JVM Seg Fault With Multithreaded JNI Code

Question

Hello all:
I'm having a problem with a multi threaded application. The application is used to bridge an exsisting C++ application with an exsisting Java application so I went with a JNI implementation. After I completed my implementation, I found that the bridge would crash after a few hours (or a few million JNI calls). At first I thought there was a memory leak but my memory usage was not going up (according to the task manager) and the heap was staying around 3 MB (according to a third party tool). I finally figured out that the problem was my JNI code. The JNI code is all in a wrapper DLL (built with MinGW) that four other DLL's call dynamically. So I made a test app to test the wrapper dll. The test app is a simple two threaded application that creates a java object and deletes the java object. CODE:
#include <process.h>
#include "JNIUtilityInterface.h"  // Link to the JNI dll (GetProcAdress, etc.)
void test(void * arg) {   
   for (int i = 0; i < 4000000; i++) {
      jobject jTest = createObjectByClass(NULL, "JNITest", "()V");
      deleteObject(jTest);
   int num = (int)arg;
   printf("done - %d\n", num);              
int main(int argc, char *argv[])
   clearJNIUtilityDll();
   initJNIUtilityDll();
   invokeJVM();         
   for (int i = 0; i < 3; i++) {
      if (i == 0) {
         _beginthread(test, 0, (void*)i);
      } else {           
         _beginthread(test, 0, (void*)i);
   Sleep(240000);
   exitJNIUtilityDll();
   return 0;
}createObjectByClass code:
jobject JNIUtility_API createObjectByClass(jboolean * hasException, 
                                           const char * className, 
                                           const char * descriptor, va_list args) {                                                                                                                    
   jclass clazz;
   jmethodID mid;
   jobject obj;
   JNIEnv * env;
   env = getEnv();
   if (env != NULL) {  
      if (env->EnsureLocalCapacity(2) == JNI_OK) {
         jclass clazz = env->FindClass(className);                                                
         jmethodID mid = env->GetMethodID(clazz, "<init>", descriptor);       
         if (mid) {
            obj = env->NewObjectV(clazz, mid, args);                                             
         env->DeleteLocalRef(clazz);          
      if (hasException) {
         *hasException = env->ExceptionCheck();
   env = NULL;
   return obj;
} getEnv code:
JNIEnv * getEnv() {          
   JNIEnv * env;
   jint result;
   result = jvm->GetEnv((void **)&env, JNI_VERSION_1_4);
   if (result == JNI_EDETACHED) {
      if (jvm->AttachCurrentThreadAsDaemon((void **)&env, NULL) == JNI_OK) {
         //printf("Attached current thread!\n");
   return env;
}deleteObject code:
void JNIUtility_API deleteObject(jobject obj) {
   JNIEnv * env;
   env = getEnv();
   if (env != NULL) {                                                                                                                                               
      env->DeleteLocalRef(obj);
   env = NULL;                             
}The other code is not important since I know this is where the seg fault is happening. Again this is just test code but when I run this code with only one thread, it works fine (with the 4000000 for loop). The JavaVM is globally cached as well.
If anybody nows how to properly run two or more native threads that use JNI methods, please let me know or if you have any tips, that will be good too! If you want any more code to try for yourself, I'm willing to send it to you.
Thanks!!

Answer 1

Nevermind, I found out that if you build you code using jni.h from 1.5.0_06 and run you code using 1.5.0_04 JRE, you run into problems with multi threading! Don't do this because it leads to headaches and high blood pressure :)