HibernateJpaDialect's HibernateConnectionHandle does not call close() for Hibernate 4.x anymore, since on 4.2, the exposed Connection handle isn't a "borrowed connection" wrapper but rather the actual underlying Connection - with a close() call immediately returning the Connection to the pool and making the local handle invalid for further use within the transaction.
Issue: SPR-10395
Use LinkedHashMaps/Sets wherever exposed to users, and code tests defensively in terms of expected Map/Set ordering. Otherwise, there'll be runtime order differences between JDK 7 and JDK 8 due to internal HashMap/Set implementation differences.
Issue: SPR-9639
(cherry picked from commit 9c09a0a)
In particular, avoid accidental usage of ASM for core JDK types - which will fail in case of a new bytecode version in the JDK, even if the application itself has been compiled with an earlier bytecode target.
Issue: SPR-10292
(cherry picked from commit d3a4068)
Rework the reference documentation to better distinguish the differences
between @Bean methods used in @Comonent vs @Configuration classes. The
'Using the @Bean annotation' section now only covers concepts applicable
when using @Bean methods in @Configuration _or_ @Component classes.
Information only applicable to @Configuration classes has been moved to
a new 'Using the @Configuration annotation' section.
An additional sidebar section attempts to explain the differences
between the two approaches.
Issue: SPR-9425
Also, RequestBindingInterceptor now obtains HttpServletRequest from locally passed-in request handle and delegates to buildRequestAttributes as well now.
Issue: SPR-10342
Just AsyncAnnotationBeanPostProcessor switches "beforeExistingAdvisors" to "true" by default. So effectively, MethodValidation/PersistenceExceptionTranslationPostProcessor apply after existing advisors by default again, fixing the 3.1->3.2 regression.
Issue: SPR-10309
Assuming reference comparisons are much quicker than checking
likely-nested logger levels, perform the former first.
Also, since the reference can match only one of the instances,
use "else if" to short-circuit the search.
The problem was that clocks go forward *at* 2am, so
2am doesn't exist once a year. Users might be surprised
that their cron trigger doesn't go off one night, but that
is arguably correct (and what happens now). The test can be
modified if we decide to change the trigger behaviour.
