Re: [hibernate-dev] Differences in ORM exception handling in 5.1 vs 5.3

I've been looking at differences in Hibernate exception handling for applications that uses "native" (non-JPA) Hibernate when moving from 5.1 to 5.3. As you know, exception handling changed in 5.2 when hibernate-entitymanager was merged into hibernate-core. This change is documented in the 5.2 migration guide. [1] Here is the relevant text: "org.hibernate.HibernateException now extends javax.persistence.PersistenceExceptions. Hibernate methods that "override" methods from their JPA counterparts now will also throw various JDK defined RuntimeExceptions (such as IllegalArgumentException, IllegalStateException, etc) as required by the JPA contract." While digging into this, I see that a HibernateException thrown when using 5.1 may, in 5.3, be unwrapped, or may be wrapped by a PersistenceException (or a subclass), IllegalArgumentException, or IllegalStateException, depending on the particular operation being performed when the HibernateException is thrown. The reason why the exceptions may be wrapped or unwrapped is because Hibernate converts exceptions (via AbstractSharedSessionContract#exceptionConverter) for JPA operations which may wrap the HibernateException or throw a different exception. Hibernate does not convert exceptions from strictly "native" operations, so those exceptions remain unwrapped. Here are a couple of examples to illustrate: 1) A HibernateException (org.hibernate.TransientObjectException) was thrown in 5.1. In 5.3, the same condition can result in org.hibernate.TransientObjectException that is either unwrapped, or wrapped by IllegalStateException. I've pushed a test to my fork to illustrate. [2] Thrown during Session#save, #saveOrUpdate In 5.1: org.hibernate.TransientObjectException (unwrapped) In 5.3: org.hibernate.TransientObjectException (unwrapped) Thrown during Session#persist, #merge, #flush In 5.1, org.hibernate.TransientObjectException (unwrapped) In 5.3, org.hibernate.TransientObjectException wrapped by IllegalStateException 2) A HibernateException thrown when using 5.1 may be wrapped by a PersistenceException, even though HibernateException already extends PersistenceException. For example, see TransactionTimeoutTest#testTransactionTimeoutFailure: https://github.com/hibernate/hibernate-orm/blob/master/hibernate-core/src... The exception thrown by the test indicates that the transaction timed out. This exception is important enough that an application might retry the operation, or at least log for future investigation. Thrown during Session#persist (or when changed to use Session#merge or Session#flush): In 5.1: org.hibernate.TransactionException (unwrapped) In 5.3: org.hibernate.TransactionException wrapped by javax.persistence.PersistenceException Thrown if the test is changed to use Session#save or #saveOrUpdate instead: In 5.1: org.hibernate.TransactionException (unwrapped) In 5.3: org.hibernate.TransactionException (unwrapped) Similarly, by adding some logging, I see that HibernateException objects can be wrapped by PersistenceException when running the 5.3 hibernate-core unit tests. Depending on the context, I see that some of the following exceptions can be wrapped or unwrapped. org.hibernate.exception.ConstraintViolationException org.hibernate.exception.DataException org.hibernate.exception.GenericJDBCException org.hibernate.exception.SQLGrammarException org.hibernate.id.IdentifierGenerationException org.hibernate.loader.custom.NonUniqueDiscoveredSqlAliasException org.hibernate.PersistentObjectException org.hibernate.PropertyAccessException org.hibernate.PropertyValueException org.hibernate.TransactionException You can see an example using org.hibernate.exception.ConstraintViolationException at [3]. In order to deal with these differences, an application could change the following (which was appropriate for 5.1): try { ... } catch (HibernateException ex) { procressHibernateException( ex ); } to the following for 5.3: try { ... } catch (PersistenceException | IllegalStateException | IllegalArgumentException ex) { if ( HibernateException.class.isInstance( ex ) ) { handleHibernateException( (HibernateException) ex ); } else if ( HibernateException.class.isInstance( ex.getCause() ) ) { handleHibernateException( (HibernateException) ex.getCause() ); } } IMO, it's a little clumsy having to deal with both wrapped and unwrapped exceptions. It would be better if exceptions were consistently wrapped, or consistently unwrapped. I haven't had much of a chance to think about how we can deal with this, but one thing that comes to mind is to allow an application to choose a particular ExceptionConverter implementation. Hibernate would need to be changed to always convert exceptions (including for strictly native operations) before returning to the application. For example, a property, hibernate.exception_converter could be added with the following values: jpa - default for JPA applications (org.hibernate.internal.ExceptionConverterImpl) native (or legacy?) - default for native applications that does not wrap HibernateException fully-qualified class name that implements ExceptionConverter If users have to make changes to exception handling for 5.3, do you think they would be willing to change their application to use JPA exceptions (i.e., hibernate.exception_converter=jpa)? Comments? Regards, Gail [1] https://github.com/hibernate/hibernate-orm/blob/5.2/migration-guide.adoc [2] https://github.com/gbadner/hibernate-core/blob/exception-compatibility/or... [3] https://github.com/gbadner/hibernate-core/blob/exception-compatibility/or...