Java records can not - by design - inherit from another object (see Why Java records do not support inheritance?). So I wonder what would be the best way to achieve the following.
Given my JSON data contains objects that have some common data + unique data. For example, type, width and height are in all shapes, but depending on the type, they can have additional fields:
{
"name": "testDrawing",
"shapes": [
{
"type": "shapeA",
"width": 100,
"height": 200,
"label": "test"
},
{
"type": "shapeB",
"width": 100,
"height": 200,
"length": 300
},
{
"type": "shapeC",
"width": 100,
"height": 200,
"url": "www.test.be",
"color": "#FF2233"
}
]
}
In "traditional" Java you would do this with
BaseShape with width and height
ShapeA extends BaseShape with label
ShapeB extends BaseShape with length
ShapeC extends BaseShape with URL and color
But I'm a bit stubborn and really would like to use records.
My solution now looks like this:
@JsonIgnoreProperties(ignoreUnknown = true)
@JsonInclude(JsonInclude.Include.NON_NULL)
public record Drawing(
@JsonProperty("name")
String name,
@JsonProperty("shapes")
@JsonDeserialize(using = TestDeserializer.class)
List<Object> shapes // I don't like the Objects here...
) {
}
@JsonIgnoreProperties(ignoreUnknown = true)
@JsonInclude(JsonInclude.Include.NON_NULL)
public record ShapeA (
@JsonProperty("type") String type,
@JsonProperty("width") Integer width,
@JsonProperty("height") Integer height,
@JsonProperty("label") String label
) {
}
@JsonIgnoreProperties(ignoreUnknown = true)
@JsonInclude(JsonInclude.Include.NON_NULL)
public record ShapeB(
@JsonProperty("type") String type,
@JsonProperty("width") Integer width,
@JsonProperty("height") Integer height,
@JsonProperty("length") Integer length
) {
}
@JsonIgnoreProperties(ignoreUnknown = true)
@JsonInclude(JsonInclude.Include.NON_NULL)
public record ShapeC(
@JsonProperty("type") String type,
@JsonProperty("width") Integer width,
@JsonProperty("height") Integer height,
@JsonProperty("url") String url,
@JsonProperty("color") String color
) {
}
I don't like repeated code and it's a bad practice... But in the end I can get this loaded with this helper class:
public class TestDeserializer extends JsonDeserializer {
ObjectMapper mapper = new ObjectMapper();
@Override
public List<Object> deserialize(JsonParser jsonParser, DeserializationContext deserializationContext) throws IOException {
List<Object> rt = new ArrayList<>();
JsonNode node = jsonParser.getCodec().readTree(jsonParser);
if (node instanceof ArrayNode array) {
for (Iterator<JsonNode> it = array.elements(); it.hasNext(); ) {
JsonNode childNode = it.next();
rt.add(getShape(childNode));
}
} else {
rt.add(getShape(node));
}
return rt;
}
private Object getShape(JsonNode node) {
var type = node.get("type").asText();
switch (type) {
case "shapeA":
return mapper.convertValue(node, ShapeA.class);
case "shapeB":
return mapper.convertValue(node, ShapeB.class);
case "shapeC":
return mapper.convertValue(node, ShapeC.class);
default:
throw new IllegalArgumentException("Shape could not be parsed");
}
}
}
And this test proves to be working OK:
@Test
void fromJsonToJson() throws IOException, JSONException {
File f = new File(this.getClass().getResource("/test.json").getFile());
String jsonFromFile = Files.readString(f.toPath());
ObjectMapper mapper = new ObjectMapper();
Drawing drawing = mapper.readValue(jsonFromFile, Drawing.class);
String jsonFromObject = mapper.writeValueAsString(drawing);
System.out.println("Original:\n" + jsonFromFile.replace("\n", "").replace(" ", ""));
System.out.println("Generated:\n" + jsonFromObject);
assertAll(
//() -> assertEquals(jsonFromFile, jsonFromObject),
() -> assertEquals("testDrawing", drawing.name()),
() -> assertTrue(drawing.shapes().get(0) instanceof ShapeA),
() -> assertTrue(drawing.shapes().get(1) instanceof ShapeB),
() -> assertTrue(drawing.shapes().get(2) instanceof ShapeC)
);
}
What would be the best way to achieve this with the Jackson library and Java Records?
Extra sidenote: I will also need to be able to write back to JSON in the same format as the original.
Records cannot inherit because they are intended to be a solid contract, but they can implement an interface. So you can do something like this with JasonSubTypes with Jackson 2.12 or above:
Models
@JsonIgnoreProperties(ignoreUnknown = true)
@JsonInclude(JsonInclude.Include.NON_NULL)
public record Drawing(
String name,
List<BaseShape> shapes
) { }
// added benefit of interface here is it reminds you to have the default fields
@JsonTypeInfo(use = JsonTypeInfo.Id.DEDUCTION)
@JsonSubTypes({
@JsonSubTypes.Type(ShapeA.class),
@JsonSubTypes.Type(ShapeB.class),
@JsonSubTypes.Type(ShapeC.class)
})
public interface BaseShape {
Integer width();
Integer height();
}
@JsonIgnoreProperties(ignoreUnknown = true)
@JsonInclude(JsonInclude.Include.NON_NULL)
public record ShapeA (
Integer width,
Integer height,
String label
) implements BaseShape { }
@JsonIgnoreProperties(ignoreUnknown = true)
@JsonInclude(JsonInclude.Include.NON_NULL)
public record ShapeB(
Integer width,
Integer height,
Integer length
) implements BaseShape { }
@JsonIgnoreProperties(ignoreUnknown = true)
@JsonInclude(JsonInclude.Include.NON_NULL)
public record ShapeC(
Integer width,
Integer height,
String url,
String color
) implements BaseShape { }
Test Class
@Slf4j
class DemoTest {
private ObjectMapper objectMapper = ObjectMapperBuilder.getObjectMapper();
@Test
void test() throws JsonProcessingException {
final String testString = objectMapper
.writerWithDefaultPrettyPrinter()
.writeValueAsString(
new Drawing(
"happy",
List.of(
new ShapeA(1, 1, "happyShape"),
new ShapeB(2, 2, 3),
new ShapeC(2, 2, "www.shape.com/shape", "blue"
)
)
)
);
log.info("From model to string {}", testString);
Drawing drawing = objectMapper.readValue(testString, Drawing.class);
log.info(
"Captured types {}",
drawing
.shapes()
.stream()
.map(s -> s.getClass().getName())
.collect(Collectors.toSet())
);
log.info(
"From string back to model then again to string {}",
objectMapper
.writerWithDefaultPrettyPrinter()
.writeValueAsString(drawing)
);
}
}
Here's the test log output:
17:06:41.293 [Test worker] INFO com.demo.DemoTest - From model to string {
"name" : "happy",
"shapes" : [ {
"width" : 1,
"height" : 1,
"label" : "happyShape"
}, {
"width" : 2,
"height" : 2,
"length" : 3
}, {
"width" : 2,
"height" : 2,
"url" : "www.shape.com/shape",
"color" : "blue"
} ]
}
17:06:41.353 [Test worker] INFO com.demo.DemoTest - Captured types [com.demo.DemoTest$ShapeB, com.demo.DemoTest$ShapeA, com.demo.DemoTest$ShapeC]
17:06:41.354 [Test worker] INFO com.demo.DemoTest - From string back to model then again to string {
"name" : "happy",
"shapes" : [ {
"width" : 1,
"height" : 1,
"label" : "happyShape"
}, {
"width" : 2,
"height" : 2,
"length" : 3
}, {
"width" : 2,
"height" : 2,
"url" : "www.shape.com/shape",
"color" : "blue"
} ]
}
Note that you can add the type field as a name property of the @JsonSubTypes.Type annotation, but this works with or without discriminator as long as the fields in your records are never exactly the same.
You can read more about JsonSubtypes here.