[blindmute]

It's a great benefit for a component to be able to specify what data it needs from remote without worrying about if an ancestor or sibling also needed it. React query will only fetch once the unique queries in the whole tree.

Without this feature, every component needing remote data D must share a parent ancestor fetching D for all of them, even when the children are not conceptually related. Adding another component higher in the tree means you have to hoist the fetcher up to that level, etc. React query is an incredible upgrade to anything else I've used over the years.

[gofreddygo]

> must share a parent ancestor fetching D for all of them

What happened to the whole "service" pattern? Isn't this the core problem it solved for Java in the 90's?

[blindmute]

An external service fetcher doesn't have integration with React because the entire tree potentially needs to know when the data changed. Once you write the code to subscribe to those changes where necessary, you've basically invented another state library.

[gofreddygo]

* Service fetcher does fetching and updating store

* Component does subscribing

* React does updating component tree with changes.

If a component wants changes it must subscribe. If not, it doesn't.

If there is a change, subscribed components get updates. If not, they don't.

Responses are (conditionally) cached in the store by its respective service. Any component that wants data just asks the service. Service fetches from the store or remote. I can customize and unit test the the service.

I don't see the overlap in responsibility between any state lib (react or whatever in this case) and the purpose built service.

What is react query's role here ?

[erikpukinskis]

You’re hand-waving away “subscribing” as if it’s not a time bomb.

The set of possible events any given query might need to subscribe to is massive. Yes, you can build a large app with a pub/sub model but the number of subscriptions runs away from you and it gets really hard to know which of 1000 subscriptions you need to refresh when an association between two random pieces of data is created/deleted.

Associations are where you really get screwed IMO.

That said, despite exploring lots of possibilities I don’t think there’s any perfect way to do this automatically. So while I think you’re hand-waving away a hard problem, I also think your proposed solution is probably one of the better ones.

Still, I don’t think you should pooh-pooh people trying to solve the general problem. It’s a worthy one.

[gofreddygo]

That's a sensible response, so thanks for that.

Being hard to define and manage subscriptions is a problem, a bigger deeper problem that react query or any other lib isn't solving.

By "set of possible events any given query might need to subscribe to is massive" i assume you meant "component" and not query, if not then i don't know what that means, if yes, that component needs refactoring.

I'm interested in libraries not because of what it can do for my code or me or my team but what problem it solved for the team that needed it in the first place, bad enough to write a lib for it. I want to avoid those problems.

I start with importing libs for quick TAT and slowly replace them with a few hundred lines of code that i can understand, modify, debug, monitor, unit and integration test. I usually end up redefining or better defining the problem and rescoping the issue to not needing to solve said problems.

Most libs do NOT provide that level of valuable solutions to worthy problems to go keep them around as deps. Deps are not free and have no liability to me or my code. React query is one more in that long list. IMHO.

[erikpukinskis]

I do mean “query”. I am presuming you want every query in your app to live update when the results change. The problem I’m indicate is: how do you know which queries are invalid after any given event?

[blindmute]

> What is react query's role here ?

Its role is to do everything you just said in one package, plus other networking features. It fetches, updates the store, subscribes the components, caches the responses, performantly rerenders. It allows configuration of when to refetch, how to cache, polling, pagination, infinite scroll, etc etc via a simple API.

Nothing is stopping you from writing all this yourself, but libs exist for a reason. It's a terribly useful networking package. If you just use a generic store and write a fetcher yourself, you have to at the very least write logic for when to (re)fetch and for persisting the responses to the store.