VM not started if invoke is used in a C++ Daemon

Question

Hi,
I'm trying to load the VM (JDK 1.4) in a C++ program which is using fork() to create a new process. The JNI_CreateJavaVM Call is working as long as it is used in the main process, if I do it in the forked background proc it is failing without any message or dump.
Since this JDK there is a new lib structure. There is a libjvm.so in a subdir named client and another one in a subdir named server. I have not found any documentation on why this is so or what lib I need to load for what purpose.
Any help is appreciated.
Thanks,
J-P

Answer 1

Hi J-P,
Thanks for the clarification. Guess I went on about something that turned out not to be your situation. But this is good - we've cleared things up now.
However
threading is not always the solution for everything, I
need a fork to spawn a child process first and this
child process should be able to cleanly load and start
a full VM from scratch.This is actually just what the daemons I have inherited now do. Each starts running, forks off a child which will be set up to actually be the daemon process, and then the child continues by loading and creating the JVM via the JNI invocation calls. Then the child can make calls into Java.
Crashing
1. Start Process
2. Fork
3. Continue parent without Java
4. Load VM in childOk, so the only difference in the overall process model that I have is that my parent processes don't continue on. They are terminated immediately after the child is off and running.
I used -verbose:jni to see what happens and it crashes
away immediately when the first call to
RegisterNatives comes up. Ah - now I just realized that my daemons don't actually call RegisterNatives, though I've used that method in other non-daemon processes...
As for differences with forked processes versus those created from your typical shell process, yes, I think there are some, but I can't say what they are offhand. Have you tried reading the man pages for "fork" on your OS? Which OS are you using by the way? I am running my stuff on HP-UX 11 and Solaris 8, and someone else has done a Linux port. All are working now except that we have some other elusive problems only on Solaris 8, ironically... we are using JRE 1.3.1.
in my opinion it is a bug somewhere in the VMWell, maybe, but I've learned to always second guess myself with JNI, because there are so many things that can go wrong in your native code, it's incredible. Nonetheless it does seem odd that the same thing runs fine when the fork factor is removed.
I think there's another option you can give the JVM to check all the arguments of your JNI calls... if only I could remember what that was. Try hunting for the Hotspot VM options; a search term on java.sun.com I've used before is 'Hotspot VM documentation'. Also check out any documentation of the supplier of the JVM and/or OS you are running to see if there are any special compiler switches you need - sometimes there are.
Finally, here's a real silly question... you've got any environment vars that need to be exported (to the child process)? I don't know what, maybe something like LD_LIBRARY_PATH.
- Chris