JNI wchar_t problem (unix platforms)

Question

Hi all,
i have such kind of broblem on unix platforms. I have unicode string in JNI and when i create jstring and return it to java the string is not completed. So,
wchar_t name[50] = L"Ob als Sch�ler";
jstring objectName = NULL;
len = wcslen(name);
objectName = (*env)->NewString(env, name, len);
return objectName;
I receive in Java "Ob als". I receive warning when compile the c code, theat argument #2 is not compatible with prototype. This happened because the second parameter is unsigned short, and wchar_t, which i passed, is four bytes. How can i fix this problem with wide char string? Is it a good idea to convert wchar_t string to char string and after that to create NewStringUTF? I noticed that when the len is twice bigger i.e. len = 2*wcslen(name) the string which is created is full. Is it correct to use this function with twice more length?
Thanks,
Valentin

Answer 1

The problem happens when the size of wchar_t mismatches the size of jchar. Under Linux, wchar_t has size 4 while jchar has size 2. Using the short-wchar option for GCC can surely set wchar_t to have size 2, but problems will occur with any library compiled without the option; this will include all standard libraries.
I have successfully passed strings from Java to JNI using the following workaround. My solution worked with Linux x64, but there is no reason it won't work in 32 bits.
Suppose we have a class with a getString native method returning a string from the C world, and a print method printing a string coming from Java using C' wprintf.
Here is the code for the getString method:
JNIEXPORT jstring JNICALL Java_TestJNI_getString
(JNIEnv * env, jclass cls) {
   const wchar_t* str = L"Here is a quite long and useless string. Note that this even works with non-ASCII characters, like in Allô les élèves.";
   size_t len = wcslen(str);
   if (sizeof(wchar_t) != sizeof(jchar)) {
      // In Linux, we get there.
      // First, allocate a buffer for the string to pass to Java, not forgetting the \0
      jchar* str2 = (jchar*)malloc((len+1)*sizeof(jchar));
      int i;
      for (i = 0; i < len; i++)
         // This discards two bytes in str, but these should be 0 in UTF-16
str2[i] = str[i];
str2[len] = 0;
// Now we have a compatible string!
jstring js = (*env)->NewString(env, str2, len);
// And we can free the buffer, to prevent memory leaks
free(str2);
return js;
else
// Under Windows, we don't need such a hack
return (*env)->NewString(env, str, len);
And now the code for print:JNIEXPORT void JNICALL Java_TestJNI_print
(JNIEnv * env, jclass cls, jstring js) {
const jchar* str = (*env)->GetStringChars(env, js, NULL);
if (sizeof(wchar_t) != sizeof(jchar)) {
// Size mismatch, so translate the string
// Get the length using Java (strlen or wcslen should return the wrong length, or even segfault)
jint len = (*env)->GetStringLength(env, js);
// Allocate a buffer, not forgetting the \0 terminating character
wchar_t* str2 = (wchar_t*)malloc((len+1)*sizeof(wchar_t));
int i;
for (i = 0; i < len; i++)
str2[i] = str[i];
str2[len] = 0;
wprintf(L"%ls\n", str2);
free(str2);
else {
// Under Windows, we don't need the hack
wprintf(L"%ls\n", str);
(*env)->ReleaseStringChars(env, js, str);