I’ve got a C++ program that internally uses java (via my C++ dll that wrapps the WebLogic jsmc.dll that internally uses jvm.dll).
When I set CLASSPATH before running my program, all JAR libraries are found and the program works properly.
When I do not set the CLASSPATH before running my program, the JARs are not found, which of course is expected.
Now, when I set the CLASSPATH before running my program, but clear this CLASSPATH environment variable inside the program code before loading my dll that uses java, a strange thing happens: all JARs are still found and the program works as if everything was OK. I have verified by several ways that the CLASSPATH is really deleted from the env variables (e.g. by using ProcessExplorer or by printing its value).
QUESTION:
Can you explain this behaviour to me? I’m not wondering why java ignores the CLASSPATH I set, but how is it possible that java sees the old CLASSPATH value, not the current one? I emphasize that it isn’t possible for java to store the old CLASSPATH value somehow because java was not loaded at the the time the old value was available.
How can I make the java to respect the changes in the process env variables?
DETAILS:
The problem above is just a simplification I’ve made to explore my real problem. I’m trying to set the CLASSPATH from within the program and avoid to have it set externally. But the java uses the externally set CLASSPATH, not the one I set inside the program.
I read and set the env variables values using the Windows API (GetEnvironmentVariableA, SetEnvironmentVariableA). I have verified that the program process environment variables really change after setting them this way. I even printed the CLASSPATH value from the dll that uses java, before calling any java method. I checked using the ProcessMonitor that jvm.dll is really loaded after the CLASSPATH is deleted. I also tried to exclude the possibility that the CLASSPATH is read from the parent process. Now I’m pretty sure that at the time jvm.dll is being loaded, the CLASSPATH is already deleted from the process environment.
I have tried both a Visual C++ 2010 testing program and the HP LoadRunner C-compiler (mmdrv.exe) vuser script, with the same result. LoadRunner is the main reason why I need to solve this problem.
The problem was caused by that the C-runtime somehow caches the environment variables. While I was trying to modify the CLASSPATH using the system function SetEnvironmentVariableA(), jmsc.dll read CLASSPATH from the C-runtime cache. The C-runtime tries to synchronize its cache with the real values in the process environment, but evidently not very successfully. It was necessary for me to replace the system call to SetEnvironmentVariableA() with the call to _putenv() from the C-runtime in order to change the CLASSPATH.
But there was another problem. There were various versions of C-runtime used by my code, each having its own environment cache. My VC code was linked against msvcr100.dll, while jmsc.dll (that instantiates the Java VM) uses msvcrt.dll. The solution was to link my code to msvcrt.dll too, so that my code sets CLASSPATH using _putenv() from the same C-runtime that jmsc.dll reads.
Thanks to Harry Johnston for the crucial hint, and Peter Cetinski for valuable information.