Selects

This page describes how to write SELECT statements with drift's dart_api. To make examples easier to grasp, they're referencing two common tables forming the basis of a todo-list app:

class TodoItems extends Table {
  IntColumn get id => integer().autoIncrement()();
  TextColumn get title => text().withLength(min: 6, max: 32)();
  TextColumn get content => text().named('body')();
  IntColumn get category => integer().nullable().references(Categories, #id)();
}

@DataClassName('Category')
class Categories extends Table {
  IntColumn get id => integer().autoIncrement()();
  TextColumn get name => text()();
}

For each table you've specified in the @DriftDatabase annotation on your database class, a corresponding getter for a table will be generated. That getter can be used to run statements:

@DriftDatabase(tables: [TodoItems, Categories])
class MyDatabase extends _$MyDatabase {

  // the schemaVersion getter and the constructor from the previous page
  // have been omitted.

  // loads all todo entries
  Future<List<TodoItem>> get allTodoItems => select(todoItems).get();

  // watches all todo entries in a given category. The stream will automatically
  // emit new items whenever the underlying data changes.
  Stream<List<TodoItem>> watchEntriesInCategory(Category c) {
    return (select(todos)..where((t) => t.category.equals(c.id))).watch();
  }
}

Drift makes writing queries easy and safe. This page describes how to write basic select queries, but also explains how to use joins and subqueries for advanced queries.

Simple selects

You can create select statements by starting them with select(tableName), where the table name is a field generated for you by drift. Each table used in a database will have a matching field to run queries against. Any query can be run once with get() or be turned into an auto-updating stream using watch().

Where

You can apply filters to a query by calling where(). The where method takes a function that should map the given table to an Expression of boolean. A common way to create such expression is by using equals on expressions. Integer columns can also be compared with isBiggerThan and isSmallerThan. You can compose expressions using a & b, a | b and a.not(). For more details on expressions, see this guide.

Limit

You can limit the amount of results returned by calling limit on queries. The method accepts the amount of rows to return and an optional offset.

Future<List<TodoItem>> limitTodos(int limit, {int? offset}) {
  return (select(todoItems)..limit(limit, offset: offset)).get();
}

Ordering

You can use the orderBy method on the select statement. It expects a list of functions that extract the individual ordering terms from the table. You can use any expression as an ordering term - for more details, see this guide.

Future<List<TodoItem>> sortEntriesAlphabetically() {
  return (select(todoItems)
        ..orderBy([(t) => OrderingTerm(expression: t.title)]))
      .get();
}

You can also reverse the order by setting the mode property of the OrderingTerm to OrderingMode.desc.

Single values

If you know a query is never going to return more than one row, wrapping the result in a List can be tedious. Drift lets you work around that with getSingle and watchSingle:

Stream<TodoItem> entryById(int id) {
  return (select(todoItems)..where((t) => t.id.equals(id))).watchSingle();
}

If an entry with the provided id exists, it will be sent to the stream. Otherwise, null will be added to stream. If a query used with watchSingle ever returns more than one entry (which is impossible in this case), an error will be added instead.

Mapping

Before calling watch or get (or the single variants), you can use map to transform the result.

Stream<List<String>> contentWithLongTitles() {
  final query = select(todoItems)
    ..where((t) => t.title.length.isBiggerOrEqualValue(16));

  return query.map((row) => row.content).watch();
}

Deferring get vs watch

If you want to make your query consumable as either a Future or a Stream, you can refine your return type using one of the Selectable abstract base classes;

// Exposes `get` and `watch`
MultiSelectable<TodoItem> pageOfTodos(int page, {int pageSize = 10}) {
  return select(todoItems)..limit(pageSize, offset: page);
}

// Exposes `getSingle` and `watchSingle`
SingleSelectable<TodoItem> selectableEntryById(int id) {
  return select(todoItems)..where((t) => t.id.equals(id));
}

// Exposes `getSingleOrNull` and `watchSingleOrNull`
SingleOrNullSelectable<TodoItem> entryFromExternalLink(int id) {
  return select(todoItems)..where((t) => t.id.equals(id));
}

These base classes don't have query-building or map methods, signaling to the consumer that they are complete results.

Joins

Drift supports sql joins to write queries that operate on more than one table. To use that feature, start a select regular select statement with select(table) and then add a list of joins using .join(). For inner and left outer joins, a ON expression needs to be specified.

// We define a data class to contain both a todo entry and the associated
// category.
class EntryWithCategory {
  EntryWithCategory(this.entry, this.category);

  // The classes are generated by drift for each of the tables involved in the
  // join.
  final TodoItem entry;
  final Category? category;
}

// in the database class, we can then load the category for each entry
Stream<List<EntryWithCategory>> entriesWithCategory() {
  final query = select(todoItems).join([
    leftOuterJoin(categories, categories.id.equalsExp(todoItems.category)),
  ]);

  return query.watch().map((rows) {
    return rows.map((row) {
      return EntryWithCategory(
        row.readTable(todoItems),
        row.readTableOrNull(categories),
      );
    }).toList();
  });
}

Of course, you can also join multiple tables:

/// Searches for todo entries in the same category as the ones having
/// `titleQuery` in their titles.
Future<List<TodoItem>> otherTodosInSameCategory(String titleQuery) async {
  // Since we're adding the same table twice (once to filter for the title,
  // and once to find other todos in same category), we need a way to
  // distinguish the two tables. So, we're giving one of them a special name:
  final otherTodos = alias(todoItems, 'inCategory');

  final query = select(otherTodos).join([
    // In joins, `useColumns: false` tells drift to not add columns of the
    // joined table to the result set. This is useful here, since we only join
    // the tables so that we can refer to them in the where clause.
    innerJoin(categories, categories.id.equalsExp(otherTodos.category),
        useColumns: false),
    innerJoin(todoItems, todoItems.category.equalsExp(categories.id),
        useColumns: false),
  ])
    ..where(todoItems.title.contains(titleQuery));

  return query.map((row) => row.readTable(otherTodos)).get();
}

Parsing results

Calling get() or watch on a select statement with join returns a Future or Stream of List<TypedResult>, respectively. Each TypedResult represents a row from which data can be read. It contains a rawData getter to obtain the raw columns. But more importantly, the readTable method can be used to read a data class from a table.

In the example query above, we've read the todo entry and the category from each row like this:

return query.watch().map((rows) {
  return rows.map((row) {
    return EntryWithCategory(
      row.readTable(todoItems),
      row.readTableOrNull(categories),
    );
  }).toList();
});

Note: readTable will throw an ArgumentError when a table is not present in the row. For instance, todo entries might not be in any category. To account for that, we use row.readTableOrNull to load categories.

Custom columns

Select statements aren't limited to columns from tables. You can also include more complex expressions in the query. For each row in the result, those expressions will be evaluated by the database engine.

Future<List<(TodoItem, bool)>> loadEntries() {
  // assume that an entry is important if it has the string "important" somewhere in its content
  final isImportant = todoItems.content.like('%important%');

  return select(todoItems).addColumns([isImportant]).map((row) {
    final entry = row.readTable(todoItems);
    final entryIsImportant = row.read(isImportant)!;

    return (entry, entryIsImportant);
  }).get();
}

Note that the like check is not performed in Dart - it's sent to the underlying database engine which can efficiently compute it for all rows.

Aliases

Sometimes, a query references a table more than once. Consider the following example to store saved routes for a navigation system:

class GeoPoints extends Table {
  IntColumn get id => integer().autoIncrement()();
  TextColumn get name => text()();
  TextColumn get latitude => text()();
  TextColumn get longitude => text()();
}

class Routes extends Table {

  IntColumn get id => integer().autoIncrement()();
  TextColumn get name => text()();

  // contains the id for the start and destination geopoint.
  IntColumn get start => integer()();
  IntColumn get destination => integer()();
}

Now, let's say we wanted to also load the start and destination GeoPoint object for each route. We'd have to use a join on the geo-points table twice: For the start and destination point. To express that in a query, aliases can be used:

class RouteWithPoints {
  final Route route;
  final GeoPoint start;
  final GeoPoint destination;

  RouteWithPoints({this.route, this.start, this.destination});
}

// inside the database class:
Future<List<RouteWithPoints>> loadRoutes() async {
  // create aliases for the geoPoints table so that we can reference it twice
  final start = alias(geoPoints, 's');
  final destination = alias(geoPoints, 'd');

  final rows = await select(routes).join([
    innerJoin(start, start.id.equalsExp(routes.start)),
    innerJoin(destination, destination.id.equalsExp(routes.destination)),
  ]).get();

  return rows.map((resultRow) {
    return RouteWithPoints(
      route: resultRow.readTable(routes),
      start: resultRow.readTable(start),
      destination: resultRow.readTable(destination),
    );
  }).toList();
}

The generated statement then looks like this:

SELECT
    routes.id, routes.name, routes.start, routes.destination,
    s.id, s.name, s.latitude, s.longitude,
    d.id, d.name, d.latitude, d.longitude
FROM routes
    INNER JOIN geo_points s ON s.id = routes.start
    INNER JOIN geo_points d ON d.id = routes.destination

ORDER BY and WHERE on joins

Similar to queries on a single table, orderBy and where can be used on joins too. The initial example from above is expanded to only include todo entries with a specified filter and to order results based on the category's id:

Stream<List<EntryWithCategory>> entriesWithCategory(String entryFilter) {
  final query = select(todos).join([
    leftOuterJoin(categories, categories.id.equalsExp(todos.category)),
  ]);
  query.where(todos.content.like(entryFilter));
  query.orderBy([OrderingTerm.asc(categories.id)]);
  // ...
}

As a join can have more than one table, all tables in where and orderBy have to be specified directly (unlike the callback on single-table queries that gets called with the right table by default).

Group by

Sometimes, you need to run queries that aggregate data, meaning that data you're interested in comes from multiple rows. Common questions include

• how many todo entries are in each category?
• how many entries did a user complete each month?
• what's the average length of a todo entry?

What these queries have in common is that data from multiple rows needs to be combined into a single row. In sql, this can be achieved with "aggregate functions", for which drift has builtin support.

Additional info: A good tutorial for group by in sql is available here.

To write a query that answers the first question for us, we can use the count function. We're going to select all categories and join each todo entry for each category. What's special is that we set useColumns: false on the join. We do that because we're not interested in the columns of the todo item. We only care about how many there are. By default, drift would attempt to read each todo item when it appears in a join.

Future<void> countTodosInCategories() async {
  final amountOfTodos = todoItems.id.count();

  final query = select(categories).join([
    innerJoin(
      todoItems,
      todoItems.category.equalsExp(categories.id),
      useColumns: false,
    )
  ]);
  query
    ..addColumns([amountOfTodos])
    ..groupBy([categories.id]);

  final result = await query.get();

  for (final row in result) {
    print('there are ${row.read(amountOfTodos)} entries in'
        '${row.readTable(categories)}');
  }
}

To find the average length of a todo entry, we use avg. In this case, we don't even have to use a join since all the data comes from a single table (todos). That's a problem though - in the join, we used useColumns: false because we weren't interested in the columns of each todo item. Here we don't care about an individual item either, but there's no join where we could set that flag. Drift provides a special method for this case - instead of using select, we use selectOnly. The "only" means that drift will only report columns we added via "addColumns". In a regular select, all columns from the table would be selected, which is what you'd usually need.

Stream<double> averageItemLength() {
  final avgLength = todoItems.content.length.avg();
  final query = selectOnly(todoItems)..addColumns([avgLength]);

  return query.map((row) => row.read(avgLength)!).watchSingle();
}

Using selects as inserts

In SQL, an INSERT INTO SELECT statement can be used to efficiently insert the rows from a SELECT statement into a table. It is possible to construct these statements in drift with the insertFromSelect method. This example shows how that method is used to construct a statement that creates a new category for each todo entry that didn't have one assigned before:

Future<void> createCategoryForUnassignedTodoEntries() async {
  final newDescription = Variable<String>('category for: ') + todoItems.title;
  final query = selectOnly(todoItems)
    ..where(todoItems.category.isNull())
    ..addColumns([newDescription]);

  await into(categories).insertFromSelect(query, columns: {
    categories.name: newDescription,
  });
}

The first parameter for insertFromSelect is the select statement statement to use as a source. Then, the columns map maps columns from the table in which rows are inserted to columns from the select statement. In the example, the newDescription expression as added as a column to the query. Then, the map entry categories.description: newDescription is used so that the description column for new category rows gets set to that expression.

Subqueries

Starting from drift 2.11, you can use Subquery to use an existing select statement as part of more complex join.

This snippet uses Subquery to count how many of the top-10 todo items (by length of their title) are in each category. It does this by first creating a select statement for the top-10 items (but not executing it), and then joining this select statement onto a larger one grouping by category:

Future<List<(Category, int)>> amountOfLengthyTodoItemsPerCategory() async {
  final longestTodos = Subquery(
    select(todoItems)
      ..orderBy([(row) => OrderingTerm.desc(row.title.length)])
      ..limit(10),
    's',
  );

  // In the main query, we want to count how many entries in longestTodos were
  // found for each category. But we can't access todos.title directly since
  // we're not selecting from `todos`. Instead, we'll use Subquery.ref to read
  // from a column in a subquery.
  final itemCount = longestTodos.ref(todoItems.title).count();
  final query = select(categories).join(
    [
      innerJoin(
        longestTodos,
        // Again using .ref() here to access the category in the outer select
        // statement.
        longestTodos.ref(todoItems.category).equalsExp(categories.id),
        useColumns: false,
      )
    ],
  )
    ..addColumns([itemCount])
    ..groupBy([categories.id]);

  final rows = await query.get();

  return [
    for (final row in rows) (row.readTable(categories), row.read(itemCount)!),
  ];
}

Any statement can be used as a subquery. But be aware that, unlike subquery expressions, full subqueries can't use tables from the outer select statement.

JSON support

sqlite3 has great support for JSON operators that are also available in drift (under the additional 'package:drift/extensions/json1.dart' import). JSON support is helpful when storing a dynamic structure that is best represented with JSON, or when you have an existing structure (perhaps because you're migrating from a document-based storage) that you need to support.

As an example, consider a contact book application that started with a JSON structure to store contacts:

import 'dart:convert';

import 'package:drift/drift.dart';
import 'package:drift/extensions/json1.dart';
import 'package:json_annotation/json_annotation.dart';

@JsonSerializable()
class ContactData {
  final String name;
  final List<String> phoneNumbers;

  ContactData(this.name, this.phoneNumbers);

  factory ContactData.fromJson(Map<String, Object?> json) =>
      _$ContactDataFromJson(json);

  Map<String, Object?> toJson() => _$ContactDataToJson(this);
}

To easily store this contact representation in a drift database, one could use a JSON column:

class _ContactsConverter extends TypeConverter<ContactData, String> {
  @override
  ContactData fromSql(String fromDb) {
    return ContactData.fromJson(json.decode(fromDb) as Map<String, Object?>);
  }

  @override
  String toSql(ContactData value) {
    return json.encode(value.toJson());
  }
}

class Contacts extends Table {
  IntColumn get id => integer().autoIncrement()();
  TextColumn get data => text().map(_ContactsConverter())();

  TextColumn get name => text().generatedAs(data.jsonExtract(r'$.name'))();
}

Note the name column as well: It uses generatedAs with the jsonExtract function to extract the name field from the JSON value on the fly. The full syntax for JSON path arguments is explained on the sqlite3 website.

To make the example more complex, let's look at another table storing a log of phone calls:

class Calls extends Table {
  IntColumn get id => integer().autoIncrement()();
  BoolColumn get incoming => boolean()();
  TextColumn get phoneNumber => text()();
  DateTimeColumn get callTime => dateTime()();
}

Let's say we wanted to find the contact for each call, if there is any with a matching phone number. For this to be expressible in SQL, each contacts row would somehow have to be expanded into a row for each stored phone number. Luckily, the json_each function in sqlite3 can do exactly that, and drift exposes it:

Future<List<(Call, Contact)>> callsWithContact() async {
  final phoneNumbersForContact =
      contacts.data.jsonEach(this, r'$.phoneNumbers');
  final phoneNumberQuery = selectOnly(phoneNumbersForContact)
    ..addColumns([phoneNumbersForContact.value]);

  final query = select(calls).join(
      [innerJoin(contacts, calls.phoneNumber.isInQuery(phoneNumberQuery))]);

  return query
      .map((row) => (row.readTable(calls), row.readTable(contacts)))
      .get();
}

Selects without tables

Some queries don't need a FROM clause at all and instead just select some expressions directly. An example for this may be a select that just uses subquery expressions, like here to query whether any rows exist in a table:

Future<bool> hasTodoItem() async {
  final todoItemExists = existsQuery(select(todoItems));
  final row = await selectExpressions([todoItemExists]).getSingle();
  return row.read(todoItemExists)!;
}

The selectExpressions API is similar to selectOnly, except that it doesn't require any table at all. Instead, the expressions in the list passed to selectExpressions are evaluated in a standalone select statement and can be parsed from the TypedResult class returned when evaluating the query.