String width calculation is a hot path in terminal rendering — it's called on every line of output in tools like Ink, cli-truncate, wrap-ansi, and log-update. With the growth of AI-powered terminal agents that stream large amounts of text, this is becoming even more performance-sensitive.

I profiled string-width and found two low-hanging optimizations: stripAnsi runs unconditionally even when no ANSI codes are present, and Intl.Segmenter runs on every call even for pure ASCII input where width = length.

ASCII + ANSI benchmarks (Apple M2 Max, Node 25.2.1)

Non-ASCII inputs (CJK, emoji, mixed) are unaffected — they fall through to the existing Intl.Segmenter path.

PRs

• perf: add ASCII fast path and codepoint iteration to avoid Intl.Segmenter #72 — ASCII fast path + stripAnsi guard
• perf: add fast path for strings without ANSI codes chalk/strip-ansi#54 — adds string.includes('\x1B') fast path to strip-ansi
• perf: use packed arrays with binary search for lookups get-east-asian-width#14 — packed arrays with binary search (3.5x faster, -25% bundle size)

Optimizations

1. ASCII fast path — regex /^[\u0020-\u007E]*$/ detects pure printable ASCII, returns string.length directly. Skips Intl.Segmenter, regex, and EAW lookup entirely. 61x faster for short ASCII, the most common input type.

2. stripAnsi guard — check string.includes('\x1B') || string.includes('\x9B') before calling stripAnsi. Avoids the overhead of strip-ansi's regex on strings that contain no ANSI escapes. 14x faster for ANSI-free short strings.