Re: [eclipselink-dev] PersistenceProvider.createContainerEntityManagerFa

[Tom Ware <tom.ware@xxxxxxxxxx>]

Hi Philipp,

  When createContainerEntityManagerFactory() is used some assumptions will be 
made about the environment that it is invoked in - based on what the JPA spec 
allows.  For instance:

  - PersistenceUnitInfo will be already be properly constructed - rather than 
constructed by the provider
  - persistenceUnitInfo.getNewTempClassLoader() will provide a classloader that 
conforms with the spec rules for that classloader
  - the environment will be setup so that all classes can be discovered by the 
provider (some guarantees can be made about the default behavior of 
exclude-unlisted-classes)

  Basically we assume we are specifically in a Java EE container.  How have you 
built your PersistenceUnitInfo?  Does Spring's JPA integration help deal with 
this in OSGi?

  Given that we are not aware of a Java EE Container that provides the hooks we 
need in an OSGi way, we have not done any of the investigation necessary to 
bring this to a usable level.  Given that a number of application servers are 
moving towards OSGi, it is certainly something we will want to investigate.

  You should be able to use the javax.persistence.jtaDataSource and 
javax.persistence.nonJtaDataSource properties in the map provided to 
createEntityManagerFactory(String, Map) method to pass in the datasources.

-Tom

Philipp Nanz wrote:
> Hi folks,
>
> I am currently investigating venues to setup JPA w/ EclipseLink in a
> 100% OSGi project on top of Spring DM.
>
> Now what puzzles me is that 
> PersistenceProvider.createContainerEntityManagerFactory is
> explicitly disabled in the OSGi environment.
>
> This is problematic because it is effectively the only way to inject a
> datasource into the mix without JNDI.
>
> Is there a specific reason for this? Because I have subclassed the
> non-OSGi PersistenceProvider analogous to the OSGi one, but kept that
> method intact -- and it seems to work fine.
>
> Also from reading the JPA 1.0 spec (EJB 3.0 Persistence) I could not
> discover any conflict of this method with the OSGi environment in general.
>
> It would be great if anyone could clear this up!
>
> Kind regards,
> Philipp
>
> PS: This issue is also discussed in Bug #252263
> _______________________________________________
> eclipselink-dev mailing list
> eclipselink-dev@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/eclipselink-dev