Permalink
Fetching contributors…
Cannot retrieve contributors at this time
134 lines (87 sloc) 6.07 KB

5.3 Migration Guide

This guide discusses migration from Hibernate ORM version 5.2 to version 5.3. For migration from earlier versions, see any other pertinent migration guides as well.

Background

Hibernate 5.3 adds support for the JPA 2.2 specification

Known changes

Changes to positional query parameter handling

This really breaks down into 2 related changes:

  • Support for JDBC-style parameter declarations in HQL/JPQL queries has been removed. This feature has been deprecated since 4.1 and removing it made implementing the second change, so we decided to remove that support. JDBC-style parameter declaration is still supported in native-queries.

  • Since JPA positional parameters really behave more like named parameters (they can be repeated, declared in any order, etc.) Hibernate used to treat them as named parameters - it relied on Hibernate’s JPA wrapper to interpret the JPA setParameter calls and properly handle delegating to the named variant. This is actually a regression in 5.2 as it causes javax.persistence.Parameter#getPosition to report null.

For JDBC-style parameter declarations in native queries, we have also moved to using one-based instead of zero-based parameter binding to be consistent with JPA. That can temporarily be reverted by setting the hibernate.query.sql.jdbc_style_params_base setting to true which reverts to expecting zero-based binding.

Change in the @TableGenerator stored value

In order to be compliant with the JPA specification, the sequence value stored by Hibernate 5.3 in the database table used by the javax.persistence.TableGenerator is the last generated value. Previously, Hibernate stored the next sequence value.

For backward compatibility, a new setting called hibernate.id.generator.stored_last_used was introduced, which gives you the opportunity to fall back to the old Hibernate behavior.

Note

Existing applications migrating to 5.3 and using the @TableGenerator have to set the hibernate.id.generator.stored_last_used configuration property to false.

Second-level cache provider SPI changes

Hibernate’s second-level cache SPI needed to be redesigned to match with expectations between Hibernate and the various caching providers. And the SPI did not really clarify the intention. These mis-matched expectations certainly had the potential to lead to bugs. Although it was originally slated for 6.0 to delay the SPI changes, we decided to back-port the work to 5.3 to address a number of bugs that could have been avoided with a clarified SPI.

The changes also allow the caching providers to perform serious optimizations based on the users configuration of domain data caching in Hibernate.

Details can be seen on the HHH-11356 Jira issue

One potential upgrade concern is any custom org.hibernate.cache.spi.QueryCacheFactory implementations. org.hibernate.cache.spi.QueryCacheFactory was meant as a means to allow service providers the ability to define query result caching, generally with more stale-tolerant query result invalidation handling. However, the contract also bound it to the old second level cache contracts so they had to be updated. Its responsibilities also changed so we also decided to "rename it" to org.hibernate.cache.spi.TimestampsCacheFactory

Another specific change to be aware of is accessing cache entries via SecondLevelCacheStatistics#getEntries and NaturalIdCacheStatistics#getEntries. These methods have been deprecated, however the new caching SPI does not really require caching providers to support this. As of 5.3 these methods inherently return an empty Map (Collections#emptyMap). This has always been something that providers did not implement "correctly" anyway

Statistics changes

The change for HHH-11356 required changes in its consumers. One such consumer is the Hibernate Statistics system…​.

Drop hibernate-infinispan module

Support for using Infinispan as a Hibernate 2nd-level cache provider has been moved to the Infinispan project so the hibernate-infinispan module has been dropped.

A relocation pom which is pointing to org.infinispan:infinispan-hibernate-cache dependency is still generated, therefore, avoiding the need of updating any library dependency.

The relocation pom may be dropped in a future release.

EnhancementTask changes

The API of the org.hibernate.tool.enhance.EnhancementTask Ant task was changed, specifically the #addFileset method was dropped in favor of #setBase and #setDir

See details on the HHH-11795 Jira issue.

The main gist is that EnhancementTask was fixed (through a contribution) to actually work with Enhancer from BytecodeProvider. Previously it had not. And part of fixing that required this change.

Many-to-one association in embeddable collection elements and composite IDs

A bug introduced in 4.3 caused many-to-one associations in embeddable collection elements and composite IDs to be eagerly fetched, even when explicitly mapped as lazy.

This bug does not affect many-to-one associations that are not in a composite ID or embeddable collection element.

In 5.3.2, this bug was fixed. As a result, such associations will be fetched as specified by their mappings.

Many-to-one associations mapped by using native HBM xml are lazy by default. In order to keep the associations eager in 5.3.2 and later, mappings will need to explicitly specify that they are non-lazy.

When mapped with annotations, many-to-one associations use FetchType.EAGER by default. Starting in 5.3.2, if an association is mapped with FetchType.LAZY, the assocation will be lazily fetched, as expected.

See details on the HHH-12687 Jira issue.

5.3 → 6.0 compatibility changes

The original driving force behind these series of changes is an effort to be as proactive as possible about designing compatibility between 5.3 and 6.0.

Type system changes

Use of NavigableRole, back-ported from 6.0 rather than plain String