fix(signals): a source pending on a superseding re-ask keeps its transaction blocked (#3462)

[brenelz]

Fixes #3462.

The gap

const details = createMemo(() => delay(2000, a()));
const selected = createMemo(() => delay(2000, a() + b() + details()));
<p>Show: {String(show())}</p>
<p>Panel: {show() ? selected() : "hidden"}</p>
// click: setB(1); 500ms; setShow(true); 500ms; setA(1); 500ms; setShow(true)

setB(1) re-asks selected while nothing observes it, so the flight is in the air with no hold. setShow(true) flips the conditional onto the pending selected: the reveal opens a transaction holding show = true, blocked on selected (INV-3). setA(1) re-asks details; selected is now pending on that flight instead of its own. Then the second setShow(true): on next, Show: true publishes at once beside Panel: hidden, and Panel says 3 only when the whole chain lands seconds later. Without the repeated write, both publish together at the landing.

Traced: transitionComplete judged a reporter's source by its own flight alone, the self entry in _pendingSources (#3375 chose that over _error.source, which a later-pending input overwrites). The details re-ask superseded selected's own flight and retired the self entry, leaving selected pending on details; the verdict flipped to complete while selected's reader still could not render. The transaction was parked, so nothing re-judged it. The repeated same-value write re-entered it through setSignal's initTransition(el._transition) (which runs before the equality check), the flush trusted the verdict and committed show = true, and the panel effect, replayed against a selected whose inputs had been published beneath its first flight (CONFIG_INPUTS_PUBLISHED), refused the carve-out and parked on the chain with no transaction of its own.

Fix

A registered source blocks while it is still pending on anything, not only on its own flight:

if (sourceObserved(transition, source) && source._x?._pendingSources?.size) {

This is the verdict A15 already implies (async work observed by a reader settles as one unit with the writes that asked it) and matches what the landing does anyway: enterWaiting folds every transaction with a live reporter on the source into the landing's transaction. The only state the change touches is a source whose own flight a superseding re-ask retired; the #3375 boundary-reset case is decided earlier by reporterBlocksSource's fallback prune and still passes. The reporter set is unchanged and nothing entangles earlier than before; the transaction simply stays parked until the chain lands.

before: Show: false | Panel: hidden -> Show: true | Panel: hidden -> Show: true | Panel: 3
after:  Show: false | Panel: hidden -> Show: true | Panel: 3

Known edge not addressed: a parked transaction whose blocking source later settles synchronously (its re-run after the upstream landing returns a plain value, so there is no landing of its own to fold the waiter in) has no wake, before and after this change; the fold at a landing is the only re-entry the reporter path has.

Verification

• @solidjs/signals: 180 files / 2038 tests (2037 green, 1 skipped as on next; the timing-only heap-mark-incremental "linear in N" case failed once in the full run and passes in isolation), including the new tests/superseded-source-blocks-3462.test.ts (the issue shape on deferred gates, plus the no-repeat control pinned to the same two frames).
• @solidjs/web: 88 files / 793 tests green, including the new test/superseded-source-blocks-3462.spec.tsx, a jsdom port of the playground (real click, DOM sampled through the hold, delays scaled to 100ms) with the same control; the issue shape fails on next, passes here.
• In-package minified core floor 23,752 -> 23,750 (-2 B). Brotli caps measured against a pristine next (a8a8949) build: core floor 8,849 -> 8,851 (+2 B), which tips a cap that sat 1 B under the baseline, so it is ratcheted 8.85 -> 8.90 KB with the measurement in the note; + createStore is pre-existing red on next by 9 B (15,809 vs 15,800, from 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 reads 15,770 here; every other scenario is within noise of the baseline and under its cap.
• Docs: INTERNALS-ASYNC-STATE.md's transitionComplete bullet restated (judged by the set, not the self entry); rules index regenerated (--check clean). Changeset included.

🤖 Generated with Claude Code

[changeset-bot]

🦋 Changeset detected

Latest commit: 2a3ae8c

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/superseded-source-blocks-3462 (2a3ae8c) with next (0da94f9)}