5:11 a.m.
It seems we're discussing this issue in multiple places,
so to let you all know I'll repeat it hare:
I think shading is a really really bad idea :)
Could we try to have the enhanced entities to not need Javassist in in
their *direct* classloader; we can still have a normal Javassist as a
module dependency of Hibernate?
That would require to just make sure the generated bytecode doesn't
directly refer to Javassist types but uses an indirection controlled
by Hibernate code.. which in turn can use Javassist or even
alternatives in future, if we'd like to experiment.
I'm not familiar enough with Javassist to know if that's an option
as-is but we can either improve Javassist to allow such a thing or use
some alternatives, like Gunnar and Hardy also suggested on the
hibernate-dev mailing list.
To summarize, I agree with Stuart and would hope that Scott's branch
can be improved by minimizing the amount of Javassist code which
actually needs to be copied by using some simple delegation to
Hibernte types, which in turn can use a private, non-shaded Javassist
taking advantage of the isolation provided by JBoss Modules.
--Sanne
On 12 February 2016 at 03:19, Scott Marlow <smarlow(a)redhat.com> wrote:
What if Javassist packaged these same (proxy/runtime) classes in a
separate javassist-runtime jar and we shaded only the proxy/runtime
classes? That way we only repackage the same classes that we included
for this hack test (e.g.
org.hibernate.bytecode.internal.javassist.proxy.*).
Early testing results of the hack test look good
(https://gist.github.com/scottmarlow/ad878968c5a7c6fbbfb7).
Scott
On 02/11/2016 09:04 PM, Stuart Douglas wrote:
> It depends if you are going to shade all the javassist classes or just
> the "javassist.util.proxy" package (not sure if this is actually
> possible with the shade plugin).
>
> The main advantage is that you can upgrade javassist to get fixes to
> issues that affect bytecode generation. So if JDK9 comes out with new
> bytecodes that the current version of Javassist does not understand then
> upgrading javassist will allow the older version of hibernate to work
> with classes compiled against the newer JDK version. If all of javassist
> is shaded into hibernate then that version of hibernate will never work
> with the newer bytecodes.
>
> I think this is less of an issue if you are still publishing the
> non-Javassist shaded hibernate as well as a shaded version, but if the
> only published artifact has javassist shaded in then it may limit
> forward compatibility.
>
> Stuart
>
>
> On Fri, 12 Feb 2016 at 12:53 Steve Ebersole <steve(a)hibernate.org
> <mailto:steve@hibernate.org>> wrote:
>
> Ugh. That is an awful lot of classes copied over. What exactly was
> the benefit of this over shading again? I mean both case lose the
> ability to simply drop in fixes from upstream Javassist. So what
> does this "clone" approach gain versus shadowing?
>
>
>
> On Thu, Feb 11, 2016 at 7:13 PM Scott Marlow <smarlow(a)redhat.com
> <mailto:smarlow@redhat.com>> wrote:
>
> >>
> >> On 02/11/2016 03:02 PM, Stuart Douglas wrote:
> >> > Have you considered a 3rd alternative, which is to
> use a custom
> >> > ProxyFactory instead of javassists built in one?
> >> >
> >> > AFAIK the main issue is that javassist proxies
> require access to the
> >> >
'javassist.util.proxy.MethodHandler|RuntimeSupport'
> classes. You
> >> could
> >> > create a similar org.hibernate interface, and a
> proxy factory
> >> that uses
> >> > this method handler instead.
> >> >
> >> > Basically you just copy the code from
> javassist.util.proxy into
> >> > hibernate. This is a relatively small amount of
> code, so it
> >> should not
> >> > really add any maintenance burden.
> >>
> >> We talked about this as well via [1]. I understand the
> concept but have
> >> not tried doing this. I like this approach as well, if
> it works. One
> >> of the cons with cloning that Steve Ebersole pointed
> out (see response
> >> on Feb-03-2016 9:01am), is that that users lose the
> ability to drop a
> >> different version of Javassist in (since we maintain
> our own cloned copy
> >> of the Javassist proxy/runtime code).
> >>
> >>
> >> The proxy code is a relatively small part of javassist, so
> unless a bug
> >> is in the proxy code itself this should not be that big a deal.
> >
> > Thanks for the encouragement to go down this path. :)
> >
>
> Started a hack attempt at the clone via
> https://github.com/scottmarlow/hibernate-orm/tree/javassistproxy.
> Seems
> to pass the Hibernate ORM unit tests.
>
> Scott
>
> _______________________________________________
> wildfly-dev mailing list
> wildfly-dev(a)lists.jboss.org <mailto:wildfly-dev@lists.jboss.org>
>
> https://lists.jboss.org/mailman/listinfo/wildfly-dev
>
_______________________________________________
wildfly-dev mailing list
wildfly-dev(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/wildfly-dev