fix(signals): isPending() inside a memo reports a hold whose staged node is unstamped (#3457)

[brenelz]

Fixes #3457.

The gap

const slow = createMemo(() => delay(1000, count()));
const copy = createMemo(() => count());
const pending = createMemo(() => isPending(copy));
<p>Slow: {slow()}</p>
<p>Direct pending: {String(isPending(copy))}</p>
<p>Memo pending: {String(pending())}</p>

setCount(1) holds the write for a second. Observed on next (jsdom port, 100ms delay):

t=0..80:  Count: 0 | Slow: 0 | Direct pending: true  | Memo pending: false
t=100:    Count: 1 | Slow: 1 | Direct pending: false | Memo pending: false

Expected Memo pending: true beside Direct pending: true for the whole hold.

Traced: slow recomputes first, pends, and queue notification opens the transaction (the ambient batch is adopted; currentBatch becomes the transaction). copy recomputes next, with the transaction already active: queuePendingNode pushes its staged 1 straight into the transaction's pending nodes, but nothing stamps copy._transition on that path (stamping happens at batch adoption and at stash). syncCompanions (#3413) correctly flips copy's pending companion to true, which OPT-dirties pending on the companion's lane.

pending recomputes under that lane (A31) and probes copy. A memo is not a stale reader, so laneReadsCommitted declines and the probe reads the fresh staged 1. recordFreshRead asks heldAwaitingAsync(copy): t resolves to activeTransition, whose _asyncReporters already lists slow as pending, but copy carries no stamp yet and the scan was gated on if (!et) return false. So copy lands in freshReads, the A10 pairing rule mutes the companion's true, and the memo caches false. It is registered in suppressedProbes, but wakeSuppressedProbes only fires when a reporter is added, and slow registered before copy ever recomputed, so nothing corrects it. The direct render effect is stale, reads the committed 0 under the lane, is not a fresh read, and reports true. Hence the disagreement for the whole hold.

The stamp gate was added by #3078 alongside the t = activeTransition fallback, purely to keep the reporter scan's previous reach; it carries no rule of its own. #3413 (companion sync) and #3445 (A31 lane posture) both hold here: the companion flips, the memo runs under its own lane. The failure is in the verdict pairing, not the value read.

Fix

Drop the stamp gate in heldAwaitingAsync so the reporter scan also runs for an unstamped node. This is sound because whenever activeTransition is non-null, currentBatch === activeTransition, so a node with a staged _pendingValue and no stamp necessarily belongs to that transaction: t resolved to the transaction that owns its staged write. The A10 pairing rule is unchanged for what it was ruled for, a LANDED answer awaiting reveal: the scan returns true only while a source in the transaction is still pending with reporters, so a transaction with nothing computing still mutes the verdict for a reader that saw the fresh value, stamped or not. The #3078 action carve-out is untouched.

Trade-off stated plainly: the still-computing carve-out now also applies in the mid-flush window between a transaction's adoption and its stash. The alternative, stamping _transition in queuePendingNode when the batch is a transaction, would touch the scheduler hot path and the core byte floor for the same outcome.

Timeline after the fix (same jsdom port):

t=0..80:  Count: 0 | Slow: 0 | Direct pending: true  | Memo pending: true
t=100:    Count: 1 | Slow: 1 | Direct pending: false | Memo pending: false

Verification

• @solidjs/signals: 180 files / 2036 tests passed, 1 skipped, including the new tests/ispending-memo-unstamped-hold-3457.test.ts (A10 / #3457 isPending(copy) inside a memo reports the hold like the direct probe; fails on next, passes here). 2.0.0-rc.8 isPending reports false for a held synchronous memo #3413, Adding combined isPending breaks atomic publication #3442 and 2.0.0-rc.3 | didn't get the memo (isPending memo reads inconsistently in a transition) #3078 pins still green.
• @solidjs/web: 88 files / 792 tests passed, including the new test/ispending-memo-unstamped-hold-3457.spec.tsx (A10 / #3457 both probes read true for the whole hold, false after the reveal; the playground repro with a real click and the DOM sampled through the hold; fails on next, passes here).
• tests/treeshake.test.ts green; verdict.ts is shed from the core floor bundle, so the floor is unchanged.
• Brotli caps (scripts/size, measured against a pristine a8a89497 build): + isPending/latest 11,258 -> 11,253 B (-5 B); every other scenario byte-identical to the baseline (core floor 8,849; simple-app 11,563; hydrating 18,991 / 28,768; CSR 14,429; observe 15,807 / 27,335; frames 11,370). No cap ratcheted. Note: signals: + createStore is pre-existing red on next by 9 B (15,809 vs the 15,800 cap, from the store changes in perf(signals,web,universal,html): merge/omit are always lazy views; consumers read the leaves #3454/test(signals): store oracle + four signal states; latest() seed leak and body-end visibility fixes #3455) and is unchanged by this PR (+0 B).
• Spec: A10's mechanism note now states the still-computing carve-out and the unstamped case; verdict.ts cites A10 at the site; rules index regenerated (--check clean; the one unresolved test id is pre-existing on next). Changeset included.

🤖 Generated with Claude Code

[changeset-bot]

🦋 Changeset detected

Latest commit: 6861fa5

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

This PR includes changesets to release 11 packages

Name	Type
@solidjs/signals	Patch
test-integration	Patch
@solidjs/web	Patch
@solidjs/babel-plugin	Patch
@solidjs/compiler	Patch
@solidjs/diagnostics	Patch
@solidjs/element	Patch
@solidjs/h	Patch
@solidjs/html	Patch
solid-js	Patch
@solidjs/universal	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]

Merging this PR will not alter performance

✅ 172 untouched benchmarks

---

_{Comparing brenelz:fix/ispending-memo-unstamped-hold-3457 (6861fa5) with next (5f7da9d)}