This guide discusses migration to Hibernate ORM version 6.2. For migration from earlier versions, see any other pertinent migration guides as well.

DDL type changes

OffsetTime mapping changes

OffsetTime now depends on @TimeZoneStorage and the hibernate.timezone.default_storage setting. Since the default for this setting is now TimeZoneStorageType.DEFAULT, this means that the DDL expectations for such columns changed.

If the target database supports time zone types natively like H2, Oracle, SQL Server and DB2 z/OS, the type code SqlTypes.TIME_WITH_TIMEZONE is now used, which maps to the DDL type time with time zone.

Due to this change, schema validation errors could occur on existing databases.

The migration to time with time zone requires a migration expression like cast(old as time with time zone) which will interpret the previous time as local time and compute the offset for the time with time zone based on the current date and time zone settings of your database session.

If the target database does not support time zone types natively, Hibernate behaves just like before.

To retain backwards compatibility, configure the setting hibernate.timezone.default_storage to NORMALIZE.

UUID mapping changes on MariaDB

On MariaDB, the type code SqlTypes.UUID now by default refers to the DDL type uuid, whereas before it was using binary(16). Due to this change, schema validation errors could occur on existing databases.

The migration to uuid requires a migration expression like cast(old as uuid).

To retain backwards compatibility, configure the setting hibernate.type.preferred_uuid_jdbc_type to BINARY.

UUID mapping changes on SQL Server

On SQL Server, the type code SqlTypes.UUID now by default refers to the DDL type uniqueidentifier, whereas before it was using binary(16). Due to this change, schema validation errors could occur on existing databases.

The migration to uuid requires a migration expression like cast(old as uuid).

To retain backwards compatibility, configure the setting hibernate.type.preferred_uuid_jdbc_type to BINARY.

JSON mapping changes on Oracle

On Oracle 12.1+, the type code SqlTypes.JSON now by default refers to the DDL type blob and on 21+ to json, whereas before it was using clob. Due to this change, schema validation errors could occur on existing databases.

The migration to blob and json requires a migration expression like cast(old as blob) and cast(old as json) respectively.

To get the old behavior, annotate the column with @Column(definition = "clob").

This change was done because blob and json are way more efficient and because we don’t expect wide usage of SqlTypes.JSON yet.

JSON mapping changes on H2

On H2 1.4.200+, the type code SqlTypes.JSON now by default refers to the DDL type json, whereas before it was using clob. Due to this change, schema validation errors could occur on existing databases.

The migration to json requires a migration expression like cast(old as json). Note that this change in behavior is backwards compatible and you do not need to change your schema, unless you are running into schema validation errors and want to fix them.

To get the old behavior, annotate the column with @Column(definition = "clob").

This change was done because the native json type is more efficient and because we don’t expect wide usage of SqlTypes.JSON yet.

Datatype for enums

Hibernate 6.1 changed the implicit SQL datatype for mapping enums from TINYINT to SMALLINT to account for Java supporting up to 32K enum entries which would overflow a TINYINT. However, almost no one is developing enums with that many entries. Starting in 6.2, the choice of implicit SQL datatype for storing enums is sensitive to the number of entries defined on the enum class. Enums with more than 128 entries are stored as SMALLINT implicitly, otherwise TINYINT is used.

NOTE

On MySQL, enums are now stored using the ENUM datatype by default

Timezone and offset storage

hibernate.timezone.default_storage now defaults to DEFAULT, meaning:

  • if the database/dialect supports it, time zones of date/time values are stored by using the timestamp with time zone SQL column type;

  • otherwise, time zones of date/time values are not stored, and date/time values are normalized to UTC.

In Hibernate ORM 5, time zones were not stored, but normalized to the time zone set in hibernate.jdbc.time_zone, the JVM time zone by default.

This discrepancy might lead to incorrect date/time being loaded from the database for properties of type OffsetDateTime and ZonedDateTime if your application was migrated from Hibernate ORM 5 and was setting hibernate.jdbc.time_zone to a non-UTC timezone.

To revert to Hibernate ORM 5’s behavior, set the configuration property hibernate.timezone.default_storage to NORMALIZE.

Byte[]/Character[] mapping changes

Hibernate historically allowed mapping Byte[] and Character[] in a domain model as basic values to VARBINARY and (N)VARCHAR SQL types.

Strictly speaking, this is an inaccurate mapping. Because the Java wrapper types (Byte and Character) are used, null elements are allowed. However, it is not possible to store such domain values as VARBINARY and (N)VARCHAR SQL types. In fact, attempting to store such values leads to errors on previous versions. The legacy support has an implicit contract that the Byte[] and Character[] types are handled exactly the same as the byte[] and char[] variants.

Building on the ability to use structured SQL types (ARRAY, SQLXML, …​) for storing basic values, 6.2 makes it configurable how to handle mappings of this type:

DISALLOW

(default) Throw an informative and actionable error

ALLOW

Allows the use of the wrapper arrays stored as structured SQL types (ARRAY, SQLXML, …​) to maintain proper null element semantics.

LEGACY

Allows the use of the wrapper arrays stored as VARBINARY and VARCHAR, disallowing null elements.

The main idea here is for applications using these types in the domain model to make a conscious decision about how these values are stored.

Some mappings are considered implicit opt-in to the legacy behavior; e.g. using @Lob or @Nationalized

For those using such mappings, there are a few options -

  1. Migrate the domain model to use byte[] and char[] instead.

  2. Specify hibernate.type.wrapper_array_handling=legacy to enable the legacy behavior.

  3. Specify @JavaType(ByteArrayJavaType.class) or @JavaType(CharacterArrayJavaType.class) attribute-by-attribute

  4. Specify hibernate.type.wrapper_array_handling=allow. If the schema is legacy, migrate the database schema to use a structured SQL type. E.g.

    1. Execute alter table tbl rename column array_col to array_col_old to have the old format available

    2. Execute alter table tbl add column array_col DATATYPE array to add the column like the new mapping expects it to be

    3. Run the query select t.primary_key, t.array_col_old from table t to extract byte[] or String

    4. For every result, load the Hibernate entity by primary key and set the field value to transformed result Byte[] or Character[]

    5. Finally, drop the old column alter table tbl drop column array_col_old

Check constraints for boolean and enum mappings

Check constraints now correctly generated for boolean and enum mappings

UNIQUE constraint for optional one-to-one mappings

Previous versions of Hibernate did not create a UNIQUE constraint on the database for logical[1] one-to-one associations marked as optional. That is not correct from a modeling perspective as the foreign-key should be constrained as unique. Starting in 6.2, those UNIQUE constraints are now created.

Often the association can also be remapped using @ManyToOne + @UniqueConstraint instead.

Column type inference for number(n,0) in native SQL queries on Oracle

Since Hibernate 6.0, columns of type number with scale 0 on Oracle were interpreted as boolean, tinyint, smallint, int, or bigint, depending on the precision.

Now, columns of type number with scale 0 are interpreted as int or bigint depending on the precision.

Removal of support for legacy database versions

This version introduces the concept of minimum supported database version for most of the database dialects that Hibernate supports. This implies that the legacy code for versions that are no longer supported by their vendors, has been removed from the hibernate-core module. It is, however, still available in the hibernate-community-dialects module, just under a different package, namely org.hibernate.community.dialect instead of org.hibernate.dialect. Note that this also includes version specific dialects like PostgreSQL81Dialect, MariaDB102Dialect etc.

The minimum supported dialect versions are as follows:

Dialect Minimum supported version

MySQL

5.7

SQL Server 2008

10.0

DB2

10.5

DB2i

7.1

DB2z

12.1

MariaDB

10.3

H2

1.4.197

Derby

10.14.2

Sybase

16.0

CockroachDB

21.1

PostgreSQL

10.0

Oracle

11.2

HSQLDB

2.6.1

Changes to CDI handling

When CDI is available and configured, Hibernate can use the CDI BeanManager to resolve various bean references. JPA explicitly defines support for this for both attribute-converters and entity-listeners.

Hibernate also has the ability to resolve some of its extension points using the CDI BeanManager. Version 6.2 adds a new boolean hibernate.cdi.extensions setting to control this:

true

indicates to use the CDI BeanManager to resolve these extensions

false

(the default) indicates to not use the CDI BeanManager to resolve these extensions

The previous behavior was to always load the extensions from CDI if it was available. However, this can sometimes lead to timing issues with the BeanManager not being ready for use when we need those extension beans. Starting with 6.2, these extensions will only be resolved from the CDI BeanManager if hibernate.cdi.extensions is set to true.

Change enhancement defaults and deprecation

The enableLazyInitialization and enableDirtyTracking enhancement tooling options in the ANT task, Maven Plugin and Gradle Plugin, as well as the respective hibernate.enhancer.enableLazyInitialization and hibernate.enhancer.enableDirtyTracking configuration settings, switched their default values to true and the settings are now deprecated for removal without replacement. See HHH-15641 for details.

The global property hibernate.bytecode.use_reflection_optimizer switched the default value to true and the setting is now deprecated for removal without replacement. See HHH-15631 for details.

API / SPI / Internal distinction

Dating back to Hibernate 5.x, we have been cleaning up packages to make the distinction between contracts which are considered an API, SPI and internal. We’ve done some more work on that in 6.2 as well.

org.hibernate.cfg package

The org.hibernate.cfg package has been especially egregious in mixing APIs and internals historically. The only true API contracts in this package include org.hibernate.cfg.AvailableSettings and org.hibernate.cfg.Configuration which have been left in place.

Additionally, while it is considered an internal detail, org.hibernate.cfg.Environment has also been left in place as many applications have historically used it rather than org.hibernate.cfg.AvailableSettings.

A number of contracts are considered deprecated and have been left in place.

The rest have been moved under the org.hibernate.boot package where they more properly belong.

org.hibernate.loader package

Most of the org.hibernate.loader package is really an SPI centered around org.hibernate.loader.ast which supports loading entities and collections by various types of keys - primary-key, unique-key, foreign-key and natural-key. org.hibernate.loader.ast has already been previously well-defined in terms of SPI / internal split.

Changes in integration contracts (SPIs)

SPI is a category of interfaces that we strive to maintain with more stability than internal APIs, but which might change from minor to minor upgrades as the project needs a bit of flexibility.

These are not considered public API so should not affect end-user (application developer’s) code but such changes might break integration with other libraries which integrate with Hibernate ORM.

During the development of Hibernate ORM 6.2 the following SPIs have seen some modifications:

EntityPersister#lock

Changed from EntityPersister#lock(Object, Object, Object, LockMode, SharedSessionContractImplementor) to EntityPersister#lock(Object, Object, Object, LockMode, EventSource). This should be trivial to fix as EventSource and SharedSessionContractImplementor are both contracts of the SessionImpl; to help transition we recommend using the methods isEventSource and asEventSource, available on the `SharedSessionContractImplementor`contract.

N.B. method asEventSource will throw an exception for non-compatible type; but because of previous restrictions all invocations to lock actually had to be compatible: this is now made cleared with the signature change.

EntityPersister#multiLoad

The same change was applieed to multiLoad(Object[] ids, SharedSessionContractImplementor session, MultiIdLoadOptions loadOptions), now migrated to multiLoad(Object[] ids, EventSource session, MultiIdLoadOptions loadOptions)

The same conversion can be safely applied.

Executable#afterDeserialize

As in the previous two cases, the parameter now accepts EventSource instead of SharedSessionContractImplementor.

The same conversion can be safely applied.

JdbcType#getJdbcRecommendedJavaTypeMapping()

The return type of JdbcType#getJdbcRecommendedJavaTypeMapping() was changed from BasicJavaType to JavaType. Even though this is a source compatible change, it breaks binary backwards compatibility. We decided that it is fine to do this though, as this is a new minor version.

Query Path comparison

As of 6.2, comparisons of paths are type checked early. This means that a comparison predicate in HQL or JPA Criteria might fail to construct if the types of the left and right hand side are not compatible.

In general, two types T1 and T2 are considered compatible if

  • T1 == T2

  • T1 instanceof T2 or T2 instanceof T1

  • T1 is temporal and T2 is temporal

  • T1 or T2 is unknown

  • T1 can be widened/coerced to T2, or the other way around

Widening/Coercion usually refers to e.g. widening an integer to a long, but can also mean that a string constant can be interpreted as enum when comparing against an enum attribute.

Note that a comparison of a temporal attribute against a string literal worked before

from MyEntity e where e.temporalAttribute > '2020-01-01'

but has to be changed to the proper temporal literal now

from MyEntity e where e.temporalAttribute > date 2020-01-01

Batch Fetching and LockMode

When LockMode is greater than READ Hibernate does not execute the batch fetching so existing uninitialized proxies will not be initialized. This because the lock mode is different from the one of the proxies in the batch fetch queue.

E.g.

` MyEntity proxy = session.getReference( MyEntity.class, 1 ); MyEntity myEntity = session.find(MyEntity.class, 2, LockMode.WRITE); ` only the entity with id equals to 2 will be loaded but the proxy will not be initialized.

Integrating Static Metamodel Generation

The integration of static metamodel generation in a project has changed; the recommended way to do this now is by harnessing the annotation processor classpath. This is true for both gradle and maven.

Check out the specific sections in the User Guide for a guideline on how to do this for Gradle or Maven.


1. A "true" one-to-one mapping is one in which both sides use the same primary-key value and the foreign-key is defined on the primary-key column to the other primary-key column. A "logical" one-to-one is really a many-to-one with a UNIQUE contraint on the key-side of the foreign-key. See link:https://docs.jboss.org/hibernate/orm/6.2/userguide/html_single/Hibernate_User_Guide.html#associations for more information