I have bunch of model classes which have fields of type List<X> where X is one of many things (e.g. String, Integer, but also some of my own types). I'm using GSON to parse JSON representations of these models.
My problem is that the server I'm dealing with (which is beyond my control) somehow removed singleton arrays and replaces them by the contained object.
For example, instead of returning:
{
"foo": [ "bar"],
"bleh": [ { "some": "object" } ]
}
It returns:
{
"foo": "bar",
"bleh": { "some": "object" }
}
Now assume that the Java model class look something like this:
public class Model {
private List<String> foo;
private List<SomeObject> bleh;
}
Currently this causes GSON to throw an exception because it finds BEGIN_STRING or BEGIN_OBJECT where it expects BEGIN_ARRAY.
For arrays or lists of Strings this is easily solved using a TypeAdapter<List<String>>. But the problem is I have Lists with many different element types and I don't want to write a separate TypeAdapter for each case. Nor have I been able to a generic TypeAdapter<List<?>>, because at some point you need to know the type.
So is there another way to configure GSON to be smart enough to turn single objects or values into arrays/lists? Or in other words, just "pretend" that the [ and ] are there where it expects to find them although they aren't there?
But the problem is I have Lists with many different element types and I don't want to write a separate TypeAdapter for each case. Nor have I been able to a generic TypeAdapter>, because at some point you need to know the type.
This is what type adapter factories are designed for: you can control every type in Gson instance configuration.
final class AlwaysListTypeAdapterFactory<E>
implements TypeAdapterFactory {
// Gson can instantiate it itself
private AlwaysListTypeAdapterFactory() {
}
#Override
public <T> TypeAdapter<T> create(final Gson gson, final TypeToken<T> typeToken) {
// If it's not a List -- just delegate the job to Gson and let it pick the best type adapter itself
if ( !List.class.isAssignableFrom(typeToken.getRawType()) ) {
return null;
}
// Resolving the list parameter type
final Type elementType = resolveTypeArgument(typeToken.getType());
#SuppressWarnings("unchecked")
final TypeAdapter<E> elementTypeAdapter = (TypeAdapter<E>) gson.getAdapter(TypeToken.get(elementType));
// Note that the always-list type adapter is made null-safe, so we don't have to check nulls ourselves
#SuppressWarnings("unchecked")
final TypeAdapter<T> alwaysListTypeAdapter = (TypeAdapter<T>) new AlwaysListTypeAdapter<>(elementTypeAdapter).nullSafe();
return alwaysListTypeAdapter;
}
private static Type resolveTypeArgument(final Type type) {
// The given type is not parameterized?
if ( !(type instanceof ParameterizedType) ) {
// No, raw
return Object.class;
}
final ParameterizedType parameterizedType = (ParameterizedType) type;
return parameterizedType.getActualTypeArguments()[0];
}
private static final class AlwaysListTypeAdapter<E>
extends TypeAdapter<List<E>> {
private final TypeAdapter<E> elementTypeAdapter;
private AlwaysListTypeAdapter(final TypeAdapter<E> elementTypeAdapter) {
this.elementTypeAdapter = elementTypeAdapter;
}
#Override
public void write(final JsonWriter out, final List<E> list) {
throw new UnsupportedOperationException();
}
#Override
public List<E> read(final JsonReader in)
throws IOException {
// This is where we detect the list "type"
final List<E> list = new ArrayList<>();
final JsonToken token = in.peek();
switch ( token ) {
case BEGIN_ARRAY:
// If it's a regular list, just consume [, <all elements>, and ]
in.beginArray();
while ( in.hasNext() ) {
list.add(elementTypeAdapter.read(in));
}
in.endArray();
break;
case BEGIN_OBJECT:
case STRING:
case NUMBER:
case BOOLEAN:
// An object or a primitive? Just add the current value to the result list
list.add(elementTypeAdapter.read(in));
break;
case NULL:
throw new AssertionError("Must never happen: check if the type adapter configured with .nullSafe()");
case NAME:
case END_ARRAY:
case END_OBJECT:
case END_DOCUMENT:
throw new MalformedJsonException("Unexpected token: " + token);
default:
throw new AssertionError("Must never happen: " + token);
}
return list;
}
}
}
Now you just have to tell Gson which fields are not well-formed.
Of course, you might configure the whole Gson instance to accept such lists, but let it be more precise using the #JsonAdapter annotation:
final class Model {
#JsonAdapter(AlwaysListTypeAdapterFactory.class)
final List<String> foo = null;
#JsonAdapter(AlwaysListTypeAdapterFactory.class)
final List<SomeObject> bleh = null;
#Override
public String toString() {
return "Model{" + "foo=" + foo + ", bleh=" + bleh + '}';
}
}
final class SomeObject {
final String some = null;
#Override
public String toString() {
return "SomeObject{" + "some='" + some + '\'' + '}';
}
}
Test data:
single.json
{
"foo": "bar",
"bleh": {"some": "object"}
}
list.json
{
"foo": ["bar"],
"bleh": [{"some": "object"}]
}
Example:
private static final Gson gson = new Gson();
public static void main(final String... args)
throws IOException {
for ( final String resource : ImmutableList.of("single.json", "list.json") ) {
try ( final JsonReader jsonReader = getPackageResourceJsonReader(Q43412261.class, resource) ) {
final Model model = gson.fromJson(jsonReader, Model.class);
System.out.println(model);
}
}
}
And the output:
Model{foo=[bar], bleh=[SomeObject{some='object'}]}
Model{foo=[bar], bleh=[SomeObject{some='object'}]}
You can simply write your own JsonDeserializer where you check whether your bleh or foo are JsonObjects or JsonArrays.
To check if a JsonElement is an array or an object:
JsonElement element = ...;
if (element.isJsonObject()) {
//element is a JsonObject
} else if (element.isJsonArray()) {
//element is a JsonArray
}
One solution to this would be to write a custom TypeAdapterFactory which creates an adapter which peeks at the JSON data. If it encounters something other than a JSON array (or JSON null) it wraps it inside a JSON array before deserializing it:
// Only intended for usage with #JsonAdapter on fields
class SingleValueOrListAdapterFactory implements TypeAdapterFactory {
#Override
public <T> TypeAdapter<T> create(Gson gson, TypeToken<T> type) {
// Note: Cannot use getDelegateAdapter due to https://github.com/google/gson/issues/1028
TypeAdapter<T> listAdapterDelegate = gson.getAdapter(type);
TypeAdapter<JsonElement> jsonElementAdapter = gson.getAdapter(JsonElement.class);
return new TypeAdapter<T>() {
#Override
public void write(JsonWriter out, T value) throws IOException {
listAdapterDelegate.write(out, value);
}
#Override
public T read(JsonReader in) throws IOException {
JsonToken peeked = in.peek();
if (peeked == JsonToken.NULL || peeked == JsonToken.BEGIN_ARRAY) {
return listAdapterDelegate.read(in);
} else {
// Wrap JSON element in a new JSON array before deserializing it
JsonElement jsonElement = jsonElementAdapter.read(in);
JsonArray jsonArray = new JsonArray();
jsonArray.add(jsonElement);
return listAdapterDelegate.fromJsonTree(jsonArray);
}
}
};
}
}
The above implementation is designed only for usage with #JsonAdapter on fields, for example:
#JsonAdapter(SingleValueOrListAdapterFactory.class)
private List<MyClass> myField;
Compared to the currently accepted answer this provides the following advantages because it simply delegates the actual deserialization to listAdapterDelegate:
Custom List (or Collection) subclasses are supported because creation of them is delegated to Gson
Gson's default type resolution logic is used to determine the element type and to deserialize it
But it also has the following disadvantage:
Decreased performance because if the data is not already in a JSON array it is first deserialized to a JsonElement before the actual deserialization is performed
When using the GSON library, you could just check whether or not the following token is an object or an array. This of course requires you to go more fine grained while parsing the XML, but it allows you full control of what do you want to get from it. Sometimes we are not under control of the XML, and it could come handy.
This is an example to check if the next token is an object or an array, using the JsonReader class to parse the file:
if (jsonReader.peek() == JsonToken.BEGIN_ARRAY) {
jsonReader.beginArray()
} else if (jsonReader.peek() == JsonToken.BEGIN_OBJECT) {
jsonReader.beginObject()
}
And at the end of the array / object, you could do the same, but for the end tokens:
if (jsonReader.peek() == JsonToken.END_ARRAY) {
jsonReader.endArray()
} else if (jsonReader.peek() == JsonToken.END_OBJECT) {
jsonReader.endObject()
}
This way, you could have identical code (adding an extra check, to verify if you are on an array or on an object) to parse your array of objects, or a single object.
I had this same problem consuming xml / json from a vendor - they certainly weren't going to change their code for me :) There were several resources on the web that I used before changing adapting them to my own version This SO answer was very helpful. I spent some time looking at the gson code and finding a lot of private variable that I wanted access to. So, essentially what my custom collection adapter does is peek to see if the next element is an object. If not, we just delegate the read to the previous adapter (that we have overridden).
If the next element is an object, we use gson to process that. We then convert that to an array of one object. Use gson to write that to a string, then pass that string as a JsonReader to the underlying adapter. This can then create an instance of the underlying list and add the one element we have.
Here's the AdapterTypeFactory:
public enum ListSingleObjectAdapterFactory implements TypeAdapterFactory {
INSTANCE; // Josh Bloch's Enum singleton pattern
#SuppressWarnings({ "unchecked", "rawtypes" })
#Override
public <T> TypeAdapter<T> create(final Gson gson, final TypeToken<T> typeToken) {
Class<? super T> rawType = typeToken.getRawType();
if (!Collection.class.isAssignableFrom(rawType)) {
return null;
}
TypeAdapter collectionAdapter = gson.getDelegateAdapter(this, typeToken);
Class genericClass = (Class) ((ParameterizedType) typeToken.getType())
.getActualTypeArguments()[0];
return new SingleObjectOrCollectionAdapter(
gson, collectionAdapter, genericClass);
}
}
Then the type adapter I have is:
public class SingleObjectOrCollectionAdapter<T> extends TypeAdapter<Collection<T>> {
private Class<T> adapterclass;
private Gson gson;
private TypeAdapter arrayTypeAdapter;
public SingleObjectOrCollectionAdapter(Gson gson, TypeAdapter<T> collectionTypeAdapter, Class<T> componentType) {
arrayTypeAdapter = collectionTypeAdapter;
this.gson = gson;
adapterclass = componentType;
}
#Override
public Collection<T> read(JsonReader reader) throws IOException {
Collection<T> collection;
JsonReader myReader = reader;
if (reader.peek() == JsonToken.BEGIN_OBJECT) {
T inning = gson.fromJson(reader, adapterclass);
String s = gson.toJson(new Object[]{inning});
myReader = new JsonReader(new StringReader(s));
}
collection = (Collection)arrayTypeAdapter.read( myReader );
return collection;
}
#Override
public void write(JsonWriter writer, Collection<T> value) throws IOException {
arrayTypeAdapter.write(writer, value);
}
}
Finally, we need to register the adapter factory:
GsonBuilder gb = new GsonBuilder().registerTypeAdapterFactory(ListSingleObjectAdapterFactory.INSTANCE);
So far, it seems to be working well handling both single and multiple objects - although I wouldn't be surprised if it needs some tweaking down the road.
Related
Here is the json schema:
As you can see, rated can be both boolean and object.
I am using Retrofit 2 and Gson converter. How should I create my model for this schema?
Here's how I solved this issue:
Create a custom type adapter in your model and parse rated manually;
public class AccountState {
//#SerializedName("rated") //NOPE, parse it manually
private Integer mRated; //also don't name it rated
public Integer getRated() {
return mRated;
}
public void setRated(Integer rated) {
this.mRated = rated;
}
public static class AccountStateDeserializer implements JsonDeserializer<AccountState> {
#Override
public AccountState deserialize(JsonElement json, Type typeOfT, JsonDeserializationContext context) throws JsonParseException {
AccountState accountState = new Gson().fromJson(json, AccountState.class);
JsonObject jsonObject = json.getAsJsonObject();
if (jsonObject.has("rated")) {
JsonElement elem = jsonObject.get("rated");
if (elem != null && !elem.isJsonNull()) {
if(elem.isJsonPrimitive()){
accountState.setRated(null);
}else{
accountState.setRated(elem.getAsJsonObject().get("value").getAsInt());
}
}
}
return accountState ;
}
}
}
Here you create your gson with this custom adapter:
final static Gson gson = new GsonBuilder()
.registerTypeAdapter(AccountState.class, new AccountState.AccountStateDeserializer())
.create();
Add it to retrofit like that:
Retrofit retrofit = new Retrofit.Builder()
.baseUrl(BuildConfig.ENDPOINT)
.addConverterFactory(GsonConverterFactory.create(gson))
.client(okHttpClient)
.build();
TADADADADADADADDAD!
You can make it work without having to implement a custom converter.
All you have to do is put a general "Object" type for the variable and then you just check which data type it is by doing this:
if(object.getClass == YourClass.class){
Whatever we = ((YourClass) object).getWhatever();
} else if(object.getClass == YourOtherClass.class){
String name = ((YourOtherClass) object).getName();
}
You can add as many data types to this variable as you like.
You can also use the java types "String.class", "Boolean.class" or whatever you like.
Gson has a nice feature allowing to inject a custom type adapter or a type adapter factory to a certain field therefore letting Gson to manage the host object and the latter's fields (de)serialization. So, you can be sure that AccountState could be still deserialized with ReflectiveTypeAdapterFactory and ReflectiveTypeAdapterFactory.Adapter so all deserialization strategies defined in GsonBuilder could be applied.
final class AccountState {
// This is what can make life easier. Note its advantages:
// * PackedBooleanTypeAdapterFactory can be reused multiple times
// * AccountState life-cycle can be managed by Gson itself,
// so it can manage *very* complex deserialization automatically.
#JsonAdapter(PackedBooleanTypeAdapterFactory.class)
final Boolean rated = null;
}
Next, how PackageBooleanTypeAdapterFactory is implemented:
final class PackedBooleanTypeAdapterFactory
implements TypeAdapterFactory {
// Gson can instantiate this itself, no need to expose
private PackedBooleanTypeAdapterFactory() {
}
#Override
public <T> TypeAdapter<T> create(final Gson gson, final TypeToken<T> typeToken) {
// Check if it's the type we can handle ourself
if ( typeToken.getRawType() == Boolean.class ) {
final TypeAdapter<Boolean> typeAdapter = new PackedIntegerTypeAdapter(gson);
// Some Java "unchecked" boilerplate here...
#SuppressWarnings("unchecked")
final TypeAdapter<T> castTypeAdapter = (TypeAdapter<T>) typeAdapter;
return castTypeAdapter;
}
// If it's something else, let Gson pick a downstream type adapter on its own
return null;
}
private static final class PackedIntegerTypeAdapter
extends TypeAdapter<Boolean> {
private final Gson gson;
private PackedIntegerTypeAdapter(final Gson gson) {
this.gson = gson;
}
#Override
public void write(final JsonWriter out, final Boolean value) {
throw new UnsupportedOperationException();
}
#Override
public Boolean read(final JsonReader in)
throws MalformedJsonException {
// Pick next token as a JsonElement
final JsonElement jsonElement = gson.fromJson(in, JsonElement.class);
// Note that Gson uses JsonNull singleton to denote a null
if ( jsonElement.isJsonNull() ) {
return null;
}
if ( jsonElement.isJsonPrimitive() ) {
return jsonElement
.getAsJsonPrimitive()
.getAsBoolean();
}
if ( jsonElement.isJsonObject() ) {
return jsonElement
.getAsJsonObject()
.getAsJsonPrimitive("value")
.getAsBoolean();
}
// Not something we can handle
throw new MalformedJsonException("Cannot parse: " + jsonElement);
}
}
}
Demo:
public static void main(final String... args) {
parseAndDump("{\"rated\":null}");
parseAndDump("{\"rated\":true}");
parseAndDump("{\"rated\":{\"value\":true}}");
}
private static void parseAndDump(final String json) {
final AccountState accountState = gson.fromJson(json, AccountState.class);
System.out.println(accountState.rated);
}
Output:
null
true
true
Note that JsonSerializer and JsonDeserializer both have some performance and memory cost due to its tree model design (you can traverse JSON trees easily as long as they are in memory). Sometimes, for simple cases, a streaming type adapter may be preferable. Pros: consumes less memory and works faster. Cons: hard to implement.
final class AccountState {
#JsonAdapter(PackedBooleanTypeAdapter.class)
final Boolean rated = null;
}
Note that the rated field accepts a type adapter directly because it does not need Gson instances to build JSON trees (JsonElements).
final class PackedBooleanTypeAdapter
extends TypeAdapter<Boolean> {
// Gson still can instantiate this type adapter itself
private PackedBooleanTypeAdapter() {
}
#Override
public void write(final JsonWriter out, final Boolean value) {
throw new UnsupportedOperationException();
}
#Override
public Boolean read(final JsonReader in)
throws IOException {
// Peeking the next JSON token and dispatching parsing according to the given token
final JsonToken token = in.peek();
switch ( token ) {
case NULL:
return parseAsNull(in);
case BOOLEAN:
return parseAsBoolean(in);
case BEGIN_OBJECT:
return parseAsObject(in);
// The below might be omitted, since some code styles prefer all switch/enum constants explicitly
case BEGIN_ARRAY:
case END_ARRAY:
case END_OBJECT:
case NAME:
case STRING:
case NUMBER:
case END_DOCUMENT:
throw new MalformedJsonException("Cannot parse: " + token);
// Not a known token, and must never happen -- something new in a newer Gson version?
default:
throw new AssertionError(token);
}
}
private Boolean parseAsNull(final JsonReader in)
throws IOException {
// null token still has to be consumed from the reader
in.nextNull();
return null;
}
private Boolean parseAsBoolean(final JsonReader in)
throws IOException {
// Consume a boolean value from the reader
return in.nextBoolean();
}
private Boolean parseAsObject(final JsonReader in)
throws IOException {
// Consume the begin object token `{`
in.beginObject();
// Get the next property name
final String property = in.nextName();
// Not a value? Then probably it's not what we're expecting for
if ( !property.equals("value") ) {
throw new MalformedJsonException("Unexpected property: " + property);
}
// Assuming the property "value" value must be a boolean
final boolean value = in.nextBoolean();
// Consume the object end token `}`
in.endObject();
return value;
}
}
This one should work faster. The output remains the same. Note that Gson does not require a GsonBuilder for both cases. As far as I remember how Retrofit 2 works, GsonConverterFactory is still required (not sure, Gson is not the default serializer in Retrofit 2?).
One of the fields in my JSON response is a String[] when containing more than one element, and a String when it's just one. Like this:
"assets": [
"0901d196804adc1c",
"0901d196804ebd93",
"0901d196804ea5e2"
]
"assets": "0901d196804adc1c"
Ideally, I would like to get a String[] always, so if the JSON type of the element is String, convert it to a String[] with one element.
How can I do that?
If you cannot edit the response on the server side: Please refer to this question and answers, looks quite similar to your situation.
If you can edit the response, just reply always with String arrays (f.e. "assets": ["0901d196804adc1c"]).
You have two options:
1) Implement custom type adapter for gson to handle such situations (preferable solution).
2) Define field of type Object and cast it to the appropriate type at runtime
public static class AssetsContainer{
private Object assets;
public List<String> getAssets() {
if(assets instanceof List<?>) {
return (List<String>) assets;
} else if(assets instanceof String){
return Arrays.asList((String) assets);
} else {
//TODO: handle
return null;
}
}
}
How about using the TypeAdapter API?
Gson gson = new GsonBuilder()
.registerTypeAdapterFactory(SingletonListTypeAdapter.FACTORY)
.create();
The following will check for non-array JSON types when expecting array types and try to make them into singleton Java Lists. (Note that this uses Lists, not arrays. You can adapt it if you want, but Effective Java notes that application-layer code should prefer the Collections APIs over arrays.)
final class SingletonListTypeAdapter<T> extends TypeAdapter<List<T>> {
static final TypeAdapterFactory FACTORY = new TypeAdapterFactory() {
#SuppressWarnings("unchecked")
#Override public <T> TypeAdapter<T> create(Gson gson, TypeToken<T> type) {
if (type.getRawType() != List.class) {
return null;
}
TypeToken<?> collectionElementType = TypeToken.get(
getCollectionElementType((ParameterizedType) type.getType()));
TypeAdapter<List<Object>> delegate = (TypeAdapter<List<Object>>)
gson.getDelegateAdapter(this, collectionElementType);
return (TypeAdapter<T>) new SingletonListTypeAdapter<>(delegate);
}
};
private final TypeAdapter<T> delegate;
SingletonListTypeAdapter(TypeAdapter<T> delegate) {
this.delegate = delegate;
}
#Override public void write(JsonWriter out, List<T> value) throws IOException {
out.beginArray();
for (int i = 0, size = value.size(); i < size; i++) {
delegate.write(out, value.get(i));
}
out.endArray();
}
#Override public List<T> read(JsonReader in) throws IOException {
if (in.peek() != BEGIN_ARRAY) {
return Collections.singletonList(delegate.read(in));
}
in.beginArray();
List<T> expanding = new ArrayList<>();
while (in.hasNext()) {
expanding.add(delegate.read(in));
}
in.endArray();
return Collections.unmodifiableList(expanding);
}
static Type getCollectionElementType(ParameterizedType type) {
Type[] types = type.getActualTypeArguments();
Type paramType = types[0];
if (paramType instanceof WildcardType) {
return ((WildcardType) paramType).getUpperBounds()[0];
}
return paramType;
}
}
Whenever I get error, the error body is as follows:
[
{
"errorCode": 10001,
"resource": null,
"resourceId": null,
"field": null,
"parameter": null,
"header": null,
"allowedValues": null,
"maxLength": null,
"minLength": null
}
]
The error body is an array. I have different bodies for success of many API methods, but the error array response is standardized. I tried doing many things
making wrapper class with generic type success response and array of error response and made deserializer for that, but I can't deserialize from type variable and from paramaterized class.
made a ErrorDeserializer but I have no idea how can I make Retrofit use it for error responses.
I could definitely just serialize raw string everytime on every callback for all my api methods, but I have so many of them, I need generalized solution. If I didn't explain myself properly, please ask.
I'll add examples of what I tried (they will be incomplete however):
Response wrap class:
public class ResponseWrap<T> {
#Nullable
private final T response;
#Nullable
private final List<ErrorResponse> errorResponses;
public ResponseWrap(#Nullable T response, #Nullable List<ErrorResponse> errorResponses) {
this.response = response;
this.errorResponses = errorResponses;
}
}
Error response class:
public class ErrorResponse {
private int errorCode;
private String resource;
private String resourceId;
private String field;
private String parameter;
private String header;
private String allowedValues;
private int maxLength;
private int minLength;
// getters and setters
}
Error deserializer:
public class ErrorDeserializer implements JsonDeserializer<ArrayList<ErrorResponse>> {
#Override
public ArrayList<ErrorResponse> deserialize(JsonElement json, Type typeOfT, JsonDeserializationContext context) {
Gson gson = new Gson();
Type listType = new TypeToken<ArrayList<ErrorResponse>>(){}.getType();
ArrayList<ErrorResponse> list = new Gson().fromJson(json, listType);
final JsonArray jsonArray = json.getAsJsonArray();
for (int i = 0; i < jsonArray.size(); i++) {
ErrorResponse error = new ErrorResponse();
JsonObject jsonObject = jsonArray.get(i).getAsJsonObject();
error.setErrorCode(jsonObject.get("errorCode").getAsInt());
error.setResource(jsonObject.get("resource").getAsString());
error.setResourceId(jsonObject.get("resourceId").getAsString());
error.setField(jsonObject.get("field").getAsString());
error.setParameter(jsonObject.get("parameter").getAsString());
error.setHeader(jsonObject.get("header").getAsString());
error.setAllowedValues(jsonObject.get("allowedValues").getAsString());
error.setMaxLength(jsonObject.get("maxLength").getAsInt());
error.setMinLength(jsonObject.get("minLength").getAsInt());
list.add(error);
}
return list;
}
}
Response wrap deserializer - it's not working, 2 errors:
List error = new Gson().fromJson(jsonObject.getAsJsonObject("error"), ArrayList.class); // Can't select from parameterized class
T success = new Gson().fromJson(jsonObject, T.class); // Can't select from type variable
public class ResponseWrapDeserializer<T> implements JsonDeserializer<ResponseWrap<T>> {
#Override
public ResponseWrap<T> deserialize(JsonElement json, Type typeOfT, JsonDeserializationContext context) throws JsonParseException {
// Get JsonObject
final JsonObject jsonObject = json.getAsJsonObject();
if (jsonObject.has("error")) {
Gson gson = new GsonBuilder()
.registerTypeAdapter(typeOfT, new ErrorDeserializer())
.setDateFormat("yyyy-MM-dd'T'HH:mm:ss")
.create();
List<ErrorResponse> error = new Gson().fromJson(jsonObject.getAsJsonObject("error"), ArrayList<ErrorResponse>.class);
return new ResponseWrap<T>(null, error);
} else {
T success = new Gson().fromJson(jsonObject, T.class);
return new ResponseWrap<T>(success, null);
}
}
}
The idea was to use them all like this:
#POST("Login")
Call<ResponseWrap<AccessTokenResponse>> Login(#Body LoginRequest request);
But I can't because of above mentioned reasons.
The question is: How to process error responses in a generic way that are in an array using Retrofit2?
You cannot write T.class -- this is illegal in Java. In order to overcome this limitation you must either pas a Type instance yourself somehow or resolve generic types parameters from what Gson gives you. In the first case you'd need dozen JSON deserializers to bind various ResponseWrap<T> parametization; whilst in the second case can simply resolve the actual type parameter yourself. At the call site you can use TypeTokens -- a special Gson mechanism to define a type parameter via a type parameterization. Also note that you don't have to instantiate internal Gson instances: this might be relatively expensive (especially in sequence) and disrespect the Gson configuration the current deserializer is bound for - use JsonDeserializationContext since it can give you all you need (except downstream type adapters).
The following JSON deserializer uses the second approach as I find it more convenient.
final class ResponseWrapJsonDeserializer<T>
implements JsonDeserializer<ResponseWrap<T>> {
// This deserializer holds no state, so we can hide its instantiation details away
private static final JsonDeserializer<ResponseWrap<Object>> responseWrapJsonDeserializer = new ResponseWrapJsonDeserializer<>();
// Type instances from TypeToken seems to be fully immutable and can be treated as value types, thus we can make them static final to re-use (it's safe)
private static final Type errorResponseListType = new TypeToken<List<ErrorResponse>>() {
}.getType();
private ResponseWrapJsonDeserializer() {
}
// Just cheating the call site: we always return the same instance if the call site requests for a specially typed deserializer (it's always the same instance however, this is just how Java generics work)
static <T> JsonDeserializer<ResponseWrap<T>> getResponseWrapJsonDeserializer() {
#SuppressWarnings({ "rawtypes", "unchecked" })
final JsonDeserializer<ResponseWrap<T>> cast = (JsonDeserializer) responseWrapJsonDeserializer;
return cast;
}
#Override
public ResponseWrap<T> deserialize(final JsonElement jsonElement, final Type type, final JsonDeserializationContext context)
throws JsonParseException {
// Checking if jsonElement looks like an error (I'm not sure if it's possible to check HTTP statuses delegating them to request/response converters in Retrofit)
if ( isError(jsonElement) ) {
final List<ErrorResponse> errorResponses = context.deserialize(jsonElement, errorResponseListType);
return new ResponseWrap<>(null, errorResponses);
}
// If it does not look an error, then:
// * resolve what's the actual T in the given ResponseWrap<T>
// * deserialize the JSON tree as an instance of T -- it's like we're stripping the wrapper and then instantiate the wrap due to our rules
final T response = context.deserialize(jsonElement, resolveTypeParameter0(type));
return new ResponseWrap<>(response, null);
}
private static Type resolveTypeParameter0(final Type type) {
// The given type does not have parameterization?
if ( !(type instanceof ParameterizedType) ) {
// Then it's raw, simply <Object> or <?>
return Object.class;
}
// If it's parameterized, let's take it's first parameter as ResponseWrap is known to a have a single type parameter only
return ((ParameterizedType) type).getActualTypeArguments()[0];
}
// Some AI party here, he-he
private static boolean isError(final JsonElement jsonElement) {
if ( !jsonElement.isJsonArray() ) {
return false;
}
final JsonArray jsonArray = jsonElement.getAsJsonArray();
for ( final JsonElement innerJsonElement : jsonArray ) {
if ( !innerJsonElement.isJsonObject() ) {
return false;
}
final JsonObject innerJsonObject = innerJsonElement.getAsJsonObject();
final boolean looksLikeErrorObject = innerJsonObject.has("errorCode");
if ( !looksLikeErrorObject ) {
return false;
}
}
return true;
}
}
Next, register the deserializer for your Gson instance:
private static final Gson gson = new GsonBuilder()
.registerTypeAdapter(ResponseWrap.class, getResponseWrapJsonDeserializer())
.create();
And test it with
success.json
{
"foo": [1, 2, 3]
}
failure.json
[
{"errorCode": 10001},
{"errorCode": 10002}
]
// It's a constant
// Also, ResponseWrap<Map<String,List<Integer>>>.class is illegal in Java
private static final Type type = new TypeToken<ResponseWrap<Map<String, List<Integer>>>>() {
}.getType();
public static void main(String... args)
throws IOException {
final String successJson = getPackageResourceString(Q43525433.class, "success.json");
final String failureJson = getPackageResourceString(Q43525433.class, "failure.json");
final ResponseWrap<Map<String, List<Integer>>> success = gson.fromJson(successJson, type);
final ResponseWrap<Map<String, List<Integer>>> failure = gson.fromJson(failureJson, type);
System.out.println("SUCCESS: " + success.response);
for ( final ErrorResponse response : failure.errorResponses ) {
System.out.println("FAILURE: " + response.errorCode);
}
}
The output:
SUCCESS: {foo=[1, 2, 3]}
FAILURE: 10001
FAILURE: 10002
And yes, don't forget to add gson to Retrofit using GsonConverterFactory.create(gson).
Also, you might be interested in Json response parser for Array or Object that describe the almost the same issue but from another perspective.
I have an API that can return JSON arrays or objects. Example JSON object
{
"id": 1,
"name": "name"
}
JSON array:
[
{
"id": 1,
"name": "name"
},
{
"id": 1,
"name": "name"
}
]
When mapping a JSON object response to a POJO I use:
MyEntity myEntity = new Gson().fromJson(jsonString, MyEntity.class);
When mapping a JSON array response to an array of POJOs I use:
MyEntity[] myEntity = new GSON().fromJson(jsonString, MyEntity[].class);
How can I convert those two responses to the appropriate types dynamically?
NOTE: I can't modify the server response, this is a public API.
Thank you!
EDIT:
I am trying to implement a method that does this automatically but I am missing something. The method
public <T> T convertResponseToEntity(Class<T> classOfT)
{
JsonElement jsonElement = this.gson.fromJson(getResponseAsString(), JsonElement.class);
if (jsonElement.isJsonArray()) {
Type listType = new TypeToken<T>(){}.getType();
return this.gson.fromJson(getResponseAsString(), listType);
}
return this.gson.fromJson(getResponseAsString(), (Type) classOfT);
}
It returns a list of LinkedTreeMaps. How can I modify the code to return the same content as Object[]?
How can I convert those 2 responses dynamically to the appropriate type?
It depends on how to interpret the "appropriate type" here because it would lead to instanceof or visitor pattern to get the appropriate type once you try to handle the parsed-from-JSON object every time you need it. If you can't change the API, you can smooth the way you use it. One of possible options here is handling such response as if everything is a list. Even a single object can be handled as a list with one element only (and many libraries work with sequences/lists only having that fact: Stream API in Java, LINQ in .NET, jQuery in JavaScript, etc).
Suppose you have the following MyEntity class to handle the elements obtained from the API you need:
// For the testing purposes, package-visible final fields are perfect
// Gson can deal with final fields too
final class MyEntity {
final int id = Integer.valueOf(0); // not letting javac to inline 0 since it's primitive
final String name = null;
#Override
public String toString() {
return id + "=>" + name;
}
}
Next, let's create a type adapter that will always align "true" lists and single objects as if it were a list:
final class AlwaysListTypeAdapter<T>
extends TypeAdapter<List<T>> {
private final TypeAdapter<T> elementTypeAdapter;
private AlwaysListTypeAdapter(final TypeAdapter<T> elementTypeAdapter) {
this.elementTypeAdapter = elementTypeAdapter;
}
static <T> TypeAdapter<List<T>> getAlwaysListTypeAdapter(final TypeAdapter<T> elementTypeAdapter) {
return new AlwaysListTypeAdapter<>(elementTypeAdapter);
}
#Override
#SuppressWarnings("resource")
public void write(final JsonWriter out, final List<T> list)
throws IOException {
if ( list == null ) {
out.nullValue();
} else {
switch ( list.size() ) {
case 0:
out.beginArray();
out.endArray();
break;
case 1:
elementTypeAdapter.write(out, list.iterator().next());
break;
default:
out.beginArray();
for ( final T element : list ) {
elementTypeAdapter.write(out, element);
}
out.endArray();
break;
}
}
}
#Override
public List<T> read(final JsonReader in)
throws IOException {
final JsonToken token = in.peek();
switch ( token ) {
case BEGIN_ARRAY:
final List<T> list = new ArrayList<>();
in.beginArray();
while ( in.peek() != END_ARRAY ) {
list.add(elementTypeAdapter.read(in));
}
in.endArray();
return unmodifiableList(list);
case BEGIN_OBJECT:
return singletonList(elementTypeAdapter.read(in));
case NULL:
return null;
case END_ARRAY:
case END_OBJECT:
case NAME:
case STRING:
case NUMBER:
case BOOLEAN:
case END_DOCUMENT:
throw new MalformedJsonException("Unexpected token: " + token);
default:
// A guard case: what if Gson would add another token someday?
throw new AssertionError("Must never happen: " + token);
}
}
}
Gson TypeAdapter are designed to work in streaming fashion thus they are cheap from the efficiency perspective, but not that easy in implementation. The write() method above is implemented just for the sake of not putting throw new UnsupportedOperationException(); there (I'm assuming you only read that API, but don't know if that API might consume "either element or a list" modification requests). Now it's necessary to create a type adapter factory to let Gson pick up the right type adapter for every particular type:
final class AlwaysListTypeAdapterFactory
implements TypeAdapterFactory {
private static final TypeAdapterFactory alwaysListTypeAdapterFactory = new AlwaysListTypeAdapterFactory();
private AlwaysListTypeAdapterFactory() {
}
static TypeAdapterFactory getAlwaysListTypeAdapterFactory() {
return alwaysListTypeAdapterFactory;
}
#Override
public <T> TypeAdapter<T> create(final Gson gson, final TypeToken<T> typeToken)
throws IllegalArgumentException {
if ( List.class.isAssignableFrom(typeToken.getRawType()) ) {
final Type elementType = getElementType(typeToken);
// Class<T> instances can be compared with ==
final TypeAdapter<?> elementTypeAdapter = elementType == MyEntity.class ? gson.getAdapter(MyEntity.class) : null;
// Found supported element type adapter?
if ( elementTypeAdapter != null ) {
#SuppressWarnings("unchecked")
final TypeAdapter<T> castTypeAdapter = (TypeAdapter<T>) getAlwaysListTypeAdapter(elementTypeAdapter);
return castTypeAdapter;
}
}
// Not a type that can be handled? Let Gson pick a more appropriate one itself
return null;
}
// Attempt to detect the list element type
private static Type getElementType(final TypeToken<?> typeToken) {
final Type listType = typeToken.getType();
return listType instanceof ParameterizedType
? ((ParameterizedType) listType).getActualTypeArguments()[0]
: Object.class;
}
}
And how it's used after all:
private static final Type responseItemListType = new TypeToken<List<MyEntity>>() {
}.getType();
private static final Gson gson = new GsonBuilder()
.registerTypeAdapterFactory(getAlwaysListTypeAdapterFactory())
.create();
public static void main(final String... args) {
test("");
test("{\"id\":1,\"name\":\"name\"}");
test("[{\"id\":1,\"name\":\"name\"},{\"id\":1,\"name\":\"name\"}]");
test("[]");
}
private static void test(final String incomingJson) {
final List<MyEntity> list = gson.fromJson(incomingJson, responseItemListType);
System.out.print("LIST=");
System.out.println(list);
System.out.print("JSON=");
gson.toJson(list, responseItemListType, System.out); // no need to create an intermediate string, let it just stream
System.out.println();
System.out.println("-----------------------------------");
}
The output:
LIST=null
JSON=null
-----------------------------------
LIST=[1=>name]
JSON={"id":1,"name":"name"}
-----------------------------------
LIST=[1=>name, 1=>name]
JSON=[{"id":1,"name":"name"},{"id":1,"name":"name"}]
-----------------------------------
LIST=[]
JSON=[]
-----------------------------------
Just parse it into JsonElement and check actual element type:
Gson g = new Gson();
JsonParser parser = new JsonParser();
JsonElement e = parser.parse( new StringReader(jsonString) );
if(e instanceof JsonObject) {
MyEntity myEntity = g.fromJson(e, MyEntity.class);
} else {
MyEntity[] myEntity = g.fromJson(e, MyEntity[].class);
}
I have never done much with serialization, but am trying to use Google's gson to serialize a Java object to a file. Here is an example of my issue:
public interface Animal {
public String getName();
}
public class Cat implements Animal {
private String mName = "Cat";
private String mHabbit = "Playing with yarn";
public String getName() {
return mName;
}
public void setName(String pName) {
mName = pName;
}
public String getHabbit() {
return mHabbit;
}
public void setHabbit(String pHabbit) {
mHabbit = pHabbit;
}
}
public class Exhibit {
private String mDescription;
private Animal mAnimal;
public Exhibit() {
mDescription = "This is a public exhibit.";
}
public String getDescription() {
return mDescription;
}
public void setDescription(String pDescription) {
mDescription = pDescription;
}
public Animal getAnimal() {
return mAnimal;
}
public void setAnimal(Animal pAnimal) {
mAnimal = pAnimal;
}
}
public class GsonTest {
public static void main(String[] argv) {
Exhibit exhibit = new Exhibit();
exhibit.setAnimal(new Cat());
Gson gson = new Gson();
String jsonString = gson.toJson(exhibit);
System.out.println(jsonString);
Exhibit deserializedExhibit = gson.fromJson(jsonString, Exhibit.class);
System.out.println(deserializedExhibit);
}
}
So this serializes nicely -- but understandably drops the type information on the Animal:
{"mDescription":"This is a public exhibit.","mAnimal":{"mName":"Cat","mHabbit":"Playing with yarn"}}
This causes real problems for deserialization, though:
Exception in thread "main" java.lang.RuntimeException: No-args constructor for interface com.atg.lp.gson.Animal does not exist. Register an InstanceCreator with Gson for this type to fix this problem.
I get why this is happening, but am having trouble figuring out the proper pattern for dealing with this. I did look in the guide but it didn't address this directly.
Here is a generic solution that works for all cases where only interface is known statically.
Create serialiser/deserialiser:
final class InterfaceAdapter<T> implements JsonSerializer<T>, JsonDeserializer<T> {
public JsonElement serialize(T object, Type interfaceType, JsonSerializationContext context) {
final JsonObject wrapper = new JsonObject();
wrapper.addProperty("type", object.getClass().getName());
wrapper.add("data", context.serialize(object));
return wrapper;
}
public T deserialize(JsonElement elem, Type interfaceType, JsonDeserializationContext context) throws JsonParseException {
final JsonObject wrapper = (JsonObject) elem;
final JsonElement typeName = get(wrapper, "type");
final JsonElement data = get(wrapper, "data");
final Type actualType = typeForName(typeName);
return context.deserialize(data, actualType);
}
private Type typeForName(final JsonElement typeElem) {
try {
return Class.forName(typeElem.getAsString());
} catch (ClassNotFoundException e) {
throw new JsonParseException(e);
}
}
private JsonElement get(final JsonObject wrapper, String memberName) {
final JsonElement elem = wrapper.get(memberName);
if (elem == null) throw new JsonParseException("no '" + memberName + "' member found in what was expected to be an interface wrapper");
return elem;
}
}
make Gson use it for the interface type of your choice:
Gson gson = new GsonBuilder().registerTypeAdapter(Animal.class, new InterfaceAdapter<Animal>())
.create();
Put the animal as transient, it will then not be serialized.
Or you can serialize it yourself by implementing defaultWriteObject(...) and defaultReadObject(...) (I think thats what they were called...)
EDIT See the part about "Writing an Instance Creator" here.
Gson cant deserialize an interface since it doesnt know which implementing class will be used, so you need to provide an instance creator for your Animal and set a default or similar.
#Maciek solution works perfect if the declared type of the member variable is the interface / abstract class. It won't work if the declared type is sub-class / sub-interface / sub-abstract class unless we register them all through registerTypeAdapter(). We can avoid registering one by one with the use of registerTypeHierarchyAdapter, but I realize that it will cause StackOverflowError because of the infinite loop. (Please read reference section below)
In short, my workaround solution looks a bit senseless but it works without StackOverflowError.
#Override
public JsonElement serialize(T object, Type interfaceType, JsonSerializationContext context) {
final JsonObject wrapper = new JsonObject();
wrapper.addProperty("type", object.getClass().getName());
wrapper.add("data", new Gson().toJsonTree(object));
return wrapper;
}
I used another new Gson instance of work as the default serializer / deserializer to avoid infinite loop. The drawback of this solution is you will also lose other TypeAdapter as well, if you have custom serialization for another type and it appears in the object, it will simply fail.
Still, I am hoping for a better solution.
Reference
According to Gson 2.3.1 documentation for JsonSerializationContext and JsonDeserializationContext
Invokes default serialization on the specified object passing the specific type information. It should never be invoked on the element received as a parameter of the JsonSerializer.serialize(Object, Type, JsonSerializationContext) method. Doing so will result in an infinite loop since Gson will in-turn call the custom serializer again.
and
Invokes default deserialization on the specified object. It should never be invoked on the element received as a parameter of the JsonDeserializer.deserialize(JsonElement, Type, JsonDeserializationContext) method. Doing so will result in an infinite loop since Gson will in-turn call the custom deserializer again.
This concludes that below implementation will cause infinite loop and cause StackOverflowError eventually.
#Override
public JsonElement serialize(Animal src, Type typeOfSrc,
JsonSerializationContext context) {
return context.serialize(src);
}
I had the same problem, except my interface was of primitive type (CharSequence) and not JsonObject:
if (elem instanceof JsonPrimitive){
JsonPrimitive primitiveObject = (JsonPrimitive) elem;
Type primitiveType =
primitiveObject.isBoolean() ?Boolean.class :
primitiveObject.isNumber() ? Number.class :
primitiveObject.isString() ? String.class :
String.class;
return context.deserialize(primitiveObject, primitiveType);
}
if (elem instanceof JsonObject){
JsonObject wrapper = (JsonObject) elem;
final JsonElement typeName = get(wrapper, "type");
final JsonElement data = get(wrapper, "data");
final Type actualType = typeForName(typeName);
return context.deserialize(data, actualType);
}