perf: optimize signals hot paths

[brenelz]

Summary

Four targeted optimizations to @solidjs/signals hot paths, driven by the CodSpeed flamegraphs from run 6a2a5421 plus local --cpu-prof profiles of the same workloads.

1. O(1) dependency revalidation (graph.ts, types.ts, core.ts)

When a computation re-reads a dependency it already saw earlier in the same recompute pass non-consecutively (e.g. doc.meta.author then doc.meta.timestamp — doc.meta read twice), link() fell back to isValidLink, which scans the dep list from the head to check membership. With large dep lists this is O(n²): it was 51% of total time in the reconcile: deep tree, all ~12k paths subscribed bench.

Links now carry a _gen stamp of the subscriber's _depGen recompute counter, set on creation and on in-order reuse. The membership question "was this dep already revalidated this pass?" becomes one integer compare. Equivalence: a link is inside the [head.._depsTail] prefix iff it was created or (B)-reused during the current pass, which is exactly what the stamp records.

2. Skip redundant subscriber walks on batched writes (core.ts, heap.ts, effect.ts)

Writing the same signal N times in one batch walked the full subscriber list N times — after the first walk, every subscriber is already queued, so the remaining walks are provable no-ops (the updateSignals:update1to1000 flamegraph is ~95% insertSubs + insertIntoHeap).

setSignal now skips the walk while a global notifyEpoch is unchanged. The epoch advances whenever a queued notification is consumed — any heap removal (deleteFromHeap, which all recompute paths go through) or a tracked-effect run draining _modified — and a signal's cached epoch is invalidated when it gains a new subscriber (link()) and on optimistic writes, which mutate subscriber lane state and always walk.

3. Reconcile allocation reduction (reconcile.ts)

• getAllKeys allocated [...keys, ...nextKeys] + a Set + Array.from for every tracked node in the tree; it now returns the existing keys array when key sets match (the common shape), and merges without the spread copies otherwise.
• unwrap did $TARGET symbol lookups on primitive leaves, forcing boxing; primitives now return immediately.

4. Small allocation cleanups (store.ts, utils.ts, core.ts)

• getKeys only wraps Object.keys in an untrack closure when the source is itself a wrapped store (plain objects can't trigger traps).
• snapshotImpl's no-override walk reads each property once via the descriptor instead of three times.
• Effects reuse one bound runner instead of allocating _runEffect.bind(null, el) per update (runEffect already no-ops on stale _modified, so re-enqueueing the same function is safe).

Results (local, dev build, same workloads as the CodSpeed suite)

Benchmark	Before	After	Δ
reconcile: all ~12k paths subscribed	27.5 ms	5.6 ms	4.9x
reconcile: single deep() effect	13.4 ms	5.6 ms	2.4x
updateSignals:update1to1000	1.44 ms	0.06 ms	24x
reconcile: 10 sparse paths	0.43 ms	0.38 ms	~12%

Expected CodSpeed movement: update1to1000 (183.8 ms), reconcile all paths (84.2 ms), and reconcile deep() (76.6 ms) should drop substantially. The DOM-lane benches won't move much — their flamegraphs are 75–95% jsdom internals.

Trade-offs

• +1 field per link (_gen), +2 fields per node (_depGen, _notifyEpoch): ~8–16 bytes each, may show as a sub-percent tick on the creation benches.
• notifyEpoch is a monotonically increasing double — no wraparound concern in practice.

Testing

• 749 solid-signals + 293 solid-web + 415 solid tests pass
• tsc -p tsconfig.build.json clean
• Verified the epoch-skip invalidation paths: heap pops, tracked-effect runs, sync pull-based recomputes, suppressComputedRecompute, dispose, and mid-batch new subscribers

🤖 Generated with Claude Code

[changeset-bot]

🦋 Changeset detected

Latest commit: 91d1fd9

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

This PR includes changesets to release 9 packages

Name	Type
@solidjs/signals	Patch
test-integration	Patch
solid-js	Patch
babel-preset-solid	Patch
@solidjs/web	Patch
@solidjs/html	Patch
@solidjs/h	Patch
@solidjs/universal	Patch
@solidjs/element	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-hq]

Merging this PR will improve performance by 76.5%

⚠️ 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

⚡ 7 improved benchmarks
❌ 4 regressed benchmarks
✅ 107 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	createComputations:create1to1	81.5 ms	97.4 ms	-16.33%
❌	read	25.3 µs	28.2 µs	-10.4%
❌	read	32 µs	34.9 µs	-8.27%
❌	construct	56.7 µs	60.5 µs	-6.25%
⚡	updateSignals:update1to1000	183.8 ms	1.9 ms	×97
⚡	propagation:diamond	2.7 ms	1.2 ms	×2.3
⚡	propagation:avoidable	2 ms	1.2 ms	+70.03%
⚡	reconcile: deep tree, single deep() effect	76.6 ms	54.6 ms	+40.44%
⚡	reconcile: deep tree, all ~12k paths subscribed	84.2 ms	65.6 ms	+28.39%
⚡	hasAllowed	26.5 µs	24.4 µs	+8.76%
⚡	ownKeys	163.8 µs	152 µs	+7.75%

Tip

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

---

_{Comparing brenelz:perf/signals-hot-paths (91d1fd9) with next (b26bc04)}

[brenelz]

Used the new Fable 5 model to test so see if it had any perf suggestions and it came up with these.

I more or less wanted to see how codspeed would show up for these perf improvements.

Feel free to take any of them or discard them