Hibernate 6.0 Final

Posted by    |       Hibernate ORM Releases

It has been years in the making, but ORM 6.0 Final has finally been released!

This announcement will discuss the major changes, as well as give insight into why certain choices were made.

We will also be following up with a series of more focused posts targeting specific improvements or cool new features. Stay tuned!

APIs and SPIs

While many things have changed in 6.0, we strove to minimize changes to APIs to help mitigate migration costs.

See https://hibernate.org/community/compatibility-policy/ for a discussion of what we consider an API versus an SPI.

Applications which use only the Jakarta Persistence APIs will be source compatible within the discussion in Jakarta Persistence.

Applications using Hibernate APIs will generally be bytecode and source compatible, aside from the removal of deprecated stuff. There are a few one-off changes that break bytecode and/or source compatibility; these are covered in the migration guide.

One specific change to note is that many of these contracts have been better defined with type parameters. Theses were inconsistently and sometimes poorly defined in previous versions.

Quite a few SPI contracts have changed to support many of the topics discussed here as well as in the migration guide. Many will also be the subject of the mentioned follow-up posts.

Jakarta Persistence

Java Persistence has become Jakarta Persistence as part of the overall move of Java EE to Jakarta. Various legal requirements forced the changing of the javax namespace - for persistence, that means changing from javax.persistence to jakarta.persistence for package names as well as property and hint names.

This is clearly an unfortunate and invasive change, but beyond our control. Luckily Jakarta have developed a transformer to help with these migrations. We actually used this tool to migrate Hibernate’s own source code. It works well-ish.

For those using Maven, you are in luck (well, within the bounds of actually using Maven) in that Jakarta themselves provide a Maven plugin to integrate this transformer.

For those using Gradle, you can use the tasks we developed to transform Hibernate’s source code.

There is also a command-line form. See the transformer docs for details.


A few years ago, around the 5.4 timeframe, we worked with the amazing performance team at Red Hat to squeeze even more great performance out of Hibernate ORM.

This work was part of a larger effort to improve the performance of WildFly. Ultimately, the limiting factor to additional improvements within Hibernate was our approach of reading values from a JDBC ResultSet by name rather than by position. For every JDBC driver out there, reading by name is slower.

It quickly became obvious that minimal changes would not be enough, and so this work led to many changes. A great analogy is to consider migrating a Map-based solution to List-based. There is the obvious impact of changing calls to accept an int rather than a String as well as internally keeping track of the positions of each selected value within the ResultSet. There is also the perhaps not-so-obvious impact of changing the callers and consumers of those contracts to keep track of positions.

These changes have led to improvements on a number of fronts:

  1. As mentioned, reading by position is significantly faster than reading by name which leads to performance improvements.

  2. Historically Hibernate generated SQL select queries with a defined pattern of named column aliases which were later used to access the specific result. We’ve all seen these "ugly" aliases. With these changes, those select-clause aliases are no longer needed resulting in much more readable generated SQL.

  3. Although we implemented some improved support for limiting needed joins within an entity mapping (joined inheritance, secondary tables) in 5.x, 6.0 allows even better opportunity for this. In fact, the support for this in 5.x was conceptually back-ported from the 6.0 work.

  4. (2) and (3) combined results in much smaller SQL needing to be sent to the server which can have an impact on network communication. Every bit helps.

This was by far the biggest force behind 6.0 initially.

Mapping Model

The mapping model is an SPI and as such will not be seen by all users. But it is a major development and impacts many users providing extensions.

The main driving force behind this mapping model work was Read-by-position, and we had a number of design goals in developing it:

  • support positional processing of attributes

  • make it object-oriented

  • make it user friendly

This model can be accessed though RuntimeMetamodelsImplementor which provides access to both:


Historically, Hibernate’s annotations grew directly from its hbm.xml mappings. These old annotations are String-based just like XML, providing all the cons of XML and really none of the benefits of annotations.

6.0 redesigns Hibernate’s annotations with type-safety in mind, as well as better leveraging the benefits of annotations. Most are also usable as meta-annotations.

Annotations for mapping basic values saw the most change. The User Guide contains the details.

The mapping of embeddables has also changed. Again, see the User Guide for details. Embeddables now support constructor injection

Query API

  • Parameter binding

  • Session#createSelectionQuery

  • Session#createMutationQuery

  • others???

Semantic Query Model

Hibernate’s Semantic Query Model (SQM) is its semantic representation of HQL and Criteria queries. HQL is interpreted into SQM; Hibernate’s Criteria implementations are SQM nodes.

6.0 implements quite a few changes shared between HQL and Criteria. Most of these are covered in HQL and Criteria chapters of the User Guide.

Some specific changes include

  • Automatic de-duplication of single entity results in a Query. See the Migration Guide for details

  • Set operations (union, intersect, except)

  • Set aggregations (listagg, e.g.)

  • Window operations (over, e.g.)

  • Vastly improved function support. See the User Guide for details.

  • ILIKE operator

  • Improved temporal support (arithmetic, etc)


Previous versions of Hibernate used Antlr 2 for parsing. 6.0 updates to Antlr 4 for a few reasons:

  • Antlr 2 is no longer supported, and has not for years

  • Antlr 4 is faster than Antlr 2

  • Antlr 4 grammars are easier to maintain, while the previous Antlr 2 grammars were poorly defined (largely as a function of Antlr 2 itself) and difficult to maintain.


Hibernate’s legacy Criteria API has been deprecated for many years and has been fully removed in 6.0. Support for Criteria queries is now offered solely through the Jakarta Persistence APIs plus extensions.

As mentioned in Semantic Query Model, Hibernate’s SQM model is the implementation of the Jakarta Persistence Criteria node APIs. This offers significantly better performance in terms of execution compared to previous versions which essentially converted the Criteria to HQL and translated the HQL.

6.0 also adds a new setting related to Criteria performance - hibernate.criteria.copy_tree. The Jakarta Persistence specification requires that a copy be made of the Criteria tree passed to EntityManager#createQuery. This obviously has a performance impact, but is intended for safety. hibernate.criteria.copy_tree allows Hibernate to not make a copy of the tree which results in better performance. Just be sure to not mutate the tree after the call to #createQuery.


6.0 goes all-in in terms of modeling queries as trees. We discussed above how that works for HQL and Criteria queries, but we also now model SQL queries as trees.

This has quite a few benefits, but the main one is direct Dialect involvement. The tree acts as an API to a contract which translates the AST into JDBC calls, which allows much more powerful involvement by the Dialect in this process.

Dialect initialization

In previous versions, Dialect was essentially static details about the database being used. This meant that Dialect implementations could not incorporate version-specific deviations, which is why Hibernate had so many version-specific subclasses.

6.0 changes the way Dialects are created to allow them to initialize themselves based on the version of the database/driver being used.


6.0 also introduces a new @Incubating annotation which is intended to notify users that a particular contract may change in the future. These are typically new contracts which we may need to change in response to additional use cases or clarification of existing use cases.

Think of it as a "use at your own risk" kind of notice. Obviously we will strive to not change such contracts, but this gives us the flexibility to do so if needed and communicating that this could potentially happen to the user.

Sometimes settings are considered incubating. These are indicated by @Incubating on the corresponding org.hibernate.cfg.AvailableSettings field and are also logged using the org.hibernate.orm.incubating category.

We also generate a documentation report.

Release artifacts

Starting with 6.0 we will no longer be publishing zip and tgz bundles to SourceForge.

Starting in 6.0 we now publish additional documentation artifacts, such as:

Over the next few weeks we will also begin re-evaluating both:

Much of this content is either out-of-date or incorporated elsewhere.


For additional details, see:

To get in touch, use the usual channels as discussed on the website.

Back to top