Search this site


 Home  JNI Intro  JNI data types  JNI overhead

Search this site:
Threads Database Profiling Regular expressions Random numbers Compression Exceptions C Equivalents in Java

 What do you think of this article? Did it help you? Found a mistake? Feedback and suggestions here

Getting started with JNI

Having introduced the principle of the Java Native Interface, we'll see how to write our first native method. The stages are essentially as follows:

  • create a C/C++ project that will build to a DLL (or dynamic library on the platform in question);
  • ensure the project includes the directories containing the JNI headers in its build path;
  • write the Java method signatures for our native methods;
  • run the javah tool to create C header files from those method signatures and add the headers to our project;
  • based on those headers, write the native implementations of our methods (in C/C++);
  • compile the DLL or dynamic library;
  • in the Java code, call System.loadLibrary() to link in the library at runtime, then call our native methods.

Creating the DLL/library project

In Visual Studio, create a new Win32 Project. Then, in the Application Wizard that appears, click Next to display the Application Settings, from where you should select DLL as the application type, as illustrated below:

Selecting an application type of DLL from the Visual Studio Application Settings panel of the Win32 Application Wizard.

Then, we need to tell the compiler where to find the JNI headers. These live inside a directory called jdk/include inside the directory in which your JDK has been installed. You need to include jdk/include and also any subdirectory for the given platform, e.g. jdk/include/win32. To add these to the compiler's search path in Visual Studio, right-click on the project name in the project tree, then select Properties (see illustration). Then, in the C/C++ General section, add the aforementioned directories to Additional Include Directories as illustrated.

Adding the JNI headers to the compiler's include path.

Writing the Java method signature

Next, we write the signature of the method that we're going to implement natively. This generally looks like a regular Java method signature, but with the native modifier. Superficially, the method is also like an abstract method in the sense that the signature ends in a semicolon and there is no method body. (But note that strictly an abstract method is one that will be overridden.)

In principle, native methods can have the same signature as plain old Java methods. For example, you can pass in Strings, arrays and other objects to the native method. In practice, it is worth bearing in mind that accessing object fields is slightly fiddly (though perfectly possible) from native code; accessing primitives is generally easier, and a good strategy for keeping the native code simple can be to extract object fields from the Java side and pass them to the native method as primitives. As our first example, we're going to implement a native method that takes a number, doubles it and returns the result back to Java. Not terribly exciting, but it will illustrate a few important concepts.

For our simple example, we create a Java project with a single class Test inside the package test. To this class, we add a native static method called getDouble() that will take an integer and return one:

package test;

public class Test {
  public native static int getDouble(int n);
}

Generating the header file

Now, we need to make sure that our DLL has a function whose signature corresponds in a particular way to the Java method. As you get more experienced with JNI programming, you can predict what the name will need to be. But to start with, it's easier to use the javah tool supplied with the JDK.

The javah tool takes a given Java class and generates a C header file that contains functions with the correct signatures for all of the native methods defined in that class. For example, from inside our Java project's directory, we can run the following command (obviously supplying the appropriate path to the JDK as necessary):

C:\Java\JDK5\jdk\bin\javah -classpath . test.Test

If all is well, this will generate a file called test_Test.h that contains among other things the following line:

JNIEXPORT jint JNICALL Java_test_Test_getDoubled(JNIEnv *, jclass, jint);

The header file will also contain some boilerplate code to include the JNI headers and to ensure that if the compiler is running in C++ mode, it doesn't mangle the function names in the resulting DLL or library. For this reason, you should generally not try and bypass the stage of including the header file in your C/C++ project. To add the header file to the C/C++ project, copy it into the latter's directory, and then add the header to the project in your IDE as appropriate. For example, in Visual Studio, you right-click on Headers in the project tree, then select Add > Existing Item.

Write the native method implementation and compile the DLL/library

Now, we need to write the C function that will be our native method implementation. To do this, we first copy the function signature into a C/C++ source file. If you're using an IDE such as Visual Studio, it will probably have created a default source file to which you can add the function. In my example project called "TestDLL", for example, Visual Studio added a blank source file called TestDLL.cpp. So I copy the above line into TestDLL.cpp, adding names for the variables and a method body. I also include the header file, because this in turn includes the JNI headers, which we'll need to access JNI functions and data types. The result looks as follows:

#include "stdafx.h"
#include "test_Test.h"

JNIEXPORT jint JNICALL Java_test_Test_getDoubled(JNIEnv *env,
           jclass clz, jint n) {
	return n * 2;
}

Note that the very first line (to include required standard headers) was added by the IDE. The gubbinry of our function is the simple line to return double the parameter n. Note that two other parameters are passed into our function. We ignore these for now, but potentially they provide various facilities such as a route to accessing fields of defined Java objects and other parts of the running "environment".

With a bit of luck, we should now be able to Build the project, resulting in a shiny DLL file (or other library file depending on the platform).

Including the library and calling the native method

Now finally from Java, we can call our native method. To do so, we need to:

  • make sure that the DLL/library can be found by the JVM;
  • add a line of Java to include the library at runtime.

Placing the DLL/library in either the classpath or in a standard place for the platform in question (such as C:\Windows\System or /usr/lib/) is usually sufficient for the JVM to be able to find it. Then, in Java, our main method looks as follws:

public static void main(String[] args) {
  System.loadLibrary("TestDLL");
  for (int n = 1; n <= 20; n++) {
    System.out.println(n + " x 2 = " + getDoubled(n));
  }
}

Note the call to System.loadLibrary(), which effectively takes the file name of the library minus any platform-dependent extension such as .DLL or .so.

With all the previous steps in place, our native method can then be called just like any other Java method.


Written by Neil Coffey. Copyright © Javamex UK 2009. All rights reserved.