All the examples are fetching data from a server, and in such cases I think tanstack query already does all the hard part. I feel like people under-use react query and put too much state in their FE. This might be relevant if your app has some really complicated interactions, but for most apps they should really be a function of server, not client, state. Of course this exact reasoning is why I moved off react altogether and now use htmx in most of my projects
It's not just react query, you can make a quick useFetch and useMutation hooks (or claude can), it's not that complex. If you don't need more advanced features (eg caching), you can easily cut down on 3rd party dependencies.
The problems OP tries to address are unfortunately a deep design flaw in mainstream frameworks like React and Vue. This is due to 2 properties they have:
1. They marry view hierarchy to state hierarchy
2. They make it very ergonomic to put state in components
I've been through this endless times. There are significant ways to reduce this friction, but in the end there's a tight ceiling.
This is why this kind of work feels like chasing a moving target. You always end up ruining something inherent to the framework in a pursuit to avoid the tons of footguns it's susceptible to.
It's also why I moved to Gleam and Lustre (elm architecture) and bid those PITAs farewell
Is this because Elm forces you to separate the model computations from the view computations, which then lets you compose the model shape in one place and the view shape in the other, or some other property of the framework that I'm not aware of?
Vue provides a computed feature that acts as a buffer layer between the view and the state, so the view and the state are not necessarily strictly bound
You can use React in an MVC framework, with React used to implement the 'V' in MVC. You can use class components, and the code becomes extremely simple. Business logic is moved to 'M' layer (models) and the 'C' layer (controllers) coordinates everything. No hooks or other messy stuff needed.
In fact, React was originally designed to be used this way (as the V in MVC)!
You can do this pretty simply with hooks and reducers, where the reducer/dispatch are the model/controller.
Class components have their own pitfalls when you start messing around with componentDidUpdate, which was part of the motivation behind functional components IIRC.
I will say that I used to really not like hooks and that’s because I think the migration documentation and messaging was handled pretty poorly.
That being said, if you’ve worked with state management like Redux, I would argue that hooks are pretty similar to how you hook up that kind of data store.
Sorry for being pedantic, but the first example could be rewritten to extract the pattern into a higher level hook, eg useNotifications. One way to simplify components before reaching for store libraries. The reusable hook now contains all the state and effects and logic, and the component is more tidy.
function Dashboard() {
const { user } = useAuth();
const {loading, error, notifications, undreadCount, markAsRead} = useNotifications(user);
if (loading) return <Skeleton />;
if (error) return <p>Failed to load: {error}</p>;
return (
<div>
<h1>Dashboard ({unreadCount} unread)</h1>
<StatsCards stats={stats} />
<NotificationList items={notifications} onRead={markAsRead} />
</div>
);
}
Working with multiple teams in a large project, hooks can be a nightmare to maintain.
I see 7x layers deep of hooks, with no test cases to support them. Some of the side effects are not properly tested, and mocks that abstract away the whole implementation means the test case only works for certain scenarios.
FWIW this scenario might be an outlier in large projects, considering how some developers prefer to "just wrap the hook in another hook, and not worry about its internals".
That's a valid concern. I've seen some hard to grok hooks with polling, async stuff, hard to follow logic, etc. Like with anything, need to have taste, it's easy to dump too much into one hook and like you mention, it gets hard to follow what gets triggered when.
I was wondering if I was crazy for thinking "how is what he's suggesting different than just putting that 'class' into a hook function?" I'm glad to see someone already wrote it up, kudos.
Javascript and classes go together like toothpaste and orange juice. All good JS programmers I know essentially pretend that classes don't exist in the language (or if they use them, they only do so rarely, for very niche cases).
JS does not have classical OOP built in! It has Brandon Eich's prototypal inheritance system (which has some key differences), along with a 2015 addition to the language to pretend it has OOP (but really that's just lipstick on the underlying prototypal pig).
If you use classes in JS, you're bound to be disappointed at some point when they don't behave like classical OOP. Most devs accept that and use more functional approaches (like factory functions) instead of OOP.
Counterpoint: classes are a great way to bundle state and logic - which is exactly what UI components are - and components models should use classes more, not less.
React's "functional" components are simply poor approximations of classes. Instead of easy to read and reason about class fields, you put state in useState() function classes that are effectively named by their appearance order in source and who's current state you can't introspect with dev tools!
The component function mixes one-time setup (which should be a class constructor) with repeated render calls (which should be a method), so those of course have to be separated by putting the one-time work inside of a closure inside of the repeated should-have-been-a-callback function. Event listeners and other callbacks are another huge mess. Don't forget useMemo() or useCallback() (but which?).
It's actually quite mad.
And the differences between classical and prototypal inheritance basically don't even pop up under normal class usage in JS: just use class fields, don't mess with the prototype chain, and don't dynamically add or delete properties - all things that are how classical inheritance works - and things just work like you expect.
They're modeling reactivity, not classes. It's a well established pattern in functional programming
The one time setup mixed with repeated render calls is odd, but it's a design decision they made. It reduces boiler plate, though I don't necessarily agree with it because it is a leaky abstraction
I have noticed that inheritance is largely ignored by experienced developers but it's a hard argument to make that "all good JS programmers do this".
Classes are invaluable and are an extremely efficient and ergonomic way to manage state in GUI applications.
That said, avoiding classes was published in some blog post at some point and the JS hype machine went crazy with FP. As a consequence, I have yet to observe a maintainable React codebase. Good looking and performant React applications are even fewer and farther between.
Personally, writing idiomatic React has me focus too much on render cycles that I think less about how the application looks & feels. Appropriate abstractions become more difficult to conceptualize and any non-trivial application ends up a 5mb bundle with no multi-threading or optimizations. This is also what I have observed "the best JS devs" do in the wild.
I'm not sure if it would support inheriting from a custom store very well. It might get tricky with the templating. But the author of this seems to have done a good job of not ignoring inheritance.
Personally this library isn't to my taste - but I successfully use classes (without inheritance) along with reactivity primitives to create beautiful, tiny and high performance React applications
I think the biggest issue with classes is subclassing, it looks like a good feature to have, but ends up being a problem.
If one avoids subclassing, I think classes can be quite useful as a tool to organize code and to "name" structures. In terms of performance, they offer some good optimizations (hidden class, optimized instantiation), not to mention using the memory profiler when all your objects are just instances of "Object" can be a huge pain.
Essentially `new Foo()`, where `Foo` can be a subclass of `Bar` that inherits properties in the same way we all learned in our Java (or whatever actual OOP) language.
JavaScript gives you a class syntax that lets you make classes and extend them from each other, and for the most part they will work the same way as a class from a language like Java ... but some things won't.
You can either become an expert on prototype chains, and how JS actually implements OOP (differently from Java) ... or you can just write factory functions instead of classes.
Don’t most OO languages have similar differences? My understanding is that even Java is quite different from, say, Smalltalk (which arguably is “more OG”).
I went through the code, and it is roughly equivalent to doing it without classes. The nesting level in the example isn't deep - everything seems to be beneath a SnapStore or a SnapFormStore, without inheriting from user defined classes. I think the use of classes is fine, and the way that it's introduced in the blog post is good, it says "plain TypeScript classes". It is used as a means for ergonomically writing and understanding the code more than it is for setting up invariants or complex polymorphism.
All the examples are fetching data from a server, and in such cases I think tanstack query already does all the hard part. I feel like people under-use react query and put too much state in their FE. This might be relevant if your app has some really complicated interactions, but for most apps they should really be a function of server, not client, state. Of course this exact reasoning is why I moved off react altogether and now use htmx in most of my projects
It's not just react query, you can make a quick useFetch and useMutation hooks (or claude can), it's not that complex. If you don't need more advanced features (eg caching), you can easily cut down on 3rd party dependencies.
Yeah, fetching data from a server in useEffect is a widely acknowledged and documented antipattern.
It's a widely documented anti-pattern, while also not giving a convenient alternative.
Is it? I have never had any issues doing it this way.
This post explains it pretty well: https://dev.to/deveshsangwan/stop-using-useeffect-for-data-f...
The problems OP tries to address are unfortunately a deep design flaw in mainstream frameworks like React and Vue. This is due to 2 properties they have:
1. They marry view hierarchy to state hierarchy
2. They make it very ergonomic to put state in components
I've been through this endless times. There are significant ways to reduce this friction, but in the end there's a tight ceiling.
This is why this kind of work feels like chasing a moving target. You always end up ruining something inherent to the framework in a pursuit to avoid the tons of footguns it's susceptible to.
It's also why I moved to Gleam and Lustre (elm architecture) and bid those PITAs farewell
Is this because Elm forces you to separate the model computations from the view computations, which then lets you compose the model shape in one place and the view shape in the other, or some other property of the framework that I'm not aware of?
Yes. In the Elm architecture it is possible (and ergonomic) to model your state separately from the view
Vue provides a computed feature that acts as a buffer layer between the view and the state, so the view and the state are not necessarily strictly bound
Computed is only a concept because you need a band-aid for that lack of separation. You can read more about my efforts ditching Vue here:
https://blog.nestful.app/s/the-tech-behind-nestful
You can use React in an MVC framework, with React used to implement the 'V' in MVC. You can use class components, and the code becomes extremely simple. Business logic is moved to 'M' layer (models) and the 'C' layer (controllers) coordinates everything. No hooks or other messy stuff needed.
In fact, React was originally designed to be used this way (as the V in MVC)!
See https://github.com/Rajeev-K/mvc-router
You can do this pretty simply with hooks and reducers, where the reducer/dispatch are the model/controller.
Class components have their own pitfalls when you start messing around with componentDidUpdate, which was part of the motivation behind functional components IIRC.
Not everyone likes hooks. Class components may have issues, but if hooks are the solution I'll stay with class components, thank you very much!
I will say that I used to really not like hooks and that’s because I think the migration documentation and messaging was handled pretty poorly.
That being said, if you’ve worked with state management like Redux, I would argue that hooks are pretty similar to how you hook up that kind of data store.
Yeah, Redux is another big mistake I stayed away from. "Line after line of throat clearing and pseduostructure", as one developer put it [1]
[1] https://news.ycombinator.com/item?id=15342764
Sorry for being pedantic, but the first example could be rewritten to extract the pattern into a higher level hook, eg useNotifications. One way to simplify components before reaching for store libraries. The reusable hook now contains all the state and effects and logic, and the component is more tidy.
Working with multiple teams in a large project, hooks can be a nightmare to maintain.
I see 7x layers deep of hooks, with no test cases to support them. Some of the side effects are not properly tested, and mocks that abstract away the whole implementation means the test case only works for certain scenarios.
FWIW this scenario might be an outlier in large projects, considering how some developers prefer to "just wrap the hook in another hook, and not worry about its internals".
That's a valid concern. I've seen some hard to grok hooks with polling, async stuff, hard to follow logic, etc. Like with anything, need to have taste, it's easy to dump too much into one hook and like you mention, it gets hard to follow what gets triggered when.
I was wondering if I was crazy for thinking "how is what he's suggesting different than just putting that 'class' into a hook function?" I'm glad to see someone already wrote it up, kudos.
@OP: PEBKAC, respectfully.
Far cleaner, how is testability though?
Very easy - mock the useNotifications and you can easily see all the behaviour by changing three properties.
Javascript and classes go together like toothpaste and orange juice. All good JS programmers I know essentially pretend that classes don't exist in the language (or if they use them, they only do so rarely, for very niche cases).
JS does not have classical OOP built in! It has Brandon Eich's prototypal inheritance system (which has some key differences), along with a 2015 addition to the language to pretend it has OOP (but really that's just lipstick on the underlying prototypal pig).
If you use classes in JS, you're bound to be disappointed at some point when they don't behave like classical OOP. Most devs accept that and use more functional approaches (like factory functions) instead of OOP.
Counterpoint: classes are a great way to bundle state and logic - which is exactly what UI components are - and components models should use classes more, not less.
React's "functional" components are simply poor approximations of classes. Instead of easy to read and reason about class fields, you put state in useState() function classes that are effectively named by their appearance order in source and who's current state you can't introspect with dev tools!
The component function mixes one-time setup (which should be a class constructor) with repeated render calls (which should be a method), so those of course have to be separated by putting the one-time work inside of a closure inside of the repeated should-have-been-a-callback function. Event listeners and other callbacks are another huge mess. Don't forget useMemo() or useCallback() (but which?).
It's actually quite mad.
And the differences between classical and prototypal inheritance basically don't even pop up under normal class usage in JS: just use class fields, don't mess with the prototype chain, and don't dynamically add or delete properties - all things that are how classical inheritance works - and things just work like you expect.
They're modeling reactivity, not classes. It's a well established pattern in functional programming
The one time setup mixed with repeated render calls is odd, but it's a design decision they made. It reduces boiler plate, though I don't necessarily agree with it because it is a leaky abstraction
Reactivity and functional programming are orthogonal though.
And in most functional systems the one-time setup function would return the render function so the render function can close over state.
Which is pretty much what a class with fields, a constructor, and a render method give you.
I have noticed that inheritance is largely ignored by experienced developers but it's a hard argument to make that "all good JS programmers do this".
Classes are invaluable and are an extremely efficient and ergonomic way to manage state in GUI applications.
That said, avoiding classes was published in some blog post at some point and the JS hype machine went crazy with FP. As a consequence, I have yet to observe a maintainable React codebase. Good looking and performant React applications are even fewer and farther between.
Personally, writing idiomatic React has me focus too much on render cycles that I think less about how the application looks & feels. Appropriate abstractions become more difficult to conceptualize and any non-trivial application ends up a 5mb bundle with no multi-threading or optimizations. This is also what I have observed "the best JS devs" do in the wild.
Inheritance is used here:
https://github.com/thalesfp/snapstate/blob/ba8a8d7ce25d6a4ef...
I'm not sure if it would support inheriting from a custom store very well. It might get tricky with the templating. But the author of this seems to have done a good job of not ignoring inheritance.
Personally this library isn't to my taste - but I successfully use classes (without inheritance) along with reactivity primitives to create beautiful, tiny and high performance React applications
Encapsulation, inheritance and polymorphism all work fine with JavaScript classes. OOP works just fine.
What doesn't work in JavaScript is functional programming.
I don’t see how? You can do all sorts of FP in js and it even has some of it in its built in APIs. .then comes from FP
Yes, but promises are (unfortunately) _not_ monads!
https://rybicki.io/blog/2023/12/23/promises-arent-monads.htm...
You are right, there are some FP features in JavaScript. But the way it is used in the predominant UI framework (namely React) is definitely breaks FP, see https://mckoder.medium.com/why-react-is-not-functional-b1ed1...
JS is primarily a functional language, it is built around functions as first class objects and closures.
The issue is the bar has been raised for what people call "functional" now. Everyone is picky "OMG not pure so it isn't really functional!!"
Yeesh.
> Javascript and classes go together like toothpaste and orange juice.
Even HN has been taken over by shills from Big Mint.
I think the biggest issue with classes is subclassing, it looks like a good feature to have, but ends up being a problem.
If one avoids subclassing, I think classes can be quite useful as a tool to organize code and to "name" structures. In terms of performance, they offer some good optimizations (hidden class, optimized instantiation), not to mention using the memory profiler when all your objects are just instances of "Object" can be a huge pain.
For clarity, what do you call "classical OOP"?
(disclaimer: FP all the way, regardless)
Essentially `new Foo()`, where `Foo` can be a subclass of `Bar` that inherits properties in the same way we all learned in our Java (or whatever actual OOP) language.
JavaScript gives you a class syntax that lets you make classes and extend them from each other, and for the most part they will work the same way as a class from a language like Java ... but some things won't.
You can either become an expert on prototype chains, and how JS actually implements OOP (differently from Java) ... or you can just write factory functions instead of classes.
Can you give examples of how they are different? I've only done OOP in JS so I'm not aware of what I'm missing or what's supposed to be different.
Don’t most OO languages have similar differences? My understanding is that even Java is quite different from, say, Smalltalk (which arguably is “more OG”).
Potato potato. Js classes are just closure sugar. So what?
Syntax makes sense and improves readability in business logic. Readability is good
I went through the code, and it is roughly equivalent to doing it without classes. The nesting level in the example isn't deep - everything seems to be beneath a SnapStore or a SnapFormStore, without inheriting from user defined classes. I think the use of classes is fine, and the way that it's introduced in the blog post is good, it says "plain TypeScript classes". It is used as a means for ergonomically writing and understanding the code more than it is for setting up invariants or complex polymorphism.
We have a similar style of react state manager that we use at Aha! https://github.com/aha-app/mvc
I think the intent is very similar even though there are some structural differences: move the state and state logic out of the view to classes.
Same: https://github.com/alshdavid-public/mvvm/blob/main/examples/...
Just to sanity-check my reading of this:
- Zustand exposes itself as a hook.
- MobX does that observer-wrapper thing
- Snapstate instead has an explicit writing step (`scoped()`) at the bottom of a component
If so, I really quite like that. Kudos!
What exactly abut MobX do you find too magical/implicit?
Still waiting for "I was tired of AI titles using the format 'I was tired of x, so I built y', so I built ..."
Seems like a solution in search of a problem.
I wish I'd never used current-day React as well