Re: [PROPOSAL] New Virtual Packages and way to handle Classpath
I think we should define an interface in the Java policy, then write
wrappers for the different compilers, and install the wrappers as the
alternative (this is already done for jikes-* at least and for the gij
JVM). So we can indeed handle -cp in the wrappers.
What interface do you suggest? The discussion her pretty much showed,
that you can't rely on anything. This would mean:
* -classpath is probably save (-cp already not...)
A precise interface should be discussed together. Off my head:
-classpath, -cp, -sourcepath, -O, -d, -g, -deprecation
If a feature is not implemented at all by a compiler, the wrapper should
print a warning, and proceed (in particular, -g, -O and -deprecation are
pretty harmless to ignore). We would also submit a feature request
upstream for that feature.
I agree that we cannot just say " if they are command-line compatible
with Sun's JDK javac", as this is a moving target. The list of options
must be in the Java policy.
rt.jar could be defined by default, but overridable with -bootclasspath
(handled in the wrapper if necessary). This matches Sun's javac
behaviour, so it should make more existing scripts work out of the box
* setting up rt.jar? in javac or in the calling script? The script
doesn't know the location, so we would be better off with
/usr/lib/java-dev-env, which would be relative easy to maintain.
Thanks for this reply!
Thanks for your efforts to improve Java on Debian!