2.0.0-rc.0 | royal flush - inconsistent read/write in onSettled #3006

Activity

1. ryansolid

   added a commit that references this issue on Aug 17, 2026

   fix(signals): unify effect-phase write visibility — callbacks never o…

   ...

   a7780c7
2. ryansolid commented on Aug 17, 2026

   

   ryansolid

   on Aug 17, 2026

   Member

   More actions

   Thanks for the careful report — you found a genuine inconsistency, and it's now fixed on next (a7780c7).

   What was happening: onSettled (and createTrackedEffect) callbacks ran with a tracked context internally, which made them observe their own staged writes immediately — unlike the effect half of createEffect and event handlers, which read committed values until the graph settles. That asymmetry was an accident of the read path, not a design choice.

   The fix: all effect-phase scopes now share one contract:

   • Reads return the settled graph — a callback never observes its own unsettled write (same as setState in a React effect).
   • Writes are queued into the same flush's continuation, so dependent memos/effects still update before the flush returns — nothing is lost or deferred to another tick.
   • Functional setters still compose: set(v => v + 1) twice increments twice.

   On flush(): it genuinely cannot run inside these callbacks — the flush that invokes them is already in progress, so it's inherently non-reentrant (React's flushSync has the identical restriction in effects). Dev now reports this loudly everywhere (onSettled/createTrackedEffect throw; createEffect callbacks warn instead of silently no-op'ing), and both messages point at the escape hatch: queueMicrotask(() => flush()). The JSDoc for onSettled and createTrackedEffect now documents all of this.
3. ryansolid

   closed this as completedon Aug 17, 2026
4. ryansolid

   added 3 commits that reference this issue on Aug 18, 2026

   chore(size): bump scenario limits for the rc batch (flatten 66accfb, #…

   ...

   b36f896

   store rewrite: mirror core's #3006 committed-visibility rule

   ...

   b1d7ee8

   Store rewrite: single-home storage model replaces the legacy store (#…

   ...

   0797215
5. ryansolid

   added a commit that references this issue on Aug 27, 2026

   Fix projection stores leaking unsettled writes to effect-phase readers (

   ...

   de9e3cb
6. ryansolid

   added a commit that references this issue on Sep 6, 2026

   fix(signals): wake tracked effects when a value they read stale commi…

   ...

   3af0f6b
7. ryansolid

   added 2 commits that reference this issue on Nov 2, 2026

   fix(signals): tracked effect re-runs for a same-pass write (#4598) (#…

   ...

   aeca8da

   fix(signals): tracked effect re-runs for a memo's same-pass write and…

   ...

   7ea8706