perf: improve the performance of timezone

[tmkx]

it's 20% faster in a 10,000-invocation loop test

❯ node ./node.test.mjs ✔ dev (166.546375ms) ✔ perf (133.484125ms)

import assert from 'node:assert'
import test from 'node:test'

import dayjs from './dayjs.min.js'
import utc from './plugin/utc.js'
import timezone from './plugin/timezone.js'

import dayjsPerf from './perf/dayjs.min.js'
import utcPerf from './perf/plugin/utc.js'
import timezonePerf from './perf/plugin/timezone.js'

dayjs.extend(utc)
dayjs.extend(timezone)

dayjsPerf.extend(utcPerf)
dayjsPerf.extend(timezonePerf)

test('dev', () => {
  for (let i = 0; i < 10000; i++) {
    const offset = dayjs
      .tz('2024-01-03 00:00:00', 'America/New_York')
      .utcOffset()
    assert.equal(offset, -300)
  }
})

test('perf', () => {
  for (let i = 0; i < 10000; i++) {
    const offset = dayjsPerf
      .tz('2024-01-03 00:00:00', 'America/New_York')
      .utcOffset()
    assert.equal(offset, -300)
  }
})

[tmkx]

It can significantly improve the first paint time in a heavy-dependent app like Calendar.

[tmkx]

any feedback?

[buza-me]

I'm not the library author, but: Your test is made in one script. You need to run separate scripts for benchmarks. 10k iterations on 'dev' version gives everything a decent warm-up. I'm saying "everything", because the difference may even come from node:assert optimized by v8. I believe that removing string concatenation improves the script speed, but still your benchmark is not reliable imo. Second thing is: How can someone run your test snippet, if they do not have the files?

import dayjs from './dayjs.min.js'
import utc from './plugin/utc.js'
import timezone from './plugin/timezone.js'

import dayjsPerf from './perf/dayjs.min.js'
import utcPerf from './perf/plugin/utc.js'
import timezonePerf from './perf/plugin/timezone.js'

You should probably link some repo that contains those files.

[tmkx]

I agree with you that the number is not very precise, it's affected by many factors, including warm-up, execution order, cpu occupation, etc. I showed this just for simplicity and sorry for the confusion.

I've set up a benchmark using Vitest, you can run npm run bench with https://github.com/tmkx/dayjs/tree/perf/timezone-bench, and here is my result:

 ✓ bench/timezone.bench.ts (2) 1236ms
     name             hz     min     max    mean     p75     p99    p995    p999     rme  samples
   · perf      40,160.84  0.0230  0.5659  0.0249  0.0245  0.0385  0.0592  0.1507  ±0.45%    20081   fastest
   · baseline  33,307.05  0.0276  1.5496  0.0300  0.0292  0.0548  0.0710  0.1799  ±0.75%    16654  

import { bench } from 'vitest'

bench('perf', () => {
  dayjsPerf.tz('2024-01-03 00:00:00', 'America/New_York').utcOffset()
})

bench('baseline', () => {
  dayjs.tz('2024-01-03 00:00:00', 'America/New_York').utcOffset()
})

the baseline and perf files are the output of npm run build in two branches.