Low level synchronization in legacy library
1 posts in topic
Flat View  Flat View
Older Post First |   Reply to Topic
TOPIC ACTIONS:
 

Posted By:   Gary_Feldman
Posted On:   Friday, September 7, 2001 10:43 AM

I have an existing C++ library that I need to make available from Java. The library itself is not multithreaded, but it has some global data that it protects using Posix (i.e. native) mutexes. I have no problem with the JNI layer making the library callable, with the exception of the synchronization. As near as I can tell, the right way to integrate this is to use MonitorEnter/Exit calls. The problem with this is that I don't have the JNIEnv available at that point in the code, and to make it directly available, I'd have to propagate it through many layers of calls. The one possible workaround that I have is to use JNI_GetCreatedJavaVMs to get the VM pointer, and then use GetEnv to get the JNIEnv.   More>>

I have an existing C++ library that I need to make available from Java. The library itself is not multithreaded, but it has some global data that it protects using Posix (i.e. native) mutexes. I have no problem with the JNI layer making the library callable, with the exception of the synchronization.


As near as I can tell, the right way to integrate this is to use MonitorEnter/Exit calls. The problem with this is that I don't have the JNIEnv available at that point in the code, and to make it directly available, I'd have to propagate it through many layers of calls.


The one possible workaround that I have is to use JNI_GetCreatedJavaVMs to get the VM pointer, and then use GetEnv to get the JNIEnv. This seems to rely on the current constraint of only one VM per process. However, I don't know if that constraint applies to platforms other than Sun's (with the obvious example, for me at least, being WebLogic). Also, this seems to be a temporary solution, as the design of JNI_GetCreatedJavaVMs clearly shows an intent to have more than one VM in the future, but with no easy global way to find the current VM (or JNIEnv, for that matter).


Any suggestions or corrections would be welcome.


Gary

   <<Less
Report | Quote This | Send private message | Reply | Print

Re: Low level synchronization in legacy library

Posted By:   Christopher_Konvalin  
Posted On:   Tuesday, September 18, 2001 04:16 PM

I believe you could do the synchronization via the 'synchronized' keyword in Java. Even though it's a call via JNI, the synchronization is enforced at the Java method.


Unfortunately, however, this is object-level synchronization, so if you need a finer-grain lock I know there's a trick similar to the following that would work (from memory, so you'll want to verify):



private short a;
private short b;

public someMethodA(...)
{

  synchronized(a)
  {
    someJNIMethodA(...);
  }

}

public someMethodB(...)
{
  synchronized(b)
  {
    someJNIMethodB(...);
  }
}



This effectively allows two threads to access the same object, but lock with a finer grain. It's kind of ugly, and could take longer to debug, but if you absolutely need this kind of solution it apparently works. If I remember right this is outlined in the 'Effective Java' book published by Addison Wesley.


Hope this helps.


Chris

Report | Quote This | Send private message | Reply | Print
About | Sitemap | Contact