fix(signals): refresh() lifts a stale manual-write mask inside a transaction

[brenelz]

Fixes #3026

The bug

refresh(store) on a derived store (createStore(fn, init)) never re-ran the source function when called inside an action() — but only when a setStore had happened earlier in the same transaction (in the issue repro, the setStore right after act1() is absorbed into the still-ambient action transition).

Root cause: the derived-store setter calls suppressComputedRecompute, which sets REACTIVE_MANUAL_WRITE so a manual write wins over a queued recompute for the tick (#2692 / core R31). That mask is only cleared by the pending-node drain (commitPendingNode) — which inside a transaction doesn't run until the action commits. refresh() silently bailed on the mask, so any setStore early in an action swallowed every later refresh() for the rest of the transaction. That's why deferring the refresh in a setTimeout (after commit) worked.

The fix

suppressComputedRecompute now stamps the current clock tick on the node (_manualWriteTime), and refresh() distinguishes the two cases:

• Same synchronous tick: the manual write still wins over the refresh, in both orders — the pinned Solid v2.0.0-beta.10 batched parent & derived signals manual update is neglected for derived signals #2692 contract ("manual memo writes win over refresh within the same tick", createMemo.test.ts) is unchanged.
• Mask stamped in an earlier tick (only possible when a transaction is holding the drain open): the explicit refresh lifts the mask, the source re-runs, and its result reconciles over the transaction draft at commit — one atomic reveal.

When lifting the mask, no REACTIVE_REASK (quiet re-ask) mark is set, since the batch carries a real manual value change.

Tests

New tests in action.test.ts ("refresh(store) inside an action") cover the issue shape and the re-fetch shape (fresh object per derive — reconciled result wins over the draft at commit), and store/createStore.test.ts ("derived store manual writes") pins both #2692 precedence directions (same-tick setStore-then-refresh and refresh-then-setStore both keep the manual write) plus the cross-tick refresh. Full solid-signals and solid suites pass with no regressions.

One note on the issue's exact repro: the derive there returns the same object identity the store already adopted as its raw state, so the re-run's reconcile is an identity no-op (owned-raw model, pre-existing behavior). The source function re-running — the reported symptom — is fixed; realistic re-fetch derives that return fresh data get the full expected reconcile.

🤖 Generated with Claude Code

[changeset-bot]

🦋 Changeset detected

Latest commit: d76914a

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

This PR includes changesets to release 9 packages

Name	Type
@solidjs/signals	Patch
test-integration	Patch
solid-js	Patch
babel-preset-solid	Patch
@solidjs/web	Patch
@solidjs/html	Patch
@solidjs/h	Patch
@solidjs/universal	Patch
@solidjs/element	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

[codspeed-hq]

Merging this PR will degrade performance by 24.3%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

❌ 1 regressed benchmark
✅ 126 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	omit	235.8 µs	311.6 µs	-24.3%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing brenelz:fix/refresh-in-action-3026 (d76914a) with next (77d1348)}