Passing (byref) String from Java to C++ via JNI

Question

I wish to pass a string to a C++ Dll as a parameter in a function. The problem is that I don't know howto receive back the data after it is filled in the C++ dll. I am trying to do what is called passing parameters by reference.
Java Code:
public class ABKeyBoard {
public native long leerBanda(int pista, String datos);
public static void main(String[] args) {
String datos=new String();
System.loadLibrary("ABKeyBoard");
new ABKeyBoard().leerBanda(1,datos);
System.out.println(datos); //the content of datos here is empty.
C++ Code:
Java_ABKeyBoard_leerBanda(JNIEnv *env, jobject obj,jint pista, jstring datos)
     char buffer[2024];
     memset(buffer,     0x00,     sizeof(buffer));
     strcpy(buffer, "xxxx");
     datos = env->NewStringUTF(buffer);
return;
Thanks for your help.

Answer 1

In java every parameter are always passed by value.
The datos parameter is a local copy of the string
reference you pass to the method.This is wrong. The String passed to the native method is the same String object you use in Java. Although everything is passed by value in Java, what is actually passed by value is the reference to the String. This means that you can modify the object you pass, but you are not allowed to change the reference to point to a totally different object. That is where the problem is coming in.
The trouble is that it is illegal to modify a String, even from native code. If you need to make changes in-place to the text, pass an array of chars (if your native code uses Unicode), an array of bytes (if it uses normal 8-bit characters) or a StringBuffer. You can legally modify any of these data structures with the new data. But the StringBuffer object is the only one whose length can be changed after it is created. Unfortunately it is also the hardest to use from JNI.
Generally I think you should always pass arrays of bytes/chars to native code instead of Strings when possible. They can be modified in place, and you can use String's methods to get a byte-array in the platform's proper encoding. Using the GetStringUTFChars method is problematic because UTF only maps directly onto ASCII in the case of characters which are in the set of 7-bit ASCII characters. Your code will do wrong things if your String happens to contain some other character, unless your native code expects UTF format strings.
The good news is that C(++) functions which return results in their arguments do not ordinarily change the length. So you should be able to allocate a byte[] or char[] ahead of time of the appropriate size (don't forget to add the trailing null, which is not a component of Java strings). I think byte[] or char[] is the best answer because you can easily map those onto C-style arrays with Get[Primitive]ArrayRegion; the return of that is suitable for passing directly to native code, as long as you have remembered the null-terminator. For instance you could do (*env)->GetByteArrayRegion(env, javaArray, 0, arrayLength, CArray) and then your CArray would be changed to point at the contents of the javaArray (note: it does not copy data into CArray, it changes CArray to point at the array contents, so do not allocate memory for CArray first). Then when you do ReleaseByteArrayRegion the results will be propagated back to Java.