A pure-Odin modern text engine — parsing, itemization, shaping, line-breaking, rasterization. Built first to replace fontstash + stb_truetype in Skald, designed to be useful for any Odin project that needs production-quality text.

Status: v1.2.1 — bounded shape cache (LRU) on top of v1.1's Latin autohinter and v1.0's complex-script punch list. UAX #9 bidi at 100 %, UAX #29 graphemes at 100 %, CFF2 variable instances, COLRv1 emoji with the full 28 W3C composite blend modes + linear / radial / sweep gradients, GPOS mark-to-ligature, frozen API (API.md), per-script shaping for Devanagari, Bengali, Gujarati, Kannada, Odia, Tamil, Telugu, Malayalam, Gurmukhi, Thai, Lao, Khmer, Myanmar verified against HarfBuzz byte-for-byte, and a Thai word-break dictionary so Thai paragraphs reflow at word boundaries rather than as one giant unbreakable word. Full scoreboard in CHANGELOG.md.

_{Latin · Cyrillic · Greek · Arabic (joined + RTL) · Hebrew (RTL) · Chinese · Japanese · colour emoji · OpenType ligatures.}

The current Odin text-rendering story is vendor:fontstash plus stb_truetype — a glyph atlas and basic kerning. That covers Latin text at small scales but cannot:

• Shape complex scripts (Arabic joining, Devanagari conjuncts, Thai clusters, bidi).
• Break lines per Unicode UAX #14.
• Render colour emoji (COLRv0 / COLRv1 / CBDT / sbix tables).
• Apply OpenType features (ligatures, GPOS kerning, contextual alternates, stylistic sets).
• Subpixel-position glyphs.

runa is the long-term fix — a modern text engine in idiomatic Odin with zero C dependencies at v1.0.

All four items closed:

• Thai word-break dictionary — embedded PyThaiNLP corpus, longest-match trie, ~62 k entries. ✓
• Complex Khmer multi-consonant clusters — pre-base reorder fixed to move to syllable start, not just before base. ✓
• Full Myanmar shaping — joined the Indic pipeline; medial YA / medial RA / IPC=Top_And_Bottom_And_Left all reorder correctly. ✓
• Two remaining bidi BidiCharacterTest mismatches — ISR formation now tunnels through all-BN higher-level intervening runs. 100.00 % conformance. ✓

Per-feature deliverables, conformance numbers, and tracked patch-level work in CHANGELOG.md.

The library is a set of plain Odin packages — no build script needed.

odin check . -no-entry-point          # type-check the library
odin test  tests/parse                # parser tests
odin test  tests/raster               # rasterizer smoke tests
odin test  tests/runa                 # facade integration tests

The runnable demos:

# PGM-output, no GUI dependency — shapes Latin + emoji, writes a PPM
# image you can open in any viewer.
odin run examples/hello_world -- \
    tests/fonts/Roboto-Regular.ttf \
    tests/fonts/Twemoji-Mozilla.ttf \
    /tmp/hello.ppm

# Live raylib window — shows OpenType shaping, kerning, ligatures,
# and tinted text running through raylib's stock DrawTexturePro.
# ~50 lines of glue between runa and the renderer; demonstrates that
# runa is renderer-agnostic — works in any Odin project that can
# sample a texture and draw a quad (sokol_gfx, custom Vulkan / Metal,
# even a pure CPU pixel buffer).
odin run examples/raylib

Test fonts are not committed — fetch them into tests/fonts/ per tests/fonts/README.md, or let CI fetch them. Tests that need a missing font skip with an INFO log so the synthetic suite still runs on a fresh clone.

runa is licensed under the zlib license — see LICENSE. Permissive, GPL-compatible, the same licence Odin's own standard library uses.

Embedded Unicode UCD data files (Scripts.txt, LineBreak.txt, DerivedBidiClass.txt) ship under the Unicode-DFS-2016 licence. Test fonts are not committed — see tests/fonts/README.md for per-font sources and licences.

API is frozen (API.md) — bidi at 100 %, line break at 99.91 %, graphemes at 100 %, UAX #29 word boundaries at 100 %, UAX #15 normalization (NFC / NFD / NFKC / NFKD) at 100 %, all 28 COLR composite blend modes, CFF2 variations, and the full complex-script set (9 Brahmic — Devanagari, Bengali, Tamil, Telugu, Kannada, Malayalam, Gurmukhi, Gujarati, Odia — plus 4 SEA — Thai, Lao, Myanmar, Khmer) all real. See CONTRIBUTING.md for build / test instructions and the open-work pointer list (long-tail Khmer cluster polish, performance hardening, COLRv1 sweep gradients).