Edit Source File Issue

JSON and serialization

It is hard to think of a mobile app that doesn’t need to communicate with a web server or easily store structured data at some point. When making network-connected apps, the chances are that we need to consume some good old JSON, sooner or later.

In this tutorial, we look into ways of using JSON with Flutter. We go over what JSON solution to use in different scenarios and why.

Which JSON serialization method is right for me?

This article covers two general strategies for working with JSON:

  • Manual serialization and deserialization
  • Automated serialization and deserialization via code generation

Different projects come with different complexities and use cases. For smaller proof-of-concept projects or quick prototypes, using code generators might be overkill. For apps with several JSON models with more complexity, doing serialization by hand can quickly become tedious, repetitive and lend itself to lots of small errors.

Use manual serialization for smaller projects

Manual JSON serialization refers to using the built-in JSON decoder in dart:convert. It involves passing the raw JSON string to the json.decode() method, and then looking up the values you need in the Map<String, dynamic> the method returns. It has no external dependencies or particular setup process, and it is good for quick proof of concepts.

Where the manual serialization does not perform well is when your project becomes bigger. Writing the serialization logic by hand can become hard to manage and error-prone. If you have a typo when accessing an unexisting JSON field, your code throws an error during runtime.

If you do not have many JSON models in your project and are looking to test a concept quickly, manual serialization might be the way you want to start. For an example of manual serialization, see here.

Use code generation for medium to large projects

JSON serialization with code generation means having an external library generate the serialization boilerplate for you. They involve some initial setup and running a file watcher that generates the code from your model classes. For example, json_serializable and built_value are these kinds of libraries.

This approach scales well for a larger project. There is no hand-written boilerplate needed, and typos when accessing JSON fields are caught at compile-time. The downside with code generation is that it involves some initial setup. Also, the generated source files may produce visual clutter in your project navigator

You might want to use generated code for JSON serialization when you have a medium or a larger project. To see an example of code generation based JSON serialization, see here.

Is there a GSON/Jackson/Moshi equivalent in Flutter?

The simple answer is no.

Such a library would require using runtime reflection, which is disabled in Flutter. Runtime reflection interferes with tree shaking, which Dart has supported for quite a long time. With tree shaking, we can “shake off” unused code from our release builds. This allows us to optimize the size of our applications significantly.

Since reflection makes all code implicitly used by default, it makes tree shaking difficult. The tools cannot know what parts are unused at runtime, so the redundant code is hard to strip away. App sizes cannot be easily optimized when using reflection.

Although we cannot use runtime reflection with Flutter, some libraries give us similarly easy to use APIs but are based on code generation instead. This approach is covered in more detail in the code generation libraries section.

Serializing JSON manually using dart:convert

Basic JSON serialization in Flutter is very simple. Flutter has a built-in dart:convert library, which includes a straightforward JSON encoder and decoder.

Here is an example JSON for a simple user model.

{
  "name": "John Smith",
  "email": "john@example.com"
}

With dart:convert, we can serialize this JSON model in two ways. Let’s have a look at both.

Serializing JSON inline

By looking at the dart:convert JSON documentation, we see that we can decode the JSON by calling the json.decode method, with our JSON string as the method argument.

Map<String, dynamic> user = json.decode(json);

print('Howdy, ${user['name']}!');
print('We sent the verification link to ${user['email']}.');

Unfortunately, json.decode() merely returns a Map<String, dynamic>, meaning that we do not know the types of the values until runtime. With this approach, we lose most of the statically typed language features: type safety, autocompletion and most importantly, compile-time exceptions. Our code can become instantly more error-prone.

For example, whenever we access the name or email fields, we could quickly introduce a typo. A typo which our compiler does not know of since our entire JSON merely lives in a map structure.

Serializing JSON inside model classes

We can combat the previously mentioned problems by introducing a plain model class, which we call User. Inside the User class, we have:

  • a User.fromJson constructor, for constructing a new User instance from a map structure
  • a toJson method, which converts a User instance into a map.

This way, the calling code can now have type safety, autocompletion for the name and email fields and compile-time exceptions. If we make typos or treat the fields as ints instead of Strings, our app will not even compile, instead of crashing at runtime.

user.dart

class User {
  final String name;
  final String email;

  User(this.name, this.email);

  User.fromJson(Map<String, dynamic> json)
      : name = json['name'],
        email = json['email'];

  Map<String, dynamic> toJson() =>
    {
      'name': name,
      'email': email,
    };
}

Now the responsibility of the serialization logic is moved inside the model itself. With this new approach, we can deserialize a user quite easily.

Map userMap = json.decode(json);
var user = new User.fromJson(userMap);

print('Howdy, ${user.name}!');
print('We sent the verification link to ${user.email}.');

To serialize a user, we just pass the User object to the json.encode method. We don’t need to call the toJson method here, since json.encode already does it for us.

String json = json.encode(user);

This way, the calling code does not have to worry about JSON serialization at all. However, the model class still definitely has to. In a production app, we would want to be sure that the serialization works properly. In practice, the User.fromJson and User.toJson methods both need to have unit tests in place to verify correct behavior.

Also, real-world scenarios are not usually that simple. It is unlikely that we can get by with such small JSON responses. Nested JSON objects are not that uncommon either.

It would be nice if there were something that handled the JSON serialization for us. Luckily, there is!

Serializing JSON using code generation libraries

Although there are other libraries available, in this tutorial, we use the json_serializable package. It is an automated source code generator that can generate the JSON serialization boilerplate for us.

Since the serialization code is not handwritten and maintained by us anymore, we minimize the risk of having JSON serialization exceptions at runtime.

Setting up json_serializable in a project

To include json_serializable in our project, we need one regular and two dev dependencies. In short, dev dependencies are dependencies that are not included in our app source code.

The latest versions of these required dependencies can be seen by following this link.

pubspec.yaml

dependencies:
  # Your other regular dependencies here
  json_annotation: ^0.2.3

dev_dependencies:
  # Your other dev_dependencies here
  build_runner: ^0.8.0
  json_serializable: ^0.5.0

Run flutter packages get inside your project root folder (or click “Packages Get” in your editor) to make these new dependencies available in your project.

Creating model classes the json_serializable way

Let’s see how to convert our User class to a json_serializable one. For the sake of simplicity, we use the dumbed-down JSON model from the previous samples.

user.dart

import 'package:json_annotation/json_annotation.dart';

/// This allows our `User` class to access private members in 
/// the generated file. The value for this is *.g.dart, where 
/// the star denotes the source file name.
part 'user.g.dart';

/// An annotation for the code generator to know that this class needs the 
/// JSON serialization logic to be generated.
@JsonSerializable()

/// Every json_serializable class must have the serializer mixin. 
/// It makes the generated toJson() method to be usable for the class. 
/// The mixin's name follows the source class, in this case, User.
class User extends Object with _$UserSerializerMixin {
  User(this.name, this.email);

  String name;
  String email;

  /// A necessary factory constructor for creating a new User instance
  /// from a map. We pass the map to the generated _$UserFromJson constructor. 
  /// The constructor is named after the source class, in this case User.
  factory User.fromJson(Map<String, dynamic> json) => _$UserFromJson(json);
}

With this setup, the source code generator will generate code for serializing the name and email fields from JSON and back.

If needed, it is also easy to customize the naming strategy. For example, if the API we are working with returns objects with snake_case, and we want to use lowerCamelCase in our models, we can use the @JsonKey annotation with a name parameter:

/// Tell json_serializable that "registration_date_millis" should be
/// mapped to this property.
@JsonKey(name: 'registration_date_millis')
final int registrationDateMillis;

Running the code generation utility

When creating json_serializable classes the first time, you will get errors similar to the image below.

IDE warning when the generated code for a model class does not exist yet.

These errors are entirely normal and are simply because the generated code for the model class does not exist yet. To resolve this, we must run the code generator that generates the serialization boilerplate for us.

There are two ways of running the code generator.

One-time code generation

By running flutter packages pub run build_runner build in our project root, we can generate json serialization code for our models whenever needed. This triggers a one-time build which goes through our source files, picks the relevant ones and generates the necessary serialization code for them.

While this is pretty convenient, it would nice if we did not have to run the build manually every time we make changes in our model classes.

Generating code continuously

A watcher can make our source code generation process more convenient. It watches changes in our project files and automatically builds the necessary files when needed. We can start the watcher by running flutter packages pub run build_runner watch in our project root.

It is safe to start the watcher once and leave it running in the background.

Consuming json_serializable models

To deserialize a JSON string json_serializable way, we do not have actually to make any changes to our previous code.

Map userMap = json.decode(json);
var user = new User.fromJson(userMap);

Same goes for serialization. The calling API is the same as before.

String json = json.encode(user);

With json_serializable, we can forget any manual JSON serialization in the User class. The source code generator creates a file called user.g.dart, which has all the necessary serialization logic. Now we do not necessarily have to write automated tests to be sure that the serialization works - it is now the library’s responsibility to make sure the serialization works appropriately.

Further references