I have some questions about the Java packaging guidelines with respect to native libraries [1]. JNI using packages must put the JAR and .so in %{_libdir}/%{name}, and use System.load() with a full path instead of using System.loadLibrary(). I guess this is for multilib situations, so you can have multiple JARs, each pointing to the correct .so. Since that full path can start with either /usr/lib or /usr/lib64, how have others generated that full path?
Wouldn't it be better to have a single JAR in /usr/share/java, and choose the correct path with the os.arch system property (which, annoyingly, is "amd64" with OpenJDK and "x86_64" with gcj on my machine)? We have to pass a full path to System.load() anyway, so I don't see the advantage of having the JAR under %{_libdir} in that case. Has there been any thoughts around providing a Fedora-specific JAR to hide the details of native library loading? (Maybe "Fedora.loadLibrary()"?)
Also, I know how to generate a .so with GCJ using aot-compile-rpm. How is this supposed to work for applications? Say I have a JAR file with a main method, and a wrapper script to invoke it in %{_bindir}. On a system that uses GCJ primarily, does that wrapper script do the right thing (invoke the gcj-compiled code)?
Finally, does anybody know why "gcj --target-help" sends this to stderr?
cc1: warning: command line option "-fbootclasspath=./:/usr/share/java/libgcj-4.3.0.jar" is valid for Java but not for C
References: [1] https://fedoraproject.org/wiki/Packaging/Java#Packaging_JAR_file
Thanks,