Moshi

Moshi is a modern JSON library for Android and Java. It makes it easy to parse JSON into Java objects:

String json = ...;

Moshi moshi = new Moshi.Builder().build();
JsonAdapter<BlackjackHand> jsonAdapter = moshi.adapter(BlackjackHand.class);

BlackjackHand blackjackHand = jsonAdapter.fromJson(json);
System.out.println(blackjackHand);

And it can just as easily serialize Java objects as JSON:

BlackjackHand blackjackHand = new BlackjackHand(
    new Card('6', SPADES),
    Arrays.asList(new Card('4', CLUBS), new Card('A', HEARTS)));

Moshi moshi = new Moshi.Builder().build();
JsonAdapter<BlackjackHand> jsonAdapter = moshi.adapter(BlackjackHand.class);

String json = jsonAdapter.toJson(blackjackHand);
System.out.println(json);

Built-in Type Adapters

Moshi has built-in support for reading and writing Java’s core data types:

• Primitives (int, float, char...) and their boxed counterparts (Integer, Float, Character...).
• Arrays, Collections, Lists, Sets, and Maps
• Strings
• Enums

It supports your model classes by writing them out field-by-field. In the example above Moshi uses these classes:

class BlackjackHand {
  public final Card hidden_card;
  public final List<Card> visible_cards;
  ...
}

class Card {
  public final char rank;
  public final Suit suit;
  ...
}

enum Suit {
  CLUBS, DIAMONDS, HEARTS, SPADES;
}

to read and write this JSON:

{
  "hidden_card": {
    "rank": "6",
    "suit": "SPADES"
  },
  "visible_cards": [
    {
      "rank": "4",
      "suit": "CLUBS"
    },
    {
      "rank": "A",
      "suit": "HEARTS"
    }
  ]
}

The Javadoc catalogs the complete Moshi API, which we explore below.

Custom Type Adapters

With Moshi, it’s particularly easy to customize how values are converted to and from JSON. A type adapter is any class that has methods annotated @ToJson and @FromJson.

For example, Moshi’s default encoding of a playing card is verbose: the JSON defines the rank and suit in separate fields: {"rank":"A","suit":"HEARTS"}. With a type adapter, we can change the encoding to something more compact: "4H" for the four of hearts or "JD" for the jack of diamonds:

class CardAdapter {
  @ToJson String toJson(Card card) {
    return card.rank + card.suit.name().substring(0, 1);
  }

  @FromJson Card fromJson(String card) {
    if (card.length() != 2) throw new JsonDataException("Unknown card: " + card);

    char rank = card.charAt(0);
    switch (card.charAt(1)) {
      case 'C': return new Card(rank, Suit.CLUBS);
      case 'D': return new Card(rank, Suit.DIAMONDS);
      case 'H': return new Card(rank, Suit.HEARTS);
      case 'S': return new Card(rank, Suit.SPADES);
      default: throw new JsonDataException("unknown suit: " + card);
    }
  }
}

Register the type adapter with the Moshi.Builder and we’re good to go.

Moshi moshi = new Moshi.Builder()
    .add(new CardAdapter())
    .build();

Voilà:

{
  "hidden_card": "6S",
  "visible_cards": [
    "4C",
    "AH"
  ]
}

Another example

Note that the method annotated with @FromJson does not need to take a String as an argument. Rather it can take input of any type and Moshi will first parse the JSON to an object of that type and then use the @FromJson method to produce the desired final value. Conversely, the method annotated with @ToJson does not have to produce a String.

Assume, for example, that we have to parse a JSON in which the date and time of an event are represented as two separate strings.

{
  "title": "Blackjack tournament",
  "begin_date": "20151010",
  "begin_time": "17:04"
}

We would like to combine these two fields into one string to facilitate the date parsing at a later point. Also, we would like to have all variable names in CamelCase. Therefore, the Event class we want Moshi to produce like this:

class Event {
  String title;
  String beginDateAndTime;
}

Instead of manually parsing the JSON line per line (which we could also do) we can have Moshi do the transformation automatically. We simply define another class EventJson that directly corresponds to the JSON structure:

class EventJson {
  String title;
  String begin_date;
  String begin_time;
}

And another class with the appropriate @FromJson and @ToJson methods that are telling Moshi how to convert an EventJson to an Event and back. Now, whenever we are asking Moshi to parse a JSON to an Event it will first parse it to an EventJson as an intermediate step. Conversely, to serialize an Event Moshi will first create an EventJson object and then serialize that object as usual.

class EventJsonAdapter {
  @FromJson Event eventFromJson(EventJson eventJson) {
    Event event = new Event();
    event.title = eventJson.title;
    event.beginDateAndTime = eventJson.begin_date + " " + eventJson.begin_time;
    return event;
  }

  @ToJson EventJson eventToJson(Event event) {
    EventJson json = new EventJson();
    json.title = event.title;
    json.begin_date = event.beginDateAndTime.substring(0, 8);
    json.begin_time = event.beginDateAndTime.substring(9, 14);
    return json;
  }
}

Again we register the adapter with Moshi.

Moshi moshi = new Moshi.Builder()
    .add(new EventJsonAdapter())
    .build();

We can now use Moshi to parse the JSON directly to an Event.

JsonAdapter<Event> jsonAdapter = moshi.adapter(Event.class);
Event event = jsonAdapter.fromJson(json);

Adapter convenience methods

Moshi provides a number of convenience methods for JsonAdapter objects:

• nullSafe()
• nonNull()
• lenient()
• failOnUnknown()
• indent()
• serializeNulls()

These factory methods wrap an existing JsonAdapter into additional functionality. For example, if you have an adapter that doesn't support nullable values, you can use nullSafe() to make it null safe:

String dateJson = "\"2018-11-26T11:04:19.342668Z\"";
String nullDateJson = "null";

// Hypothetical IsoDateDapter, doesn't support null by default
JsonAdapter<Date> adapter = new IsoDateDapter();

Date date = adapter.fromJson(dateJson);
System.out.println(date); // Mon Nov 26 12:04:19 CET 2018

Date nullDate = adapter.fromJson(nullDateJson);
// Exception, com.squareup.moshi.JsonDataException: Expected a string but was NULL at path $

Date nullDate = adapter.nullSafe().fromJson(nullDateJson);
System.out.println(nullDate); // null

In contrast to nullSafe() there is nonNull() to make an adapter refuse null values. Refer to the Moshi JavaDoc for details on the various methods.

Parse JSON Arrays

Say we have a JSON string of this structure:

[
  {
    "rank": "4",
    "suit": "CLUBS"
  },
  {
    "rank": "A",
    "suit": "HEARTS"
  }
]

We can now use Moshi to parse the JSON string into a List<Card>.

String cardsJsonResponse = ...;
Type type = Types.newParameterizedType(List.class, Card.class);
JsonAdapter<List<Card>> adapter = moshi.adapter(type);
List<Card> cards = adapter.fromJson(cardsJsonResponse);

Fails Gracefully

Automatic databinding almost feels like magic. But unlike the black magic that typically accompanies reflection, Moshi is designed to help you out when things go wrong.

JsonDataException: Expected one of [CLUBS, DIAMONDS, HEARTS, SPADES] but was ANCHOR at path $.visible_cards[2].suit
  at com.squareup.moshi.JsonAdapters$11.fromJson(JsonAdapters.java:188)
  at com.squareup.moshi.JsonAdapters$11.fromJson(JsonAdapters.java:180)
  ...

Moshi always throws a standard java.io.IOException if there is an error reading the JSON document, or if it is malformed. It throws a JsonDataException if the JSON document is well-formed, but doesn’t match the expected format.

Built on Okio

Moshi uses Okio for simple and powerful I/O. It’s a fine complement to OkHttp, which can share buffer segments for maximum efficiency.

Borrows from Gson

Moshi uses the same streaming and binding mechanisms as Gson. If you’re a Gson user you’ll find Moshi works similarly. If you try Moshi and don’t love it, you can even migrate to Gson without much violence!

But the two libraries have a few important differences:

• Moshi has fewer built-in type adapters. For example, you need to configure your own date adapter. Most binding libraries will encode whatever you throw at them. Moshi refuses to serialize platform types (java.*, javax.*, and android.*) without a user-provided type adapter. This is intended to prevent you from accidentally locking yourself to a specific JDK or Android release.
• Moshi is less configurable. There’s no field naming strategy, versioning, instance creators, or long serialization policy. Instead of naming a field visibleCards and using a policy class to convert that to visible_cards, Moshi wants you to just name the field visible_cards as it appears in the JSON.
• Moshi doesn’t have a JsonElement model. Instead it just uses built-in types like List and Map.
• No HTML-safe escaping. Gson encodes = as \u003d by default so that it can be safely encoded in HTML without additional escaping. Moshi encodes it naturally (as =) and assumes that the HTML encoder – if there is one – will do its job.

Custom field names with @Json

Moshi works best when your JSON objects and Java objects have the same structure. But when they don't, Moshi has annotations to customize data binding.

Use @Json to specify how Java fields map to JSON names. This is necessary when the JSON name contains spaces or other characters that aren’t permitted in Java field names. For example, this JSON has a field name containing a space:

{
  "username": "jesse",
  "lucky number": 32
}

With @Json its corresponding Java class is easy:

class Player {
  String username;
  @Json(name = "lucky number") int luckyNumber;

  ...
}

Because JSON field names are always defined with their Java fields, Moshi makes it easy to find fields when navigating between Java and JSON.

Alternate type adapters with @JsonQualifier

Use @JsonQualifier to customize how a type is encoded for some fields without changing its encoding everywhere. This works similarly to the qualifier annotations in dependency injection tools like Dagger and Guice.

Here’s a JSON message with two integers and a color:

{
  "width": 1024,
  "height": 768,
  "color": "#ff0000"
}

By convention, Android programs also use int for colors:

class Rectangle {
  int width;
  int height;
  int color;
}

But if we encoded the above Java class as JSON, the color isn't encoded properly!

{
  "width": 1024,
  "height": 768,
  "color": 16711680
}

The fix is to define a qualifier annotation, itself annotated @JsonQualifier:

@Retention(RUNTIME)
@JsonQualifier
public @interface HexColor {
}

Next apply this @HexColor annotation to the appropriate field:

class Rectangle {
  int width;
  int height;
  @HexColor int color;
}

And finally define a type adapter to handle it:

/** Converts strings like #ff0000 to the corresponding color ints. */
class ColorAdapter {
  @ToJson String toJson(@HexColor int rgb) {
    return String.format("#%06x", rgb);
  }

  @FromJson @HexColor int fromJson(String rgb) {
    return Integer.parseInt(rgb.substring(1), 16);
  }
}

Use @JsonQualifier when you need different JSON encodings for the same type. Most programs shouldn’t need this @JsonQualifier, but it’s very handy for those that do.

Omit fields with transient

Some models declare fields that shouldn’t be included in JSON. For example, suppose our blackjack hand has a total field with the sum of the cards:

public final class BlackjackHand {
  private int total;

  ...
}

By default, all fields are emitted when encoding JSON, and all fields are accepted when decoding JSON. Prevent a field from being included by adding Java’s transient keyword:

public final class BlackjackHand {
  private transient int total;

  ...
}

Transient fields are omitted when writing JSON. When reading JSON, the field is skipped even if the JSON contains a value for the field. Instead it will get a default value.

Default Values & Constructors

When reading JSON that is missing a field, Moshi relies on the Java or Android runtime to assign the field’s value. Which value it uses depends on whether the class has a no-arguments constructor.

If the class has a no-arguments constructor, Moshi will call that constructor and whatever value it assigns will be used. For example, because this class has a no-arguments constructor the total field is initialized to -1.

public final class BlackjackHand {
  private int total = -1;
  ...

  private BlackjackHand() {
  }

  public BlackjackHand(Card hidden_card, List<Card> visible_cards) {
    ...
  }
}

If the class doesn’t have a no-arguments constructor, Moshi can’t assign the field’s default value, even if it’s specified in the field declaration. Instead, the field’s default is always 0 for numbers, false for booleans, and null for references. In this example, the default value of total is 0!

public final class BlackjackHand {
  private int total = -1;
  ...

  public BlackjackHand(Card hidden_card, List<Card> visible_cards) {
    ...
  }
}

This is surprising and is a potential source of bugs! For this reason consider defining a no-arguments constructor in classes that you use with Moshi, using @SuppressWarnings("unused") to prevent it from being inadvertently deleted later:

public final class BlackjackHand {
  private int total = -1;
  ...

  @SuppressWarnings("unused") // Moshi uses this!
  private BlackjackHand() {
  }

  public BlackjackHand(Card hidden_card, List<Card> visible_cards) {
    ...
  }
}

Composing Adapters

In some situations Moshi's default Java-to-JSON conversion isn't sufficient. You can compose adapters to build upon the standard conversion.

In this example, we turn serialize nulls, then delegate to the built-in adapter:

class TournamentWithNullsAdapter {
  @ToJson void toJson(JsonWriter writer, Tournament tournament,
      JsonAdapter<Tournament> delegate) throws IOException {
    boolean wasSerializeNulls = writer.getSerializeNulls();
    writer.setSerializeNulls(true);
    try {
      delegate.toJson(writer, tournament);
    } finally {
      writer.setLenient(wasSerializeNulls);
    }
  }
}

When we use this to serialize a tournament, nulls are written! But nulls elsewhere in our JSON document are skipped as usual.

Moshi has a powerful composition system in its JsonAdapter.Factory interface. We can hook in to the encoding and decoding process for any type, even without knowing about the types beforehand. In this example, we customize types annotated @AlwaysSerializeNulls, which an annotation we create, not built-in to Moshi:

@Target(TYPE)
@Retention(RUNTIME)
public @interface AlwaysSerializeNulls {}

@AlwaysSerializeNulls
static class Car {
  String make;
  String model;
  String color;
}

Each JsonAdapter.Factory interface is invoked by Moshi when it needs to build an adapter for a user's type. The factory either returns an adapter to use, or null if it doesn't apply to the requested type. In our case we match all classes that have our annotation.

static class AlwaysSerializeNullsFactory implements JsonAdapter.Factory {
  @Override public JsonAdapter<?> create(
      Type type, Set<? extends Annotation> annotations, Moshi moshi) {
    Class<?> rawType = Types.getRawType(type);
    if (!rawType.isAnnotationPresent(AlwaysSerializeNulls.class)) {
      return null;
    }

    JsonAdapter<Object> delegate = moshi.nextAdapter(this, type, annotations);
    return delegate.serializeNulls();
  }
}

After determining that it applies, the factory looks up Moshi's built-in adapter by calling Moshi.nextAdapter(). This is key to the composition mechanism: adapters delegate to each other! The composition in this example is simple: it applies the serializeNulls() transform on the delegate.

Composing adapters can be very sophisticated:

• An adapter could transform the input object before it is JSON-encoded. A string could be trimmed or truncated; a value object could be simplified or normalized.

• An adapter could repair the output object after it is JSON-decoded. It could fill-in missing data or discard unwanted data.

• The JSON could be given extra structure, such as wrapping values in objects or arrays.

Moshi is itself built on the pattern of repeatedly composing adapters. For example, Moshi's built-in adapter for List<T> delegates to the adapter of T, and calls it repeatedly.

Precedence

Moshi's composition mechanism tries to find the best adapter for each type. It starts with the first adapter or factory registered with Moshi.Builder.add(), and proceeds until it finds an adapter for the target type.

If a type can be matched multiple adapters, the earliest one wins.

To register an adapter at the end of the list, use Moshi.Builder.addLast() instead. This is most useful when registering general-purpose adapters, such as the KotlinJsonAdapterFactory below.

Kotlin

Moshi is a great JSON library for Kotlin. It understands Kotlin’s non-nullable types and default parameter values. When you use Kotlin with Moshi you may use reflection, codegen, or both.

Reflection

The reflection adapter uses Kotlin’s reflection library to convert your Kotlin classes to and from JSON. Enable it by adding the KotlinJsonAdapterFactory to your Moshi.Builder:

val moshi = Moshi.Builder()
    .addLast(KotlinJsonAdapterFactory())
    .build()

Moshi’s adapters are ordered by precedence, so you should use addLast() with KotlinJsonAdapterFactory, and add() with your custom adapters.

The reflection adapter requires the following additional dependency:

<dependency>
  <groupId>com.squareup.moshi</groupId>
  <artifactId>moshi-kotlin</artifactId>
  <version>1.11.0</version>
</dependency>

implementation("com.squareup.moshi:moshi-kotlin:1.11.0")

Note that the reflection adapter transitively depends on the kotlin-reflect library which is a 2.5 MiB .jar file.

Codegen

Moshi’s Kotlin codegen support is an annotation processor. It generates a small and fast adapter for each of your Kotlin classes at compile time. Enable it by annotating each class that you want to encode as JSON:

@JsonClass(generateAdapter = true)
data class BlackjackHand(
        val hidden_card: Card,
        val visible_cards: List<Card>
)

The codegen adapter requires that your Kotlin types and their properties be either internal or public (this is Kotlin’s default visibility).

Kotlin codegen has no additional runtime dependency. You’ll need to enable kapt and then add the following to your build to enable the annotation processor:

<dependency>
  <groupId>com.squareup.moshi</groupId>
  <artifactId>moshi-kotlin-codegen</artifactId>
  <version>1.11.0</version>
  <scope>provided</scope>
</dependency>

kapt("com.squareup.moshi:moshi-kotlin-codegen:1.11.0")

You must also have the kotlin-stdlib dependency on the classpath during compilation in order for the compiled code to have the required metadata annotations that Moshi's processor looks for.

Limitations

If your Kotlin class has a superclass, it must also be a Kotlin class. Neither reflection or codegen support Kotlin types with Java supertypes or Java types with Kotlin supertypes. If you need to convert such classes to JSON you must create a custom type adapter.

The JSON encoding of Kotlin types is the same whether using reflection or codegen. Prefer codegen for better performance and to avoid the kotlin-reflect dependency; prefer reflection to convert both private and protected properties. If you have configured both, generated adapters will be used on types that are annotated @JsonClass(generateAdapter = true).

Download

Download the latest JAR or depend via Maven:

<dependency>
  <groupId>com.squareup.moshi</groupId>
  <artifactId>moshi</artifactId>
  <version>1.11.0</version>
</dependency>

or Gradle:

implementation("com.squareup.moshi:moshi:1.11.0")

Snapshots of the development version are available in Sonatype's snapshots repository.

R8 / ProGuard

Moshi contains minimally required rules for its own internals to work without need for consumers to embed their own. However if you are using reflective serialization and R8 or ProGuard, you must add keep rules in your proguard configuration file for your reflectively serialized classes.

License

Copyright 2015 Square, Inc.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

   http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.