Skip unchanged {{#each}} item subtrees during updates

[NullVoxPopuli-ai-agent]

Re-cut of #21512 (closed during the #21520 spike consolidation) onto current main, now with the full-suite verification that was outstanding: 9443 tests, 0 failures locally.

During revalidation, each {{#each}} item collects the tags its subtree consumed (via a tracking-frame finalizer on the item's try-frame); on later passes, an item whose collected tag validates clean is skipped entirely instead of walking every opcode of its subtree. A triviality gate (≤2 opcodes and no nested block) keeps the bookkeeping off items too small to profit. Measured standalone at ~1.9x on an 8x-throttled dbmon (walk-heavy, many clean items per frame); no observable behavior change — the same subtrees re-render for the same reasons, cheaper.

This was the campaign's first lever and its most reviewed artifact; it needs no RFC, no flag, and composes with (but does not require) the scheduler work in #21493/#21520.

🤖 Generated with Claude Code

[NullVoxPopuli-ai-agent]

Isolated same-batch measurement (today's main vs this branch, 8x throttle, mutation-anchored, zero render failures) — confirmed strong win, no regressions:

bench	main	this PR
DB Monitor	7.3 fps (7–11)	17.1 fps (13–18) — 2.3x, spreads fully separated
1k items, 1 update each (seq, async)	1823ms	756ms (2.4x)
1k items, 25% random (async)	577ms	317ms (1.8x)
1k items, 5% random (async)	187ms	122ms
everything else	—	neutral (±5%, overlapping spreads)

🤖 Generated with Claude Code

[NullVoxPopuli]

By benches on my machine:

pnpm bench 1

duration phase no difference [-21ms to 23ms]
renderEnd phase no difference [0ms to 0ms]
render1000Items1End phase no difference [-1ms to 1ms]
clearItems1End phase no difference [0ms to 2ms]
render1000Items2End phase estimated improvement -1ms [-3ms to 0ms] OR -2.57% [-6.4% to -0.25%]
clearItems2End phase no difference [0ms to 0ms]
render10000Items1End phase no difference [-6ms to 3ms]
clearManyItems1End phase no difference [-1ms to 1ms]
render10000Items2End phase no difference [-3ms to 10ms]
clearManyItems2End phase no difference [-2ms to 1ms]
render1000Items3End phase no difference [0ms to 1ms]
append1000Items1End phase no difference [0ms to 6ms]
append1000Items2End phase no difference [-2ms to 2ms]
updateEvery10thItem1End phase no difference [-3ms to 1ms]
updateEvery10thItem2End phase no difference [-3ms to 1ms]
selectFirstRow1End phase no difference [0ms to 1ms]
selectSecondRow1End phase no difference [0ms to 1ms]
removeFirstRow1End phase estimated improvement -1ms [-3ms to 0ms] OR -5.68% [-12.69% to -0.45%]
removeSecondRow1End phase no difference [-3ms to 0ms]
swapRows1End phase no difference [-2ms to 1ms]
swapRows2End phase no difference [-1ms to 2ms]
clearItems4End phase estimated regression +1ms [0ms to 2ms] OR +6.78% [2.08% to 11.88%]
paint phase no difference [0ms to 2ms]

pnpm bench 2

duration phase no difference [-31ms to 6ms]
renderEnd phase no difference [0ms to 0ms]
render1000Items1End phase no difference [-1ms to 1ms]
clearItems1End phase no difference [-1ms to 2ms]
render1000Items2End phase no difference [-1ms to 1ms]
clearItems2End phase no difference [0ms to 0ms]
render10000Items1End phase estimated improvement -4ms [-9ms to -1ms] OR -0.91% [-2.03% to -0.11%]
clearManyItems1End phase no difference [-1ms to 1ms]
render10000Items2End phase no difference [-9ms to 2ms]
clearManyItems2End phase no difference [-2ms to 0ms]
render1000Items3End phase no difference [-2ms to 0ms]
append1000Items1End phase no difference [-4ms to 2ms]
append1000Items2End phase no difference [-3ms to 5ms]
updateEvery10thItem1End phase no difference [-2ms to 1ms]
updateEvery10thItem2End phase no difference [-2ms to 1ms]
selectFirstRow1End phase no difference [-1ms to 1ms]
selectSecondRow1End phase no difference [-1ms to 0ms]
removeFirstRow1End phase no difference [-2ms to 0ms]
removeSecondRow1End phase estimated improvement -1ms [-3ms to 0ms] OR -6.55% [-12.91% to -0.28%]
swapRows1End phase no difference [-2ms to 0ms]
swapRows2End phase no difference [-1ms to 0ms]
clearItems4End phase no difference [-1ms to 0ms]
paint phase no difference [-1ms to 1ms]

and my rere-benchmark comparison: overall very improved:

[NullVoxPopuli]

all these changes are for the list op

[NullVoxPopuli-ai-agent]

Correct — the entire mechanism lives in the list opcodes (ListItemOpcode/ListBlockOpcode). That's deliberate: the VM's existing skip machinery (cache groups / JumpIfNotModified) exists only at component boundaries, and list items are exactly where large clean-subtree walks concentrate without any component boundary to catch them — dbmon's rows being the canonical case. Non-list opcodes keep their behavior untouched, which is also what keeps the blast radius reviewable.

[NullVoxPopuli]

Ran a few timees frm diffrnt machine

[NullVoxPopuli-ai-agent]

The two suites agreeing-by-disagreeing is the mechanism working as designed: the skip's win scales with (walk size) × (clean-item fraction) × (update frequency).

• tracerbench neutral = the no-regression result. Its phases are dominated by initial render, teardown, and order changes — paths the skip doesn't touch — and its update phases run unthrottled, where walking 900 clean subtrees costs well under the ±1–3ms confidence intervals. dbmon at 8x with continuous bursts is the same mechanism magnified ~30x.
• selectFirstRow/selectSecondRow neutral is expected for a subtler reason: selection compares each row against a shared selectedId, so every item's collected tag includes that tag — dirtying it invalidates all items and the skip correctly stands down (same walk as before, no penalty). Worth stating in the description: lists whose items all consume a common tag see no benefit, by design, and no cost — likewise @index-keyed lists under insertion.
• clearItems4End +6.78% (run 1, CI excludes zero): doesn't reproduce in run 2 ([-1ms, 0ms]) and its clearManyItems siblings are neutral in both runs — most likely noise. Candidate mechanism if it ever reproduces: items now retain a subtreeTag + finalizer closure, marginally more to collect at teardown. A deep-sample run of just the clear phases would settle it if desired.

🤖 Generated with Claude Code