sqlite3_jieba #

A jieba-backed FTS5 tokenizer for SQLite, shipped as a native asset. It adds a jieba tokenizer and a jieba_cut() SQL function to package:sqlite3 — no extra files to bundle, no dictionary to extract at runtime.

Chinese has no spaces, so SQLite's built-in unicode61 tokenizer indexes a whole sentence as one term and full-text search effectively stops working. trigram works but triples the index and throws away relevance. This package segments Chinese properly instead.

CREATE VIRTUAL TABLE notes USING fts5(title, body, tokenize='jieba');

Install #

dependencies:
  sqlite3_jieba: ^0.1.0

Native assets are built by a Dart build hook, so you need rustup on the machine that compiles your app (not on your users' machines). Nothing else — no CMake, no NDK setup beyond what Flutter already requires.

Use #

import 'package:sqlite3/sqlite3.dart';
import 'package:sqlite3_jieba/sqlite3_jieba.dart';

void main() {
  // Registers via sqlite3_auto_extension, which is process-level state: call it
  // BEFORE opening the database you want to search. Calling it twice is fine.
  loadJiebaTokenizer();

  final db = sqlite3.openInMemory();
  db.execute("CREATE VIRTUAL TABLE notes USING fts5(body, tokenize='jieba')");
  db.execute('INSERT INTO notes(body) VALUES (?)', ['今天天气很好，我去公园散步了']);

  for (final row in db.select(
    "SELECT highlight(notes, 0, '[', ']') AS hit FROM notes WHERE notes MATCH ?",
    ['"公园"'],
  )) {
    print(row['hit']); // 今天天气很好，我去[公园]散步了
  }
}

jieba_cut(text) #

Returns the distinct tokens of text as a JSON array. Use it to build MATCH strings from user input, so the query terms come from the same vocabulary the index was built with:

final tokens = (jsonDecode(
  db.select('SELECT jieba_cut(?) AS t', [query]).single['t'] as String,
) as List).cast<String>();

final match = tokens.map((t) => '"${t.replaceAll('"', '""')}"').join(' OR ');
db.select('SELECT rowid FROM notes WHERE notes MATCH ?', [match]);

With drift #

Register before drift opens its connection — including the background isolate and read pool, which inherit the process-level registration:

loadJiebaTokenizer();
final executor = NativeDatabase.createInBackground(File(path), readPool: 3);

With highlight() / snippet() #

Tokens carry byte offsets into the original text, so both functions return marked-up source rather than index terms. They need the text to be reachable, which means a normal or external content table — on a content='' contentless table SQLite returns NULL silently.

Behaviour #

• Input is split into whitespace-delimited runs, and again wherever Han meets non-Han.
• Han runs go through jieba in search mode: the long word and its dictionary sub-words are both indexed, the sub-words marked FTS5_TOKEN_COLOCATED so they sit at the same position. 今天天气 is findable as 今天天气, 今天 and 天气.
• Everything else is segmented with Unicode UAX#29 word boundaries and lowercased. No stemming: searching runs does not match running.
• Runs without a letter or digit (punctuation, symbols) are dropped, matching unicode61.
• Documents and queries take the identical path, which is what keeps phrase queries aligned.

Language support #

For Japanese or Thai, SQLite's built-in trigram tokenizer is the practical fallback.

How it works #

An FTS5 tokenizer has to be a C-ABI loadable extension: fts5_api is only reachable from inside one, through SELECT fts5(?1) plus sqlite3_bind_pointer. So this package is a Rust cdylib with a ~100-line C shim for the SQLite plumbing; the shim's entry point is re-exported by Rust as sqlite3_jieba_init, and Dart hands that address to sqlite3_auto_extension through package:sqlite3's SqliteExtension.

The jieba dictionary (5 MB of text) is DEFLATE-compressed into the binary by jieba-rs, so there are no asset files to ship or unpack.

License #

MIT. Bundles jieba-rs (MIT) and its dictionary, and unicode-segmentation (MIT/Apache-2.0).