Java implementation of JSON API specification v1.0 for moshi.
JsonAdapter.Factory jsonApiAdapterFactory = ResourceAdapterFactory.builder()
.add(Article.class)
.add(Person.class)
.add(Comment.class)
// ...
.build();
Moshi moshi = new Moshi.Builder()
.add(jsonApiAdapterFactory)
// ...
.build();
You're now ready to serialize/deserialize JSON API objects with cool Moshi interface!
String json = "...";
Type type = Types.newParameterizedType(Document.class, Article.class); // Type of Document<Article>
JsonAdapter<Document<Article>> adapter = ((JsonAdapter<Document<Article>>) moshi.adapter(type));
ArrayDocument<Article> articles = adapter.fromJson(json).asArrayDocument();
for (Article article : articles) {
System.out.println(article.title);
}
Simply add a retrofit converter and you get all the cool stuff in Retrofit!
public interface MyAPI {
@GET("posts")
Call<Post[]> listPosts();
@GET("posts/{id}")
Call<Post> getPost(@Path("id") String id);
@GET("posts/{id}/comments")
Call<Comment[]> getComments(@Path("id") String id);
@POST("posts/{id}/comments")
Call<Document> addComment(@Path("id") String id, @Body Comment comment);
@DELETE("posts/{id}/relationships/comments")
Call<Document> removeComments(@Path("id") String id, @Body ResourceIdentifier[] commentIds);
@GET("posts/{id}/relationships/comments")
Call<ResourceIdentifier[]> getCommentRels(@Path("id") String id);
}
No annoying Call<Document<RESOURCE>>
declaration required and Document
is wrap/unwrapped automatically by the converter.
And use that declaration when need Document
to collecting errors or any other information.
Extend a Resource
class to create a model for resource object.
@JsonApi(type = "people")
class Person extends Resource {
@Json(name="first-name") String firstName;
@Json(name="last-name") String lastName;
String twitter;
}
@JsonApi(type = ...)
annotation identifies each model by type
as is mentioned in specification.
There are two kinds of relationship defined in JSON API specification. Defining these relationship in resource object is quite simple:
@JsonApi(type = "articles")
public class Article extends Resource {
public String title;
public HasOne<Person> author;
public HasMany<Comment> comments;
}
Relationships can be resolved to resource object in a Document
:
Person author = article.author.get(article.getDocument());
You can use Resource.getDocument()
to access the Document
object the Resource
be added/included in.
Further more, with a little bit encapsulation:
@JsonApi(type = "articles")
public class Article extends Resource {
private String title;
private HasOne<Person> author;
private HasMany<Comment> comments;
public String getTitle() {
return title;
}
public void setTitle(String title) {
this.title = title;
}
public Person getAuthor() {
return author.get(getDocument());
}
public List<Comment> getComments() {
return comments.get(getDocument());
}
}
Document<Article> document = new ObjectDocument<>();
document.set(article);
document.include(author);
// Serialize
JsonAdapter<Document<Article>> adapter = moshi.adapter(document.getType());
System.out.println(adapter.toJson(document));
// => {
// data: { "type": "articles", "relationships": { "author": { "data": "type": "people", id: "1" } } },
// included: [
// { "type": "people", "attributes": { "first-name": "Yuki", "last-name": "Kiriyama", "twitter": "kamikat_bot" } }
// ]
// }
// Deserialize
Document<Article> document2 = adapter.fromJson(...);
assert document2.get() instanceof Article
assert document2.get().getDocument() == document2
All resources added/included in a Document
will keep a reference which can be accessed from Resource.getDocument
.
Deserialization will fail when processing an unknown type of resource.
Create a default
typed model to avoid this problem and parses all unknown type of resource object into the default model.
@JsonApi(type = "default")
class Unknown extends Resource {
// nothing...
}
You'd like to access meta
/links
/jsonapi
value on Document
for example.
Document<Article> document = ...;
document.getMeta() // => JsonBuffer
As meta
and links
can contain a variant of objects, they are not been parsed when access with getMeta
and getLinks
.
You will get a JsonBuffer
and you're expected to implement your JsonAdapter<T>
to read/write these objects.
In gradle build script:
repositories {
jcenter()
}
dependencies {
implementation 'com.squareup.moshi:moshi:1.4.0'
implementation 'moe.banana:moshi-jsonapi:<version>'
}
For library version >= 3.6, moshi is removed from runtime dependencies of the library to become a peer dependency.
Use snapshot version:
repositories {
maven { url "https://jitpack.io" }
}
dependencies {
implementation 'com.squareup.moshi:moshi:1.4.0'
implementation 'moe.banana:moshi-jsonapi:master-SNAPSHOT'
}
(you may be asked to clean gradle library cache to access the latest snapshot)
For moshi-jsonapi:
-keepattributes Signature
-keepclassmembers public abstract class moe.banana.jsonapi2.** {
*;
}
For moshi, if you use a custom JSON adapter (e.g. for Enum types):
-keepclassmembers class ** {
@com.squareup.moshi.FromJson *;
@com.squareup.moshi.ToJson *;
}
Feature | Supported | Note |
---|---|---|
Serialization | Yes | |
Deserialization | Yes | |
Custom-named fields | Yes | With @Json |
Top level errors | Yes | |
Top level metadata | Yes | |
Top level links | Yes | |
Top level JSON API Object | Yes | |
Resource metadata | Yes | |
Resource links | Yes | |
Relationships | Yes | HasOne and HasMany |
Inclusion of related resources | Yes | |
Resource IDs | Yes |
3.x supports all features supported by JSON API specification. And the interface changed a lot especially in serialization/deserialization.
More object oriented features are added to new API. If you're using the library with Retrofit, migration should be a lot easier by using a
special Converter
adapts Document<Article>
to Article[]
and backward as well (see retrofit section). Migration should be
easy if you use latest 2.x API with some OO features already available. Otherwise, it can take hours to migrate to new API.
2.x abandoned much of seldomly used features of JSON API specification and re-implement the core of JSON API without AutoValue since AutoValue is considered too verbose to implement a clean model.
And the new API no longer requires a verbose null check since you should take all control over the POJO model's nullability check.
Another major change is that the new API is not compatible with AutoValue any more. Means that one have to choose 1.x implementation if AutoValue is vital to bussiness logic.
(The MIT License)