Regression in v2: parent() returns old data even after the parent was invalidated using invalidate()

[kyngs]

Describe the bug

This is hard to explain, but I'll do my best.
Imagine the following route structure:

routes/
├─ +page.svelte
├─ +layout.ts
├─ [id]/
│  ├─ +page.svelte
│  ├─ +page.ts
├─ add/
│  ├─ +page.svelte

The root page shows a basket of fruits, which is loaded from an API in the root +layout.ts. It also includes a button that links to the add page.
The add page includes a button that adds fruit to the basket by posting to the API. After doing so, it invalidates the root +layout.ts load function and then navigates to the root page.
The [id] page only verifies whether a fruit with such an ID exists in the basket. It does that using the +page.ts load function, which fetches the parent using await parent() and then observes, whether a fruit with such an ID is in the parent data.

Now, let's imagine I do this:

• I navigate to the add page
• I press the button to add a fruit
• I get redirected to the root page, where the fruit is now listed.
• I click the fruit, which redirects me to /[id] with the fruit id.
• However, the page tells me that no such fruit exists.
• I click back and navigate to the root page, which now does not list the fruit at all.
• I refreshed the page, and now everything works.

Here's a video where I do that:

simplescreenrecorder-2024-01-20_22.15.00.mp4

However, when I downgrade to sveltekit v1, or use invalidateAll() instead of invalidate(), everything works correctly, and the parent() function returns correct data.

Reproduction

SvelteKit v2: https://stackblitz.com/edit/stackblitz-starters-dp4avt?file=src%2Froutes%2F%5Bid%5D%2F%2Bpage.ts
The issue here is easily reproducible as shown in the video

SvelteKit v1: https://stackblitz.com/edit/stackblitz-starters-et87cn?file=src%2Froutes%2F%5Bid%5D%2F%2Bpage.ts
The issue is not reproducible, everything works as expected. See the video:

simplescreenrecorder-2024-01-20_22.24.28.mp4

And as a bonus, SvelteKit v2 using invalidateAll(): https://stackblitz.com/edit/stackblitz-starters-4rh3gc?file=src%2Froutes%2Fadd%2F%2Bpage.svelte
The issue is also not reproducible, everything works as expected.

Logs

Not applicable

System Info

System:
    OS: Linux 6.6 Arch Linux
    CPU: (24) x64 AMD Ryzen 9 5900X 12-Core Processor
    Memory: 35.83 GB / 62.70 GB
    Container: Yes
    Shell: 3.7.0 - /usr/bin/fish
  Binaries:
    Node: 18.18.2 - /usr/bin/node
    Yarn: 1.22.21 - /usr/bin/yarn
    npm: 10.2.5 - /usr/bin/npm
    pnpm: 8.14.0 - /usr/bin/pnpm
    bun: 1.0.18 - /usr/bin/bun
  Browsers:
    Chromium: 120.0.6099.199
  npmPackages:
    @sveltejs/adapter-auto: ^3.0.0 => 3.1.1 
    @sveltejs/kit: ^2.0.0 => 2.4.1 
    @sveltejs/vite-plugin-svelte: ^3.0.0 => 3.0.1 
    svelte: ^4.2.7 => 4.2.9 
    vite: ^5.0.3 => 5.0.12

Severity

blocking an upgrade

Additional Information

PS: Even though using invalidateAll is a workaround, I've decided to use the "blocking an upgrade" severity as rerunning all the load functions is not acceptable.
I hope I explained the issue well, if you have any questions, ask me.

[gterras]

This bug is very real but hasn't been acknowledged by the team yet, see #11268 (comment).

And also :

#11663
#11446
#10209
#10123
#10457
#10359

[kyngs]

This bug is very real but hasn't been acknowledged by the team yet, see #11268 (comment).

And also :

#11663 #11446 #10209 #10123 #10457 #10359

Thanks for the information! I hope they tackle this soon as this is a very severe issue for me.
I'm not really experienced in nodejs, so I have no idea where to start if I were to fix this on my own.

[kyngs]

I'm kinda spooked this is a half year old issue tho

[gterras]

I'm kinda spooked this is a half year old issue tho

Partial invalidation has never been fully functional in SvelteKit, even pre 1.0. I suspect none of the maintainers know how to fix it.

[kyngs]

I'm kinda spooked this is a half year old issue tho

Partial invalidation has never been fully functional in SvelteKit, even pre 1.0. I suspect none of the maintainers know how to fix it.

Hasn't it? Tbh I've never had any issues with it in v1

[gterras]

Hasn't it? Tbh I've never had any issues with it in v1

You most likely had the double invalidation bug with SK1 without realizing it #10359 since they seem heavily related, but at least it wasn't app breaking.

[kyngs]

This bug is very real but hasn't been acknowledged by the team yet, see #11268 (comment).

And also :

#11663 #11446 #10209 #10123 #10457 #10359

@Rich-Harris I'm sorry for mentioning you, as I'm sure you have a lot of work. But you're the author of the original PR which @gterras suspects to be the culprit. Unfortunately, the issues with invalidating have been here for almost half a year, and they seem to progressively get worse. Could you please take a look into this?

[kyngs]

This bug is very real but hasn't been acknowledged by the team yet, see #11268 (comment).

And also :

#11663 #11446 #10209 #10123 #10457 #10359

I can confirm that the PR is causing the issue, removing the invalidated.length = 0; does fix my issue, but it might cause a new one.

[kyngs]

@stalkerg hey! Seeing that you liked the issue, do you have the same problem?

[stalkerg]

@kyngs not yet; I did it to improve visibility. Unfortently I have no time now to go deeper and fix it. If nobody fixes it next week, I will try to help.

[hmnd]

I'm experiencing the same issue. In my case, I have a page that calls parent() and an invalidate call in the layout that redirects to /login when an auth check fails. Previously that worked just fine, but now it gets stuck showing the same page, but loaded with /login's page data.

[kyngs]

@kyngs not yet; I did it to improve visibility. Unfortently I have no time now to go deeper and fix it. If nobody fixes it next week, I will try to help.

Seems like noone fixed it yet, could you please look into it?

[denjpeters]

Is there any movement on this? This is a fairly serious regression, and is preventing us from moving to SK v2.

I did create a simple GitHub project that demonstrates the problem, if that helps... https://github.com/denjpeters/SK2Test

[kyngs]

Is there any movement on this? This is a fairly serious regression, and is preventing us from moving to SK v2.

I did create a simple GitHub project that demonstrates the problem, if that helps... https://github.com/denjpeters/SK2Test

Unfortunately, there is no movement on this.

[gterras]

Tried to start something here #11819

[dummdidumm]

Closing as duplicate of #11446