2.0.0-rc.4 | optimistic store, read in an effect, changes the visible rendering. and has inconsistent value when logged and when rendered.

[mizulu]

Describe the bug

block enabled

block enabled visible render

    // Renders
    //1,2,3
    //1,2,3,1111
    //3,2,1,5,666

block enabled logs

[2026-08-30T18:10:39.253Z] [log] Initial  [1,2]
[2026-08-30T18:10:39.254Z] [log] 1 yield:pre [1,2]

[2026-08-30T18:10:39.259Z] [log]             EFFECT: [1,2,3]
[2026-08-30T18:10:39.259Z] [log]             EFFECT: [1,2,3]
[2026-08-30T18:10:39.260Z] [log] 1 yield:post [1,2,3]
[2026-08-30T18:10:39.265Z] [log] start
[2026-08-30T18:10:39.267Z] [log] act 0
[2026-08-30T18:10:41.255Z] [log] 2  yield:pre [1,2,3]
[2026-08-30T18:10:41.258Z] [log] 2  yield:post [3,2,1]
[2026-08-30T18:10:42.261Z] [log] <<<<<<<<<<<
[2026-08-30T18:10:42.265Z] [log]             EFFECT: [3,2,1,1111]
[2026-08-30T18:10:43.266Z] [log] 3 mutation:pre [3,2,1]
[2026-08-30T18:10:43.270Z] [log] 3 mutation:post [3,2,1,5]
[2026-08-30T18:10:44.276Z] [log] 4 mutation:pre [3,2,1,5]
[2026-08-30T18:10:44.282Z] [log] 4 mutation:post [3,2,1,5,666]
[2026-08-30T18:10:45.280Z] [log] generator end
[2026-08-30T18:10:49.269Z] [log] <act 0
[2026-08-30T18:10:49.271Z] [log]             EFFECT: [3,2,1,5,666]
[2026-08-30T18:10:49.272Z] [log]             EFFECT: [3,2,1,5,666]

block disabled

---

block disabled visible renders

    // Renders
    //1,2,3
    //3,2,1
    //3,2,1,5,6

block disabled logs

[2026-08-30T18:14:27.326Z] [log] Initial  [1,2]
[2026-08-30T18:14:27.327Z] [log] 1 yield:pre [1,2]
[2026-08-30T18:14:27.328Z] [log] 1 yield:post [1,2,3]
[2026-08-30T18:14:27.330Z] [log] start
[2026-08-30T18:14:27.330Z] [log] act 0
[2026-08-30T18:14:29.320Z] [log] 2  yield:pre [1,2,3]
[2026-08-30T18:14:29.321Z] [log] 2  yield:post [3,2,1]
[2026-08-30T18:14:30.343Z] [log] <<<<<<<<<<<
[2026-08-30T18:14:31.324Z] [log] 3 mutation:pre [3,2,1]
[2026-08-30T18:14:31.324Z] [log] 3 mutation:post [3,2,1,5]
[2026-08-30T18:14:32.325Z] [log] 4 mutation:pre [3,2,1,5]
[2026-08-30T18:14:32.326Z] [log] 4 mutation:post [3,2,1,5,666]
[2026-08-30T18:14:33.338Z] [log] generator end
[2026-08-30T18:14:37.329Z] [log] <act 0

Your Example Website or App

https://s.olid.uk/id/ArFAHr5-Qv2P_vA6KctNiA

Steps to Reproduce the Bug or Issue

see reproduction code, disable and enable block by flipping the flag

Expected behavior

expecting consistent render, regardless if we have an effect or not

Screenshots or Videos

No response

Platform

.

Additional context

import { render } from '@solidjs/web';
import {  createOptimisticStore, Loading, action, createRenderEffect, deep } from 'solid-js';



var actCounter = 0
const act = action(function* (ms = 2000) {
  var i = actCounter++
  console.log("act", i)
  yield new Promise(r => setTimeout(r, ms))
  console.log("<act", i)
})
export default function App() {
  const [s, ss] = createOptimisticStore(async function* (s) {

    console.log("Initial ", JSON.stringify(s))       // offset:0 


    console.log("1 yield:pre", JSON.stringify(s))
    yield [1, 2, 3]
    console.log("1 yield:post", JSON.stringify(s))

    await new Promise(r => setTimeout(r, 2000));
    console.log("2  yield:pre", JSON.stringify(s)) // offset: 2000
    yield [3, 2, 1]
    console.log("2  yield:post", JSON.stringify(s))



    await new Promise(r => setTimeout(r, 2000));  // offset: 4000
    console.log("3 mutation:pre", (JSON.stringify(s)))
    s.push(5)  // not going to render 
    console.log("3 mutation:post", (JSON.stringify(s)))

    await new Promise(r => setTimeout(r, 1000)); // offset: 5000

    console.log("4 mutation:pre", (JSON.stringify(s)))
    s.push(666)
    console.log("4 mutation:post", (JSON.stringify(s)))

    await new Promise(r => setTimeout(r, 1000)); // offset: 6000

    console.log("generator end")
  }, [1, 2] )


  setTimeout(() => {
    console.log("start")
    act(10000);  // 10s transition 

    setTimeout(() => {
      console.log("<<<<<<<<<<<") // this will run prior to "3 mutation:pre" between the 2000..4000 offset 
      ss(draft => draft.push(1111))
    }, 3000)  // offset: 3000
  }, 0)


  // The application renders differently depends if we run this block or not.
  const  enableBlock = true
  if (enableBlock) {
    // Renders: 
    //1,2,3
    //1,2,3,1111
    //3,2,1,5,666
    createRenderEffect(() => {
      return deep(s)

    }, (s) => {
      console.log("            EFFECT:", JSON.stringify(s))
    })

  } else {
    // Renders
    //1,2,3
    //3,2,1
    //3,2,1,5,6
  }


  return (
    <div class="p-2">
      <Loading>
        {JSON.stringify(s)}
      </Loading>
    </div>
  );
}


if (typeof document !== 'undefined') {
  render(() => <App />, document.getElementById('root')!);
}

block disabled playground
block enabled playground

The optimistic write is sometimes visible sometimes it does not depends on our conditional effect read.

also the rendering and the render effect do not align in values for example

when enabled the rendered dom shows:
1,2,3,1111

but the render effect logs:
3,2,1,1111

if not a mistake in reproduction this might be considered an additional symptom

may be related in part or in full to the root cause of the issue in

#3123

[ryansolid]

Root-caused and fixed in #3145. The optimistic write and the store's landing were being captured by the unrelated act() transaction: a transaction opened without any writes (an action that only awaits) never scheduled the flush that parks it, so it stayed ambient across the async gap and adopted whatever arrived next — the store's yield froze until the action's 10s settled, and the deep()-observer run diverged from the DOM. The fix enforces the documented containment ruling (A26): entering a transaction now guarantees a flush, so the ambient window closes regardless of whether the transaction wrote anything. Deterministic regression pins landed at both the signals and web layers, plus a faithful port of your playground. Follow-ups on the deeper store-architecture questions: #3146, #3147. Thanks for the exceptionally reproducible report — the playground link is what made the timing analysis possible.

[ryansolid]

Fixed on next (2023daa, PR #3145). The root cause was a transaction whose ambient window never closed: parking is flush-driven, and an action whose first statements only await opens a transaction with no writes, so nothing scheduled the flush that parks it — activeTransition stayed armed across the async gap and adopted your store's authoritative landing into a transaction it had nothing to do with. That is why the mid-flight yield would not render until the unrelated action settled, why the optimistic push composed over the stale base, and why deep() observers and the DOM disagreed — the passive observer changed adoption timing, not truth. initTransition now guarantees a flush, closing the ambient window in one flush regardless of whether the transaction wrote anything. Pinned by three test layers, including a faithful port of your playground program on real timers with the deep() observer both present and absent. Thank you for the unusually precise reproduction — the observer-changes-rendering observation was the tell.