Native Drift (cross-platform)

Run drift on both mobile and desktop

Supported platforms

The drift/native.dart library uses the sqlite3 package to send queries. At the moment, that package supports iOS, macOS and Android out of the box. Most Linux Distros have sqlite available as a shared library, those are supported as well.

If you're shipping apps for Windows and Linux, it is recommended that you bundle a sqlite3.so and sqlite3.dll file with your app. You can then make drift support your setup by running this code before opening the database:

import 'dart:ffi';
import 'dart:io';
import 'package:sqlite3/sqlite3.dart';
import 'package:sqlite3/open.dart';

void main() {
  open.overrideFor(OperatingSystem.linux, _openOnLinux);

  final db = sqlite3.openInMemory();
  db.dispose();
}

DynamicLibrary _openOnLinux() {
  final script = File(Platform.script.toFilePath());
  final libraryNextToScript = File('${script.path}/sqlite3.so');
  return DynamicLibrary.open(libraryNextToScript.path);
}
// _openOnWindows could be implemented similarly by opening `sqlite3.dll`

For Flutter apps, using the drift_flutter package as suggested in the setup instructions takes care of these steps.

Drift-managed background isolates

Being a C library, SQLite runs SQL statements synchronously, blocking the thread issuing a statement for IO or computations necessary to run statements. Especially in mobile apps, this blocking nature means that the database should not be accessed on the UI isolate directly, as this can cause dropped frames or other UI issues.

When using NativeDatabase.createInBackground instead of the raw NativeDatbase constructor, drift will set up a background isolate responsible for hosting the database:

QueryExecutor openDatabase() {
  return NativeDatabase.createInBackground(
    File('path/to/database.db'),
    isolateSetup: () {
      open.overrideFor(OperatingSystem.linux, _openOnLinux);
    },
  );
}

You can use the returned QueryExecutor with the constructor of your database class. This means that the usage of the database doesn't change at all, only the setup code needs to be adapted to use a background isolate.

Using multiple read isolates

Using a single background isolate to host the database is sufficient for most applications. In some cases though, it may be beneficial to use more than one background isolate for the database:

  1. Using multiple isolates can improve startup performance of your application if it runs a lot of queries when starting up.
  2. If you have a mix of "expensive" reads (e.g. due to large data sizes in some tables) and small/faster reads, distributing them across multiple isolates ensures long-running reads don't impact others as much.
  3. With a single thread, long-running writes or transactions block reads. This is not the case when using multiple isolates.

Efficiently using multiple isolates requires the use of write-ahead logging (WAL), which allows a single writer and multiple readers to operate on the same database file in parallel. Not using WAL will cause "database is locked" errors when multiple isolates access the same database.

An additional pool of readers can be enabled with the readPool argument on NativeDatabase.createInBackground:

QueryExecutor openMultiThreadedDatabase() {
  return NativeDatabase.createInBackground(
    File('path/to/database.db'),
    isolateSetup: () {
      open.overrideFor(OperatingSystem.linux, _openOnLinux);
    },
    setup: (database) {
      // This is important, as accessing the database across threads otherwise
      // causes "database locked" errors.
      // With write-ahead logging (WAL) enabled, a single writer and multiple
      // readers can operate on the database in parallel.
      database.execute('pragma journal_mode = WAL;');
    },
    readPool: 4,
  );
}

In this snippet, drift will spawn five isolates to host the database: One for writes, and four additional ones only used for reads. Note that transactions and exclusively blocks on the database will always use the write isolate.

Using native drift with an existing database

If your existing sqlite database is stored as a file, you can just use NativeDatabase(thatFile) - no further changes are required.

If you want to load databases from assets or any other source, you can use a LazyDatabase. It allows you to perform some async work before opening the database:

// before
NativeDatabase(File('...'));

// after
LazyDatabase(() async {
  final file = File('...');
  if (!await file.exists()) {
    // copy the file from an asset, or network, or any other source
  }
  return NativeDatabase(file);
});

Using existing databases is explained in more detail in this example.

Used compile options on Android

On Android, iOS and macOs, depending on sqlite3_flutter_libs will include a custom build of sqlite instead of using the one from the system. The chosen options help reduce binary size by removing features not used by drift. Important options are marked in bold.

  • We use the -O3 performance option
  • SQLITE_DQS=0: This will make sqlite not accept double-quoted strings (and instead parse them as identifiers). This matches the behavior of drift and compiled queries
  • SQLITE_THREADSAFE=0: Since the majority of Flutter apps only use one isolate, thread safety is turned off. Note that you can still use the isolate api for background operations. As long as all database accesses happen from the same thread, there's no problem.
  • SQLITE_DEFAULT_MEMSTATUS=0: The sqlite3_status() interfaces are not exposed by drift, so there's no point of having them.
  • SQLITE_MAX_EXPR_DEPTH=0: Disables maximum depth when sqlite parses expressions, which can make the parser faster.
  • SQLITE_OMIT_AUTHORIZATION, SQLITE_OMIT_DECLTYPE, SQLITE_OMIT_DEPRECATED, SQLITE_OMIT_GET_TABLE, SQLITE_OMIT_LOAD_EXTENSION, SQLITE_OMIT_PROGRESS_CALLBACK, SQLITE_OMIT_SHARED_CACHE, SQLITE_OMIT_TCL_VARIABLE, SQLITE_OMIT_TRACE: Disables features not supported by drift.
  • SQLITE_USE_ALLOCA: Allocate temporary memory on the stack
  • SQLITE_UNTESTABLE: Remove util functions that are only required to test sqlite3
  • SQLITE_HAVE_ISNAN: Use the isnan function from the system instead of the one shipped with sqlite3.
  • SQLITE_ENABLE_FTS5: Enable the fts5 engine for full-text search.
  • SQLITE_ENABLE_JSON1: Enable the json1 extension for json support in sql query.

For more details on sqlite compile options, see their documentation.

Drift-only functions

The NativeDatabase includes additional sql functions not available in standard sqlite:

  • pow(base, exponent) and power(base, exponent): This function takes two numerical arguments and returns base raised to the power of exponent. If base or exponent aren't numerical values or null, this function will return null. This function behaves exactly like pow in dart:math.
  • sqrt, sin, cos, tan, asin, acos, atan: These functions take a single argument. If that argument is null or not a numerical value, returns null. Otherwise, returns the result of applying the matching function in dart:math.
  • regexp: Wraps the Dart RegExp apis, so that foo REGEXP bar is equivalent to RegExp(bar).hasMatch(foo). Note that we have to create a new RegExp instance for each regexp sql call, which can impact performance on large queries.
  • current_time_millis: Returns the current unix timestamp as milliseconds. Equivalent to DateTime.now().millisecondsSinceEpoch in Dart.

Note that NaN, -infinity or +infinity are represented as NULL in sql.

When enabling the moor_ffi module in your build options, the generator will allow you to use those functions in drift files or compiled queries.

To use those methods from Dart, you need to import package:drift/extensions/native.dart. You can then use the additional functions like this:

import 'package:drift/drift.dart';
// those methods are hidden behind another import because they're only available with a NativeDatabase
import 'package:drift/extensions/native.dart';

class Coordinates extends Table {
  RealColumn get x => real()();
  RealColumn get y => real()();
}

// Can now be used like this:
Future<List<Coordinate>> findNearby(Coordinate center, int radius) {
  return (select(coordinates)..where((other) {
    // find coordinates where sqrt((center - x)² + (center.y - y)²) < radius
    final distanceSquared = sqlPow(center.x - row.x, 2) + sqlPow(center.y - row.y, 2);
    return sqlSqrt(distanceSquared).isLessThanValue(radius);
  })).get();
}

All the other functions are available under a similar name (sqlSin, sqlCos, sqlAtan and so on). They have that sql prefix to avoid clashes with dart:math.

Migrating from moor_flutter to drift/native

First, adapt your pubspec.yaml: You can remove the moor_flutter dependency and instead add both the drift and sqlite3_flutter_libs dependencies:

dependencies:
 drift: ^2.19.1+1
 sqlite3_flutter_libs:
 sqflite: ^1.1.7 # Still used to obtain the database location
dev_dependencies:
 drift_dev: ^2.19.1

Adapt your imports:

  • In the file where you created a FlutterQueryExecutor, replace the moor_flutter import with package:drift/native.dart.
  • In all other files where you might have imported moor_flutter, just import package:drift/drift.dart.

Replace the executor. This code:

FlutterQueryExecutor.inDatabaseFolder(path: 'db.sqlite')

can now be written as

import 'package:sqflite/sqflite.dart' show getDatabasesPath;
import 'package:path/path.dart' as p;

LazyDatabase(() async {
  final dbFolder = await getDatabasesPath();
  final file = File(p.join(dbFolder, 'db.sqlite'));
  return NativeDatabase(file);
})

Note: If you haven't shipped a version with moor_flutter to your users yet, you can drop the dependency on sqflite. Instead, you can use path_provider which works on Desktop. Please be aware that FlutterQueryExecutor.inDatabaseFolder might yield a different folder than path_provider on Android. This can cause data loss if you've already shipped a version using moor_flutter. In that case, using getDatabasePath from sqflite is the suggested solution.