Cascade all-delete-orphan of constrained one-to-one association throws InvalidCastException on explicit flush

[nkreipke]

When trying to delete an orphaned child entity in a one-to-one association, an InvalidCastException is thrown if session.Flush() is called explicitly.

Example (full solution at https://github.com/nkreipke/NHOneToOneBug):

<class name="Entity1">
    <id name="Id">
        <generator class="guid"/>
    </id>
    <one-to-one name="Child" cascade="all-delete-orphan" constrained="true" />
</class>
<class name="Entity2">
    <id name="Id">
        <generator class="foreign">
            <param name="property">Parent</param>
        </generator>
    </id>
    <one-to-one name="Parent" />
</class>

var entity = session.Query<Entity1>().First(x => x.Id == id);

// Child entity should be deleted due to all-delete-orphan cascading
entity.Child = null;

session.Update(entity);

// !!! This is required for the bug to trigger
session.Flush();

// !!! InvalidCastException
transaction.Commit();

This will throw the following exception at transaction.Commit():

System.InvalidCastException: Unable to cast object of type 'NHOneToOneBug.Entity2' to type 'NHOneToOneBug.Entity1'.
   at (Object )
   at NHibernate.Bytecode.Lightweight.AccessOptimizer.GetSpecializedPropertyValue(Object target)
   at NHibernate.Bytecode.AccessOptimizerExtensions.GetSpecializedPropertyValue(IAccessOptimizer optimizer, Object target)
   at NHibernate.Tuple.Entity.PocoEntityTuplizer.GetIdentifierPropertyValue(Object entity)
   at NHibernate.Tuple.Entity.AbstractEntityTuplizer.GetIdentifier(Object entity)
   at NHibernate.Persister.Entity.AbstractEntityPersister.GetIdentifier(Object obj)
   at NHibernate.Persister.Entity.AbstractEntityPersister.IsTransient(Object entity, ISessionImplementor session)
   at NHibernate.Engine.ForeignKeys.IsTransientFast(String entityName, Object entity, ISessionImplementor session)
   at NHibernate.Engine.ForeignKeys.IsTransientSlow(String entityName, Object entity, ISessionImplementor session)
   at NHibernate.Event.Default.DefaultDeleteEventListener.OnDelete(DeleteEvent event, ISet`1 transientEntities)
   at NHibernate.Impl.SessionImpl.FireDelete(DeleteEvent event, ISet`1 transientEntities)
   at NHibernate.Impl.SessionImpl.Delete(String entityName, Object child, Boolean isCascadeDeleteEnabled, ISet`1 transientEntities)
   at NHibernate.Engine.Cascade.CascadeProperty(Object parent, Object child, IType type, CascadeStyle style, String propertyName, Object anything, Boolean isCascadeDeleteEnabled)
   at NHibernate.Engine.Cascade.CascadeOn(IEntityPersister persister, Object parent, Object anything)
   at NHibernate.Event.Default.AbstractFlushingEventListener.CascadeOnFlush(IEventSource session, IEntityPersister persister, Object key, Object anything)
   at NHibernate.Event.Default.AbstractFlushingEventListener.PrepareEntityFlushes(IEventSource session)
   at NHibernate.Event.Default.AbstractFlushingEventListener.FlushEverythingToExecutions(FlushEvent event)
   at NHibernate.Event.Default.DefaultFlushEventListener.OnFlush(FlushEvent event)
   at NHibernate.Impl.SessionImpl.Flush()
   at NHibernate.Impl.SessionImpl.FlushBeforeTransactionCompletion()
   at NHibernate.Impl.SessionImpl.BeforeTransactionCompletion(ITransaction tx)
   at NHibernate.Transaction.AdoTransaction.Commit()

This only happens if session.Flush() was called beforehand. Without explicit flushing, the child entity is deleted correctly.

I suspect the bug is in SessionImpl. In this line, the values of entityName (which is Entity1) and child (which is of type Entity2) are copied into the DeleteEvent fields entityName and entity, and the code might later assume they refer to the same type:

nhibernate-core/src/NHibernate/Impl/SessionImpl.cs

Line 871 in d1a7e10

FireDelete(new DeleteEvent(entityName, child, isCascadeDeleteEnabled, this), transientEntities);

[bahusoid]

Your one-to-one mapping has constrained="true" which means NHibernate assumes that one-to-one record always exists (always created and deleted along with parent record). So to me it looks like invalid mapping combination (constrained + delete orphan)

Does it fail the same with constrained="false"?

[nkreipke]

It does indeed not occur without constrained="true", although we also don't specify that in production and still get the exception. I'm however currently unable to reproduce that in a test project.

I don't think it matters though as your fix will most probably resolve our issue as well. Thank you for looking into this.