Benchmark snapshots are captured in benchmarks/chrome.json and benchmarks/safari.json. The tables below reflect the snapshot taken on 2026-03-28.
layout() is the resize hot path — call it on every container width change. prepare() is the one-time cost paid when text first appears. Optimizing resize performance means keeping layout() fast; prepare() budget is spent once per text block.
Top-level batch
Numbers over the shared 500-text batch corpus.
| Browser | prepare() | layout() | DOM batch | DOM interleaved |
|---|
| Chrome | 18.85ms | 0.09ms | 4.05ms | 43.50ms |
| Safari | 18.00ms | 0.12ms | 87.00ms | 149.00ms |
layout().
Rich line APIs
Shared corpus
| Browser | layoutWithLines() | walkLineRanges() | layoutNextLine() |
|---|
| Chrome | 0.05ms | 0.03ms | 0.07ms |
| Safari | 0.05ms | 0.03ms | 0.07ms |
| Browser | layoutWithLines() | walkLineRanges() | layoutNextLine() |
|---|
| Chrome | 4.99ms | 2.17ms | 6.39ms |
| Safari | 4.66ms | 2.37ms | 5.53ms |
prepare() is split into its two internal phases — analyze() for segmentation and glue work, measure() for canvas width measurement — so you can see which script is expensive for which reason.
| Corpus | analyze() | measure() | prepare() | layout() | segs (analyze→prepared) | lines @ 300px |
|---|
| Japanese prose (story 2) | 1.90ms | 4.00ms | 6.10ms | 0.02ms | 1,773→2,667 | 193 |
| Japanese prose | 4.10ms | 8.30ms | 12.60ms | 0.03ms | 3,606→5,044 | 380 |
| Korean prose | 2.40ms | 8.40ms | 11.40ms | 0.04ms | 5,282→9,679 | 428 |
| Chinese prose | 6.10ms | 13.10ms | 19.20ms | 0.05ms | 5,433→7,949 | 626 |
| Chinese prose (story 2) | 3.70ms | 8.10ms | 11.80ms | 0.03ms | 3,271→4,745 | 375 |
| Thai prose | 8.10ms | 5.40ms | 13.50ms | 0.05ms | 10,281→10,281 | 1,024 |
| Myanmar prose | 0.60ms | 0.70ms | 1.30ms | <0.01ms | 797→797 | 81 |
| Myanmar prose (story 2) | 0.30ms | 0.60ms | 0.90ms | <0.01ms | 498→498 | 54 |
| Urdu prose | 2.20ms | 3.30ms | 5.50ms | 0.03ms | 6,051→6,051 | 351 |
| Khmer prose | 5.80ms | 4.50ms | 10.40ms | 0.05ms | 11,109→11,109 | 591 |
| Hindi prose | 4.30ms | 6.60ms | 11.10ms | 0.04ms | 9,958→9,958 | 653 |
| Arabic prose | 29.10ms | 34.80ms | 63.50ms | 0.17ms | 37,603→37,603 | 2,643 |
segs column shows segment count going from the analysis phase into the prepared handle. Scripts like CJK expand their segment count because words are split into individual graphemes for per-character line breaking. Scripts like Thai, Myanmar, Urdu, Khmer, Hindi, and Arabic do not expand because they do not require that splitting.
Notes
- Chrome is the main maintained performance baseline. Safari snapshots are included but are noisier and warm up less predictably.
- The checked-in JSON snapshots are cold checker runs. Ad hoc page-driven numbers, especially in Safari, can differ significantly after warmup.
- Refresh
benchmarks/chrome.json and benchmarks/safari.json when a change affects the text engine hot path (src/analysis.ts, src/measurement.ts, src/line-break.ts, src/layout.ts, src/bidi.ts, or pages/benchmark.ts).