Repository navigation
Conversation
…s#3726) A derived store whose first run returned a pending promise and whose rerun landed synchronously kept STATUS_UNINITIALIZED on the projection computed until the flush commit, because recompute only clears the flag on a creation pass. The recompute-side settle walk (solidjs#3181) requires the node to be initialized, so it skipped the projection, and readers subscribed to a store node the landing left unchanged (`"length" in store` on an array seed, `Object.keys`, an unchanged `length`) had no value notification to fall back on. They stayed pending, blank inside a loading boundary, while a sibling read of the source updated. The projection's sync commit through the setter now retires the flag the way asyncWrite does after its setter landing, so the walk releases those readers in the same flush. The retirement is gated on the previous run having been pending, which leaves a born-held creation pass unchanged.
The projection computed retires STATUS_UNINITIALIZED after a synchronous landing only when the pass is mainline or its transaction is parked by nothing but the node's own flight, so a landing staged by a live action or beside another pending source keeps the seed invisible until that transaction commits. The gate is keyed on the node's own pending-source entry instead of any STATUS_PENDING, so a first run parked on an upstream async source no longer trips it. The tests assert in the flush that lands the sync value, register a length reader, and pin the transaction-held case.
🦋 Changeset detectedLatest commit: 37ac480 The changes in this PR will be included in the next version bump. This PR includes changesets to release 12 packages
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 |
Merging this PR will not alter performance
Comparing 
Footnotes
|
|
…ked readers (supersedes #3732) (#3789) * test(signals): pin #3726 on L2 — derived store sync landing wakes parked readers Port of #3732 (brenelz) onto the hold model. The PR patched `store/next/projection.ts` to retire STATUS_UNINITIALIZED after a synchronous landing so core's #3181 settle walk would release readers parked on the superseded first flight: readers whose store node the landing left unchanged (`"length" in store`, `Object.keys`, an unchanged `length`) had no value notification to fall back on and stayed blank inside a loading boundary while a sibling read of the source updated. On L2 the core already holds, with no store-side patch: a reader of a derive with a flight up is parked on the derive itself (`pullFamily` links it, as a reader of a memo with a flight), and the derive's first commit (uninitialized → a value) is a value change for every subscriber (`recompute`: `wasUninitialized` → `insertSubs`) — the walk's uninitialized exemption is moot. The PR's three tests pass unchanged; its superseded source change is dropped (`src/store/next/` is gone). Added: the report's sequence verbatim (the source write and the superseded promise's resolution in one timer callback, under a loading boundary, microtask-driven) in signals and as a `<Loading>` web spec; a verdict probe parked on the first flight; and the case the PR listed as open — a landing a transaction holds: the parked readers re-run as the transaction's work and reveal at its commit, under a boundary and out (red on the PR's base with and without its fix; green on L2). Tests only — no changeset. RULES-INDEX regenerated (A19/A25/A29 gain the test's citations). Co-authored-by: Brenley Dueck <brenleydueck@gmail.com> Co-authored-by: Cursor <cursoragent@cursor.com> * test(signals): pin GabbeV's board shape from #3726 on L2 The #3726 comment's playground, fetched: an async card source behind a memo (`createMemo(() => ready() ?? pending)`), a derived `createStore(() => source(), [])`, `createOptimisticStore` over it and a keyed `<For>`, remounted under a `<Loading>` that has already revealed; a count outside the view. Before, Activity → Board left "All cards: 1" beside an empty lane: the remounted readers were parked on the derive's first flight under the boundary's hold, and the held synchronous landing never woke them — the case PR #3732 listed as open (red on its base with and without its fix). Same root-cause family as the report, not a distinct bug. On L2 the revealed boundary holds the view switch (A29: a boundary already showing content holds like any reader) and the lane reveals with its rows as one frame — the count and the list never disagree. Pinned as the playground itself (web spec) and a signals reduction. The report's own playground, fetched too: the `Source:` hole is outside the `<Loading>` and the promise is created once — the first web test and the signals "report's playground" test now follow it exactly. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Brenley Dueck <brenleydueck@gmail.com> Co-authored-by: Cursor <cursoragent@cursor.com>
|
2 participants
Fixes #3726
What was broken
A derived
createStorewhose first run returned a pending promise, and whose rerun after a source write returned a value synchronously, left some readers pending forever. A reader whose store node the landing did not change ("length" in store,Object.keys, an unchangedlength) rendered blank inside ``, while a sibling read of the source updated.Root cause
The projection computed kept
STATUS_UNINITIALIZEDuntil the flush commit, becauserecomputeclears it only on a creation pass. The settle walk that releases readers parked on a superseded flight skips uninitialized nodes, and unchanged-node readers get no value notification to fall back on.Change
After the synchronous commit,
runProjectionComputedNextclearsSTATUS_UNINITIALIZEDwhen the node's own flight was pending and no transaction holds the landing, the same way the async landing path does after its setter commit. A landing held by a transaction (a live action, a blocked transition, or another pending source) keeps the flag, so the seed stays invisible until that transaction commits.Verification
cd packages/signals && npx vitest run tests/store/derived-presence-async-3726.test.ts: 3 passed. Two of the cases fail onnextwithout the fix.packages/signals: 4794 passed.packages/solid: 817 passed.packages/web: default, hydrate and server suites pass, except one server spec that also fails onnext.Open
recompute, which would cover every computed type.nexttoo.