On 08/04/2016 14:02, Dawid Weiss wrote:
I don't have time to look it just now but
java.annotations.common/javax.ws.rs.ext.Providers is
<module-name>/<class-name> and I wouldn't have expected to see this type
javax.ws.rs.ext.Providers in this module.
Yes, it's a mess, I agree. It comes from the patched module. Why is
this such a mess -- I've no idea, I'm just the user of the library. :)
Just so I have it clear. Someone has packaged the annotations with the
JAX-RS classes and -Xpatch:java.annotations.common=<jar> is used to
patch them into java.annotations.common module, is that right? I think
explains things as the interface will not be accessible outside of that
module and why the generated Proxy class is not accessible.
:
I remember this discussion. For now I disabled the tests that involve
RestEast on Java9 and everything else passes. What's interesting is
that our application (not the individual tests, the whole thing) works
out of the box even without all these module-related hacks -- we use a
tiny launcher that in turn collects dependencies into a single
URLClassLoader, then runs the application from there. I found this out
by accident, but I can already see how people may try to abuse this
instead of going through the struggle of finding correct JVM switches
for modules (until they become more widespread)...
URLClassLoader does parent delegation so I assume the types in package
javax.annotation are being split between loaders. This is not going to
work if there package private access is required.
-Alan
