Gson serialization ignore empty string generically

Question

Is there a generic way to tell Gson to not write empty string?

I strongly dislike having to implement a TypeAdapter which handles every field as the answer here somewhat suggests.

Answer 1

Sort of. As far as I know, Gson does not let you control over object fields much, and the only control know to me you can have it @JsonAdapter. For example,

import com.google.gson.annotations.JsonAdapter;

final class Pack {

    @JsonAdapter(EmptyStringTypeAdapter.class)
    final String foo;

    @JsonAdapter(EmptyStringTypeAdapter.class)
    final String bar;

    private Pack(final String foo, final String bar) {
        this.foo = foo;
        this.bar = bar;
    }

    static Pack of(final String foo, final String bar) {
        return new Pack(foo, bar);
    }

    @Override
    public String toString() {
        return foo + " " + bar;
    }

}

Despite it might look tedious for someone, it gives your total control over your data transfer objects giving you a choice on what to do with this or that string. The example type adapter may be as follows:

final class EmptyStringTypeAdapter
        extends TypeAdapter<String> {

    private EmptyStringTypeAdapter() {
    }

    @Override
    @SuppressWarnings("resource")
    public void write(final JsonWriter jsonWriter, @Nullable final String s)
            throws IOException {
        if ( s == null || s.isEmpty() ) {
            jsonWriter.nullValue();
        } else {
            jsonWriter.value(s);
        }
    }

    @Override
    @Nonnull
    @SuppressWarnings("EnumSwitchStatementWhichMissesCases")
    public String read(final JsonReader jsonReader)
            throws IOException {
        final JsonToken token = jsonReader.peek();
        switch ( token ) {
        case NULL:
            return "";
        case STRING:
            return jsonReader.nextString();
        default:
            throw new IllegalStateException("Unexpected token: " + token);
        }
    }

}

One caveat here is that it cannot restore empty strings from nulls (and you're caught with irreversible conversion here unfortunately), so you might also want to take a look at https://github.com/google/gson/blob/master/extras/src/main/java/com/google/gson/typeadapters/PostConstructAdapterFactory.java to restore @JsonAdapter(EmptyStringTypeAdapter.class)-annotated fields on read. Example test:

private static final Gson gson = new Gson();

private static final Type listOfStringType = new TypeToken<List<String>>() {
}.getType();

public static void main(final String... args) {
    // Single elements
    ImmutableList.of(Pack.of("", ""), Pack.of("foo", ""), Pack.of("", "bar"), Pack.of("foo", "bar"))
            .stream()
            .peek(pack -> System.out.println("Pack before: " + pack))
            .map(gson::toJson)
            .peek(json -> System.out.println("JSON: " + json))
            .map(json -> gson.fromJson(json, Pack.class))
            .peek(pack -> System.out.println("Pack after: " + pack))
            .forEach(pack -> System.out.println());
    // Multiple elements
    final List<String> stringsBefore = ImmutableList.of("", "foo", "bar");
    System.out.println(stringsBefore);
    final String stringsJson = gson.toJson(stringsBefore, listOfStringType);
    System.out.println(stringsJson);
    final List<String> stringsAfter = gson.fromJson(stringsJson, listOfStringType);
    System.out.println(stringsAfter);
}

Output:

Pack before:
JSON: {}
Pack after: null null -- [!] not "" '"

Pack before: foo
JSON: {"foo":"foo"}
Pack after: foo null -- [!] not foo ""

Pack before: bar
JSON: {"bar":"bar"}
Pack after: null bar -- [!] not "" bar

Pack before: foo bar
JSON: {"foo":"foo","bar":"bar"}
Pack after: foo bar

[, foo, bar]
["","foo","bar"]
[, foo, bar]

However, I don't think that writing sophisticated (de)serialization strategies is a good choice, and you probably might be interested in redesigning your DTOs and data (de)serialization. Moreover, "" is a value, whilst null is not -- I would never mix them and I would revise why your system is designed that way (it really looks like an empty/null values mix issue).

Answer 2

Thanks to @Lyubomyr for his answer but I found a solution that fits our use case better:

If we set all empty strings and objects to null the leftover JSON after serializing only contains nodes with actual data:

 /**
   * convert object to json
   */
  public String toJson(Object obj) {
    // Convert emtpy string and objects to null so we don't serialze them
    setEmtpyStringsAndObjectsToNull(obj);
    return gson.toJson(obj);
  }

  /**
   * Sets all empty strings and objects (all fields null) including sets to null.
   *
   * @param obj any object
   */
  public void setEmtpyStringsAndObjectsToNull(Object obj) {
    for (Field field : obj.getClass().getDeclaredFields()) {
      field.setAccessible(true);
      try {
        Object fieldObj = field.get(obj);
        if (fieldObj != null) {
          Class fieldType = field.getType();
          if (fieldType.isAssignableFrom(String.class)) {
            if(fieldObj.equals("")) {
              field.set(obj, null);
            }
          } else if (fieldType.isAssignableFrom(Set.class)) {
            for (Object item : (Set) fieldObj) {
              setEmtpyStringsAndObjectsToNull(item);
            }
            boolean setFielToNull = true;
            for (Object item : (Set) field.get(obj)) {
              if(item != null) {
                setFielToNull = false;
                break;
              }
            }
            if(setFielToNull) {
              setFieldToNull(obj, field);
            }
          } else if (!isPrimitiveOrWrapper(fieldType)) {
            setEmtpyStringsAndObjectsToNull(fieldObj);
            boolean setFielToNull = true;
            for (Field f : fieldObj.getClass().getDeclaredFields()) {
              f.setAccessible(true);
              if(f.get(fieldObj) != null) {
                setFielToNull = false;
                break;
              }
            }
            if(setFielToNull) {
              setFieldToNull(obj, field);
            }
          }
        }
      } catch (IllegalAccessException e) {
        System.err.println("Error while setting empty string or object to null: " + e.getMessage());
      }
    }
  }

  private void setFieldToNull(Object obj, Field field) throws IllegalAccessException {
    if(!Modifier.isFinal(field.getModifiers())) {
      field.set(obj, null);
    }
  }

  private boolean isPrimitiveOrWrapper(Class fieldType)  {
    return fieldType.isPrimitive()
        || fieldType.isAssignableFrom(Integer.class)
        || fieldType.isAssignableFrom(Boolean.class)
        || fieldType.isAssignableFrom(Byte.class)
        || fieldType.isAssignableFrom(Character.class)
        || fieldType.isAssignableFrom(Float.class)
        || fieldType.isAssignableFrom(Long.class)
        || fieldType.isAssignableFrom(Double.class)
        || fieldType.isAssignableFrom(Short.class);
  }

Performance whise this runs reasonably quick. If you have a lot of empty fields this saves time (and space) when serializing and sending/writing to DB.

Answer 3

Since you are interested in omitting empty strings during serialization, you can only implement JsonSerializer instead of full TypeAdapter.

public class EmptyStringSerializer implements JsonSerializer<String> {
    @Override 
    public JsonElement serialize(String src, Type typeOfSrc, JsonSerializationContext context) {
        if (src == null || src.isEmpty())
            return JsonNull.INSTANCE;
        return new JsonPrimitive(src);
    }
}

And then use it same as a TypeAdapter:

Gson gson = new GsonBuilder()
                .registerTypeAdapter(String.class, new EmptyStringSerializer())
                .create();

(This approach relies on omitting null values from serialization.)