You are viewing a plain text version of this content. The canonical link for it is here.
Posted to notifications@grails.apache.org by "ali-jawad (via GitHub)" <gi...@apache.org> on 2026/10/23 15:25:45 UTC
[I] GORM for Hibernate 5: save() no longer sets the owner on a new bidirectional one-to-one child (8.0.0 regression) [grails-core]
ali-jawad opened a new issue, #16740:
URL: https://github.com/apache/grails-core/issues/16740
### Summary
With GORM for Hibernate 5 in Grails 8.0.0, `save()` no longer sets the back-reference of a new child in a bidirectional one-to-one where the child `belongsTo` the owner. On 7.2.4 it did. The owner can be new or already persisted. Depending on the mapping, the child is then saved with a null in-memory back-reference, its insert fails (`hasOne`), or validation rejects it (a non-nullable back-reference).
### Versions
- Broken: Grails 8.0.0, `org.apache.grails:grails-hibernate5-bom:8.0.0` → `org.apache.grails.data:grails-data-hibernate5-core:8.0.0`, Hibernate ORM 5.6.15.Final (`hibernate-core-jakarta`)
- Works: Grails 7.2.4, `org.apache.grails.data:grails-data-hibernate5-core:7.2.4`, the same Hibernate ORM 5.6.15.Final
- The change first shipped in 8.0.0-M2. The `7.2.x` branch does not have it.
### Steps to reproduce
Use an app on GORM for Hibernate 5 with H2 on the test runtime classpath, and add this spec to `src/test/groovy`:
```groovy
import grails.gorm.annotation.Entity
import grails.test.hibernate.HibernateSpec
class OneToOneBackReferenceSpec extends HibernateSpec {
List<Class> getDomainClasses() { [Author, Profile] }
void "saving the owner sets the back-reference of its new one-to-one child"() {
given:
Profile profile = new Profile(bio: 'Writes about GORM')
Author author = new Author(name: 'Jane', profile: profile)
when:
author.save(flush: true)
then:
profile.author.is(author)
!author.hasErrors()
Profile.count() == 1
}
}
@Entity
class Author {
String name
Profile profile
static constraints = {
profile nullable: true
}
}
@Entity
class Profile {
String bio
static belongsTo = [author: Author]
}
```
### Expected
`save()` sets `profile.author` to `author`, as on 7.2.4, and both rows are inserted.
### Actual
`profile.author` stays `null` after `save()`. What follows depends on the mapping:
- **The shape above, with Grails 8's defaults** (properties nullable): `save()` succeeds and both rows are stored with the correct foreign key. Only the in-memory back-reference stays `null`, so `profile.author.is(author)` fails.
- **`static hasOne = [profile: Profile]`** (foreign key in the child): the flush fails with `DataIntegrityViolationException`, NULL not allowed for column `AUTHOR_ID`, on the insert into the child's table. The same happens for a persisted owner with a new child, and for `hasOne` without `belongsTo`.
- **With `grails.gorm.default.nullable = false`** (or `nullable: false` on the back-reference): the cascaded validation rejects the null `author`, so `save()` returns `null`, and `failOnError: true` throws a `ValidationException`.
Inside a Grails application, the map constructor and `bindData` set the back-reference through data binding, which hides the bug there. Plain assignment (`author.profile = profile`) shows it.
### Cause
`save()` calls `HibernateRuntimeUtils.autoAssociateBidirectionalOneToOnes()` before validation:
https://github.com/apache/grails-core/blob/081cd1c93ab753a234e111e2828f089ebdd810ee/grails-data-hibernate5/core/src/main/groovy/org/grails/orm/hibernate/AbstractHibernateGormInstanceApi.groovy#L130
For each owning-side bidirectional one-to-one, it skips the association unless `proxyHandler.isInitialized(child, '<backReference>')` is true, and then sets the back-reference if it is null. This method is the same in 7.2.4 ([L125-L134](https://github.com/apache/grails-core/blob/ea4ddc768f362a77fceb926ced14491c18524ca3/grails-data-hibernate5/core/src/main/groovy/org/grails/orm/hibernate/support/HibernateRuntimeUtils.groovy#L125-L134)) and in 8.0.0:
https://github.com/apache/grails-core/blob/081cd1c93ab753a234e111e2828f089ebdd810ee/grails-data-hibernate5/core/src/main/groovy/org/grails/orm/hibernate/support/HibernateRuntimeUtils.groovy#L125-L134
`HibernateProxyHandler.isInitialized(Object, String)` reads the property and passes its value to `isInitialized(Object)`. For a new child, that value is `null`:
- **7.2.4:** `isInitialized(Object)` delegates to `Hibernate.isInitialized(null)` ([handler L47-L50](https://github.com/apache/grails-core/blob/ea4ddc768f362a77fceb926ced14491c18524ca3/grails-data-hibernate5/core/src/main/groovy/org/grails/orm/hibernate/proxy/HibernateProxyHandler.java#L47-L50)), and that call returns `true` for `null`, which matches none of its `instanceof` branches and reaches the final `else` ([Hibernate 5.6.15 L84-L101](https://github.com/hibernate/hibernate-orm/blob/e924c27e1259b0b5915819e9521d86fcb8164a46/hibernate-core/src/main/java/org/hibernate/Hibernate.java#L84-L101)).
- **8.0.0:** `isInitialized(Object)` returns `false` for `null`:
https://github.com/apache/grails-core/blob/081cd1c93ab753a234e111e2828f089ebdd810ee/grails-data-hibernate5/core/src/main/groovy/org/grails/orm/hibernate/proxy/HibernateProxyHandler.java#L54-L73
The `null` guard came in with 70fde10c8257d029d79beef9f02446cd9376ac74 during the Hibernate 7 work, which reached `8.0.x` through #15568. `HibernateProxyHandler5Spec` pins it ([L132-L135](https://github.com/apache/grails-core/blob/081cd1c93ab753a234e111e2828f089ebdd810ee/grails-data-hibernate5/core/src/test/groovy/org/grails/orm/hibernate/proxy/HibernateProxyHandler5Spec.groovy#L132-L135)). The guard is meant to skip an uninitialized proxy, but it now also skips the one case the method acts on: a null back-reference.
No test caught it, because the one-to-one specs set the back-reference by hand. Examples are the TCK `OneToOneSpec` ([L73-L74](https://github.com/apache/grails-core/blob/081cd1c93ab753a234e111e2828f089ebdd810ee/grails-datamapping-tck/src/main/groovy/org/apache/grails/data/testing/tck/tests/OneToOneSpec.groovy#L73-L74)), `BidirectionalOneToOneWithUniqueSpec` and `UniqueWithHasOneSpec`.
The same branch also handles `static hasOne = [profile: Profile]`, because `ToOne.isOwningSide()` is true when the foreign key is in the child. Both shapes reproduce (see *Actual*).
### Suggested fix
In `autoAssociateBidirectionalOneToOnes`, treat a null back-reference as nothing to initialize. The Hibernate 7 module's copy of this method already has no such guard ([L96-L125](https://github.com/apache/grails-core/blob/081cd1c93ab753a234e111e2828f089ebdd810ee/grails-data-hibernate7/core/src/main/groovy/org/grails/orm/hibernate/support/HibernateRuntimeUtils.groovy#L96-L125)), and #16624 makes the same change to `MappingContextTraversableResolver` ("a null value has nothing to initialize"):
```diff
def otherSidePropertyName = otherSide.getName()
- if (!proxyHandler.isInitialized(inverseObject, otherSidePropertyName)) {
- continue
- }
-
def associationReflector = mappingContext.getEntityReflector(association.associatedEntity)
def propertyValue = associationReflector.getProperty(inverseObject, otherSidePropertyName)
if (propertyValue == null) {
```
A non-null back-reference is never overwritten, initialized or not, so the guard only ever changed the outcome for `null`. The earlier `isInitialized(target, propertyName)` check still keeps an uninitialized child proxy from being loaded.
Another option is to restore the 7.x contract, `isInitialized(null) == true`, which `GroovyProxyFactory.isInitialized(Object, String)` still follows ([L153-L156](https://github.com/apache/grails-core/blob/081cd1c93ab753a234e111e2828f089ebdd810ee/grails-datamapping-core/src/main/groovy/org/grails/datastore/gorm/proxy/GroovyProxyFactory.groovy#L153-L156)). That reverses behaviour a spec now pins, so the fix in the caller looks narrower. Either way, a spec that saves the owner without setting the back-reference would guard against this regression.
### Workaround
Set the back-reference before saving the owner:
```groovy
profile.author = author
author.profile = profile
author.save()
```
A fix with tests is ready: it removes that guard, adds a spec covering both shapes, subclasses, proxies and an owner that is already set, and passes the full hibernate5 core suite including the TCK. I can open a PR once this is acknowledged.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: notifications-unsubscribe@grails.apache.org.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] GORM for Hibernate 5: save() no longer sets the owner on a new bidirectional one-to-one child (8.0.0 regression) [grails-core]
Posted by "jdaugherty (via GitHub)" <gi...@apache.org>.
jdaugherty commented on issue #16740:
URL: https://github.com/apache/grails-core/issues/16740#issuecomment-6368977279
I'm taking a look at this.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: notifications-unsubscribe@grails.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] GORM for Hibernate 5: save() no longer sets the owner on a new bidirectional one-to-one child (8.0.0 regression) [grails-core]
Posted by "jdaugherty (via GitHub)" <gi...@apache.org>.
jdaugherty closed issue #16740: GORM for Hibernate 5: save() no longer sets the owner on a new bidirectional one-to-one child (8.0.0 regression)
URL: https://github.com/apache/grails-core/issues/16740
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: notifications-unsubscribe@grails.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org