~4% perf regression in select-row benchmark due to on optimization

[nolanlawson]

I did a Tachometer analysis of master vs the LWC version released for 252 (86682ec, v7.1.5).

Click to see

<style type="text/css"></style>

Benchmark	Before (low)	Before (high)	Before (avg)	After (low)	After (high)	After (avg)	Delta (low)	Delta (high)	Delta (avg)	Delta perc (low)	Delta perc (high)	Delta perc (avg)
engine-dom-append-rows-1k	17.07	17.22	17.14	16.82	16.93	16.88	-0.37	-0.17	-0.27	-0.02	-0.01	-2%
engine-dom-clear-rows-1k	4.81	4.92	4.86	4.69	4.79	4.74	-0.2	-0.05	-0.12	-0.04	-0.01	-3%
engine-dom-create-rows-10k	93.52	94.32	93.92	93.77	94.65	94.21	-0.31	0.89	0.29	0	0.01	0%
engine-dom-create-rows-1k	13.39	13.65	13.52	13.72	14	13.86	0.15	0.53	0.34	0.01	0.04	3%
engine-dom-expression	16.48	16.71	16.6	16.48	16.71	16.6	-0.16	0.16	0	-0.01	0.01	0%
engine-dom-light-slot-create-container-5k	261.04	262.23	261.64	263.85	265.1	264.47	1.97	3.7	2.84	0.01	0.01	1%
engine-dom-light-slot-update-slotted-content-5k	174.42	176.01	175.21	174.38	176.23	175.31	-1.12	1.31	0.1	-0.01	0.01	0%
engine-dom-light-styled-component-create-10k-same	91.52	91.94	91.73	92.54	92.92	92.73	0.72	1.29	1	0.01	0.01	1%
engine-dom-light-styled-component-create-1k-different	134.95	136.21	135.58	134.75	135.85	135.3	-1.11	0.56	-0.28	-0.01	0	0%
engine-dom-partial-update-1k	12.32	12.6	12.46	12.1	12.39	12.24	-0.42	-0.02	-0.22	-0.03	0	-2%
engine-dom-remove-row-1k	6.38	6.53	6.46	6.53	6.69	6.61	0.05	0.26	0.16	0.01	0.04	2%
engine-dom-replace-rows-1k	17.36	17.55	17.45	17.02	17.21	17.11	-0.47	-0.21	-0.34	-0.03	-0.01	-2%
engine-dom-select-row-1k	10.63	11.03	10.83	11.03	11.42	11.23	0.12	0.68	0.4	0.01	0.06	4%
engine-dom-ss-mutation-observer-10k	1348.58	1358.7	1353.64	1342.63	1352.78	1347.7	-13.1	1.23	-5.93	-0.01	0	0%
engine-dom-ss-slot-create-container-5k	358.68	360.76	359.72	360.66	362.57	361.62	0.49	3.3	1.9	0	0.01	1%
engine-dom-ss-slot-update-slotted-content-5k	217.65	218.87	218.26	220.02	221.27	220.65	1.51	3.26	2.39	0.01	0.01	1%
engine-dom-ss-styled-component-create-10k-same	147.12	148.04	147.58	147.9	148.84	148.37	0.13	1.45	0.79	0	0.01	1%
engine-dom-ss-styled-component-create-1k-different	153.97	155.42	154.7	153.61	154.68	154.15	-1.45	0.35	-0.55	-0.01	0	0%
engine-dom-ss-table-create-10k	139.92	140.99	140.45	139.37	140.32	139.84	-1.33	0.11	-0.61	-0.01	0	0%
engine-dom-styled-component-create-10k-same	164.93	166.31	165.62	167.57	169.08	168.33	1.68	3.73	2.71	0.01	0.02	2%
engine-dom-styled-component-create-1k-different	148.43	149.62	149.03	147.81	149.02	148.41	-1.47	0.23	-0.62	-0.01	0	0%
engine-dom-swap-rows-1k	10.54	10.83	10.68	10.71	10.98	10.84	-0.04	0.36	0.16	0	0.03	2%
engine-dom-table-create-10k	106.58	107.97	107.28	106.78	107.86	107.32	-0.84	0.93	0.04	-0.01	0.01	0%
engine-dom-tablecmp-append-1k	37.59	38.14	37.87	37.62	38.12	37.87	-0.37	0.37	0	-0.01	0.01	0%
engine-dom-tablecmp-create-10k	304.89	306.27	305.58	308.41	309.96	309.18	2.56	4.64	3.6	0.01	0.02	1%
engine-dom-tablecmp-create-1k	45.61	45.87	45.74	46.25	46.58	46.42	0.47	0.89	0.68	0.01	0.02	1%
engine-dom-trackedcmp-create-10k	271.56	272.94	272.25	274.46	275.79	275.13	1.92	3.84	2.88	0.01	0.01	1%
engine-dom-trackedcmp-update-10k	107.78	108.47	108.13	108.86	109.61	109.23	0.6	1.61	1.11	0.01	0.01	1%
engine-dom-tree-create-3k	101.53	102.06	101.8	102.95	103.48	103.22	1.05	1.8	1.42	0.01	0.02	1%
engine-dom-tree-remove-3k	47.45	47.87	47.66	47.42	47.93	47.68	-0.31	0.35	0.02	-0.01	0.01	0%
engine-dom-wireslist-create-10k	253.24	254.5	253.87	256.74	257.82	257.28	2.58	4.25	3.41	0.01	0.02	1%
engine-server-table-render-10k	56.16	56.38	56.27	55.47	55.66	55.57	-0.85	-0.56	-0.71	-0.02	-0.01	-1%
engine-server-tablecmp-render-10k	176.91	178.67	177.79	176.51	177.68	177.09	-1.76	0.36	-0.7	-0.01	0	0%
hydration-table-hydrate-1k	38.11	38.49	38.3	40.1	40.43	40.26	1.71	2.21	1.96	0.04	0.06	5%
hydration-tablecmp-hydrate-1k	39.08	39.39	39.23	41.8	42.13	41.96	2.5	2.96	2.73	0.06	0.08	7%

We've already fixed the last two (hydration) in #4996. For select-row, though, a git bisect revealed that #4468 was the source of the regression.

This is a bit surprising since this PR was supposed to be a perf improvement. Perhaps it improved perf in some benchmarks but regressed this one. We should do some analysis to see if overall this change is still worth it, or if there is some way to further optimize it.

git bisect script

#!/usr/bin/env bash

set -e
set -x

yarn
BENCHMARK_REF=86682ec8329eab989a6c9fd8a71eaf447e431eab BENCHMARK_SAMPLE_SIZE=400 BENCHMARK_TIMEOUT=60 yarn build:performance
cd packages/@lwc/perf-benchmarks

node ../../../node_modules/.bin/tach --config ./dist/__benchmarks__/engine-dom/js-framework-benchmark/select-row-1k.tachometer.json --json-file /tmp/bench.json

cd ../../..

highNum=$(cat /tmp/bench.json | jq .benchmarks[0].differences[1].percentChange.high | tr '.' ' ' | awk '{print $1}')
lowNum=$(cat /tmp/bench.json | jq .benchmarks[0].differences[1].percentChange.low | tr '.' ' ' | awk '{print $1}')

echo "Numbers are $lowNum to $highNum"

if (( $highNum >= 2 )); then
  if (( $lowNum >= 0 )); then
	echo "Fail"
    exit 1;
  fi
fi

echo "Success"

(I manually checked the git bisect run results, and yes, this script worked correctly.)