Bug: Multiple stores using the same key does not work properly

[Peter Larsson-Green]

I've encountered a bug when two different pages are open at the same time and uses the same store/key.

Reproducible instructions:

1. git clone https://github.com/PeppeL-G/svelte-persisted-bug.git
2. cd svelte-persisted-bug
3. npm install
4. Open / in a tab in the web browser (store1 is created)
5. Open /page-2/show in another tab in the web browser (store2 is created)
6. On the /page-2/show page, click on the Hide counter link (the store2 variable is deleted)
7. On the / page, click on the inc button (store1 is properly incremented by 1)
8. On the /page-2 page, click on the Show counter link. store2 is created/retrieved from your internal cache, but it does not contain the incremented number in store1 even though they use the same key

So, in one and the same document, it seems like creating a store with the same key a second time does not use the latest value from local storage.

[Bertram Madsen]

This seemed like a really weird case when I first read the description, but having cloned the repo and played with the code, it is reproducible. To understand what is happening here, I believe it is important to understand two key things about this library:

• When a persisted store is created, it is being kept track of in an internal object, indexing all persisted stores ever created by their key. This means two things:

1. When creating two persisted stores in the application layer with the same key, it is not two different stores refereeing to the same localStorage key, but instead the same store, as the second time the store is "created" in the application layer, a new store isn't created, but a reference to the earlier store with the same key is simply passed.
2. Persisted stores are never removed. This means that clicking "Hide counter" here removes the reference to the store in the application layer, but as the library retains a reference to the store it is not actually removed

• When a store is not subscribed, it doesn't monitor events from the localStorage. This is done for perfomance reasons (I assume) and is a normal practice within svelte stores.

I believe what is happening here is the following:

• You navigate to /, creating store1
• You navigate to /page-2/show, creating store2
• You navigate to /page-2, deleting the reference to store2 in the application layer (and therefore unsubscribing from it), but keeping it alive in the library layer
• You click the inc button on the / page, updating store1 and the reference in localStorage, but as you have unsubscribed from store2, this doesn't recieve the localStorage event, and therefore its value doesn't change
• You navigate to /page-2/show, and since store2 has been saved in the library layer, you retrieve this from the library cache with the old value and subscribe to it

Now, this here is a good catch, and one that requires a minor rework of the central structure of this library. I propose two fixes:

• Reading from localStorage once a store is subscribed to (just like when creating the store). This ensures consistency, and prevents scenarios like this, where a store becomes stale as a result of having a period of unsubscription.
• Removing the library cache of persisted stores. Not only would this fix this specific bug, but I think that it would lead to much more predictable behavior on the user-side, as this would allow to say, have two stores refering to the same key to have different settings. What we are currently doing with both the internal cache, and using the localStorage api is essentially two sources of truth, and I don't believe the cache is necessary, as we can just query the newest value from localStorage. If I (as an app-dev) is interested in keeping the store alive, I should do it in the application layer by importing it from a regular TS file.

I would love to draft up a PR for this, but I just wanted to check in with you @joshnuss, as these are some pretty significant changes to the library

[Joshua Nussbaum]

@bertmad3400 Thanks for the detailed analysis!

When creating two persisted stores in the application layer with the same key, it is not two different stores refereeing to the same localStorage key, but instead the same store

Correct. Since the persisted store is essentially a wrapper over a localStorage key, having many copies around seemed like a waste and possible consistency error.

Persisted stores are never removed
What if we fixed that? Then @PeppeL-G's issue would be resolved. Plus, it would fix a potential memory leak for apps that create many temporary stores.

[Bertram Madsen]

So I have done a lot of digging and experimentation, and have concluded two things: This is more complicated than I assumed, and I'm a dumbass who can't read the manual.

As it turns out, there is a really good reason for the internal store cache, but it isn't the one given by @joshnuss (which even though I fundamentally disagree with is another valid reason for it)

This (Storage Events) won't work on the same browsing context that is making the changes — it is really a way for other browsing contexts on the domain using the storage to sync any changes that are made. Browsing contexts on other domains can't access the same storage objects.

• MDN

The need for an internal cache structure is the fact that multiple stores using the same key will not automatically stay in sync on the same page, as the storage event won't propagate. While this has the unfortunate (and un-documented) side-effect that stores using the same key, but different settings will use the settings of whichever store was created first (which can be really confusing), but I see no way around this.

Secondly however, while my analysis was rather thorough it was completely wrong. Store1 and Store2 shouldn't stay in sync because of the storage events (that doesn't even work), but because they are a reference to the very same store. Why this doesn't work I have no clue, and after digging into this source code, related test environment details and web documentation for the last 4 hours, I'm completely toasted. I might have a look at it another day

[Joshua Nussbaum]

The need for an internal cache structure is the fact that multiple stores using the same key will not automatically stay in sync on the same page, as the storage event won't propagate

Is it possible to subscribe to storage event multiple times? Then each store could subscribe to the storage event, and unsusbscribe during cleanup and we wouldn't need a global mapping of keys/stores.

[Bertram Madsen]

Is it possible to subscribe to storage event multiple times? Then each store could subscribe to the storage event, and unsusbscribe during cleanup and we wouldn't need a global mapping of keys/stores.

No, that is unfortunately not possible. As described, storage event's only fire across context, and not in the same context, and while it would be possible to manually fire it, this would be incredibly messy, duplicate events (and related handling) and break standards. I think you made the right call with the caching structure, and as far as I can see, that isn't what's causing the problem in this specific issue

[Peter Larsson-Green]

I don't know if you have heard about it, nor if it will be useful for you, but instead of listening to the storage event, maybe the Broadcast Channel API can be helpful.

[Joshua Nussbaum]

@PeppeL-G, that's a great point!

If we want to keep cross-tabs working, we could always roll our own channel events.

And then the storage event could be captured

• To determine if an event happened outside Svelte (might not need to support this)
• To determine if a local storage key was deleted. (still unsure what the behaviour should be is when a key is deleted)