Crate golia_pinyin 

golia-pinyin — self-developed Mandarin Pinyin input method engine.

Engine surface ✓ (segmenter, fuzzy, FST dict, encode, session) + 919k-entry corpus-derived dict (Unihan + jieba + Leipzig + SUBTLEX) + L0 user-learning ranking (3-pick auto-pin). The published crate version stays at 0.1.0 per the publish strategy in lab8-ime ROADMAP item 35; internal milestone names (v0.2-data, v0.3-l0) refer to data + feature readiness. See workspace ROADMAP.

Sibling library: wubi — same architectural pattern (PHF static tables, FST main dict, zero-alloc hot path).

Quickstart

use golia_pinyin::{PinyinEngine, Session};
let engine = PinyinEngine::new();
let mut session = Session::new(&engine);
for c in "zhongguo".chars() {
    session.input_char(c);
}
let cands = session.candidates();
assert_eq!(cands.first().map(String::as_str), Some("中国"));

Module map

• syllable — 403 valid Mandarin syllable inventory (PHF set)
• fuzzy — toggleable fuzzy-pair expansion (z↔zh etc.)
• segmenter — DP segmentation of continuous pinyin strings
• dict — FST-backed pinyin → words lookup with L0 user-learning
• encode — char → readings reverse lookup
• engine — immutable PinyinEngine (dict + fuzzy)
• session — mutable Session holding the user’s input buffer
• ranking — L0 snapshot type for host-side persistence

Re-exports

pub use dict::PinyinDict;

pub use encode::char_to_pinyin;

pub use encode::covered_char_count;

pub use engine::PinyinEngine;

pub use fuzzy::FuzzyConfig;

pub use ranking::L0Snapshot;

pub use ranking::PROMOTE_THRESHOLD;

pub use segmenter::Segmentation;

pub use segmenter::segment;

pub use session::Session;

pub use syllable::VALID_SYLLABLES;

pub use syllable::count as syllable_count;

pub use syllable::is_valid as is_valid_syllable;

Modules

dict

FST-backed pinyin dictionary with a two-tier ranking model.

encode

Reverse lookup — char → Vec<pinyin>.

engine

PinyinEngine — immutable assembly of dict + fuzzy config.

fuzzy

Fuzzy syllable expansion — z↔zh, c↔ch, s↔sh, n↔l, f↔h, r↔l, in↔ing, en↔eng, an↔ang. Toggleable per-pair so users can match their own dialect / typing habits. Expansion happens at lookup time; the dictionary stays canonical (no bloat).

ranking

L0 ranking — user-learning layer on top of the immutable dict.

segmenter

Pinyin syllable segmentation via dynamic programming.

session

Per-input mutable state — accumulates the user’s typing buffer and exposes candidates / commit semantics.

syllable

Canonical Mandarin syllable inventory.