[ryansolid]

Requested by Ryan · project thread

Summary

This PR fixes two places where releasing an optimistic value could show a frame that is torn or can't be reconciled.

1. @solidjs/web: a sweep's slot re-emissions join the sweep's unit

A live server component could send a torn write. Each arg binding of a slot occurrence re-sends the whole slot record as it sweeps. frame-sink.ts slot() called put directly, so the first copy went out in its own message. That copy carried the other args' previous values, and it went out before the message carrying the holes from the same commit.

In the chat example, one commit went out as slot:composer { seen: <new>, busy: false } on its own, then slot:composer { …, busy: true } + hole:lh:2. The client applies each message in its own flush, so the pending bubble left the outbox before the thread showed the server's row. In Chromium the two landed 0.4 to 7 ms apart, sometimes in separate tasks.

Inside a sweep, slot() now writes into the sweep's unit (swept.r), as hole() and attr records already do (C13: one sweep, one frame). Outside a sweep, nothing changes.

2. @solidjs/signals: a shown keyed arrangement guess folds a correction instead of voiding it

Two overlapping actions push onto one keyed optimistic store, and the stream echoes the first write while the second is open. The correction dissolved the lane. For that window the store read torn: a hole at the first row's index, length taken from the guess, and a raw StoreTarget served for a row. At the commit, the confirmed row remounted twice. In the board example this destroyed an open title editor and its draft.

A keyed arrangement guess is rows by key. So when a correction differs from a guess the screen has shown, it now folds into the hold beneath the guess. RULES-DERIVED D-34 is amended to say so, and F-11 still holds: the screen keeps what it showed until the hold commits.

• The guess stays the lane's value and the screen's until the commit. At the commit, rows match by key.
• A guess the screen never showed is void at the correction, as D-34 says for any guess.
• Deriving passes outside the lane read the hold's world (each guess's truth, else what it covered) and join the hold, as for any held truth (2.0.0-rc until doesnt entangle the update resolving it with the action update #3164). latest() shows that world.
• until and resolve keep their D-31 postures. In laneRead, the fold's arm comes before the authoritative arm on purpose. It serves an until outside the lane the same truth _truth would, and also joins it to the hold, which is its posture over a foreign hold. With the arms reversed, three pinned D-31 posture tests fail, and a new test in optimistic-keyed-fold.test.ts pins the order.
• Keys born under a folded guess are held with it.
• A folded lane's list pass in a frame that already has a transaction reads that frame's stagings as the screen, as a verdict reader does. Without this, a list pass could pair the truth's length with the guess's rows and hand a mapper undefined. The check is gated on the lane engine, so the floors don't carry it.

The fold lives in store/optimistic.ts. The lane engine reaches it through the existing _laneRebase hook, which now takes an older flag and handles both the re-base and the fold. A keyed arrangement guess is recognized by its view's removed-keys set, which the setter now always gives a keyed array's view. A folded lane carries the store's reader of it as Transaction.folded. That property is deliberately not _-prefixed, because a new mangled name shifts every later one in the dist.

No existing test changed.

Size

Size-Exception: the maintainer accepted the keyed-fold size cost (2026-10-22). After merging next @ db8241c (#4213 re-recorded the floors), the live server components frozen floor carries the lane engine's fold hooks (+110 B, as + isPending/latest).

The other floors are +0 B minified. Following review, the fold was trimmed from the first version: + isPending/latest went from +142 to +110 B, and the all-stores app from +694 to +569 B. Every cap below was re-measured over next @ db8241c, has a ledger entry, and was accepted by the maintainer:

scenario	minified vs next	new cap
+ createStore	+44 B	15.04 KB
+ isPending/latest	+110 B	10.30 KB
hydrating + every store primitive family (carries the fold itself)	+569 B	30.66 KB
compiled hydrating (todo app, store)	+44 B	32.23 KB
page: live server components (frozen floor, floor-caps.json)	+125 B	37.64 KB
page: compiled live server components	+110 B	41.47 KB
page: live + router	+110 B	47.01 KB

CSR is +0 B minified and compiled CSR is +44 B; both stay under next's caps. All numbers were measured locally on Linux; earlier rounds matched CI exactly.

How did you test this change?

• test/server/frame-sweep-ops.spec.tsx: one commit produces exactly one batch, [slot, hole], with busy: true. It fails on next.
• tests/optimistic-keyed-fold.test.ts has 9 cases:
  • overlapping pushes, flat and nested;
  • another tab's row landing mid-write (it folds until settle and keeps identity);
  • overlap plus an unrelated change;
  • the second push timing out;
  • a mainline until over a folded lane, which delivers at the commit (D-31);
  • a keyed guess the screen never showed, which is void at the correction;
  • a truth longer than a reorder guess, in the action and on mainline.
• Test runs:
  • signals: 5460 passed after merging next.
  • solid: 882/882.
  • web client: 1407/1407.
  • web hydrate: 520 passed.
  • web server: 1763 passed and 1 failed. The failure is the server-functions-adapter-request "rebuild failure" test, which fails only on Node 22.22 here.
• The size gate passes locally.
• CodSpeed flagged "memo + sync render effect only" and "props chain depth 1". Neither reproduces: Callgrind on the built packages counts fewer instructions on this branch than on next for both. CodSpeed also warns that base and head ran on different runner types.
• Chromium:
  • Chat, 5 of 5 runs: one wire write.
  • Board: overlapping creates, the server applying writes after 300 ms, and O1's title editor open with a draft. Before this change, O1's <li> went #5 → #7 → #8 and focus ended on <body>. Now it keeps its identity and the editor keeps focus and its draft.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_015SnZadCVdDSeDzxBUhcMU9

[changeset-bot]

🦋 Changeset detected

Latest commit: 3cd7d4e

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

This PR includes changesets to release 14 packages

Name	Type
@solidjs/signals	Patch
@solidjs/web	Patch
test-integration	Patch
@solidjs/babel-plugin	Patch
@solidjs/diagnostics	Patch
@solidjs/element	Patch
@solidjs/h	Patch
@solidjs/html	Patch
board-example	Patch
server-dbmon-example	Patch
todos-server-example	Patch
@solidjs/compiler	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

[github-actions]

Size (brotli, eager entry chunk)

scenario	head	vs base	minified vs base	minified vs recorded	cap		lazy chunks (not counted)
signals: core floor (createSignal/Memo/Effect/Root/flush)	7.42 KB	−8 B (−0.1%)	0 B	0 B	7.44 KB	✅
signals: + createStore	15.03 KB	+33 B (+0.2%)	+44 B	0 B	15.04 KB	✅
signals: + isPending/latest	10.28 KB	+45 B (+0.4%)	+110 B	0 B	10.30 KB	✅
signals: + until/refresh (truth readers, no optimism)	8.04 KB	0 B	0 B	0 B	8.05 KB	✅
app: render + one signal (the simple-app floor)	9.73 KB	−10 B (−0.1%)	0 B	0 B	9.75 KB	✅
app: hydrating (no stores) with Show/For/Loading/Errored/lazy	18.37 KB	+2 B (+0.0%)	0 B	0 B	18.38 KB	✅	lazy-page.js 0.04 KB
app: hydrating + every store primitive family	30.64 KB	+239 B (+0.8%)	+569 B	0 B	30.66 KB	✅	lazy-page.js 0.04 KB
app: CSR with Show/For/Loading/Errored/lazy	13.32 KB	+4 B (+0.0%)	0 B	0 B	13.33 KB	✅	lazy-page.js 0.04 KB
app: CSR, observe tier (same app on the observe artifacts)	15.06 KB	0 B	0 B	0 B	15.08 KB	✅	lazy-page.js 0.04 KB
app: CSR, observe tier + attribution engine enabled	29.62 KB	0 B	0 B	0 B	29.64 KB	✅	lazy-page.js 0.04 KB
app: compiled floor (one template, one text hole, one delegated click)	9.92 KB	−2 B (−0.0%)	0 B	0 B	9.93 KB	✅
app: compiled CSR (JSX todo app: spread/merge/omit, events, class/style, keyed For, Show, Loading + lazy, store)	26.22 KB	+2 B (+0.0%)	+44 B	0 B	26.23 KB	✅	stats.js 0.18 KB
app: compiled hydrating (the same JSX todo app through hydrate(), compiled hydratable)	32.22 KB	+41 B (+0.1%)	+44 B	0 B	32.23 KB	✅	stats.js 0.20 KB
frames: eager client consumer (frames client + transport, lazy codec)	11.75 KB	0 B	0 B	0 B	11.76 KB	✅
page: base server components (hydrating + dynamic + frames + sf reference)	33.22 KB	−35 B (−0.1%)	+8 B	+8 B	33.27 KB	✅	assets.js 0.78 KB, decode.js 6.25 KB, lazy-page.js 0.04 KB, regions.js 1.06 KB, trace.js 9.10 KB, wire.js 2.78 KB
page: live server components (base + live/GET + action + isPending/latest)	37.63 KB	+115 B (+0.3%)	+125 B	0 B	37.64 KB	✅	assets.js 0.78 KB, decode.js 6.25 KB, lazy-page.js 0.04 KB, regions.js 1.06 KB, trace.js 9.09 KB, wire.js 2.78 KB
page: compiled base server components (the base page as JSX: templates with class/style/attributes/events, For/Show; no spread)	35.38 KB	+113 B (+0.3%)	+8 B	+8 B	35.30 KB	⚠️ over by 84 B, 12 B minified headroom	assets.js 0.78 KB, decode.js 6.25 KB, regions.js 1.06 KB, sc-comments.js 0.20 KB, trace.js 9.09 KB, wire.js 2.77 KB
page: compiled live server components (the compiled base page + live/GET + action + isPending/latest)	41.45 KB	+49 B (+0.1%)	+110 B	0 B	41.47 KB	✅	eager (counted): web.js 22.05 KB; assets.js 0.78 KB, decode.js 6.25 KB, regions.js 1.06 KB, sc-comments.js 0.19 KB, trace.js 9.13 KB, wire.js 2.77 KB
page: base + router (base page + @solidjs/router: createRouter, two routes, preload, useNavigate)	40.74 KB	−27 B (−0.1%)	0 B	0 B	40.78 KB	✅	assets.js 0.78 KB, decode.js 6.25 KB, lazy-page.js 0.04 KB, regions.js 1.06 KB, server.js 1.02 KB, serverForms.js 3.58 KB, trace.js 9.14 KB, wire.js 2.77 KB
page: live + router (live page + @solidjs/router: createRouter, two routes, preload, useNavigate)	46.99 KB	+51 B (+0.1%)	+110 B	0 B	47.01 KB	✅	eager (counted): client.js 27.17 KB; assets.js 0.78 KB, decode.js 6.25 KB, lazy-page.js 0.04 KB, regions.js 1.07 KB, server.js 1.02 KB, serverForms.js 3.32 KB, trace.js 9.14 KB, wire.js 2.80 KB
server: floor (getRequestEvent + isServer)	1.33 KB	0 B	0 B	0 B	1.34 KB	✅
server: renderToString (the server-render floor)	20.63 KB	0 B	0 B	0 B	20.65 KB	✅

⚠️ Over the brotli cap within the minified allowance (passes)

• page: compiled base server components (the base page as JSX: templates with class/style/attributes/events, For/Show; no spread): over brotli cap by 84 B; minified 107,884 B vs 107,876 B recorded with the cap (+8 B) — 12 B of the 20 B minified allowance left; +8 B minified over this PR's base

_{Bundled with Rolldown (what Vite ships), brotli q11, decimal KB. A scenario fails only when it is over its brotli cap and its minified size is more than 20 B over the minified recorded with the cap; over the cap within that allowance is brotli layout noise and passes with a warning. Caps and their recorded minified in scripts/size/scenarios.js; the floor and page caps in floor-caps.json are frozen (lower only, or Size-Exception: in the PR body). npm run ratchet lowers caps per RC; it never raises one (scripts/size/README.md).}

[coveralls]

Coverage Report for CI Build 39704959958

Warning

Build has drifted: This PR's base is out of sync with its target branch, so coverage data may include unrelated changes.
Quick fix: rebase this PR. Learn more →

Coverage remained the same at 77.662%

Details

• Coverage remained the same as the base build.
• Patch coverage: No coverable lines changed in this PR.
• No coverage regressions found.

Uncovered Changes

No uncovered changes found.

Coverage Regressions

No coverage regressions found.

---

Coverage Stats

Relevant Lines:	1349
Covered Lines:	1105
Line Coverage:	81.91%
Relevant Branches:	1055
Covered Branches:	762
Branch Coverage:	72.23%
Branches in Coverage %:	Yes
Coverage Strength:	27.53 hits per line

---

💛 - Coveralls

[codspeed]

Merging this PR will degrade performance by 24.38%

2 warnings

⚠️ 2 benchmarks are dominated by system calls

System calls cannot be simulated consistently and are left out of the measure. The reported value understates the real cost. Switch to the Walltime instrument to measure them.

Learn more about system calls

See affected benchmarks

⚠️ Base and head ran on different runner types

Some benchmarks with significant performance changes were compared across different runner types, so the changes might not come from the code. Run base and head on the same runner type, or use CodSpeed Macro Runners for more stable results.

Learn more about Macro Runners

Open the report in CodSpeed to investigate

⚡ 1 improved benchmark
❌ 1 regressed benchmark
✅ 194 untouched benchmarks
⏩ 5 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	build + consume	289.3 µs	596.6 µs	-51.5%
⚡	memo + sync render effect only (reference)	33.9 ms	28.7 ms	+17.9%

Tip

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

---

_{Comparing claude/project-thread-f38ibp (3cd7d4e) with next (db8241c)}

Footnotes

1. 5 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[ryansolid]

Review of 98f31fb. Merge after changes. Part 1 (frame-sink slot() joining the sweep's unit) is correct and safe for broadcast retain and replay. Part 2 fixes a real bug: 5 of the 7 new signals tests and the new frame-sweep-ops case fail on next. Suites: signals 5446 passed; web server 1763 passed, with only the known Node 22 server-functions-adapter-request failure.

1. Size. The gate fails on 10 scenarios. Minified: isPending/latest +420 B (138 B over its brotli cap), hydrating + every store primitive +704 B, live server components +499 B, createStore +138 B. The compiled floor is 3 B over its brotli cap (+10 B minified). Either cut the lane code or get the maintainer's OK on the raises, with Size-Exception: for the frozen floors.
2. Scope of the read change. core.ts:1814-1818 makes every lane pass read the screen once the frame has a transaction. That covers createOptimistic and non-keyed stores too, and it adds the +10 B on every floor. D-34 describes it as a list-pass rule. Gating it on passLane !== null && passLane._folded keeps all 7 new tests and the full suite passing. With the change removed, only the two "truth longer than reorder" tests fail.
3. until order. In lanes.ts:584-587, foldedRead runs before the CONFIG_AUTHORITATIVE check, so until readers join the hold and read covered() instead of _truth. Swapping the order also passes everything, so no test pins D-31 here. Either exclude authoritative reads or add a test that pins the intended behavior.
4. Unshown guesses. The keyed branch of laneWrite (lanes.ts:768-787) never checks l._shown, but D-34's title says a guess the screen never showed is void. Handle that case, or amend D-34 to say it isn't voided.
5. Minor. stale() is evaluated twice (lanes.ts:768-769), and lanes.ts reaches into (el as any)._host?.fam?.node. A store hook would be cleaner.

#4210 and #4213 merge cleanly together, and the signals suite passes on the merged tree.

— Claude via Claude Code

---

Generated by Claude Code

[ryansolid]

CodSpeed on f143b9c flags "props chain depth 1: build + consume" at −51%. This PR doesn't change anything that benchmark runs: it only calls merge/omit, and the diff never touches them. It also doesn't reproduce. I ran callgrind on the built dist/prod with the same chain run 20,000 times (the bench's chain(1) + consume), and this branch counts fewer instructions than next (2.968B vs 3.020B). CodSpeed also warns that base and head ran on different runner types.

Earlier flags on this PR show the same pattern:

• "memo + sync render effect only (reference)": 2.91B on this branch vs 3.08B on next over 12 mounts of 4000 rows.
• "projection derive: write one NESTED field": gone on the next head.

I don't have the means to re-run the CodSpeed check from here.

---

Generated by Claude Code