Hibernate models based on annotation processor APIs

[yrodiere]

Currently, the Hibernate ORM annotation processor (metamodel generation and repository generation) use annotation processor APIs directly to derive JPA / Hibernate ORM metadata.

If we had a way to run Hibernate models based on annotation processor APIs, the Hibernate ORM processor could use it, allowing it to (critically) merge XML mapping with annotation mapping.

(Creating this issue to track an idea we had during earlier discussions -- I'm surprised to find it in our notes, because I expect @gavinking has plenty of reasons to explain why this just cannot work, or at least not efficiently).

cc @marko-bekhta maybe it could also (only?) be useful for Hibernate Search, since that's something we discussed in the past.

[gavinking]

(Creating this issue to track an idea we had during earlier discussions -- I'm surprised to find it in our notes, because I expect @gavinking has plenty of reasons to explain why this just cannot work, or at least not efficiently).

As long as the metamodel is build lazily on demand, there's no problem.

What we need to avoid is trying to build the whole thing upfront, because that sucks for incremental compilation.

[sebersole]

The model is largely built lazily. Some parts are built immediately.

Now, this is controllable by the individual models - e.g. JdkClassDetails versus JandexClassDetails versus ... Here we'd have a new model - ProcessorClassDetails or what-have-you and it could do what it needs to do. However, keep in mind that if you delay that you need to later have access to the source of the information. E.g., JdkClassDetails holds a reference to Class to access details later for initializing its lazy bits.