Skip to content

Latest commit

 

History

History
447 lines (372 loc) · 16.6 KB

Flutter-macros.md

File metadata and controls

447 lines (372 loc) · 16.6 KB

2024 Flutter 重大更新,Dart 宏(Macros)编程开始支持,JSON 序列化有救

说起宏编程可能大家并不陌生,但是这对于 Flutter 和 Dart 开发者来说它一直是一个「遗憾」,这个「遗憾」体现在编辑过程的代码修改支持上,其中最典型的莫过于 Dart 的 JSON 序列化。

举个例子,目前 Dart 语言的 JSON 序列化高度依赖 build_runner 去生成 Dart 代码,例如在实际使用中我们需要:

  • 依赖 json_serializable ,通过注解声明一个 Event 对象
  • 运行 flutter packages pub run build_runner build 生成文件
  • 得到 Event.g.dart 文件,在项目中使用它去实现 JSON 的序列化和反序列化

这里最大的问题在于,我们需要通过命令行去生成一个项目文件,并且这个文件我们还可以随意手动修改,从开发角度来说,这并不优雅也不方便。

而宏声明是用户定义的 Dart 类,它可以实现一个或多个新的内置宏接口,Dart 中的宏是用正常的命令式 Dart 代码来开发,不存在单独的“宏语言”

大多数宏并不是简单地从头开始生成新代码,而是根据程序的现有属性去添加代码,例如向 Class 添加 JSON 序列化的宏,可能会查看 Class 声明的字段,并从中合成一个 toJson() ,将这些字段序列化为 JSON 对象。

我们首先看一段官方的 Demo , 如下代码所示,可以看到 :

  • MyState 添加了一个自定义的 @AutoDispose() 注解,这是一个开发者自己实现的宏声明,并且继承了 State 对象,带有 dispose 方法。
  • MyState 里有多个 a a2bc 三个对象,其中 a a2b 都实现了 Disposable 接口,都有 dispose 方法
  • 虽然 a a2bMyStatedispose(); 方法来自不同基类实现,但是基于 @AutoDispose() 的实现,在代码调用 state.dispose(); 时, a a2b 变量的 dispose 方法也会被同步调用
import 'package:macro_proposal/auto_dispose.dart';

void main() {
  var state = MyState(a: ADisposable(), b: BDisposable(), c: 'hello world');
  state.dispose();
}

@AutoDispose()
class MyState extends State {
  final ADisposable a;
  final ADisposable? a2;
  final BDisposable b;
  final String c;

  MyState({required this.a, this.a2, required this.b, required this.c});

  @override
  String toString() => 'MyState!';
}

class State {
  void dispose() {
    print('disposing of $this');
  }
}

class ADisposable implements Disposable {
  void dispose() {
    print('disposing of ADisposable');
  }
}

class BDisposable implements Disposable {
  void dispose() {
    print('disposing of BDisposable');
  }
}

如下图所示,可以看到,尽管 MyState 没用主动调用 a a2b 变量的 dispose 方法,并且它们和 MyStatedispose 也来自不同基类,但是最终执行所有 dispose 方法都被成功调用,这就是@AutoDispose() 的宏声明实现在编译时对代码进行了调整。

如下图所示是 @AutoDispose() 的宏编程实现,其中 macro 就是一个标志性的宏关键字,剩下的代码可以看到基本就是 dart 脚本的实现, macro 里主要是实现 ClassDeclarationsMacrobuildDeclarationsForClass方法,如下代码可以很直观看到关于 super.dispose();disposeCalls 的相关实现。

import 'package:_fe_analyzer_shared/src/macros/api.dart';

// Interface for disposable things.
abstract class Disposable {
  void dispose();
}

macro class AutoDispose implements ClassDeclarationsMacro, ClassDefinitionMacro {
  const AutoDispose();

  @override
  void buildDeclarationsForClass(
      ClassDeclaration clazz, MemberDeclarationBuilder builder) async {
    var methods = await builder.methodsOf(clazz);
    if (methods.any((d) => d.identifier.name == 'dispose')) {
      // Don't need to add the dispose method, it already exists.
      return;
    }

    builder.declareInType(DeclarationCode.fromParts([
      // TODO: Remove external once the CFE supports it.
      'external void dispose();',
    ]));
  }

  @override
  Future<void> buildDefinitionForClass(
      ClassDeclaration clazz, TypeDefinitionBuilder builder) async {
    var disposableIdentifier =
        // ignore: deprecated_member_use
        await builder.resolveIdentifier(
            Uri.parse('package:macro_proposal/auto_dispose.dart'),
            'Disposable');
    var disposableType = await builder
        .resolve(NamedTypeAnnotationCode(name: disposableIdentifier));

    var disposeCalls = <Code>[];
    var fields = await builder.fieldsOf(clazz);
    for (var field in fields) {
      var type = await builder.resolve(field.type.code);
      if (!await type.isSubtypeOf(disposableType)) continue;
      disposeCalls.add(RawCode.fromParts([
        '\n',
        field.identifier,
        if (field.type.isNullable) '?',
        '.dispose();',
      ]));
    }

    // Augment the dispose method by injecting all the new dispose calls after
    // either a call to `augmented()` or `super.dispose()`, depending on if
    // there already is an existing body to call.
    //
    // If there was an existing body, it is responsible for calling
    // `super.dispose()`.
    var disposeMethod = (await builder.methodsOf(clazz))
        .firstWhere((method) => method.identifier.name == 'dispose');
    var disposeBuilder = await builder.buildMethod(disposeMethod.identifier);
    disposeBuilder.augment(FunctionBodyCode.fromParts([
      '{\n',
      if (disposeMethod.hasExternal || !disposeMethod.hasBody)
        'super.dispose();'
      else
        'augmented();',
      ...disposeCalls,
      '}',
    ]));
  }
}

到这里大家应该可以直观感受到宏编程的魅力,上述 Demo 来自 dart-languagemacros/example/auto_dispose_main ,其中 bin/ 目录下的代码是运行的脚本示例,lib/ 目录下的代码是宏编程实现的示例:

https://github.com/dart-lang/language/tree/main/working/macros/example

当然,因为现在是实验性阶段,API 和稳定性还有待商榷,所以想运行这些 Demo 还需要一些额外的处理,比如版本强关联,例如上述的 auto_dispose_main 例子:

另外,还有一个第三方例子是来自 millsteedmacros ,这是一个简单的 JSON 序列化实现 Demo ,并且可以直接不用额外下载 dark-sdk,通过某个 flutter 内置 dart-sdk 版本就可以满足条件:3.19.0-12.0.pre

在本地 Flutter 目录下,切换到 git checkout 3.19.0-12.0.pre ,然后执行 flutter doctor 初始化 dark sdk 即可。

代码的实现很简单,首先看 bin 下的示例,通过 @Model() GetUsersResponseUser 声明为 JSON 对象,然后在运行时,宏编程会自动添加 fromJsontoJson 方式。

import 'dart:convert';

import 'package:macros/model.dart';

@Model()
class User {
  User({
    required this.username,
    required this.password,
  });

  final String username;
  final String password;
}

@Model()
class GetUsersResponse {
  GetUsersResponse({
    required this.users,
    required this.pageNumber,
    required this.pageSize,
  });

  final List<User> users;
  final int pageNumber;
  final int pageSize;
}

void main() {
  const body = '''
    {
      "users": [
        {
          "username": "ramon",
          "password": "12345678"
        }
      ],
      "pageNumber": 1,
      "pageSize": 30
    }
  ''';
  final json = jsonDecode(body) as Map<String, dynamic>;
  final response = GetUsersResponse.fromJson(json);
  final ramon = response.users.first;
  final millsteed = ramon.copyWith(username: 'millsteed', password: '87654321');
  final newResponse = response.copyWith(users: [...response.users, millsteed]);
  print(const JsonEncoder.withIndent('  ').convert(newResponse));
}

Model 的宏实现就相对复杂一些,但是实际上就是将类似 freezed/ json_serializable 是实现调整到宏实现了,而最终效果就是,开发者使用起来更加优雅了。

// ignore_for_file: depend_on_referenced_packages, implementation_imports

import 'dart:async';

import 'package:_fe_analyzer_shared/src/macros/api.dart';

macro class Model implements ClassDeclarationsMacro {
  const Model();

  static const _baseTypes = ['bool', 'double', 'int', 'num', 'String'];
  static const _collectionTypes = ['List'];

  @override
  Future<void> buildDeclarationsForClass(
    ClassDeclaration classDeclaration,
    MemberDeclarationBuilder builder,
  ) async {
    final className = classDeclaration.identifier.name;

    final fields = await builder.fieldsOf(classDeclaration);

    final fieldNames = <String>[];
    final fieldTypes = <String, String>{};
    final fieldGenerics = <String, List<String>>{};

    for (final field in fields) {
      final fieldName = field.identifier.name;
      fieldNames.add(fieldName);

      final fieldType = (field.type.code as NamedTypeAnnotationCode).name.name;
      fieldTypes[fieldName] = fieldType;

      if (_collectionTypes.contains(fieldType)) {
        final generics = (field.type.code as NamedTypeAnnotationCode)
            .typeArguments
            .map((e) => (e as NamedTypeAnnotationCode).name.name)
            .toList();
        fieldGenerics[fieldName] = generics;
      }
    }

    final fieldTypesWithGenerics = fieldTypes.map(
      (name, type) {
        final generics = fieldGenerics[name];
        return MapEntry(
          name,
          generics == null ? type : '$type<${generics.join(', ')}>',
        );
      },
    );

    _buildFromJson(builder, className, fieldNames, fieldTypes, fieldGenerics);
    _buildToJson(builder, fieldNames, fieldTypes);
    _buildCopyWith(builder, className, fieldNames, fieldTypesWithGenerics);
    _buildToString(builder, className, fieldNames);
    _buildEquals(builder, className, fieldNames);
    _buildHashCode(builder, fieldNames);
  }

  void _buildFromJson(
    MemberDeclarationBuilder builder,
    String className,
    List<String> fieldNames,
    Map<String, String> fieldTypes,
    Map<String, List<String>> fieldGenerics,
  ) {
    final code = [
      'factory $className.fromJson(Map<String, dynamic> json) {'.indent(2),
      'return $className('.indent(4),
      for (final fieldName in fieldNames) ...[
        if (_baseTypes.contains(fieldTypes[fieldName])) ...[
          "$fieldName: json['$fieldName'] as ${fieldTypes[fieldName]},"
              .indent(6),
        ] else if (_collectionTypes.contains(fieldTypes[fieldName])) ...[
          "$fieldName: (json['$fieldName'] as List<dynamic>)".indent(6),
          '.whereType<Map<String, dynamic>>()'.indent(10),
          '.map(${fieldGenerics[fieldName]?.first}.fromJson)'.indent(10),
          '.toList(),'.indent(10),
        ] else ...[
          '$fieldName: ${fieldTypes[fieldName]}'
                  ".fromJson(json['$fieldName'] "
                  'as Map<String, dynamic>),'
              .indent(6),
        ],
      ],
      ');'.indent(4),
      '}'.indent(2),
    ].join('\n');
    builder.declareInType(DeclarationCode.fromString(code));
  }

  void _buildToJson(
    MemberDeclarationBuilder builder,
    List<String> fieldNames,
    Map<String, String> fieldTypes,
  ) {
    final code = [
      'Map<String, dynamic> toJson() {'.indent(2),
      'return {'.indent(4),
      for (final fieldName in fieldNames) ...[
        if (_baseTypes.contains(fieldTypes[fieldName])) ...[
          "'$fieldName': $fieldName,".indent(6),
        ] else if (_collectionTypes.contains(fieldTypes[fieldName])) ...[
          "'$fieldName': $fieldName.map((e) => e.toJson()).toList(),".indent(6),
        ] else ...[
          "'$fieldName': $fieldName.toJson(),".indent(6),
        ],
      ],
      '};'.indent(4),
      '}'.indent(2),
    ].join('\n');
    builder.declareInType(DeclarationCode.fromString(code));
  }

  void _buildCopyWith(
    MemberDeclarationBuilder builder,
    String className,
    List<String> fieldNames,
    Map<String, String> fieldTypes,
  ) {
    final code = [
      '$className copyWith({'.indent(2),
      for (final fieldName in fieldNames) ...[
        '${fieldTypes[fieldName]}? $fieldName,'.indent(4),
      ],
      '}) {'.indent(2),
      'return $className('.indent(4),
      for (final fieldName in fieldNames) ...[
        '$fieldName: $fieldName ?? this.$fieldName,'.indent(6),
      ],
      ');'.indent(4),
      '}'.indent(2),
    ].join('\n');
    builder.declareInType(DeclarationCode.fromString(code));
  }

  void _buildToString(
    MemberDeclarationBuilder builder,
    String className,
    List<String> fieldNames,
  ) {
    final code = [
      '@override'.indent(2),
      'String toString() {'.indent(2),
      "return '$className('".indent(4),
      for (final fieldName in fieldNames) ...[
        if (fieldName != fieldNames.last) ...[
          "'$fieldName: \$$fieldName, '".indent(8),
        ] else ...[
          "'$fieldName: \$$fieldName'".indent(8),
        ],
      ],
      "')';".indent(8),
      '}'.indent(2),
    ].join('\n');
    builder.declareInType(DeclarationCode.fromString(code));
  }

  void _buildEquals(
    MemberDeclarationBuilder builder,
    String className,
    List<String> fieldNames,
  ) {
    final code = [
      '@override'.indent(2),
      'bool operator ==(Object other) {'.indent(2),
      'return other is $className &&'.indent(4),
      'runtimeType == other.runtimeType &&'.indent(8),
      for (final fieldName in fieldNames) ...[
        if (fieldName != fieldNames.last) ...[
          '$fieldName == other.$fieldName &&'.indent(8),
        ] else ...[
          '$fieldName == other.$fieldName;'.indent(8),
        ],
      ],
      '}'.indent(2),
    ].join('\n');
    builder.declareInType(DeclarationCode.fromString(code));
  }

  void _buildHashCode(
    MemberDeclarationBuilder builder,
    List<String> fieldNames,
  ) {
    final code = [
      '@override'.indent(2),
      'int get hashCode {'.indent(2),
      'return Object.hash('.indent(4),
      'runtimeType,'.indent(6),
      for (final fieldName in fieldNames) ...[
        '$fieldName,'.indent(6),
      ],
      ');'.indent(4),
      '}'.indent(2),
    ].join('\n');
    builder.declareInType(DeclarationCode.fromString(code));
  }
}

extension on String {
  String indent(int length) {
    final space = StringBuffer();
    for (var i = 0; i < length; i++) {
      space.write(' ');
    }
    return '$space$this';
  }
}

目前宏还处于试验性质的阶段,所以 API 还在调整,这也是为什么上面的例子需要指定 dart 版本的原因,另外宏目前规划里还有一些要求,例如

  • 所有宏构造函数都必须标记为 const
  • 所有宏必须至少实现其中一个 Macro 接口
  • 宏不能是抽象对象
  • 宏 class 不能由其他宏生成
  • 宏 class 不能包含泛型类型参数
  • 每个宏接口都需要声明宏类必须实现的方法,例如,在声明阶段应用的 ClassDeclarationsMacro 及其buildDeclarationsForClass 方法。

未来规划里,宏 API 可能会作为 Pub 包提供,通过库 dart:_macros 来提供支持 ,具体还要等正式发布时 dart 团队的决策。

总的来说,这对于 dart 和 flutter 是一个重大的厉害消息,虽然宏编程并不是什么新鲜概念,该是 dart 终于可以优雅地实现 JSON 序列化,并且还是用 dart 来实现,这对于 flutter 开发者来说,无疑是最好的新年礼物。

所以,新年快乐~我们节后再见~