-
-
Notifications
You must be signed in to change notification settings - Fork 4.5k
Svelte 5: Allow classes to opt-in to deep reactivity #11846
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Comments
Is |
It is under my control. Is there a better alternative I'm missing? |
I guess the question then is, whether there is a reason not to make the fields of the class stateful? Given the example thing: class Thing {
- foo: number;
+ foo = $state<number>();
- bar: string;
+ bar = $state<string>();
} (Would need assertions if you want to prevent |
The reason not to go that route is that One might say this is a userland problem and could be solved pretty robustly with a simple helper. Something like: let reactiveObj = $state(deeplyMakeStateful(plainObj)) where And that does work, but now we're walking the object tree twice (worse performance) and adding boilerplate (unintuitive) wherever |
How is it possible with I'm using some libs (not Svelte specific) that works a lot with classes. let myObj = $state<MyObj | undefined>()
const refresh = async (_slug: string) => {
const p = await repo(Package).findFirst({ slug: _slug }, { include: { items: true } })
if (p) {
// TODO: make it generic...
myObj = {
...p,
items: (p.items?? [])?.map((i) => {
return { ...i }
}),
}
} else {
goto('/', { replaceState: true })
throw "Doesn't exist!"
}
}
$effect(() => {
refresh($page.params.slug)
}) Would you share your Thx for you help. |
@jycouet You can e.g. use a definite assignment assertion: class Counter {
value: number = $state()!; // note `!` at the end
constructor(initial: number) {
this.value = initial;
}
} |
Hi, I am trying to migrate a UI project that also uses Classes and inheritance. I am running into this issue. When I try what you suggest in my Class field, at runtime I get an error from the source code export class BasicController extends PrecisController {
currentValue = $state<number>(0);
taper: Taper
id = $state<string | undefined>();
constructor() {
super()
} which seems to get re-written to File: src/lib/PrecisControllers.svelte.ts:197:19
113 | constructor() {
114 | super();
115 | this.currentValue = $state(0);
^
116 | this.id = $state();
117 | } I don't really understand why this doesn't work for me? .. later OK, I thought about it, and tried this way. To make a function in the base class that returns the reactive state unwrapped, as described in the docs. private getValue( widget: BasicController ): { mapped: number, normalised: number } {
let current = $state(widget.currentValue);
return {
get mapped(): number {
return remap(current / widget.height, 0, 1, widget.taper.min, widget.taper.max)
},
get normalised(): number {
return (current / widget.height)
}
}
}
getMappedValue(): number {
return this.getValue(this).mapped
}
getNormValue(): number {
return this.getValue(this).normalised
} But it still doesn't respond... stuck. |
You need a TS config that does not mess with the class definition. |
Thanks, that totally got things unstuck! |
Related #10560 |
I have just been introduced to Svelte5 and I find the lack of reactivity for instances of class something to be addressed for sure. In other frameworks (and in JS in general), classes have just been Objects on steroids, keeping all the compatibility with most operations you could perform on Objects. This is fairly common in other languages as well, such as C++ in which classes are just structures with public and private access. This means that lots of plain JS libraries chose to use classes instead of I think making some kind of helper, as proposed above, would be a convenient solution to keep compatibility with existing class-based JS libraries. |
Building on this pattern to reset state when props change, I found a strategy for deriving class properties. function computed<T, U>(
value: () => T,
transform: (input: T) => U,
start: (update: () => unknown) => void | (() => void),
) {
let stop: void | (() => void);
const _derived = $derived.by(() => {
const _value = value();
untrack(() => {
stop?.();
stop = start(
() => (_derived.current = untrack(() => transform(_value))),
);
});
let state = $state({ current: untrack(() => transform(_value)) });
return state;
});
return _derived;
} When you change a non-reactive value, you trigger an update the signal manually. let grid = $state<NonReactive>(new Grid());
let value = computed(
() => grid,
(v) => v.data.x, // derive deep `grid` value
(update) => {
const id = setInterval(() => {
grid.data.x++;
update(); // force update (i.e. value.current = grid.data.x)
}, 1000);
return () => clearInterval(id);
}
);
assert(value.current == grid.data.x) It works best if the thing triggering manual updates can be contained in the let node = $state<HTMLElement>(); // set later when DOM is ready
const height = computed(
() => node,
(node) => node?.clientHeight ?? 0,
(update) => {
if (node) {
const observer = new ResizeObserver(update);
observer.observe(node);
return () => observer.disconnect(); // cleaned up when `node` changes
}
},
); Any points for improvement or footguns with this approach? LGTM. Unrelatedly linking #14520 here for someone issue-surfing for a solution. It feels like a different problem with a similar root cause. EDIT: let node = $state<HTMLElement>(); // set later when DOM is ready
const height = computed(
() => node,
(n) => n?.clientHeight ?? 0,
);
$effect(() => {
if (!node) return;
const observer = new ResizeObserver(() => {
height.update() // force update
});
observer.observe(node);
return () => observer.disconnect(); // cleaned up when `node` changes
}) |
I'd love the feature too, |
Classes are not made deeply reactive because they are different to POJOs in that they encapsulate certain behavior and should give more control over what changes when. It's would also fail for every class that has private properties, as proxies cannot be used with them (you would get a runtime error) |
I found a bug that is more like a feature, that makes classes deeply reactive: data.svelte.js
App.svelte
If the App.svelte is not in runes mode, pressing the button How ever if you activate runes mode it doesnt work. Calling the increment method, will never result in an reaction/ increment in the UI. I am a bit confused? |
Runes mode invalidate the whole object when you mutate a property of it so it is expected |
Describe the problem
I understand the rationale for
$state
not proxying non-POJO objects, but I think having non-POJO reactive state is a very useful pattern that is worth trying to support.A very simple example (and my current use case): Part of state is a
Grid
which wraps a 2d array and provides helper methods likegrid.at(coord: Coord)
. By forcing state to be only POJO, I now have to do one of:grid[coord.row]?.[coord.col]
throughout the appGrid.at(grid, coord)
There are loads of example use-cases like this, as it's a pretty common data modeling pattern (usually a
Model
pattern is just "some raw data + some helper methods to access/modify the raw data").Describe the proposed solution
For the sake of nomenclature, let's just refer to classes and class instances, though of course classes aren't the only way to create objects with a non-
Object
prototype.It would be nice to allow classes to opt in to being "proxy-able", either by specifying that the proxy should wrap the entire object, or by specifying some subset of fields to be proxied.
A potential syntax: (exact naming TBD)
Then, when proxy checks whether to wrap the object, it can also check
prototype[$state.proxyable]
and proceed accordingly.Option 1 downsides: Users may hit errors resulting from private/internal property accesses, arguably making
$state
a slightly leakier abstraction than it currently is.Option 2 downsides: The list of proxied fields would also have to be stored in the metadata and queried during various proxy traps, which is less than ideal from a performance perspective.
Importance
nice to have
The text was updated successfully, but these errors were encountered: