injectEffect cleanup not called when atom has ttl=0

[joprice]

I have configured an atom with ttl of 0. It uses injectPromise and returns the promise's api. Is also uses injectEffect to schedule a polling updater to refresh the data. This works as expected, except when I navigate away from the page, the cleanup function is not called and the polling continues. Is this expected and do I need to configure anything else to get it to unmount? Or is there another common scenario where this can happen?

At first, since I'm using react router, I thought it might be holding routes in memory, but I checked the component tree, and do not see the page that creates the atom.

[joprice]

I inspected the ecosystem and found that the atom had a dependent on an instance returned by useAtomInstance for the same atom. Switching that for a useEcosystem and moving the usage of the instance into a callback using getInstance got the cleanup function to start be called. Is there something special about instances that they are not cleaned up on component unmount?

[joprice]

Disabling StrictMode fixes it in this case so I guess it is this fix facebook/react#25713 that is included in react 19 that's making things work.

Odd thing is, I'm not using StrictMode in the app where I'm having this issue, so maybe this is enough to exhibit the problem in this case, but not the precise case I was hitting. Is there another fix in 19 that affects useId that is unrelated to StrictMode?

[bowheart]

I don't think there's another fix in React 19. What you're possibly seeing is a bug in Zedux's useId workarounds. It's very possible Zedux 1.3 fixes the problem in the non-StrictMode app.

Ah yes, sorry, I'd forgotten a bit about this StrictMode issue. That is a beautiful, simple repro. I don't think I realized it reproduces so easily. I cloned it and plugged in/stepped through several older and newer Zedux versions. I'm back up to speed on this and then some - thanks to this simple repro, I have a fix that I was never able to quite achieve. I've plugged in React 18 and 19 and stepped through the tests. I'm pretty confident it's compatible with StrictMode and non-StrictMode in both React versions and is memory-leak-free and performant.

I've PR'd the fix in #154. I'm planning on publishing it in v1.3.1 under the latest npm dist-tag.

[joprice]

Just tried 1.3.1 and it works like a charm. I'm seeing my effect get cleaned up exactly once on routing away from a page.

[bowheart]

Beautiful, was just coming back to comment. I published v1.3.1 and plugged it into the repro repo. Even adding unstable selector refs via useAtomSelector and multiple different types of hook calls, the repro now creates the same expected dependencies in both React 18 and 19, both with and without StrictMode.

Thank you @joprice!

[joprice]

I think a similar bug might be present in v2. I'm noticing that when triggering a log out, which unmounts a significant number of components, some of the atoms are destroyed, but not all. I tried a repro here https://github.com/joprice/zedux-repro/compare/v2Gc. If you log the snapshot of the ecosystem, then toggle the component and log again, even after waiting some time, the atoms and selectors remain in memory indefinitely. The effect atom has an empty o value, which I would think means there are no observers and it should clean itself up. Am I mistaken in that assumption?

[joprice]

Oh wow - "By default, all atoms live forever." I didn't realize this. For some reason, I thought there was some other default of a few seconds. And v2 removed atomDefaults. I was using ecosystem.reset before on logout, but it's no longer feasible as it causes the ui to lose some state it needs even on the logged out screens and I have some expensive stuff I want to keep around, so I only want to reset a subgraph. But, perhaps now that I understand how this is working. If I convert more of my code to use ions, it would work well for some use cases. However, even when logged in there are some atoms where I would want to keep some expensive data around in case it's used on different screens with some time in between, but I'd want it to be immediately cleaned up upon log out. I guess I can use the function version of setTtl to compute that dynamically.

[bowheart]

ttl is still required in atoms. I pulled your example repo and changed userNameAtom, userDetailsAtom, and effectAtom to ions (which have ttl: 0 by default) and now everything cleans up when pressing Toggle (except the three nodes that are there initially)

[joprice]

Yea I've been living in a fantasy land of imaginary 5-second gc for months. Using zedux-logger made me realize things weren't behaving how I assumed. With this better observability I think I'll come up with some nice strategies to optimize things after I complete the ionization of the repo.