Inconsistent ownership warning

[benmccann]

Discussed in #15481

^{Originally posted by midzelis March 9, 2025}
Is this a bug or inconsistent behavior?
https://svelte.dev/playground/3057c1db35e745f39d4a73061a9c1e22?version=5.22.6

When passing a class with $state to a component with bindable, if the component binds to a nested property, an ownership failure is logged. However, if the component directly binds to a $state variable then no ownership failure is logged. Playground link shows bug by default. Open console under result in playground to see console logs.

In playground, the rendering/output still worked - however, in my complex real world app, it did not work as expected.

Reply from @brunnerh:

This looks like a bug to me. The warning should appear in both cases because the demo property is not actually bound (setting $bindable is just a prerequisite to that).

[paoloricciuti]

Mmm interesting...the problem arise because we only check the ownership in proxies...so the reason why the $state(0) change doesn't trigger an invalid mutation warning is because $state(0) is not proxified.

[dummdidumm]

This is arguably the correct behavior:

• declaring something public on a class means that people can reassign it from anywhere as it's considered public API
• you don't need to explicitly bind in that case because things go through the class, it's not a POJO that's mutated
• on the other hand, if you mutate the state of the class, you're now mutating something you probably shouldn't, hence the warning

One could argue it also shouldn't warn on the class mutation because it's on the class, but that might be hard to achieve.

[huntabyte]

This seems related to: #13607 as well right?

[brunnerh]

declaring something public on a class means that people can reassign it from anywhere as it's considered public API

This does not make sense to me. Properties on plain objects can also be considered their public API, I don't see why properties on classes should be treated differently in that regard. An object is being modified without binding.

[benmccann]

if you mutate the state of the class, you're now mutating something you probably shouldn't, hence the warning

How can we make that assumption?

One could argue it also shouldn't warn on the class mutation because it's on the class, but that might be hard to achieve.

I'm not sure where the warning lives, but perhaps we could skip the warning whenever we see a mutation happening to a something containing a . in the name?

[falco467]

@paoloricciuti I have finally managed to create a reproduction on the Svelte Playground:

https://svelte.dev/playground/d770d7e5b0e241758b2e24ae45ae1b32?version=5.23.0

The tricky part: The warning disappears if you reorder the two children.

[falco467]

Since there are several threads for this problem please respond if my case fits here, or if I should post it somewhere else?

[benmccann]

This needs to be <Child bind:demo />

[benmccann]

It's still confusing to me that the warning only happens when the class property is a POJO and not a literal though

[dummdidumm]

As pointed out above it works as expected, because properties are considered "you know what you're doing" terrotiry, therefore closing.