called, but unregistering in useEffect");
}
teardown();
if (isDebugMode) {
// Because `teardown.registered` will already cover the case where the effect
// handler is in charge of running the teardown, this isn't necessary. However,
// this helps us prevent impossible states.
teardown.called = true;
}
};
}, deps);
// Here, we register the "sad path". If there is an exception immediately thrown,
// then the `useEffect` cleanup handler will never be ran.
//
// We rely on the behavior that our custom `ErrorBoundary` component will always
// be rendered in the event of errors. Thus, we expect that component to call
// `runTeardowns` whenever it deems it appropriate to run our teardowns.
// Because `useTeardown` will get called multiple times, we want to ensure we only
// register the teardown once.
if (!teardown.pushed) {
teardown.pushed = true;
teardowns.push(() => {
const useEffectWillCleanUpTeardown = teardown.registered;
if (!useEffectWillCleanUpTeardown) {
if (isDebugMode) {
// If the useEffect handler was already called, there should be no way to
// re-run this teardown. The only way this impossible state can be reached
// is if a teardown is called multiple times, which should not happen during
// normal execution.
const teardownAlreadyCalled = teardown.called;
if (teardownAlreadyCalled) throw new Error("teardown already called yet running it in runTeardowns");
}
teardown();
if (isDebugMode) {
// Notify that this teardown has been called - useful for ensuring that we
// cannot reach any impossible states.
teardown.called = true;
}
}
});
}
}

It does not matter wether the returned function from useEffect gets called before or after the component unmounted: You still have access to the states valuey through the closure:
const [input, setInput] = useState(() => Store.retrieveInput());
useEffect(() => {
return () => Store.storeInput(input); // < you can access "input" here, even if the component unmounted already
}, []);
If you don't manage the input in the components state, your whole structure is broken and should be changed to manage state at the right place. In your case, you should lift the shared input state of the components to the parent.

ReactJS docs on hooks specify this:
Effects may also optionally specify how to “clean up” after them by
returning a function.
So any function you return in your useEffect hook, will be executed when the component unmounts, as well as before re-running the effect due to a subsequent render.

What is the intention of using React's useCallback hook in place of useEffect?

I'm trying to understand what the use case is for using React's useCallback hook in place of the useEffect hook.
They both appear to act as a listener for state changes of their inputs (examples taken from the React Docs):
useEffect(
() => {
const subscription = props.source.subscribe();
return () => {
subscription.unsubscribe();
};
},
[props.source],
);
const memoizedCallback = useCallback(
() => {
doSomething(a, b);
},
[a, b],
);
But, the useEffect hook gives the additional benefit of cleaning up resources where you would have previously with componentWillUnmount.
So, what is a good use case for using useCallback? And, what am I missing here?

useEffect has very specific timing aspects related to it that you can read about here. The function specified will be executed after rendering is complete and the DOM has been updated. This will happen after each rendering where any of the values specified in the second-argument array change.
useCallback doesn't automatically execute anything. It returns a function that can be executed by whatever code needs to trigger it. There is no listening to changes that causes an execution of the callback. The array values just control what instance of the function is returned. The array values do not control the timing of the function execution.
A key use case is to pass this function as a prop to a child component to use as an event handler. useCallback allows you to define an inline function to use as an event handler (thus it has access to any other variables in the context where the function is defined) without the downside of passing a unique prop to the child every render. So long as the values in the second-argument array have not changed, the same function will be returned as was returned the previous rendering. So if the child component is a pure component, it will not be forced to re-render simply because of always receiving a unique event handler function.
without useCallback
const Parent = ()=> {
const [a, setA] = useState(null);
const eventHandler = ()=> {
// every render creates a unique instance of eventHandler
// even though it always does the same thing so long as 'a' hasn't changed
doSomethingWithA(a);
}
return <Child onClick={eventHandler}/>
}
with useCallback
const Parent = ()=> {
const [a, setA] = useState(null);
const eventHandler = useCallback(()=> {
// A unique function instance is passed in to useCallback on every render, but
// eventHandler will be set to the first instance of this function
// (i.e. potentially an instance of the function that was passed to useCallback
// on a previous rendering) that was passed to useCallback
// for the current value of 'a'.
doSomethingWithA(a);
}, [a]);
return <Child onClick={eventHandler}/>
}
This article provides a bit more detail than the React docs on the use case for useCallback and other hooks.
Related answer: Trouble with simple example of React Hooks useCallback

What does useCallback/useMemo do in React?

As said in docs, useCallback
Returns a memoized callback.
Pass an inline callback and an array of inputs. useCallback will return a memoized version of the callback that only changes if one of the inputs has changed. This is useful when passing callbacks to optimized child components that rely on reference equality to prevent unnecessary renders (e.g. shouldComponentUpdate).
const memoizedCallback = useCallback(
() => {
doSomething(a, b);
},
[a, b],
);
But how does it work and where is the best to use it in React?
P.S. I think visualisation with codepen example will help everyone to understand it better. Explained in docs.

This is best used when you want to prevent unnecessary re-renders for better performance.
Compare these two ways of passing callbacks to child components taken from React Docs:
1. Arrow Function in Render
class Foo extends Component {
handleClick() {
console.log('Click happened');
}
render() {
return <Button onClick={() => this.handleClick()}>Click Me</Button>;
}
}
2. Bind in Constructor (ES2015)
class Foo extends Component {
constructor(props) {
super(props);
this.handleClick = this.handleClick.bind(this);
}
handleClick() {
console.log('Click happened');
}
render() {
return <Button onClick={this.handleClick}>Click Me</Button>;
}
}
Assuming <Button> is implemented as a PureComponent, the first way will cause <Button> to re-render every time <Foo> re-renders because a new function is created in every render() call. In the second way, the handleClick method is only created once in <Foo>'s constructor and reused across renders.
If we translate both approaches to functional components using hooks, these are the equivalents (sort of):
1. Arrow Function in Render -> Un-memoized callback
function Foo() {
const handleClick = () => {
console.log('Click happened');
}
return <Button onClick={handleClick}>Click Me</Button>;
}
2. Bind in Constructor (ES2015) -> Memoized callbacks
function Foo() {
const memoizedHandleClick = useCallback(
() => console.log('Click happened'), [],
); // Tells React to memoize regardless of arguments.
return <Button onClick={memoizedHandleClick}>Click Me</Button>;
}
The first way creates callbacks on every call of the functional component but in the second way, React memoizes the callback function for you and the callback is not created multiple times.
Hence in the first case if Button is implemented using React.memo it will always re render (unless you have some custom comparison function) because the onClick prop is different each time, in the second case, it won't.
In most cases, it's fine to do the first way. As the React docs state:
Is it OK to use arrow functions in render methods? Generally speaking,
yes, it is OK, and it is often the easiest way to pass parameters to
callback functions.
If you do have performance issues, by all means, optimize!

useCallback and useMemo are an attempt to bypass weak spots that come with the functional programming approach chosen with React hooks. In Javascript, each entity, no matter if it is a function, variable, or whatever, is created into the memory when the execution will enter the function's code block. This is a big issue for a React that will try to detect if the component needs to be rendered. The need for rerendering is deducted based on input props and contexts. Let's see a simple example without useCallback.
const Component = () => {
const [counter, setCounter] = useState(0);
const handleClick = () => {
setCounter(counter + 1);
}
return <div>
Counter:{counter}<br/>
<button onClick={handleClick}>+1</button>
</div>
}
Note that the handleClick -function instance will be created on each function call inside the block, so the event handler's address on each call will be different. The React framework will always see the event handler as changed because of this. In the example above, React will think handleClick as a new value on each call. It simply has no tools to identify it as the same call.
What useCallback does, it internally stores the first introduced version of the function and returns it to the caller, if the listed variables have not changed.
const Component = () => {
const [counter, setCounter] = useState(0);
const handleClick = useCallback(() => {
setCounter(counter + 1);
}, [])
return <div>
Counter:{counter}<br/>
<button onClick={handleClick}>+1</button>
</div>
}
Now, with the code above, React will identify the handleClick -event handler as the same, thanks to useCallback -function call. It will always return the same instance of function and React component rendering mechanism will be happy.
Storing the function internally by the useCallback will end up with a new problem. The stored instance of the function call will not have direct access to the variables of the current function call. Instead, it will see variables introduced in the initial closure call where the stored function was created. So the call will not work for updated variables. Thats why you need need tell if some used variables have changed. So that the useCallback will store the current function call instance as a new stored instance. The list of variables as the second argument of the useCallback is listing variables for this functionality. In our example, we need to tell to useCallback -function that we need to have a fresh version of counter -variable on each call. If we will not do that, the counter value after the call will be always 1, which comes from the original value 0 plus 1.
const Component = () => {
const [counter, setCounter] = useState(0);
const handleClick = useCallback(() => {
setCounter(counter + 1);
}, [counter])
return <div>
Counter:{counter}<br/>
<button onClick={handleClick}>+1</button>
</div>
}
Now we have a working version of the code that will not rerender on every call.
It is good to notice that the useState -call is here just for the same reason. Function block does not have an internal state, so hooks are using useState, useCallback and useMemo to mimic the basic functionality of classes. In this sense, functional programming is a big step back in history closer to procedural programming.
useMemo is the same kind of mechanism as useCallback but for other objects and variables. With it, you can limit the need for component rerender, as the useMemo -function will return the same values on each function call if the listed fields have not changed.
This part of the new React hooks -approach is definitely the weakest spot of the system. useCallback is pretty much counterintuitive and really error-prone. With useCallback-calls and dependencies, it is too easy to end up chasing internal loops. This caveat we did not have with the React Class approach.
The original approach with classes was more efficient after all. The useCallback will reduce the need to rerender, but it regenerates the function again every time when some of its dependant variables will change, and matching if the variables have changes itself will make overhead. This may cause more rerenders than necessary. This is not the case with React classes.

I've made a small example to help others understand better how it behaves. You can run the demo here or read the code bellow:
import React, { useState, useCallback, useMemo } from 'react';
import { render } from 'react-dom';
const App = () => {
const [state, changeState] = useState({});
const memoizedValue = useMemo(() => Math.random(), []);
const memoizedCallback = useCallback(() => console.log(memoizedValue), []);
const unMemoizedCallback = () => console.log(memoizedValue);
const {prevMemoizedCallback, prevUnMemoizedCallback} = state;
return (
<>
<p>Memoized value: {memoizedValue}</p>
<p>New update {Math.random()}</p>
<p>is prevMemoizedCallback === to memoizedCallback: { String(prevMemoizedCallback === memoizedCallback)}</p>
<p>is prevUnMemoizedCallback === to unMemoizedCallback: { String(prevUnMemoizedCallback === unMemoizedCallback) }</p>
<p><button onClick={memoizedCallback}>memoizedCallback</button></p>
<p><button onClick={unMemoizedCallback}>unMemoizedCallback</button></p>
<p><button onClick={() => changeState({ prevMemoizedCallback: memoizedCallback, prevUnMemoizedCallback: unMemoizedCallback })}>update State</button></p>
</>
);
};
render(<App />, document.getElementById('root'));

An event handler gets recreated and assigned a different address on every render by default, resulting in a changed ‘props’ object. Below, button 2 is not repeatedly rendered as the ‘props’ object has not changed. Notice how the entire Example() function runs till completion on every render.
const MyButton = React.memo(props=>{
console.log('firing from '+props.id);
return (<button onClick={props.eh}>{props.id}</button>);
});
function Example(){
const [a,setA] = React.useState(0);
const unmemoizedCallback = () => {};
const memoizedCallback = React.useCallback(()=>{},[]); // don’t forget []!
setTimeout(()=>{setA(a=>(a+1));},3000);
return (<React.Fragment>
<MyButton id="1" eh={unmemoizedCallback}/>
<MyButton id="2" eh={memoizedCallback}/>
<MyButton id="3" eh={()=>memoizedCallback}/>
</React.Fragment>);
}
ReactDOM.render(<Example/>,document.querySelector("div"));