This documentation is for the unreleased GraalVM version.Download Early Access Builds from GitHub.

Reachability Metadata

The dynamic language features of the JVM (including reflection and resource handling) compute the dynamically-accessed program elements such as invoked methods or resource URLs at runtime. The native-image tool performs static analysis while building a native binary to determine those dynamic features, but it cannot always exhaustively predict all uses. To ensure inclusion of these elements into the native binary, you should provide reachability metadata (in further text referred as metadata) to the native-image builder. Providing the builder with reachability metadata also ensures seamless compatibility with third-party libraries at runtime.

Metadata can be provided to the native-image builder in following ways:

Table of Contents #

Computing Metadata in Code #

Computing metadata in code can be achieved in two ways:

  1. By providing constant arguments to functions that dynamically access elements of the JVM. A good example of such a function is the Class.forName method. In the following code:

     class ReflectiveAccess {
         public Class<Foo> fetchFoo() throws ClassNotFoundException {
             return Class.forName("Foo");
         }
     }
    

    the Class.forName("Foo") will be computed into a constant when native binary is built and stored in its initial heap. If the class Foo does not exist, the call will be transformed into throw ClassNotFoundException("Foo").

  2. By initializing classes at build time and storing dynamically accessed elements into the initial heap of the native executable. For example:

     class InitializedAtBuildTime {
         private static Class<?> aClass;
         static {
             try {
                 aClass = Class.forName(readFile("class.txt"));
             } catch (ClassNotFoundException e) {
                 throw RuntimeException(e);
             }
         }
    
         public Class<?> fetchClass() {
             return aClass;
         }
     }
    

When metadata is computed in code, the dynamically accessed elements will be included into the native executable’s heap only if that part of the heap is reachable through an enclosing method (for example, ReflectiveAccess#fetchFoo) or a static field (for example, InitializedAtBuildTime.aClass).

Specifying Metadata with JSON #

Each dynamic Java feature that requires metadata has a corresponding JSON file named <feature>-config.json. The JSON file consists of entries that tell Native Image the elements to include. For example, Java reflection metadata is specified in reflect-config.json, and a sample entry looks like:

{
  "type": "Foo"
}

Each entry in json-based metadata should be conditional to avoid unnecessary growth in the size of the native binary. A condition is specified in the following way:

{
  "condition": {
    "typeReachable": "<fully-qualified-class-name>"
  },
  <metadata-entry>
}

An entry with a typeReachable condition is considered only when the fully-qualified class is reachable. Currently, we support only typeReachable as a condition.

Find more examples of the configuration files in the GraalVM Reachability Metadata repository.

Metadata Types #

Native Image accepts the following types of reachability metadata:

  • Java reflection (the java.lang.reflect.* API) enables Java code to examine its own classes, methods, fields, and their properties at run time.
  • JNI allows native code to access classes, methods, fields and their properties at run time.
  • Resources and Resource Bundles allow arbitrary files present in the application to be loaded.
  • Dynamic JDK Proxies create classes on demand that implement a given list of interfaces.
  • Serialization enables writing and reading Java objects to and from streams.
  • Predefined Classes provide support for dynamically generated classes.

Reflection #

Computing Reflection Metadata in Code #

Some reflection methods are treated specially and are evaluated at build time when given constant arguments. These methods, in each of the listed classes, are:

  • java.lang.Class: getField, getMethod, getConstructor, getDeclaredField, getDeclaredMethod, getDeclaredConstructor, forName, getClassLoader
  • java.lang.invoke.MethodHandles: publicLookup, privateLookupIn, arrayConstructor, arrayLength, arrayElementGetter, arrayElementSetter, arrayElementVarHandle, byteArrayViewVarHandle, byteBufferViewVarHandle, lookup
  • java.lang.invoke.MethodHandles.Lookup: in , findStatic , findVirtual , findConstructor , findClass , accessClass , findSpecial , findGetter , findSetter , findVarHandle , findStaticGetter , findStaticSetter , findStaticVarHandle , unreflect , unreflectSpecial , unreflectConstructor , unreflectGetter , unreflectSetter , unreflectVarHandle
  • java.lang.invoke.MethodType: methodType, genericMethodType, changeParameterType, insertParameterTypes, appendParameterTypes, replaceParameterTypes, dropParameterTypes, changeReturnType, erase, generic, wrap, unwrap, parameterType, parameterCount, returnType, lastParameterType

Below are examples of calls that are replaced with the corresponding metadata element:

Class.forName("java.lang.Integer")
Class.forName("java.lang.Integer", true, ClassLoader.getSystemClassLoader())
Class.forName("java.lang.Integer").getMethod("equals", Object.class)
Integer.class.getDeclaredMethod("bitCount", int.class)
Integer.class.getConstructor(String.class)
Integer.class.getDeclaredConstructor(int.class)
Integer.class.getField("MAX_VALUE")
Integer.class.getDeclaredField("value")

When passing constant arrays, the following approaches to declare and populate an array are equivalent from the point of view of the native-image builder:

Class<?>[] params0 = new Class<?>[]{String.class, int.class};
Integer.class.getMethod("parseInt", params0);
Class<?>[] params1 = new Class<?>[2];
params1[0] = Class.forName("java.lang.String");
params1[1] = int.class;
Integer.class.getMethod("parseInt", params1);
Class<?>[] params2 = {String.class, int.class};
Integer.class.getMethod("parseInt", params2);

Specifying Reflection Metadata in JSON #

Reflection metadata should be specified in a reflect-config.json file and conform to the JSON schema defined in reflect-config-schema-v1.1.0.json. The schema also includes further details and explanations how this configuration works. Here is the example of the reflect-config.json:

[
    {
        "condition": {
            "typeReachable": "<condition-class>"
        },
        "type": "<class>",
        "methods": [
            {"name": "<methodName>", "parameterTypes": ["<param-one-type>"]}
        ],
        "queriedMethods": [
            {"name": "<methodName>", "parameterTypes": ["<param-one-type>"]}
        ],
        "fields": [
            {"name": "<fieldName>"}
        ],
        "allDeclaredClasses": true,
        "allDeclaredMethods": true,
        "allDeclaredFields": true, 
        "allDeclaredConstructors": true,
        "allPublicClasses": true,
        "allPublicMethods": true,
        "allPublicFields": true,
        "allPublicConstructors": true,
        "allRecordComponents": true,
        "allNestMembers": true,
        "allSigners": true,
        "allPermittedSubclasses": true,
        "queryAllDeclaredMethods": true,
        "queryAllDeclaredConstructors": true,
        "queryAllPublicMethods": true,
        "queryAllPublicConstructors": true,
        "unsafeAllocated": true
    }
]

Java Native Interface #

Java Native Interface (JNI) allows native code to access arbitrary Java types and type members. Native Image cannot predict what such native code will lookup, write to or invoke. To build a native binary for a Java application that uses JNI, JNI metadata is most likely required. For example, the given C code:

jclass clazz = FindClass(env, "java/lang/String");

looks up the java.lang.String class, which can then be used, for example, to invoke different String methods. The generated metadata entry for the above call would look like:

{
  "type": "java.lang.String"
}

JNI Metadata In Code #

It is not possible to specify JNI metadata in code.

JNI Metadata in JSON #

JNI metadata should be specified in a jni-config.json file and conform to the JSON schema defined in jni-config-schema-v1.1.0.json. The schema also includes further details and explanations how this configuration works. The example of jni-config.json is the same as the example of reflect-config.json described above.

Resources and Resource Bundles #

Java is capable of accessing any resource on the application class path, or the module path for which the requesting code has permission to access. Resource metadata instructs the native-image builder to include specified resources and resource bundles in the produced binary. A consequence of this approach is that some parts of the application that use resources for configuration (such as logging) are effectively configured at build time.

Resource Metadata In Code #

Native Image will detect calls to java.lang.Class#getResource and java.lang.Class#getResourceAsStream in which:

  • The class on which these methods are called is constant
  • The first parameter, name, is a constant and automatically register such resources.

The code below will work out of the box, because:

  • We are using a class literal (Example.class)
  • We are using a string literal as the name parameter
    class Example {
     public void conquerTheWorld() {
         ...
         InputStream plan = Example.class.getResourceAsStream("plans/v2/conquer_the_world.txt");
         ...
     }
    }
    

Resource Metadata in JSON #

Resource metadata should be specified in a resource-config.json file and conform to the JSON schema defined in resource-config-schema-v1.0.0.json. The schema also includes further details and explanations how this configuration works. Here is the example of the resource-config.json:

{
  "resources": {
    "includes": [
      {
        "condition": {
          "typeReachable": "<condition-class>"
        },
        "pattern": ".*\\.txt"
      }
    ],
    "excludes": [
      {
        "condition": {
          "typeReachable": "<condition-class>"
        },
        "pattern": ".*\\.txt"
      }
    ]
  },
  "bundles": [
    {
      "condition": {
        "typeReachable": "<condition-class>"
      },
      "name": "fully.qualified.bundle.name",
      "locales": ["en", "de", "sk"]
    },
    {
      "condition": {
        "typeReachable": "<condition-class>"
      },
      "name": "fully.qualified.bundle.name",
      "classNames": [
        "fully.qualified.bundle.name_en",
        "fully.qualified.bundle.name_de"
      ]
    }
  ]
}

Dynamic Proxy #

The JDK supports generating proxy classes for a given interface list. Native Image does not support generating new classes at runtime and requires metadata to properly run code that uses these proxies.

Note: The order of interfaces in the interface list used to create a proxy matters. Creating a proxy with two identical interface lists in which the interfaces are not in the same order, creates two distinct proxy classes.

Code Example #

The following code creates two distinct proxies:

import java.lang.reflect.InvocationHandler;
import java.lang.reflect.Proxy;

interface IA {
}

interface IB {
}

class Example {
    public void doWork() {
        InvocationHandler handler;
        ...
        Object proxyOne = Proxy.newProxyInstance(Example.class.getClassLoader(), new Class[]{IA.class, IB.class}, handler);
        Object proxyTwo = Proxy.newProxyInstance(Example.class.getClassLoader(), new Class[]{IB.class, IA.class}, handler);
        ...
    }
}

Dynamic Proxy Metadata In Code #

The following methods are evaluated at build time when called with constant arguments:

  • java.lang.reflect.Proxy.getProxyClass
  • java.lang.reflect.Proxy.newProxyInstance

Dynamic Proxy Metadata in JSON #

Dynamic proxy metadata should be specified in a proxy-config.json file and conform to the JSON schema defined in proxy-config-schema-v1.0.0.json. The schema also includes further details and explanations how this configuration works. Here is the example of the proxy-config.json:

[
  {
    "condition": {
      "typeReachable": "<condition-class>"
    },
    "interfaces": [
      "IA",
      "IB"
    ]
  }
]

Serialization #

Java can serialize any class that implements the Serializable interface. Native Image supports serialization with proper serialization metadata registration. This is necessary as serialization usually requires reflectively accessing the class of the object that is being serialized.

Serialization Metadata Registration In Code #

Native Image detects calls to ObjectInputFilter.Config#createFilter(String pattern) and if the pattern argument is constant, the exact classes mentioned in the pattern will be registered for serialization. For example, the following pattern will register the class pkg.SerializableClass for serialization:

  var filter = ObjectInputFilter.Config.createFilter("pkg.SerializableClass;!*;")
  objectInputStream.setObjectInputFilter(proof);

Using this pattern has a positive side effect of improving security on the JVM as only pkg.SerializableClass can be received by the objectInputStream.

Wildcard patterns do the serialization registration only for lambda-proxy classes of an enclosing class. For example, to register lambda serialization in an enclosing class pkg.LambdaHolder use:

  ObjectInputFilter.Config.createFilter("pkg.LambdaHolder$$Lambda*;")

Patterns like "pkg.**" and "pkg.Prefix*" will not perform serialization registration as they are too general and would increase image size significantly.

For calls to the sun.reflect.ReflectionFactory#newConstructorForSerialization(java.lang.Class) and sun.reflect.ReflectionFactory#newConstructorForSerialization(java.lang.Class, ) native image detects calls to these functions when all arguments and the receiver are constant. For example, the following call will register SerializlableClass for serialization:

  ReflectionFactory.getReflectionFactory().newConstructorForSerialization(SerializableClass.class);

To create a custom constructor for serialization use:

  var constructor = SuperSuperClass.class.getDeclaredConstructor();
  var newConstructor = ReflectionFactory.getReflectionFactory().newConstructorForSerialization(BaseClass.class, constructor);

Proxy classes can only be registered for serialization via the JSON files.

Serialization Metadata in JSON #

Serialization metadata should be specified in a serialization-config.json file and conform to the JSON schema defined in serialization-config-schema-v1.1.0.json. The schema also includes further details and explanations how this configuration works. Here is the example of the serialization-config.json:

{
  "types": [
    {
      "condition": {
        "typeReachable": "<condition-class>"
      },
      "type": "<fully-qualified-class-name>",
      "customTargetConstructorClass": "<custom-target-constructor-class>"
    }
  ],
  "lambdaCapturingTypes": [
    {
      "condition": {
        "typeReachable": "<condition-class>"
      },
      "name": "<fully-qualified-class-name>"
    }
  ],
 "proxies": [
   {
     "condition": {
       "typeReachable": "<condition-class>"
     },
     "interfaces": ["<fully-qualified-interface-name-1>", "<fully-qualified-interface-name-n>"]
   }
 ]
}

Predefined Classes #

Native Image requires all classes to be known at build time (a “closed-world assumption”).

However, Java has support for loading new classes at runtime. To emulate class loading, the agent can trace dynamically loaded classes and save their bytecode for later use by the native-image builder. At runtime, if there is an attempt to load a class with the same name and bytecode as one of the classes encountered during tracing, the predefined class will be supplied to the application.

Note: Predefined classes metadata is not meant to be manually written.

Predefined Classes Metadata In Code #

It is not possible to specify predefined classes in code.

Predefined Classes Metadata in JSON #

Predefined classes metadata should be specified in a predefined-classes-config.json file and conform to the JSON schema defined in predefined-classes-config-schema-v1.0.0.json. The schema also includes further details and explanations how this configuration works. Here is the example of the predefined-classes-config.json:

[
  {
    "type": "agent-extracted",
    "classes": [
      {
        "hash": "<class-bytecodes-hash>",
        "nameInfo": "<class-name"
      }
    ]
  }
]

Strict Metadata Mode #

Native Image’s strict metadata mode helps ensure the correctness and composability of the Native Image metadata, by strengthening the metadata requirements for reflection queries. This mode can be activated with the -H:ThrowMissingRegistrationErrors= option and requires the following additional registrations over the default:

Reflection #

  • If a reflectively-accessed element (Class, Field, Method, etc.) is not present on the image class- or module-path, it still needs to be registered to ensure the correct exception (ClassNotFoundException or similar) is thrown. If an element is queried at runtime without having been registered, regardless of whether it is present on the class- or module-path, this query will throw a MissingReflectionRegistrationError. This change ensures that the error is not ambiguous between a non-existent element and one that was not registered for reflection in the image;
  • This rationale also requires that any query that returns a collection of class members (Class.getMethods() or similar) has to be registered in full (with "queryAllPublicMethods" in this case) to succeed at runtime. This additionally ensures that any of the registered elements can be queried individually, and non-existent elements of that type will throw the correct exception without having to be registered. However, this means that Class.getMethods() does not return the subset of methods that were registered, but throws a MissingReflectionRegistrationError if "queryAllPublicMethods" is missing.

Resources #

  • If a resource or resource bundle is not present on the image class- or module-path, it still needs to be registered to ensure the correct return value (null). If a resource is queried at runtime without having been registered, regardless of whether it is present on the class- or module-path, this query will throw a MissingResourceRegistrationError. This change ensures that the program behavior is not ambiguous between a non-existent resource and one that was not registered for runtime access;

The Native Image agent does not support custom implementations of ResourceBundle$Control or Bundles$Strategy and requires manual registrations for the reflection and resource queries that they will perform.

Transition tools #

This mode will be made the default behavior of Native Image in a future release. We encourage you to start transitioning your code as soon as possible. The Native Image agent outputs JSON files that conform to both the default and strict modes of operation. Native Image also provides some useful options for debugging issues during the transition to the strict mode:

  • To make sure that the reflection for your image is configured correctly you can add -H:ThrowMissingRegistrationErrors= to the native-image build arguments. If the resulting image fails in libraries that are not under your control, you can add a package prefix to the option to limit the errors to operations called from classes within the specified packages: -H:ThrowMissingRegistrationErrors=<package-prefix>.
  • The default behavior under -H:ThrowMissingRegistrationErrors= is to throw an error, which will potentially end the program execution. To get an overview of all places in your code where missing registrations occur, including a small stack trace, without committing to the strict behavior you can add -XX:MissingRegistrationReportingMode=Warn to the program invocation. To detect places where the application accidentally swallows a missing registration error (such as with blanket catch (Throwable t) blocks), you can add -XX:MissingRegistrationReportingMode=Exit to the program invocation. The application will then unconditionally print the error message and stack trace and exit immediately without throwing.

Further Reading #

Connect with us