A Java implementation of JMESPath
Clone or download
Latest commit d8bbfdc May 6, 2018



Build Status

If you're reading this on GitHub, please note that this is the readme for the development version and that some features described here might not yet have been released. You can find the readme for a specific version via the release tags (here is an example).

This is an implementation of JMESPath for Java and it supports searching JSON documents (via Jackson or Gson) and structures containing basic Java objects (Map, List, String, etc.) – but can also be extended to work with any JSON-like structure.


Using Maven you can add this to your dependencies:


Check the releases page for the value of ${jmespath.version}.

If you don't want both the Jackson and Gson implementations you can change it to jmespath-jackson or jmespath-gson. Some time before 1.0 the dependencies will probably be reversed so that jmespath will not depend on the specific runtimes, so you unless you have the need for multiple runtimes you should depend on the specific runtime you need.


jmespath-core has an ANTLR based parser, but the ANTLR runtime artifact has been shaded into the io.burt.jmespath package to avoid conflicts with other artifacts that may depend on ANTLR. This means that jmespath-core has no external dependencies.

jmespath-jackson obviously depends on Jackson, specifically Jackson DataBind (com.fasterxml.jackson.core:jackson-databind), but other than that it only depends on jmespath-core.

jmespath-gson depends on Gson, specifically com.google.code.gson:gson, but other than that only jmespath-core.

Basic usage

import com.fasterxml.jackson.databind.JsonNode;

import io.burt.jmespath.JmesPath;
import io.burt.jmespath.Expression;
import io.burt.jmespath.jackson.JacksonRuntime;


// The first thing you need is a runtime. These objects can compile expressions
// and they are specific to the kind of structure you want to search in.
// For most purposes you want the Jackson runtime, it can search in JsonNode
// structures created by Jackson.
JmesPath<JsonNode> jmespath = new JacksonRuntime();
// Expressions need to be compiled before you can search. Compiled expressions
// are reusable and thread safe. Compile your expressions once, just like database
// prepared statements.
Expression<JsonNode> expression = jmespath.compile("locations[?state == 'WA'].name | sort(@) | {WashingtonCities: join(', ', @)}");
// This you have to fill in yourself, you're probably using Jackson's ObjectMapper
// to load JSON data, and that should fit right in here.
JsonNode input = …;
// Finally this is how you search a structure. There's really not much more to it.
JsonNode result = expression.search(input);


jmespath-java comes in three parts: jmespath-core, jmespath-jackson, and jmespath-gson. The former contains the expression parser, core runtime, default functions and a simple runtime adapter that can search structures made up from numbers, strings, booleans, List and Map available as io.burt.jmespath.jcf.JcfRuntime (for "Java Collections Framework"). The latter contains the Jackson and Gson runtime adapters, respectively, and is what you should be using most of the time. The JCF runtime is just for internal development and testing. It primarily exists to test that there's nothing runtime-specific in the implementation.


The runtime can be configured, although there aren't many configuration options yet.

To change the behaviour when a function receives an argument of the wrong type from throwing an exception to returning null you set the silentTypeErrors configuration to true, for example like this:

import io.burt.jmespath.RuntimeConfiguration;
import io.burt.jmespath.jackson.JacksonRuntime;

RuntimeConfiguration configuration = new RuntimeConfiguration.Builder()
JmesPath<JsonNode> jmespath = new JacksonRuntime(configuration);

Many functions don't allow null and most of the time you would deal with that by checking for null and letting the result be null. This configuration makes it possible to skip all those checks and make any type error in a function call behave as if the function call resulted in null. It can also be a performance boost by avoiding throwing exceptions when you expect that an expression can sometimes fail with a type error.


jmespath-java is designed to be extensible. You can extend it in two ways: by adding new functions, and by creating different runtime adapters. These are not mutually exclusive, if you write your custom functions the right way you can use them with any runtime, and vice-versa.

Adding custom functions

In addition to the built in functions like sort, to_string, and sum you can add your own. All you need to do is to create a class that extends io.burt.jmespath.function.BaseFunction (actually implement Function from the same package, but then you'd need to do much more work yourself) and then register it with your runtime.

Here's how you add a sin function:

import java.util.List;

import io.burt.jmespath.Adapter;
import io.burt.jmespath.JmesPathType;
import io.burt.jmespath.RuntimeConfiguration;
import io.burt.jmespath.function.BaseFunction;
import io.burt.jmespath.function.FunctionArgument;
import io.burt.jmespath.function.ArgumentConstraints;

// Functions must implement Function, for example by extending BaseFunction
public class SinFunction extends BaseFunction {
  public SinFunction() {
    // This is how you tell the type checker what arguments your function accepts

  protected <T> T callFunction(Adapter<T> runtime, List<FunctionArgument<T>> arguments) {
    // Arguments can be either values or expressions, but most functions only
    // accept expressions. You don't need to do any type checking here, the
    // the runtime has made sure that if this code runs the types are correct.
    T value = arguments.get(0).value();
    // Since we want to be able to use this function with all types of inputs
    // it needs to use the runtime to convert data types.
    double n = runtime.toNumber(value).doubleValue();
    // This is the actual function, the rest is wrapping, that's the price of
    // being generic and supporting multiple implementations.
    // There are abstract classes in the io.burt.jmespath.function
    // package that can be used to avoid having to write all of the wrapping
    // for some types of functions. This function could extend MathFunction,
    // for example.
    double s = Math.sin(n);
    // We must not forget to wrap the result using the runtime.
    return runtime.createNumber(s);


import com.fasterxml.jackson.databind.JsonNode;

import io.burt.jmespath.JmesPath;
import io.burt.jmespath.function.FunctionRegistry;
import io.burt.jmespath.jackson.JacksonRuntime;

// There's a default registry that contains the built in JMESPath functions
FunctionRegistry defaultFunctions = FunctionRegistry.defaultRegistry();
// And we can create a new registry with additional functions by extending it
FunctionRegistry customFunctions = defaultFunctions.extend(new SinFunction());
// To configure the runtime with the registry we need to create a configuration
RuntimeConfiguration configuration = new RuntimeConfiguration.Builder()
// And then create a runtime with the configuration
JmesPath<JsonNode> runtime = new JacksonRuntime(configuration);
// Now the function is available in expressions
JsonNode result = runtime.compile("sin(measurements.angle)").search(input);

You can provide a name for your function, but the default is that the name will be the snake cased version of the class name, minus the "Function" suffix. SinFunction becomes sin, MyAwesomeFunction becomes my_awesome, etc.

Your function class needs to tell the runtime about what arguments it accepts. The function in the example above specifies that it accepts a single number as argument. Have a look at the existing functions and the documentation for the ArgumentConstraints DSL to see what is possible.

Creating a runtime adapter

Creating a runtime adapter is a bit more work than adding a function, but not extremely so. What you need to do is to implement the io.burt.jmespath.Adapter interface. The easiest is to start by extending io.burt.jmespath.BaseRuntime, that way you don't need to implement some of the things that are common to most runtimes, like comparing values.

Most of the work a runtime does is converting back and forth between Java types. The core runtime can't know about all of the types of structures you want to search, but it knows that they will be JSON-like, so it uses the runtime adapters to help translate. The Jackson runtime adapter, for example, translates the world of JsonNodes into Java types like List and String when asked by the core runtime.

Runtime adapters can wrap libraries like Jackson and Gson, but can also make it possible, for example, to search Java beans with JMESPath by translating JMESPath operations to reflection calls. The structure to search doesn't have to be JSON, it just has to be JSON-like.

A good starting point for writing a new runtime adapter is reading the code of the existing adapters and the docs for Adapter and BaseAdapter. There are also JUnit tests in JmesPathRuntimeTest and JmesPathComplianceTest that can be subclassed and run against any runtime, and that will help you know when your runtime is complete.

How to build and run the tests

The best place to see how to build and run the tests is to look at the .travis.yml file, but if you just want to get going run:

$ git submodule update --init --recursive
$ mvn test

And all dependencies should be installed, the code compiled and the tests run.


© 2016-2018 Burt AB, see LICENSE.txt (BSD 3-Clause).