[hibernate-dev] [Search] Redefining inheritance mapping, and why to follow same rules as ORM

[Emmanuel Bernard]

On Thu 2014-08-21 19:48, Hardy Ferentschik wrote:
> 
> On 21 Jan 2014, at 17:03, Emmanuel Bernard <emmanuel at hibernate.org> wrote:
> 
> > Your idea of reusing Jandex won't quite fully work when people put their
> > classes in different jars.
> 
> Well, the assumption of course is that we can handed a aggregated/composite index
> over all jars.
> 
> There is a caveat though, we keep talking that we get the index from a container or
> from ORM, but we need to take care of building our own index as well in case we don’t
> get a Jandex index passed. In this case we need to scan classes.

The hard part (and I did it in Hibernate EM) is
1. to find all jars involved: the discovery mechanism is pretty much
working by accident and classloader hacking
2. to scan the jar: it's easy for zip or directory based JARs, not so
easy for JARS made of different URL prototols and opaque URLs

> 
> On the other side, ORM does the Jandex building thing as well on the metamodel branch.
> So I guess that once we depend on ORM 5 we can get in all cases a Jandex index.
> Question is when that will be?

Remember, Hibernate Search is not ORM only. So the discovery mechanism cannot
fully depend on an ORM feature not as a hard dep at least.