RE: Moving system properties to gnu.classpath.*

[David Holmes]

> I'm pretty sure the check *cannot* be circumvented by a user-defined
> class loader. It may not do the check itself, but that only means it can
> load a class in another package (that happens to have the same name).
> For it to load bootstrap classes, it will need to delegate to the system
> class loader and that's the one doing the checking.

That assumes that the test is unconditionally performed in loadClass - in
which case you can't circumvent the check.

I'm not clear on exactly what the merits of this mechanism are in general,
but I guess that the primary motivation is to deny access to untrusted code,
rather than just user-code. I'm not sure what dire consequences would arise
from accessing these system properties, but it is probably much less dire
than someone accessing sun.misc.unsafe :)

Question: is the classpath security implementation at the stage where it
supports policy files etc? Do individual VM's define where the policy file
should be found?

David Holmes