On the other hand, class components have a better developer experience for components that are mostly dumb renderers that bind a lot of event handlers - `private handleFooClick = () = > { … }` is a lot nicer than `const handleFooClick = useCallback(() = > { … }, […])` and you don’t have to worry about ordering between your methods definitions in classes. Declaration ordering between hooks is annoying, but what you get in return is incremental computation. If you don’t need that stuff, sure - use a class!
I do find more and more frequently as I’m working with class components thinking “dang, this would be much easier with hooks”, especially for tricky problems. I don’t find the opposite to be true in hook components.
But for some reason, picking someone else's class component and working on it feels much easier than someone's functional component.
I 100% agree with the last statement, most problems feel easier to write with hooks than with class components.
I know I’m not typical at all but hooks kind of marked the point where I lost interest in keeping up with the changes in React. I’m lucky that my job doesn’t require me to and I’ve had a great time exploring options like Vue and Svelte. Svelte in particular feels like it takes the opposite approach to hooks-based React: rather than force you to think about the way React is processing things it lets you write code the way you think about things and transpiles it into working JS.
Beyond “everyone uses it” I can find few reasons to prefer React these days.
It didn't help that my first foray into hooks was trying to build a simple clock with `setInterval`:
https://overreacted.io/making-setinterval-declarative-with-r...
I really need to give Svelte a try.
While I love working with hooks and do so daily, I definitely felt some remorse for those now having to figure out when and how to use memoization, dependency arrays, useCallback, etc, etc etc. React in its first formulation was utterly simple and beautiful.
In defense of hooks however, it's good to keep in mind that now you can do a lot more than you could, and being able to share business logic so easily across an application is an elegant abstraction.
Hooks are better for building component abstractions than HOCs. Code which can hook into mount/unmount, can trigger re-renders via state updates, etc.
Also, dependency arrays in useEffect and useCallback are better than this kinda stuff in shouldComponentUpdate lifecycle method:
if (prevProps.commentId !== this.props.commentId) {
return true;
}
return false;
[0]: https://reactjs.org/docs/higher-order-components.htmlI generally like shaping my components as functions first and use hooks because I have to in order to do this but definitely not because I find them easier to understand than component lifecycle methods (even if they may be less verbose).
To me, hooks are the ugly part, writing component as a single function is the part that makes it worth it.
Never found HOC ugly, it's a normal function, not a magic one that hooks into React internals. Anyone familiar with OOP also regcognizes it as adapter. I think people tend to confuses syntactical simplicity with conceptual simplicity. But even syntactically, I don't find hooks simple once we pass the toy code examples.
> Hooks are better for building abstractions than HOCs
HOCs are more explicit, they change the props signature of the component. So if you like magic, you'd prefer hooks.
Hooks are not better abstraction, they are better building blocks because they compose more nicely.
A good abstraction is something you can pass around and use liberally without knowing the inner working. You can pass a HOC around and put it in conditional etc. Can't do that with hook because the inner working prohibits that and you are required to know such caveats.
https://ui.dev/why-react-hooks/
I also discussed some of the tradeoffs in a conference talk and blog post:
https://blog.isquaredsoftware.com/2019/07/blogged-answers-th...
https://blog.isquaredsoftware.com/2019/09/presentation-hooks...
That said, the React docs are pretty clear about the motivation for hooks and I think that is a good place to start: https://reactjs.org/docs/hooks-intro.html#motivation.
After that, this post explains roughly how hooks actually work (in a simplified way), and why they're a more natural fit for what's happening under the hood: https://www.netlify.com/blog/2019/03/11/deep-dive-how-do-rea...
I've read this a number of times, and none of the point make sense to me.
> It’s hard to reuse stateful logic between components
Is there some reason you can't use a `get(key)` and `set(key,value)` between class components. I don't really grok react, despite trying, so maybe there's a reason that doesn't work.
> Complex components become hard to understand
I don't understand what the hard part is supposed to be here. If you have a method that's dealing with too many concerns, you can just add methods, and then call those.
> Classes confuse both people and machines
`this` is pretty confusing in javascript, but yet that's the language react is targeting. And it's no less confusing in prototype-based objects. And after having tried to learn both hooks and `this`, it certainly doesn't seem like hooks are conceptually easier than `this`.
Probably my fundamental problem with hooks is trying to understand which state goes with which component-function invocation. It just seems like dark magic that usually works. I've tried to read the implementation to figure it out. It seems to depend on the famous DOM diff feature. It was about at that point that I lost the plot.
So, I know I'm an outlier, but class-based components just make sense to me. They can be complex, but fundamentally comprehensible. Function components seem like a fake simplicity that's masking a complexity that I've yet to figure out.
You can and the hooks way is better.
The class based components way is to wrap component in a HOC. The common state is stored in the HOC. The HOC passes the state to the component via props. This can include functions, so maybe your component can call `this.props.reloadApi()`, which changes HOC state, which results in new props sent to your component.
The hooks way is described here. [0]
What's wrong with Higher-Order Components?
One answer is that it's weird to call something a "component" which doesn't render any UI. In other words, why is an interface to access remote data a "component"?
Another answer is that HOCs break the diff algorithm without special care see. [1]
[0]: https://reactjs.org/docs/hooks-custom.html
[1]: https://reactjs.org/docs/higher-order-components.html#dont-u...
Maybe I'll try implementing this to show what I mean/learn why it wouldn't work.
If yes, the reason you can't just do that is because if that is done naively then components using `get()` won't know that they need to re-render when a `set()` is called somewhere else.
Not done naively and you do have global stores built for this, like Redux or MobX, which handle this gracefully and (at this point) use hooks under the hood and in the dev-facing API to accomplish this.
Tbh this is also now built-in to React with the `useContext()` hook, which allows you to do the same. `useState()` should really only be used for local-to-a-component level state (though technically you can obviously pass down the state/setter to any child components, but given the above I find it's frequently better to `useContext()`).
https://codepen.io/tomtheisen/pen/NWvdXyJ
I already like it better than hooks. Everything's out in the open. No magic or hidden dependencies.
// this could be made to track actual depencies with more granularity
subscribe(() => this.forceUpdate())
You're well on your way to building a decent system here...I guess if you don't pass your dependencies as an array to a function called useEffect, you aren't using hooks. Win!
And if you need a specialized function to only update 1 component instead of all of them, just make sure it isn't returned by a dastardly magical function called useState.
You joke, but yes. Also, you don't have to follow these. https://reactjs.org/docs/hooks-rules.html
And you don't have to rely on the diff remembering which state goes to which widget. To its credit, it seems to do it right at least 99% of the time.
I recognize I'm tilting at windmills here. React is a juggernaut, and many people love it. In fact, many people like the exact things that I don't. And that's ok.
Order of dispatch to subscribers seems like it create similar problems, although it would be fully deterministic, rather than a race.
Anyway, thanks for your insight.
You're largely right but this part is mostly wrong in practice because of async. You can create race-like conditions very easily with promises/async if you're not careful.
> Order of dispatch to subscribers seems like it create similar problems, although it would be fully deterministic, rather than a race.
Yes, that's closer to what I meant. Not a data race, but you could end up in a loop.
[0]https://reactjs.org/blog/2016/07/13/mixins-considered-harmfu... [1]http://raganwald.com/2016/07/16/why-are-mixins-considered-ha... [2]http://raganwald.com/2016/07/20/prefer-composition-to-inheri...
* don't put side-effect initialization code into the constructor * use state to render your components * don't set state directly, use setState, _unless_ in the constructor * if you want to use stores, declare a static field
You don't inherit classes, as there are many special methods, and overwriting them would mean calling super - before your method body? After? During?
Class based components only pay off if you have truly large and complex components, and that's an antipattern in react anyway. Hooks are concise, and super clear if you keep it simple (there are also additional rules not enforced by the compiler, like putting all side-effectful code into useEffect, but still).
I don’t think that the class-vs-function divide is the fundamental reason for the problems with the old model or the improvements with the new model. Rather, I think the learnings from the original approach informed the system and resulted in a step forward.
I find that once I need to manage a nontrivial amount of state, the functional model really feels wrong — useContext, I’m looking at you! And I wouldn’t be surprised if we end up with another turn of the crank that introduces a class-based (or prototype-based) approach with a better set of abstractions at some point down the line.