fix: update untracked writes for `BLOCK_EFFECT`s too #16110

paoloricciuti · 2025-06-08T21:14:52Z

This fixes both of this and all the test are actually green. However I'm not convinced this is the right fix because it's a bit greedy: the problem is the following.

a derived of a derived returns a component
that component is read by the component function (and is marked as CLEAN)
mounting that component update a value of that triggers the original derived
this triggers the derived of derived which is marked as MAYBE_DIRTY
since that derived was already read it's not read again so it stays MAYBE_DIRTY
when the value changes the original derived is updated but since the derived of derived is MAYBE_DIRTY his reactions are not marked/scheduled

Now including MAYBE_DIRTY in the flags to mark/schedule fixes it but i think it will schedule more than necessary and i don't know if this could have performance issues. The best action would be to mark the derived as clean if we can somehow detect that it was marked after being read but it's a bit of a weird situation (becuase if you, for example, read it again in an effect the bug it's not present)

Before submitting the PR, please make sure you do the following

It's really useful if your PR references an issue where it is discussed ahead of time. In many cases, features are absent for a reason. For large changes, please create an RFC: https://github.com/sveltejs/rfcs
Prefix your PR title with feat:, fix:, chore:, or docs:.
This message body should clearly illustrate what problems it solves.
Ideally, include a test that fails without this PR but passes with it.
If this PR changes code within packages/svelte/src, add a changeset (npx changeset).

Tests and linting

Run the tests with pnpm test and lint the project with pnpm lint

changeset-bot · 2025-06-08T21:14:56Z

🦋 Changeset detected

Latest commit: 9b3677d

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
svelte	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

github-actions · 2025-06-08T21:16:54Z

Playground

pnpm add https://pkg.pr.new/svelte@16110

dummdidumm · 2025-06-08T21:28:53Z

Looking at the benchmark the overall time spent with GC has more than doubled, so as you say this is likely not the right fix (also because what would even be the difference between maybe dirty and dirty now?)

paoloricciuti · 2025-06-08T21:42:04Z

Yeah to be fair I was even surprised no test failed but I guess it's just removing an optimization.

I guess the right fix would be to find a way to set back this kind of MAYBE_DIRTY to CLEAN even if it's not executed again.

Mmm...I guess technically the derived should be executed again since the value might change 🤔

paoloricciuti · 2025-06-08T23:16:46Z

Ok i tried a bit more, it's getting late: so I think the actual problem is that when we mark the reactions of the derived of derived the first time it's reactions are not yet populated because we are within the function call in update_reaction...so basically we update the block effect in the component function, that invalidate the derived through the $effect.pre which should mark the block for a rerun but it doesn't because block is not going to be added as a reaction until the function finishes to run.

Still need to find a proper solution.

paoloricciuti · 2025-06-09T13:26:59Z

Ok this is a different fix: it's basically extending the logic for untracked writes to BLOCK_EFFECT's too (since the get_component function runs there...not sure of the performance implication, let's see what the benchmark say.

paoloricciuti · 2025-06-09T13:33:10Z

@dummdidumm looking at the benchmarks on main i see gc_time fluctuate a bit (i've seen times up to 400)...this branch currently shows 487 which is worse but not terribly worse maybe?

I wonder if this could be the fix.

.changeset/sour-jars-sniff.md

paoloricciuti · 2025-06-09T13:41:11Z

Tbf they seems pretty similar running compare

sbench_create_signals
  time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼   403.54ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 438.36ms
  gc_time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼    178.47ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 206.05ms
sbench_create_0to1
  time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼  4.50ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 4.74ms
sbench_create_1to1
  time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼  20.19ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 20.77ms
  gc_time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼   2.29ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 2.48ms
sbench_create_2to1
  time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼  17.34ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 18.36ms
  gc_time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼   2.18ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 2.40ms
sbench_create_4to1
  time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼  16.63ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 17.17ms
  gc_time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼    2.12ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 2.46ms
sbench_create_1000to1
  time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 15.27ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 15.23ms
  gc_time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 2.34ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼   2.15ms
sbench_create_1to2
  time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼  9.51ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 10.05ms
  gc_time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼  0.50ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 0.54ms
sbench_create_1to4
  time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼   7.69ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 8.48ms
  gc_time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼    0.43ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 0.50ms
sbench_create_1to8
  time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼   6.35ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 6.96ms
sbench_create_1to1000
  time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼  5.02ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 5.28ms
kairo_avoidable_owned
  time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 273.14ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 271.49ms
  gc_time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 5.73ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼  5.39ms
kairo_avoidable_unowned
  time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 364.04ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 367.89ms
  gc_time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 5.67ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 5.66ms
kairo_broad_owned
  time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 407.80ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 405.03ms
kairo_broad_unowned
  time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 407.99ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 407.22ms
kairo_deep_owned
  time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 180.61ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 182.64ms
kairo_deep_unowned
  time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 888.63ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 891.04ms
kairo_diamond_owned
  time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 302.86ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼   274.38ms
  gc_time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 3.44ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼   3.15ms
kairo_diamond_unowned
  time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 384.47ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 383.55ms
  gc_time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 3.65ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼   3.21ms
kairo_triangle_owned
  time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 101.15ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼   89.68ms
  gc_time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 0.51ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼    0.44ms
kairo_triangle_unowned
  time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 209.01ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 207.17ms
  gc_time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼    0.52ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 0.60ms
kairo_mux_owned
  time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 307.59ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 306.99ms
  gc_time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼  1.85ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 1.96ms
kairo_mux_unowned
  time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 460.08ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 462.24ms
  gc_time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼  2.03ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 2.19ms
kairo_repeated_owned
  time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 44.79ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 44.09ms
  gc_time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 0.19ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼   0.17ms
kairo_repeated_unowned
  time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 45.48ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 44.72ms
  gc_time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 0.18ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼  0.17ms
kairo_unstable_owned
  time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 76.36ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼  74.33ms
  gc_time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 0.64ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼     0.52ms
kairo_unstable_unowned
  time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 92.72ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 92.87ms
  gc_time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼      0.61ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 0.82ms
mol_bench_owned
  time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 232.39ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 230.48ms
  gc_time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼  0.22ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 0.23ms
mol_bench_unowned
  time: fastest is a (derived-reactions-maybe-dirty-marked)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 246.80ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 247.90ms
  gc_time: fastest is b (main)
    a: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼ 0.25ms
    b: ◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼◼   0.22ms

trueadm · 2025-06-09T14:18:40Z

I'm not even sure the benchmark makes use of blocks. However, this looks like the right fix to me.

raythurnvoid · 2025-06-09T17:06:10Z

Hey I was also working on #16090 and i just found out running in production mode behaves correctly without any changes from the main branch.

paoloricciuti · 2025-06-09T19:05:46Z

Hey I was also working on #16090 and i just found out running in production mode behaves correctly without any changes from the main branch.

Yeah however the other bug this closes is still present even in production builds

raythurnvoid · 2025-06-09T20:50:49Z

I'm closing my other PR ^^.

But will spend some time investigating why prod mode fixes that thing, might be revealing something odd

tBuLi12 · 2025-06-10T02:00:03Z

running in production mode behaves correctly without any changes from the main branch

Not quite unfortunately, I've updated #16090 with info on how to observe the issue in prod as well. Hopefully this PR fixes that either way.

fix: mark reactions of MAYBE_DIRTY reactions too

02a30b5

fix: set_untracked_writes for BLOCK_EFFECT's too

20cc0fe

paoloricciuti changed the title ~~fix: mark reactions of MAYBE_DIRTY reactions too~~ fix: update untracked writes for BLOCK_EFFECTs too Jun 9, 2025

paoloricciuti commented Jun 9, 2025

View reviewed changes

.changeset/sour-jars-sniff.md Outdated Show resolved Hide resolved

chore: changeset

9b3677d

raythurnvoid mentioned this pull request Jun 9, 2025

fix: top-most derived in a chain of deriveds marked as MAYBE_DIRTY when executed from a snippet $.fallback #16111

Closed

6 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

fix: update untracked writes for `BLOCK_EFFECT`s too #16110

fix: update untracked writes for `BLOCK_EFFECT`s too #16110

paoloricciuti commented Jun 8, 2025

Uh oh!

changeset-bot bot commented Jun 8, 2025 •

edited

Loading

Uh oh!

github-actions bot commented Jun 8, 2025

Uh oh!

dummdidumm commented Jun 8, 2025

Uh oh!

paoloricciuti commented Jun 8, 2025 •

edited

Loading

Uh oh!

paoloricciuti commented Jun 8, 2025

Uh oh!

paoloricciuti commented Jun 9, 2025

Uh oh!

paoloricciuti commented Jun 9, 2025

Uh oh!

Uh oh!

paoloricciuti commented Jun 9, 2025 •

edited

Loading

Uh oh!

trueadm commented Jun 9, 2025

Uh oh!

raythurnvoid commented Jun 9, 2025 •

edited

Loading

Uh oh!

paoloricciuti commented Jun 9, 2025

Uh oh!

raythurnvoid commented Jun 9, 2025

Uh oh!

tBuLi12 commented Jun 10, 2025

Uh oh!

Uh oh!

Uh oh!

fix: update untracked writes for BLOCK_EFFECTs too #16110

Are you sure you want to change the base?

fix: update untracked writes for BLOCK_EFFECTs too #16110

Conversation

paoloricciuti commented Jun 8, 2025

Before submitting the PR, please make sure you do the following

Tests and linting

Uh oh!

changeset-bot bot commented Jun 8, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🦋 Changeset detected

Uh oh!

github-actions bot commented Jun 8, 2025

Uh oh!

dummdidumm commented Jun 8, 2025

Uh oh!

paoloricciuti commented Jun 8, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

paoloricciuti commented Jun 8, 2025

Uh oh!

paoloricciuti commented Jun 9, 2025

Uh oh!

paoloricciuti commented Jun 9, 2025

Uh oh!

Uh oh!

paoloricciuti commented Jun 9, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

trueadm commented Jun 9, 2025

Uh oh!

raythurnvoid commented Jun 9, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

paoloricciuti commented Jun 9, 2025

Uh oh!

raythurnvoid commented Jun 9, 2025

Uh oh!

tBuLi12 commented Jun 10, 2025

Uh oh!

Uh oh!

fix: update untracked writes for `BLOCK_EFFECT`s too #16110

fix: update untracked writes for `BLOCK_EFFECT`s too #16110

changeset-bot bot commented Jun 8, 2025 •

edited

Loading

paoloricciuti commented Jun 8, 2025 •

edited

Loading

paoloricciuti commented Jun 9, 2025 •

edited

Loading

raythurnvoid commented Jun 9, 2025 •

edited

Loading