[hibernate-dev] What should be the contract for PersistenceUnitInfo#addTransformer with regard to multiple persistence units mapping the same entity class?

[Luis Barreiro]

My reading on that javadoc is that you only get the transformation once 
(the first time that a class is loaded) and after that, if it's used by 
other PUs, it may not get the transformations those PUs specify.

I would like to add that the enhancer is synchronous (it only enhance 
one class at a time), hence my question about synchronization, but you 
were obviously looking at a different level.

Regards,

Luis Barreiro

Middleware Performance Team

<https://red.ht/sig>

On 06/05/2018 07:08 PM, Scott Marlow wrote:
> Thanks, I think they probably intended that the persistence provider 
> would use a marker interface like org.hibernate.engine.spi.Managed to 
> ensure that the enhancement is only done once.  Responses to your 
> questions are inline below.
>
> On Tue, Jun 5, 2018 at 12:19 PM, Luis Barreiro <lbarreiro at redhat.com 
> <mailto:lbarreiro at redhat.com>> wrote:
>
>     Hi Scott,
>
>     In the particular case of the hibernate bytecode enhancer, it adds
>     the org.hibernate.engine.spi.Managed marker interface to the
>     entity so that it only performs the enhancement once. (that
>     implies that you can't have the same entity enhanced for different
>     features in the same class loader)
>
>     Also, in theory, you could have multiple transformations applied
>     to the same entity / PU. So what is it you really want to check ?!
>
> Mostly, I'm trying to understand why the [1] javadoc requires that 
> classes are only transformed only once, as I would expect that some 
> persistence providers could want to register multiple transformers.
>
>     I don't get you last paragraph ... why ORM will be only capable of
>     registering one transformer,
>
> If we update the WildFly JPA container 
> JPADelegatingClassFileTransformer [2] to only allow the application 
> entity class to be enhanced at most once (at runtime), ORM would be 
> able to register multiple transformers but the first ORM transformer 
> that actually enhances the entity class (e.g. transformer returns 
> non-null return value), would need to terminate the "for each 
> transformer" loop in [2].
>
> Since, as you said, ORM is already using a marker interface to know 
> when a class is already enhanced, it sounds like WildFly doesn't 
> really need to enforce the [1] contract (with regard to only 
> transforming the entity class once).
>
>     and why will you need synchronization ?!
>
> If WildFly JPADelegatingClassFileTransformer [2] where modified to 
> prevent more than one transformation of an entity class, that would 
> include checking across (parallel) persistence unit deployment 
> threads, which could be done with a Java synchronization lock.
>
> Scott
>
> [1] 
> https://docs.oracle.com/javaee/7/api/javax/persistence/spi/PersistenceUnitInfo.html#addTransformer-javax.persistence.spi.ClassTransformer-
>
> [2] 
> https://github.com/wildfly/wildfly/blob/master/jpa/subsystem/src/main/java/org/jboss/as/jpa/classloader/JPADelegatingClassFileTransformer.java#L44