[ctidd]

One of the problems Redux is used to solve (as a global store) is peace of mind that you won't need to refactor large swathes of component trees to reparent state and callbacks that need to be shared or persisted across different subtrees as new business requirements arise. Hooks (especially custom hooks, which are just a composition of other hooks packaged together as a single function) make the reparenting/hoisting/anchoring of state and callbacks trivial compared to other mechanisms, providing similar peace of mind that you're not boxing yourself into an inextensible component tree. (This is regardless of context; passing down props isn't a large pain point, and is often misguided to try to solve because it leads to importable components and hidden contracts.)

Render props and HoCs solve similar problems for composition/reuse of functionality (these are all mixins at heart), but the hoistability of hooks is really the distinguishing mark in my experience.

[ricardobeat]

Replacing redux with hooks that return local state can be a road to hell, as calling the hook somewhere down the tree will duplicate the state and there is no good way to prevent developers doing that.

[nightski]

IMO that's not much different than introducing a new variable in your global state to represent the same thing because you didn't realize there was already one there.

If you have a team you need discipline, code reviews, and leadership. No state solution is going to solve this for you.

[ricardobeat]

It is not a matter of discipline. Avoiding this gotcha requires you to manually inspect your entire parent component hierarchy looking for uses of that hook - sometimes that is a plain grep away, but it is a very much invisible error. You also have to be familiar with the inner workings of the hook itself, and any other nested hooks, to verify if they use any local state or not. If it does, and you need to share that state, well, that's a big rewrite which kind of defeats the initial argument.

Maybe extra tooling could be built to avoid this, if we weren't already drowning in linter plugins...

[nightski]

Again this isn't a problem with hooks. You could easily connect a redux store and reducer to some component low in the hierarchy with it's own state. Or use a class based component with some state. A developer can introduce random state anywhere and it has nothing to do with hooks themselves. If you can't trust your team to make the right decisions about where to place state, then you need to provide more guidance.

I also disagree about the big re-write. Converting a hook's data/state to come from a prop instead is a very simple change.

[schwartzworld]

I love how every criticism of hooks is like "if you don't take care to use them well, you get bad code".

[diob]

I always try to tell folks that frameworks won't save them from themselves.

[ctidd]

It sounds like you're talking about network resolved state. I like to share this sort of state using a subscription model, where components subscribe to a property of a state root (which anchors/owns the state and how to fetch it), resolving that property's state when mounted. Where that root is in the tree determines the lifecycle of the state (e.g. it's gone when unmounted), and multiple subscribers to the same property can mount, unmount, etc. at any point during that lifecycle without duplication.

I believe this model has similarities to both Apollo and Angular services, though I don't have direct experience with either.

[ricardobeat]

No, plain useState. What you're describing can be done via Context, which gets you back to "need to refactor large swathes of component trees to reparent state and callbacks that need to be shared or persisted across different subtrees as new business requirements arise", but worse.

EDIT: quick example here https://codesandbox.io/s/unruffled-easley-ry6x2?file=/src/Ap...

This is a fairly innocuous example since the bug is immediately apparent from the button not working. Now imagine the failure mode is more subtle, and these components are about two dozen layers apart down the tree.

[ctidd]

That's not a bug with hooks, but almost a complete misunderstanding of how hooks work (*on the part of anyone who writes that code thinking it will behave otherwise). Hooks are effectively instantiated on a specific component instance. Calling useState multiple times like this is multiple instances of useState, whether or not it's wrapped in another function.

And I'd like to be very clear that I strongly advise against cutting through layers with context because I wholeheartedly agree with your assessment there. You can plumb down those handles explicitly through props and enforce they're provided in a type system.

[ricardobeat]

Of course it's not a bug, that is not in question. As I mentioned above, the problem is that even if you are fully aware of this, you still have to check that whatever `useSomeStatefulThing` hook you used to replace a global store does not use local state. So the assumption you can simply replace a global store with localized hooks to manage state is not true, and will certainly lead to bugs. You must use Context instead (or proo drilling as you suggest).

[ctidd]

Since you provided a concrete scenario, I should probably do the same to clarify what I'm referring to in re-parenting or hoisting hooks. It doesn't really map to the statement "replace a global store with localized hooks", so I'm probably doing a poor job communicating. Similarly, I do not assume "you can simply replace a global store with localized hooks", because like you say, that's not true, and I can appreciate you clearly understand how hooks work.

At the same time, unless I'm really misunderstanding the example you shared, I can't see the problem or mental overhead you're mentioning where we need to be vigilant about having the same hook used in multiple places or understanding the scope of the hook's "local" state and callbacks. That's the point of hooks as a unit of encapsulation -- just like with a class, each use of a hook is a different instance, and the scope is that of the instance (and hook instances are scoped/bound to the lifecycle of the containing component instance). The code in the problem you showed is equivalent to calling setState on the one component instance and expecting that to show up on another instance which happens to be of the same component class. If we understand how React component instances work, that's clearly not the case (state isn't broadcast across instances), and the same goes for hooks.

Looking at what I mean by the improved portabiliy/hoistability of hooks I mentioned in previous comments, let's say we have three components: App, Foo, and Bar. App renders Foo and Bar as children.

We have a business requirement that Foo has some behavior which contains multiple pieces of state and bindings to multiple React lifecycle events (e.g. mount). What we don't know is whether we'll ever need to share that behavior with Bar.

React class API: We need to write bindings for this behavior against class state and lifecycle methods, intermingled with other code. This intermingling means you know there's a good deal of refactoring to do to move the behavior upward to App if we ever need to share the behavior with Bar. As a component grows, more and more logic intermingles on those lifecycle methods and makes refactoring more challenging.

  class Foo extends React.Component {
    state: {
      behaviorState: {...},
      unrelatedState: {...},
    };

    componentDidMount() {
      initBehavior(this.props.input);
      initUnrelated();
    }

    ...
  }

React hook API: We can easily write this behavior in a custom hook, encapsulating the behavior in some combination of useState, useEffect, and other React hooks. Now we have a self-contained useBehavior hook, by definition isolated from any local state, which you can choose to use within Foo.

  const Foo = ({ input }) => {
    const behaviorState = useBehavior(input);
    const unrelatedState = useUnrelated();
    return ...;
  };

At this point, we're not so worried if we need to hoist that one function call up to App and pass the hook's returned handle (behaviorState) back down to Foo and Bar at some point in the future.

If this example sounds trivial, I've packaged up 500+ lines of functionality in a single self-contained hook before, and that's all exposed as a single function that consumers can treat as a full encapsulation of all of that functionality, while not worrying if they need to hoist that function call up and pass the output down at some point in the future.

I'm not talking about calling the same hook in Foo and Bar (these are different instances), nor sharing local state sideways across different invocations of hooks, but rather having full confidence it won't be hard to re-anchor that hook upward in the component tree if need be.