With dart, creating data classes can be tedious. You have the choice between:
- Writing them by hand, which is error-prone and time-consuming.
- Using code generation tools like
build_runner
withfreezed
, which become slow for large projects.
genq cuts through the wait and generates data classes for your Dart projects in milliseconds, not minutes. ⚡️
Inspired by freezed, genq offers a familiar syntax for defining data classes, but with a focus on lightning-fast performance.
The above image shows the difference in performance between build_runner
and genq
for generating 2500 data classes. See more details in the benchmarks section.
build_runner + freezed 🐌 |
genq 🚀 |
---|---|
In this benchmark (located in ./benchmarks
), count is the number of files in the benchmark, containing 250 annotated classes each. So for example, count=1 means 250 classes, count=2 means 500 classes, and so on. For count 10, build_runner
and freezed
took around 14.9 seconds, while genq
took 0.11 seconds. This is a >100x speedup!
- Never trust a benchmark you didn't falsify yourself.
- genq is optimized to perform one task and one task only - data class generation, whereas build_runner is built to do a lot more. Take this into account when choosing between the two.
Install genq via brew using the following command:
brew install jankuss/genq/genq
Or download the latest release from the releases page.
dependencies:
# ... Other dependencies ...
genq: ^0.3.0
import 'package:genq/genq.dart';
part 'user.genq.dart';
@genq
class User with _$User {
factory User({
required String name,
required int age,
}) = _User;
}
Read more about defining your data classes here.
Run the genq command in your project directory, and you will have your desired data classes generated in no time:
genq
To define a data class, you need to annotate the class with @genq
and provide a factory constructor with named parameters.
import 'package:genq/genq.dart'; // <- Import genq
part 'user.genq.dart'; // <- Add a part directive to the generated file
@genq // <- Annotate the class with @genq
class User with _$User { // <- Add the mixin _$<ClassName>
factory User({ // <- Define a factory constructor
required String name, // <- Define fields as named parameters
required int age,
}) = _User; // <- Redirecting constructor, _<ClassName>
}
The generated class will have the following methods:
copyWith
: Create a copy of the data class with modified fields.toString
: Generate a human-readable string representation of the data class.==
: Compare two data classes for equality.
To generate JSON serialization/deserialization code, you need to use the @Genq(json: true)
annotation instead of @genq
.
import 'package:genq/genq.dart';
part 'user.genq.dart';
@Genq(json: true)
class User with _$User {
factory User({
@JsonKey(name: 'full_name')
required String name,
required int age,
}) = _User;
}
This will generate two public functions, which you can use to serialize/deserialize the data class to/from JSON:
$UserFromJson(Map<String, dynamic> json) => /* ... */;
$UserToJson(User value) => /* ... */;
You can customize the generated JSON serialization/deserialization code for fields using the @JsonKey
annotation.
import 'package:genq/genq.dart';
part 'user.genq.dart';
@Genq(json: true)
class User with _$User {
factory User({
// Customizing the JSON key for the field 'name'. When deserializing, the value of 'full_name' will be assigned to the 'name' field.
@JsonKey(name: 'full_name')
required String name,
// Providing a default value for the field 'age'. If the field is not present in the JSON, the default value will be used.
@JsonKey(defaultValue: 99)
required int age,
}) = _User;
}
You can also provide custom fromJson
and toJson
functions for a field using the fromJson
and toJson
parameters of the @JsonKey
annotation.
import 'package:genq/genq.dart';
part 'user.genq.dart';
class UserName {
final String value;
UserName(this.value);
static UserName fromJson(String value) {
return UserName(value);
}
static String toJson(UserName value) {
return value.value;
}
}
@Genq(json: true)
class User with _$User {
factory User({
@JsonKey(
fromJson: UserName.fromJson,
toJson: UserName.toJson,
)
required UserName name,
required int age,
}) = _User;
}
You can provide a value for unknown enum values using the unknownEnumValue
parameter of the @JsonKey
annotation.
When deserializing and encountering an unknown value, the unknownEnumValue
will be used instead of throwing an exception.
import 'package:genq/genq.dart';
part 'user.genq.dart';
@GenqJsonEnum()
enum Role {
admin,
user,
unknown,
}
@Genq(json: true)
class User with _$User {
factory User({
@JsonKey(unknownEnumValue: Role.unknown)
required Role role,
}) = _User;
}
Enums are also supported for JSON serialization/deserialization. They need to be annotated with @GenqJsonEnum
.
import 'package:genq/genq.dart';
part 'user.genq.dart';
@GenqJsonEnum()
enum Role {
// You can annotate the enum values with @JsonValue to customize the JSON serialization/deserialization.
// For example, the string 'ADMIN' will get deserialized to the Role.admin value and vice versa.
// If you don't provide a value for @JsonValue, the enum key is used.
@JsonValue('ADMIN')
admin,
@JsonValue('USER')
user,
}
Similary to the @Genq(json: true) annotation, this will generate two public functions, which you can use to serialize/deserialize the enum to/from JSON:
$RoleFromJson(Object json) => /* ... */;
$RoleToJson(Role value) => /* ... */;
The fundamental idea behind the JSON codegen for it to be fast and efficient is to publicly expose the generated functions. Based on this, genq can assume for a type
T
that the functions $TFromJson
and $TToJson
are available, thus avoiding unnecessary traversal of other files.
genq uses its own subset parser of the dart language and generates code directly from the parsed AST. This allows genq to generate code much faster than build_runner
, which uses the analyzer
package. Code generation is also done in parallel for each file, which further speeds up the process.
The subset parser is written for the specific structures of data classes as defined here. Thus, there may be parsing errors if the code does not follow the expected structure. While the parser is generally robust when encountering unparsable code, there may be cases where it fails to parse the code correctly. If you encounter such a case, please open an issue with the code that caused the error.
One great thing: you don't have to choose! You can use both in your project. A good guidline would be: Use genq
for data class generation in your day-to-day development, and build_runner
for more complex code generation tasks.
If your project is sufficiently small, you might not even need genq
. However, if you find yourself or your team spending a lot of time waiting for build_runner
to generate code, genq
might be a good alternative.
- Editor support (VSCode, IntelliJ)
- Extensibility