Gunnar, In a standard bootstrap, the map parameter in the call to createContainerEntityManagerFactory(PersistenceUnitInfo info, Map map), should have the ValidatorFactory. For the path that your debugging (2 phase bootstrap), org.hibernate.jpa.boot.spi.EntityManagerFactoryBuilder.withValidatorFactory(Object) should of been called. If the Hibernate ORM version is 4.3 or 5.0, we should be calling EntityManagerFactoryBuilder.withValidatorFactory, could you check if that was called or not? Scott On 09/23/2015 07:38 AM, Gunnar Morling wrote: > > Scott, > > Upgrading the Hibernate Validator integration tests to run on WF 10 > CR1, two tests fail for me DefaultValidatorFactoryInPersistenceUnitIT > and CustomValidatorFactoryInPersistenceUnitIT. > > These are about accessing the validator factory from the entity > manager factory through the "javax.persistence.validation.factory" > property. The look-up returns null, though. I tried to debug that a > bit but I am a bit lost in WF's PersistenceUnitServiceImpl. The code > in start() populates the property in a map, but that map is not used > for the EMF actually bootstrapped (which happens through the > EntityManagerFactoryBuilder > phaseOnePersistenceUnitService.getEntityManagerFactoryBuilder(), which > does not get this property set). > > Could you check this one out? Maybe JipiJapa's > EntityManagerFactoryBuilder should set that property to VF given via > withValidatorFactory()? > > Thanks, > > --Gunnar > > [1] > https://github.com/hibernate/hibernate-validator/blob/master/integration/... > [2] > https://github.com/hibernate/hibernate-validator/blob/master/integration/... >

Gunnar, Regarding the "javax.persistence.validation.factory" property setting, I think it would also be passed to org.hibernate.jpa.boot.spi.Bootstrap.getEntityManagerFactoryBuilder(info, map). Scott On 09/23/2015 09:24 AM, Gunnar Morling wrote: > > Scott, > > Yes, EntityManagerFactoryBuilder#withValidatorFactory() is called (see > > https://github.com/wildfly/wildfly/blob/10.0.0.CR1/jpa/subsystem/src/main..., > I see it actually happening in the debugger), so I suppose Hibernate > ORM has the VF at its hands correctly. > > Only that it is not exposed via the > "javax.persistence.validation.factory" key via the EMF-exposed > property map any more. So the call to > > entityManagerFactory().getProperties().get( > "javax.persistence.validation.factory" ); > > returns null whereas it used to return the VF in earlier versions. > > But reading the spec again, it seems it only describes the > "javax.persistence.validation.factory" key as means of passing the VF > *to* JPA but not as means of obtaining it. Apparently that's just > something that happened to be the case in WF before, comments at the > test even states that this behaviour is not portable across > containers. > > So I suppose I'll have to rework that test in the course of the upgrade. > > --Gunnar > > > 2015-09-23 14:51 GMT+02:00 Scott Marlow <smarlow(a)redhat.com&gt;: >> >> Gunnar, >> >> In a standard bootstrap, the map parameter in the call to >> createContainerEntityManagerFactory(PersistenceUnitInfo info, Map map), >> should have the ValidatorFactory. >> >> For the path that your debugging (2 phase bootstrap), >> >> org.hibernate.jpa.boot.spi.EntityManagerFactoryBuilder.withValidatorFactory(Object) >> should of been called. If the Hibernate ORM version is 4.3 or 5.0, we >> should be calling EntityManagerFactoryBuilder.withValidatorFactory, could >> you check if that was called or not? >> >> Scott >> >> >> On 09/23/2015 07:38 AM, Gunnar Morling wrote: >>> >>> >>> Scott, >>> >>> Upgrading the Hibernate Validator integration tests to run on WF 10 >>> CR1, two tests fail for me DefaultValidatorFactoryInPersistenceUnitIT >>> and CustomValidatorFactoryInPersistenceUnitIT. >>> >>> These are about accessing the validator factory from the entity >>> manager factory through the "javax.persistence.validation.factory" >>> property. The look-up returns null, though. I tried to debug that a >>> bit but I am a bit lost in WF's PersistenceUnitServiceImpl. The code >>> in start() populates the property in a map, but that map is not used >>> for the EMF actually bootstrapped (which happens through the >>> EntityManagerFactoryBuilder >>> phaseOnePersistenceUnitService.getEntityManagerFactoryBuilder(), which >>> does not get this property set). >>> >>> Could you check this one out? Maybe JipiJapa's >>> EntityManagerFactoryBuilder should set that property to VF given via >>> withValidatorFactory()? >>> >>> Thanks, >>> >>> --Gunnar >>> >>> [1] >>> >>> https://github.com/hibernate/hibernate-validator/blob/master/integration/... >>> [2] >>> >>> https://github.com/hibernate/hibernate-validator/blob/master/integration/... >>> >>

> Yes, EntityManagerFactoryBuilder#withValidatorFactory() is called (see > https://github.com/wildfly/wildfly/blob/10.0.0.CR1/jpa/subsystem/src/main... , > I see it actually happening in the debugger), so I suppose Hibernate > ORM has the VF at its hands correctly. > > Only that it is not exposed via the > "javax.persistence.validation.factory" key via the EMF-exposed > property map any more. So the call to > > entityManagerFactory().getProperties().get( > "javax.persistence.validation.factory" ); > > returns null whereas it used to return the VF in earlier versions. I think this is related to this discussion - http://lists.jboss.org/pipermail/hibernate-dev/2015-July/013027.html and HHH-9905. The map value is indeed not passed along anymore. > But reading the spec again, it seems it only describes the > "javax.persistence.validation.factory" key as means of passing the VF > *to* JPA but not as means of obtaining it. Apparently that's just > something that happened to be the case in WF before, comments at the > test even states that this behaviour is not portable across > containers. I guess you are referring to JPA specs section 3.6.2 Providing the ValidatorFactory. There it is quite clear that the map passed should be used to pass the instance to the JPA provider. I guess as you suggest the question is whether a later call to 'entityManager.getEntityManagerFactory().getProperties()' is required to expose the same properties as the ones passed to the factory creation. Somehow, I would expect so, but I guess that is not explicitly expressed. > So I suppose I'll have to rework that test in the course of the upgrade. I guess that' the way to go. --Hardy _______________________________________________ hibernate-dev mailing list hibernate-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/hibernate-dev

Hi, On Wed, Sep 23, 2015 at 02:36:05PM +0000, Steve Ebersole wrote: > Its really a different discussion Hardy. The discussion here is > whether EntityManagerFactory().getProperties() should return values that > were not passed in the Map when building the EntityManagerFactory. Sure. Different, but related. The assumption in our test was that the properties passed via the persistence bootstrap, are also then exposed via EntityManagerFactory().getProperties(). This used to be the case and hence the test passed. > Yes the spec says that ValidatorFactory should be passed as part of the > "configuration Map" when building the EMF. However, WF does not do it that > way. Hibernate supports passing the ValidatorFactory in as part of the > "configuration Map", but it also supports this proprietary 2-phase approach > developed for WF. Sure. > All that said, there are other cases where we explicitly handle certain > settings to be returned from EntityManagerFactory().getProperties(). > Things like Connection info defined in the persistence.xml we explicitly > transfer into the Map backing EntityManagerFactory().getProperties(). So > the question is whether a similar thing makes sense for ValidatorFactory. I would think so. Being able to access the ValidatorFactory via EntityManagerFactory().getProperties() makes sense to me. I would give it a +1 --Hardy