[BUG] Thai text rendering is broken in TUI — combining characters cause layout misalignment

[kingits]

Description

Thai text renders incorrectly in the OpenCode TUI. Characters appear garbled with misaligned combining characters (vowels, tone marks), broken layout, and text overlapping. The same Thai text displays correctly in the terminal outside of OpenCode.

Steps to Reproduce

1. Open OpenCode TUI in any terminal (tested with Warp on macOS)
2. Type or receive a response containing Thai text (e.g. สวัสดีครับ ทดสอบภาษาไทย)
3. Observe garbled/misaligned Thai text rendering

Expected Behavior

Thai text should render correctly with proper positioning of:

• Combining vowels above/below consonants (สระบน/ล่าง): ิ ี ึ ื ุ ู
• Tone marks (วรรณยุกต์): ่ ้ ๊ ๋
• Leading vowels (สระหน้า): เ แ โ ไ ใ
• Compound characters: ำ

Actual Behavior

Thai combining characters are mispositioned, causing:

• Vowels and tone marks appearing in wrong positions
• Text overlapping and layout breaking
• Overall garbled/unreadable Thai output

Environment

• OpenCode version: 1.4.1
• OS: macOS (Darwin, Apple Silicon)
• Terminal: Warp (Thai renders correctly outside of OpenCode TUI)
• Locale: LANG=C.UTF-8, LC_CTYPE=UTF-8
• Shell: zsh

Root Cause Analysis

Thai script uses Unicode combining characters extensively. The TUI likely miscalculates character widths because:

1. Grapheme cluster width: Thai combining characters (tone marks, vowels above/below) have zero display width but are separate Unicode codepoints. Width calculation that sums individual codepoint widths produces incorrect results.
2. Related issues in the ecosystem:
   • charmbracelet/lipgloss#562 — Emoji/Unicode width miscalculation causes layout misalignment
   • mattn/go-runewidth#59 — Combining characters not handled correctly (PR Tried installing on 2 Ubuntu 24.04 machines and stuck #63 proposed but never merged)

Workaround

Using opencode web (browser UI) displays Thai correctly, but loses access to terminal-dependent plugins (e.g. tmux integration).

Additional Context

Thai is not the only complex script affected — any script using combining characters (Devanagari, Arabic, etc.) likely has the same issue. A proper fix would use grapheme-cluster-aware width calculation (e.g., rivo/uniseg) instead of per-rune width lookup.

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• RTL language support (Persian/Arabic) - characters appear reversed and disconnected #16875: RTL language support (Persian/Arabic) — characters appear reversed and disconnected; similar root cause of complex script / combining character handling in the TUI
• Chinese output occasionally renders duplicated characters and misaligned layout in Ghostty (macOS) #8175: Chinese output occasionally renders duplicated characters and misaligned layout — same symptom of character width miscalculation causing layout misalignment in Ghostty (macOS)