cangjie

Repo: github.com/mova-ITS/cangjie-dictionary (crate name stays cangjie).

A CangJie (CJ5) dictionary: bidirectional lookup, code ↔ hieroglyph.

• hi→co: given a character, return its Cangjie code(s) and provenance. A char can have more than one core code — ~10.6% do (e.g. 丅 → ml and xml).
• co→hi: given a code, return the character(s) that produce it.

Both directions are one-to-many. Built over a hardened, cross-verified CJ5 table (~20.9k CJK-BMP chars). Source: data/cangjie_bmp.tsv. Packed hi+co for the browser: data/cangjie.bin (cargo run --bin pack).

Data contract

IN	TSV text: char␉code␉trust␉sources (one row per core code)
OUT (hi→co)	&[Entry] — hi, co, trust, sources
OUT (co→hi)	&[char]
Query	one non-ASCII glyph → codes; otherwise → characters

trust / sources explain why a code is in the core table (CJ5 consensus first; Unihan corroborates, never overrides). The variants sidecar is not loaded here.

API

use cangjie::{bundled_tsv, Dict};

let dict = Dict::load(bundled_tsv())?;
let _ = dict.codes_of('一');        // hi → co  (&[Entry])
let _ = dict.chars_for("m");        // co → hi  (&[char])
let _ = dict.query("一");           // Hits::Codes / Hits::Chars
let _ = dict.codes_matching("v-c"); // pattern over codes

Dict::parse(&str) builds from TSV. Dict::from_bytes / load_bin load the packed indexes.

Patterns (codes_matching)

Whole code is 1–5 letters. Not edit-distance.

Atom	Meaning
letter	that unit, that place
-	exactly one letter
*	zero or more letters

Examples: A* starts with A · *A ends with A · V-C is exactly V + one + C · A-* is A + one + rest · A-*-B is the same atoms composed.

Plain letters with no -/* are an exact code (mg ≠ prefix; use mg*).

CLI

cargo run -- 一
cargo run -- m
cargo run --features brand -- m    # green mova.page wordmark

Expect 一 → ["m"] and m → ['一'] (plus other chars that share the code).

Optional second argument: path to another TSV.

WASM

Feature wasm (no std::fs in the module). JS fetches cangjie.bin, then WasmDict.parse_bin.

cargo run --bin pack
wasm-pack build --target web --features wasm

pkg/ is generated — do not commit it. JS API: parse_bin, parse (TSV), codes_of, chars_for, codes_matching, mask_for.

Design notes

• Dual index: char → Vec<Entry> and code → Vec<char>, built once.
• Parse is pure; load is a thin read_to_string wrapper.
• Core table: CJ5 consensus first (Jack Chow · RIME · base). Unihan corroborates, never overrides. Variants sidecar is not loaded.