GC-free (green) JSON parser/generator library for Java which isn't an object mapper, but aims to be:
- minimal: MIT licensed with no dependencies
- reactive styled: the parser can parse a JSON document part by part (i.e. if you have received the part from a block device), even byte by byte; you don't need to iterate over all the tokens with verbose
if
orswitch
, just handle a callback you are interested in - lightweight: the code never recurses or allocates more memory than it needs, the Flyweight patern can be used to wrap a receive/send buffer to prevent memory copying
- fast: high performance, more than comparable with other state-of-the-art parsers like Gson and Jackson (see Performance). No additional GC pauses introduced, since the code doesn't allocate new memory in its main/critical path
- IO buffering oriented: the parser can be fed with several parts of one single JSON document, even byte by byte; so it's easy to use it with buffering read
- robust: built according to Ecma-404 with some extensions for the number values (see Numbers)
Binaries for Maven, Ivy, Gradle, and others can be found at https://central.sonatype.com/search?q=green-jelly.
Example for Maven:
<dependency>
<groupId>io.github.green4j</groupId>
<artifactId>green-jelly</artifactId>
<version>${greenJelly.version}</version>
</dependency>
Just run the standard Gradle build and install process:
./gradlew
The generic pattern of the usage if the following:
- create an instance of the JsonParser
- set a listener (an instance of the
JsonParserListener
interface) to get all parsing events - call the
parse
method as many times as many parts of a JSON document you have - finish the parsing with a call of the
eoj
(End Of JSON) method (or useparseAndEoj
method) - parse next JSON document with the same instance of the parser
final JsonParser parser = new JsonParser();
parser.setListener(new JsonParserListener() {
...
});
parser.parse("\"st");
parser.parse("ring\"");
parser.eoj();
parser.parseAndEoj("[1,2,3]");
While creating an instance of the JsonParser
, an instance of the JsonStringBuilder
can be passed to JsonParser
's constructor. The responsibility of the JsonStringBuilder
is to store characters of a string value. There are two implementations of the interface included to the library:
- CopyingStringBuilder: copies the character of string values to internal buffer. Should be used when not the whole JSON document can be parsed at once (i.e. you use reusable/mutable buffer to receive the data via
receive
system call). Also, this builder supports unescaping on-the-fly. - FlyweightStringBuilder: stores the reference to a CharSequence passed to the
parse
method and knows the length of the string value. This builder prevents you from memory copying, but it requires any string value must be fitted into one instance of CharSequence. Also, unescaping isn't supported, since the length of the result string value, passed to theonStringValue(CharSequence data)
callback, must be the same as the length of the original string.
While parsing you can be notified about the following events with an instance of the JsonParserListener:
void onJsonStarted();
void onError(String error, int position);
void onJsonEnded();
boolean onObjectStarted();
boolean onObjectMember(CharSequence name);
boolean onObjectEnded();
boolean onArrayStarted();
boolean onArrayEnded();
boolean onStringValue(CharSequence data);
boolean onNumberValue(JsonNumber number);
boolean onNumberValue(JsonNumber number, boolean overflow); // optional, has a default implementation
boolean onTrueValue();
boolean onFalseValue();
boolean onNullValue();
If any of method returns false
, the parsing stops and can continue then (see Parsing with steps).
There is a default implementation of the listener in the library JsonParserListenerAdapter
. All callbacks of the adaptor are empty and just return true
.
Supported format of number values is a bit more relaxed than specified in Ecma-404:
- the numbers can start with both
+
or-
- the leading zeros are allowed for both mantissa and exponent
- mantissa is presented by java's signed long which is a bit wider than JavaScript numbers [-(2^53)+1, (2^53)-1]
To prevent memory allocation and unnecessary computations, the library doesn't implement any fixed or floating point arithmetic. Numbers are presented with the following interface:
public interface JsonNumber {
long mantissa();
int exp();
}
Feel free to use any type of arithmetic like decimal4j, which supports GC-free calculations, out-of-the-box java.math.BigDecimal
, which a bit slow and allocates new memory, etc. An example of java.math.BigDecimal
using:
JsonNumber number = ...
BigDecimal decimal = BigDecimal.valueOf(number.mantissa(), -number.exp());
Sometimes vendor sends numbers as string values. To parse such values you can use static JsonParser.parseNumber
method:
MutableJsonNumber number = new MutableJsonNumber();
JsonParser.parseNumber("134.4455", number);
System.out.println(number);
If the parser detects an error while parsing, you receive a notification with the onError(String error, int position)
callback. Also, after the control is given back from the parse
method, you can check the error with the following methods:
boolean hasError()
String getError()
int getErrorPosition()
Sometimes you may need to split the process of parsing into some steps. In this case you return false
from any of JsonEventListener
's callback. The parsing stops with an instance of the Next
returned back. You call Next.next()
until it returns null
. For example, the following code prints each number on new line:
final var valueHolder = new Object() {
long value;
}; // mutable value with the final reference
final JsonParser parser = new JsonParser().setListener(
new JsonParserListenerAdaptor() {
@Override
public boolean onNumberValue(final JsonNumber number) {
valueHolder.value = number.mantissa();
return false; // stop the parsing after each number
}
}
);
JsonParser.Next nextStep = parser.parse("[0,10,20,30,40,5");
while (nextStep != null) {
System.out.println(valueHolder.value);
nextStep = nextStep.next();
}
nextStep = parser.parse("0,60,70,80,90,100]");
while (nextStep != null) {
System.out.println(valueHolder.value);
nextStep = nextStep.next();
}
parser.eoj();
An example:
final AppendableWriter<StringBuilder> writer = new AppendableWriter<>(new StringBuilder());
final JsonGenerator generator = new JsonGenerator();
generator.setOutput(writer);
generator.startArray();
generator.stringValue("\test1", true); // the value requires to be escaped
generator.stringValue("test2");
generator.stringValue("test3");
generator.endArray();
generator.eoj();
System.out.println(writer.output().toString());
The library works over character based abstractions, so, it doesn't implement any encoding functionality. As in case of Gson, for instance, the user has to care about correct bytes-to/from-chars transformation if any required.
At the same time, there are few specific classes to support JSON generation in different encodings if required:
Json Writer | Buffer Type/Encoding |
---|---|
io.github.green4j.jelly.AsciiByteArrayWriter |
Byte array of ASCII encoded data. All characters outside 0-127 range are escaped |
io.github.green4j.jelly.Utf8ByteArrayWriter |
Byte array of UTF-8 encoded data |
io.github.green4j.jelly.CharArrayWriter |
Char array. Can be used as is, or to be converted later to any other encoding |
Each of the classes has its own internal mutable buffer which can be passed to IO routines after JSON is generated.
A JMH test, which sums all numbers in the document in streaming style, compared to Gson (v.2.8.5) and Jackson (v.2.9.7):
Benchmark Mode Cnt Score Error Units
JsonParserPerformanceComparison.greenJellyFlyweightSumTest avgt 25 17747.965 ± 257.699 ns/op
JsonParserPerformanceComparison.gsonJsonReaderSumTest avgt 25 23442.288 ± 109.931 ns/op
JsonParserPerformanceComparison.jacksonJsonParserSumTest avgt 25 18336.310 ± 220.573 ns/op
Source code of the test
import com.fasterxml.jackson.core.JsonFactory;
import com.google.gson.stream.JsonReader;
import com.google.gson.stream.JsonToken;
import java.io.IOException;
import java.io.StringReader;
import java.util.concurrent.TimeUnit;
import FlyweightStringBuilder;
import JsonNumber;
import JsonParser;
import JsonParserListenerAdaptor;
import org.openjdk.jmh.annotations.Benchmark;
import org.openjdk.jmh.annotations.BenchmarkMode;
import org.openjdk.jmh.annotations.Level;
import org.openjdk.jmh.annotations.Mode;
import org.openjdk.jmh.annotations.OutputTimeUnit;
import org.openjdk.jmh.annotations.Scope;
import org.openjdk.jmh.annotations.Setup;
import org.openjdk.jmh.annotations.State;
import org.openjdk.jmh.annotations.TearDown;
@BenchmarkMode(Mode.AverageTime)
@OutputTimeUnit(TimeUnit.NANOSECONDS)
public class JsonParserPerformanceComparison {
public static final String JSON = "[\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":true},\n"
+ " {\"property1\":100,\"property2\":200,\"property3\":300,\"property4\":[400,500,600,700],\"property5\":false}\n"
+ "]";
@State(Scope.Thread)
public static class JellyFlyweightSum extends JsonParserListenerAdaptor {
public JsonParser parser;
public long value;
@Setup(Level.Invocation)
public void doSetup() {
value = 0;
parser = new JsonParser(new FlyweightStringBuilder());
parser.setListener(this);
}
@TearDown(Level.Invocation)
public void doTearDown() {
parser.eoj();
}
@Override
public boolean onNumberValue(final JsonNumber number) {
value += number.mantissa();
return true;
}
}
@State(Scope.Thread)
public static class GsonSum {
JsonReader reader;
long value;
@Setup(Level.Invocation)
public void doSetup() {
value = 0;
reader = new JsonReader(new StringReader(JSON));
}
@TearDown(Level.Invocation)
public void doTearDown() throws IOException {
reader.close();
}
}
@State(Scope.Thread)
public static class JacksonSum {
com.fasterxml.jackson.core.JsonParser parser;
long value;
@Setup(Level.Invocation)
public void doSetup() throws Exception {
value = 0;
final JsonFactory factory = new JsonFactory();
parser = factory.createParser(JSON);
}
@TearDown(Level.Invocation)
public void doTearDown() throws Exception {
parser.close();
}
}
@Benchmark
public void greenJellyFlyweightSumTest(final JellyFlyweightSum sum) {
sum.parser.parse(JSON);
}
@Benchmark
public void gsonJsonReaderSumTest(final GsonSum sum) throws IOException {
final JsonReader reader = sum.reader;
_end:
while (true) {
final JsonToken token = reader.peek();
switch (token) {
case BEGIN_ARRAY:
reader.beginArray();
break;
case END_ARRAY:
reader.endArray();
break;
case BEGIN_OBJECT:
reader.beginObject();
break;
case END_OBJECT:
reader.endObject();
break;
case NAME:
reader.skipValue();
break;
case STRING:
reader.skipValue();
break;
case NUMBER:
sum.value += reader.nextLong();
break;
case BOOLEAN:
reader.skipValue();
break;
case NULL:
reader.skipValue();
break;
case END_DOCUMENT:
break _end;
}
}
}
@Benchmark
public void jacksonJsonParserSumTest(final JacksonSum sum) throws IOException {
final com.fasterxml.jackson.core.JsonParser parser = sum.parser;
com.fasterxml.jackson.core.JsonToken token;
while ((token = parser.nextToken()) != null) {
if (token == com.fasterxml.jackson.core.JsonToken.VALUE_NUMBER_INT) {
sum.value += parser.getLongValue();
}
}
}
}
Sometimes we don't worry about memory consumption and CPU utilization, but we need a simple way to work
with JSON documents. And we would prefer to don't introduce complex POJO structures and involve object mapping.
A good example of approach like that is the JSON.simple
library. The Green-Jelly
also provides similar
package io.github.green4j.jelly.simple
.
The io.github.green4j.jelly.simple.JsonValue
class is the main class to construct a representation of a JSON document:
final JsonValue json = JsonValue.newObject();
final JsonObject user = json.asObject();
user.putString("name", "Mike");
user.putInteger("age", 23);
final JsonArray rates = user.putArray("rates");
rates.addInteger(10);
rates.addInteger(8);
A JsonValue
can be serialized with a io.github.green4j.jelly.simple.JsonWriter
or with a java.io.Writer
:
final JsonValue json = ...
final StringBuilder output = new StringBuilder();
final JsonWriter jsonWriter = new JsonWriterGenerator(new JsonGenerator(output));
json.toJsonAndEoj(jsonWriter);
System.out.println(output);
final StringWriter stringWriter = new StringWriter();
json.toJsonAndEoj(stringWriter);
System.out.println(stringWriter);
To parse JSON and build a JsonValue
, use the JsonValueParser
class:
final JsonValueParser parser = new JsonValueParser();
final JsonValue json = parser.parseAndEoj("{ \"name\": \"Mike\", \"age\": 23 }");
System.out.println(json);
The code is available under the terms of the MIT License.