perf(signals): narrow store writes clone by spread; overlay only for wide owned backings (#3360 part 1)

[ryansolid]

Part one of #3360 (does not close it — part two, below, does).

Problem

#3360 reports a ~4× regression rc.0 → rc.7 on the per-write floor of stores: 100k one-key stores, one setStore each, one flush. Reproduced (prod): rc.0 ~61 ms, rc.7 ~265 ms. Both linear — it's a constant factor from the rc.1 immutable-backing model. Profiled three costs:

1. cloneRaw was a descriptor clone (getOwnPropertyDescriptors + Object.create(proto, descs)): ~530 ns for a one-key object. A spread is ~10 ns and produces the same result for plain-data containers.
2. The 2.0.0 | slow and steady wins the race #3044 prototype overlay loses on narrow and on unowned containers. Object.create(v) turns the committed backing into a V8 prototype (~200 ns), and every later flatten writes into a prototype object (~250 ns/key + validity-cell churn). On an unowned backing (every fresh store's first write) the flatten had to privatize-clone anyway, so that path paid create + descriptor clone + per-key descriptor copy — the reporter's 128 ms first flush.
3. Every pending backing is registered in two weak collections (~190 ns of identity-hash + ephemeron cost per write). Not addressed here — see part two.

Change

• scanAccessorsOnce grades the container (sc: 0 unscanned / 1 scanned / 2 plain data = Object.prototype with every own key an enumerable data property) and records the own-key count kc (bumped by set-trap writes of keys new to the container, so a {} that grows still graduates).
• cloneRaw spreads at grade 2.
• Overlay only when kc > 32 and the backing is owned; otherwise clone and swap. OVERLAY_MIN_KEYS sits at the measured crossover.
• Grade-2 containers flatten and take overlay first-writes by bare assignment; a non-plain defineProperty through the draft downgrades the grade.

Numbers (prod)

	rc.0	rc.7 / next	this PR
steady state, 1000 stores × 1000 write+flush rounds	214 ns/write	629 ns	340 ns
reporter's 100k fresh stores, one write + flush	61 ms	~265 ms	~90 ms

Existing store benches all improved or held: batched 2-key writes 5.6k → 10.2k ops/s, selection map 3.2k → 3.9k, input burst 3.2k → 4.2k, projection root/nested (#3352) 393k/328k → 465k/459k, commit-boundary flat. New write-floor.bench.ts pins the steady and fresh-store shapes.

Behavior note

A writable: false data property defined through a draft used to stay non-writable across later writes only because the overlay never touched it; on the clone path the descriptor walk normalizes it to writable (R51). Enumerability is preserved either way; the test pins that and documents the normalization.

Tests / gates

• tests/store/write-floor.test.ts (9): symbol keys, non-enumerable keys, custom prototypes, frozen sources, non-plain and accessor defineProperty through the draft, growth past the threshold, 500-store batch.
• Two #3352 clone-spy assertions loosened to ≤ 1 (spread is not a descriptor clone).
• signals 1707/1707; prod-tier failure count identical to next (238); solid 595; web 734.
• Size: createStore 14.77 → 14.92 KB, hydrating+stores 27.39 → 27.52 KB (~150 B of real code; caps bumped with notes).

Part two

The remaining ~190 ns is storeNextLookup.set(pb) + ownedRaw.add(pb) per draft. Removing them (incorrectly, to measure) gives 154 ns/write — under rc.0. The plan is to stamp owned raws with a non-enumerable owner symbol at creation and keep the WeakMap only for user-supplied objects; owned raws are never reachable by user code (snapshot copies them, there is no unwrap), so the surface is purely internal. Separate PR so each half is bisectable and CodSpeed shows its effect alone.

Made with Cursor

[changeset-bot]

🦋 Changeset detected

Latest commit: ffcd59e

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

[coveralls]

Coverage Report for CI Build 34572005825

Coverage remained the same at 71.842%

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:	1007
Covered Lines:	772
Line Coverage:	76.66%
Relevant Branches:	790
Covered Branches:	519
Branch Coverage:	65.7%
Branches in Coverage %:	Yes
Coverage Strength:	15.01 hits per line

---

💛 - Coveralls

[codspeed]

Merging this PR will regress 1 benchmark

⚠️ 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

⚡ 6 improved benchmarks
❌ 1 regressed benchmark
✅ 151 untouched benchmarks
🆕 2 new benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	createStore setter: delete + set one root key (#3044 overlay)	155.7 µs	440.8 µs	-64.67%
⚡	batched: same writes, single commit	3.3 ms	1.4 ms	×2.3
⚡	commit boundary: flush after every 2-key setter, no subscribers (#3044)	14.5 ms	10.4 ms	+39.36%
⚡	projection derive: write one NESTED field (reference)	262.3 µs	224 µs	+17.1%
⚡	style store object, replaced × 500	48.6 ms	44.8 ms	+8.54%
⚡	style store object, mutated in place × 500	42.8 ms	39.8 ms	+7.42%
⚡	class store object, replaced × 500	45.9 ms	43.1 ms	+6.69%
🆕	fresh stores: create + first write + first commit (reporter shape)	N/A	22.3 ms	N/A
🆕	steady state: owned backings (#3360)	N/A	11.1 ms	N/A

Tip

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

---

_{Comparing perf/store-write-floor (ffcd59e) with next (27aee36)}