gukhanmun-cdb

CDB dictionary backend for Gukhanmun. Implements HanjaDictionary over a CDB-based file format that is trivially auditable: the on-disk layout is a standard CDB header followed by CBOR metadata and trie records, with no proprietary compression.

File format

A __gukhanmun_meta__ key in the CDB holds a CBOR-encoded metadata map (source name, build date, etc.). All other keys are hanja strings in UTF-8; the corresponding value encodes the hangul reading, a 2-bit mark byte (requiring hanja or hangul annotation), and a length prefix in a compact binary layout.

Installation

[dependencies]
gukhanmun-cdb = "0.1"

Usage

use gukhanmun_cdb::CdbDictionary;

// From a file on disk:
let dict = CdbDictionary::open("stdict.gukcdb")?;

// From an owned Arc<[u8]>:
let bytes: Arc<[u8]> = std::fs::read("stdict.gukcdb")?.into();
let dict = CdbDictionary::from_bytes(bytes)?;

// Zero-copy from a static byte slice:
static BYTES: &[u8] = include_bytes!("stdict.gukcdb");
let dict = CdbDictionary::from_static_bytes(BYTES)?;

from_static_bytes holds the slice directly without copying, so there is no heap allocation beyond the CdbDictionary struct itself.

Trade-offs vs FST

CDB lookup is O(1) (two hash table probes), while FST lookup is O(key length) over a compressed automaton. For short hanja keys the difference is small. The FST format supports prefix streaming and has better locality for the lattice segmenter's inner loop; CDB is simpler to inspect and patch by hand.

License