For AI agents: the complete documentation index is available at /en/llms.txt, the full documentation bundle is available at /en/llms-full.txt, and this page is available as Markdown at /en/changelog/uax14-linebreak.md.
Hunea
  • English

      • Unified prose wrapping via UAX #14

      • Date: 2026-07-19
      • Commit: feat(tui): unify prose wrapping onto a UAX #14 line-break planner
      • Version cue: 0.12.1-alpha.1

      This is a foundational unification of line wrapping. Previously, wrap, prompt input wrapping, Markdown rendering, and syntax highlighting each maintained their own greedy tokenizer with slightly different behavior. They're now all converged onto a single line-break planner.

      What changed

      1. Wrapping follows the Unicode standard (UAX #14)
        The planner computes grapheme boundaries and UAX #14 break opportunities over the full text in one pass, then decides wrap points using terminal display width. Styling is just a projection onto already-determined byte ranges, so it no longer affects where lines break. The same text wraps identically whether or not it's highlighted.

      2. URLs are no longer split mid-token
        URL-like tokens are kept whole and avoid being broken in the middle (including a fallback for wrapping punctuation). Long links are easier to recognize and copy in narrow terminals.

      3. A visible change in wrapping behavior
        File paths and hyphenated words now wrap at valid UAX #14 break opportunities instead of being hard-cut at arbitrary columns. Long paths or words like foo-bar-baz break at more intuitive points. (Note: file paths themselves aren't given "never split" protection; they just wrap at standard breakpoints.)

      4. Better handling of CJK and special characters
        CJK line-break rules (kinsoku) are applied to avoid disallowed leading/trailing punctuation; LS / PS / NEL, lone CR, and CRLF are all treated as mandatory breaks.

      What you'll notice

      • Long links and paths wrap at more sensible points in narrow windows, without hard mid-token cuts.
      • Mixed CJK/Latin and punctuation-heavy paragraphs wrap more naturally.
      • In rare cases, if you relied on a "hard cut at a fixed column" look before, wrap points may differ slightly from the old version; this is an expected behavior change.

      This is a correctness-and-consistency refactor: it removes ~700 lines of duplicated implementation and adds a performance optimization for reusing line counts when user messages resize wider (O(1) reuse via a construction-time monotonicity guarantee instead of re-scanning the body). For you, the main payoff is that wrapping is finally consistent everywhere.