NewStringUTF() and freeing memory

The storage for the const char* argument to NewStringUTF() is entirely your responsibility: if you allocated test with malloc(), then you need to free() it. So, the snippet you posted is correct. You are corrupting the heap somewhere else.

I see conflicting opinions. Some say I should free it myself, some say the VM frees it, some say the VM doesn't free it and you should do strange voodoo magic to free it. I'm confused.

They are talking about the jstring instance returned by NewStringUTF(). That follows the confusing rules for 'local references'.

It is never an error to release this reference with DeleteLocalRef() when you are finished with it. However, the JVM performs some dubious magic if you call NewStringUTF() in the context of a JVM thread. When the native method returns to Java, any leaked local references are automatically cleaned up. So if you are sure your ultimate caller is in a Java thread, then you can safely leak the reference.

On the other hand, if you are running in the context of a native thread - say, some event reporting thread making callbacks to Java - there never is a return to Java, so you must call DeleteLocalRef() yourself on this jstring (and indeed all the other local references returned by typical JNI calls).

You just need DeleteLocalRef(), NewStringUTF() is just malloc memory on JVM, which the JVM will take care of the memory.