Json Jerk

Json Jerk is a flexible and fast JSON parser. It consists of several composable parts for tokenizing, (un)escaping, parsing, and handling of semantic actions. Furthermore it provides a light weight and type safe object model for JSON documents.

A simple example

The following example shows the simplest way to parse a JSON string:

String json = "{\"hello\":\"world\"}";
JsonObject jsonObject = FullJsonParser.parseObject(new UnescapingJsonTokenizer(json));

FullJsonParser is a utility class which provides methods for parsing (parts of) JSON documents. Its parseObject method takes a JsonTokenizer as argument and tries to parse an object from the input. In this case we've used a UnescapingJsonTokenizer instance for the tokenizer. This class unescapes all escape sequences it encounters. While this is generally the right thing to do, it can be very expensive in terms of performance. If we'd rather want the tokenizer not to unescape, we would use an instance of DefaultJsonTokenizer.

If parsing is successful, it results in a JsonObject instance. JsonObject is a subclass of JsonValue and represents a JSON object. There is also a JsonArray subclass and a JsonAtom subclass which represent JSON arrays and primitive JSON values, respectively.

The object model

To see how to use the JsonValue classes consider this more complex example:

String json =
    "{\"string\":\"value\"," +
    "\"number\":1.42," +
    "\"null\":null," +
    "\"false\":false," +
    "\"true\":true," +
    "\"nestedArray\":[[],[[],[]]]," +
    "\"array\":[1,2,3]," +
    "\"object\":{" +
        "\"size\":212," +
        "\"empty\":\"\"," +
        "\"array\":[1,2,3,{\"number\":142,\"array\":[1,2,3,{}]}]," +
        "\"object\":{}" +
    "}}";

JsonObject jsonObject = FullJsonParser.parseObject(new UnescapingJsonTokenizer(json));
StringBuilder sb = new StringBuilder();
printObject(jsonObject, sb);

The printObject method prints all members of the JSON object by calling printValue for each of them. The printValue method in turn determines the actual type of the JsonValue and calls one of the respective print methods.

private static void printValue(JsonValue jsonValue, StringBuilder sb) {
    switch (jsonValue.type()) {
        case OBJECT:
            printObject(jsonValue.asObject(), sb);
            break;
        case ARRAY:
            printArray(jsonValue.asArray(), sb);
            break;
        default:
            printAtom(jsonValue.asAtom(), sb);
    }
}

private static void printObject(JsonObject object, StringBuilder sb) {
    sb.append('{');
    String separator = "";
    for (Entry<String, JsonValue> pair : object.value().entrySet()) {
        sb.append(separator);
        separator = ",";
        sb.append('"').append(pair.getKey()).append("\":");
        printValue(pair.getValue(), sb);
    }
    sb.append('}');
}

private static void printArray(JsonArray array, StringBuilder sb) {
    sb.append('[');
    String separator = "";
    for (JsonValue value : array.value()) {
        sb.append(separator);
        separator = ",";
        printValue(value, sb);
    }
    sb.append(']');
}

private static void printAtom(JsonAtom atom, StringBuilder sb) {
    if (atom.type() == Type.STRING) {
        sb.append('"').append(JsonValue.escape(atom.value())).append('"');
    }
    else {
        sb.append(atom.value());
    }
}

Using a visitor on the object model

Another way to travers the object model is by implementing a 'JsonValue.Visitor' and passing it to the accept method of JsonValue. The following code - as the previous - translate the object model back to JSON.

final StringBuilder sb = new StringBuilder();
jsonObject.accept(new Visitor() {
    @Override
    public void visit(JsonAtom atom) {
        if (atom.type() == Type.STRING) {
            sb.append('"').append(JsonValue.escape(atom.value())).append('"');
        }
        else {
            sb.append(atom.value());
        }
    }

    @Override
    public void visit(JsonArray array) {
        sb.append('[');
        String separator = "";
        for (JsonValue value : array.value()) {
            sb.append(separator);
            separator = ",";
            value.accept(this);
        }
        sb.append(']');
    }

    @Override
    public void visit(JsonObject object) {
        sb.append('{');
        String separator = "";
        for (Entry<String, JsonValue> entry : object.value().entrySet()) {
            sb.append(separator);
            separator = ",";
            sb.append('"').append(entry.getKey()).append("\":");
            entry.getValue().accept(this);
        }
        sb.append('}');
    }
});

Both examples use the escape method of JsonValue for escaping strings.

Although these examples generate JSON from JsonValue, there is no need to do this manually in general. In fact, the toJson method of JsonValue does exactly this: it returns a JSON string.

Using call backs

The previous examples all created an object model of JsonValues from a JSON string. This is not the only way to parse a JSON document though. With Json Jerk we can also use an JsonHandler to get notified of parse events. Inside our JsonHandler we can then do whatever we want to handle these events. We could for example build a custom object model.

Suppose we want to separately collect the keys of all values, object and arrays of a JSON document. That is, we want to flat map these keys. Here is how:

String json = "{\"one\":1,\"two\":{\"three\":3,\"four\":[]}}";
final Set<String> atoms = new HashSet<String>();
final Set<String> objects = new HashSet<String>();
final Set<String> arrays = new HashSet<String>();

new JsonParser(new JsonHandler(){
    @Override
    public void atom(Token key, Token value) {
        atoms.add(key.text());
    }

    @Override
    public void object(JsonParser parser, Token key, JsonTokenizer tokenizer) {
        objects.add(key.text());
        super.object(parser, key, tokenizer);
    }

    @Override
    public void array(JsonParser parser, Token key, JsonTokenizer tokenizer) {
        arrays.add(key.text());
        super.array(parser, key, tokenizer);
    }
}).parseObject(new UnescapingJsonTokenizer(json));

The JsonParser calls the corresponding call back on the JsonHandler whenever it parses the respective JSON element. In contrast to other JSON parser, Json Jerk does not do recursive decent automatically. That is, when it encounters an object or an array, it will call the corresponding call back of the JsonHandler. It is then the handlers responsibility to parser the object or array itself. The easiest way to do this is to use the JsonParser and the JsonTokenizer passed to the handler. This is exactly what the method's super method does. So calling super is actually even easier.

Suppose we want to collect the keys as above but this time we want skip all keys from nested structures. To achieve this we need to change the parsing of nested structures: instead of calling super which effectively continues parsing with the current parser, we need to parse the nested structures with a different parser which does nothing but skip the whole structure:

String json = "{\"one\":1,\"two\":{\"three\":3,\"four\":[]}}";
final Set<String> atoms = new HashSet<String>();
final Set<String> objects = new HashSet<String>();
final Set<String> arrays = new HashSet<String>();

new JsonParser(new JsonHandler(){
    @Override
    public void atom(Token key, Token value) {
        atoms.add(key.text());
    }

    @Override
    public void object(JsonParser parser, Token key, JsonTokenizer tokenizer) {
        objects.add(key.text());
        new JsonParser(new JsonHandler()).parseObject(tokenizer);
    }

    @Override
    public void array(JsonParser parser, Token key, JsonTokenizer tokenizer) {
        arrays.add(key.text());
        new JsonParser(new JsonHandler()).parseArray(tokenizer);
    }
}).parseObject(new UnescapingJsonTokenizer(json));

In this example the super calls have been replaced by calls to new parsers which have an empty JsonHandler and thus do nothing but skip the current array or object.

In more complex scenarios, nested objects and arrays may be parsed with specific JsonHandlers. The FullJsonParser used in earlier examples to build the object models from JSON documents is implemented in this way: it uses a JsonParser and supplied it with different JsonHandler instances to build the hierarchical object model consisting of JsonValue objects.

Be lazy: level order parsing

Not having the parser to recursively descent on nested structures opens up some unexpected possibilities: we don't actually need to parse a nested object on a call to JsonHandler.object. What we could do instead, is to keep the state of the JsonTokenizer and parse it later when needed.

This is what the LevelOrderJsonParser does. From the outside, this parser look the same as the FullJsonParser. In contrast to the latter however, it does not parse nested structures right away but rather keeps a reference to a parser for them. When a nested structure is accessed through the corresponding JsonValue, that parser is invoked to parse the structure. This lazy approach results in a level order traversal of the JSON document compared to the post order traversal done by the FullJsonParser. Level order traversal can be beneficial if one needs to find all sub objects of a given objects without wanting to cope with deeper levels.