The generated CodeBlock that takes care of injecting the PersistenceContext may override an existing property value. This PR checks existing values and potentially omits creating the code block.
Related to: spring-projects/spring-data-jpa#2753 & spring-projects/spring-data-commons#2760
Comment From: sdeleuze
Worth to review with @snicoll when he is back from PTO.
@christophstrobl Is this PR ready for review for potential merge or do you plan further refinement?
Comment From: christophstrobl
@sdeleuze you can review it already. Happy to modify if necessary.
Comment From: snicoll
So it looks like something has post-processed the persistence context already and the core goes in the way. I think we need to fix this differently.
If I understand correctly, Spring Data has knowledge of something that we don't. Why isn't that reflected in the BeanDefinition. If it is, why is the skip necessary?
Comment From: christophstrobl
There's an EntityManagerBeanDefinitionRegistrarPostProcessor in data-jpa that post processes EntityManagerFactory. @odrotbohm can you please help us find the right point for the changes.
Comment From: mp911de
Spring Data is setting the property value for entityManager within the BeanDefinition. In JVM mode, the container skips @PersistenceContext post-processing for the property while AOT mode works differently.
Comment From: snicoll
InjectionMetadata#inject has a checkPropertySkipping that is meant to determine if injection should take place. I missed that while building the AOT contribution for the persistence context. It should attempt to run the same code so that no contribution takes place if it should be skipped.
Comment From: snicoll
Thanks for the PR @christophstrobl but I think we should tackle this one differently. I've created #30476