What is the preferred method for creating children objects?
Adding the child to a collection on it's parent and calling update on the parent, or
Creating the child explicitly, add the child to a collection on the parent, and updating the parent?
I'm struggling with #1. And it's hard enough that I'm thinking #2 is preferable (less "magic").
Not sure what would be the "preferred method" (it could depend on who does prefer). But I can share my way
If the parent is a root entity, then we should save all the reference tree in in one shot: session.SaveOrUpdate(parent).
E.g. Employee has collection of Educations. In such case, it should go like this
create Education give it reference to employee,
Add() it to collection employee.Educations.
Have/make the mapping inverse, cascade="all-delete-orphan" ...
NHibernate will do what we expect on session.SaveOrUpdate(parent)
Some snippets in xml:
<class name="Employee" table="..." lazy="true"
optimistic-lock="version" dynamic-update="true" batch-size="25" >
<cache usage="read-write" region="ShortTerm"/>
<id name="ID" column="Employee_ID" generator="native" />
<bag name="Educations" lazy="true" inverse="true"
batch-size="25" cascade="all-delete-orphan" >
<key column="Employee_ID" />
<one-to-many class="Education" />
</bag>
...
And the Education
<class name="Education" table="..." lazy="true" batch-size="25>
<cache usage="read-write" region="ShortTerm"/>
<id name="ID" column="Education_ID" generator="native" />
<many-to-one not-null="true" name="Employee" class="Employee"
column="Employee_ID" />
...
in fluent:
public class EmployeeMap : ClassMap<Employee>
{
public EmployeeMap()
{
BatchSize(25)
...
HasMany(x => x.Educations)
.AsBag()
.BatchSize(25)
.LazyLoad()
.Cascade.AllDeleteOrphan()
.Inverse()
.KeyColumn("Employee_ID")
public class EducationMap : ClassMap<Education>
{
public EducationMap()
{
...
References(x => x.Employee, "Employee_ID")
.Not.Nullable()
...
And now the C# relations:
// business
var employee = ...;
var education = new Education
{
Employee = employee,
...
};
employee.Educations.Add(education);
// DAO
session.SaveOrUpdate(employee);
If the parent is not root, there is just a relation (Employee has collection of Subordinates of type Employee), keep the persisting separated
Related
I am working with a brownfield database and am trying to configure a subclass map which joins to its subclasses with a column other than that of the specified id. The login table has a primary key column login_sk which I'd like to use as its id. It joins to two tables via a login_cust_id column (to make things more fun the corresponding columns in the adjoining tables are named differently). If I setup login_cust_id as the id of the UserMap it joins to its subclasses as expected. For what I hope are obvious reasons I do not want to use login_cust_id as the id for my User objects.
public class UserMap : ClassMap<IUser>
{
public UserMap()
{
Table("login");
Id(x => x.Id).Column("login_sk"); // want to setup map like this
// if used instead this works for subclass joining / mapping
// Id(x => x.Id).Column("login_cust_id");
// would prefer to only reference login_cust_id for subclass mapping
}
}
public class CustomerUserMap : SubclassMap<CustomerUser>
{
public CustomerUserMap()
{
Table("customer");
Map(c => c.DisplayName, "cust_mail_name");
Map(c => c.RecordChangeName, "cust_lookup_name");
KeyColumn("cust_id");
}
}
public class EntityUserMap : SubclassMap<EntityUser>
{
public EntityUserMap()
{
Table("entity");
Map(c => c.DisplayName, "entity_name");
KeyColumn("entity_id");
}
}
What I'd like to do is only use the login_cust_id column when joining to subclasses. Is there a fluent mapping setting that allows me to specify this? If not a fluent mapping is there a regular NHibernate XML mapping that would work? I'd prefer to not even map the column and only use it for joining if possible. If it helps there is a potential discriminator column login_holder_type which indicates which table to join to.
It did occur to me to setup an IClassConvention but after poking at the passed IClassInstance I could not determine any settings which would help me.
public class UserIdConvention : IClassConvention, IClassConventionAcceptance
{
public void Apply(IClassInstance instance)
{
// do something awesome with instance.Subclasses to
// specify the use of login_cust_id for subclass joining...
}
public void Accept(IAcceptanceCriteria<IClassInspector> criteria)
{
criteria.Expect(x => typeof(User).Equals(x.EntityType));
}
}
The lack of a populated Subclasses collection for the passed instance caused me to look for a more specific inspector which IParentInspector appears to be. Unfortunately Fluent NHibernate does not appear to have corresponding implementations for IParentInstance, IParentConvention or IParentConventionAcceptance like it does for IJoinedSubclassInspector. While I could probably implement my own before I do I wanted to ensure I wasn't barking up the wrong tree.
Is this sort of subclass id adjustment even possible? Am I missing something obvious in either my map or the Fluent NHibernate Conventions namespace? How can I map to a joined subclass with a different column/property than the id of parent?
I was able to think of three possible solution to your problem please see my findings below.
Solution 1: Discriminator based mapping with Join
My initial idea was to use a discriminator based mapping for modelling the inheritance, with each sub-class containing a join with a property ref, i.e
<class name="IUser" abstract="true" table="login">
<id name="Id" column="login_sk">
<generator class="identity"/>
</id>
<discriminator column="login_holder_type" not-null="true" type="System.String"/>
<subclass name="CustomerUser" discriminator-value="Customer">
<join table="customer" >
<key column="cust_id" property-ref="login_cust_id" />
<property name="DisplayName" column="cust_mail_name"/>
<property name="RecordChangeName" column="cust_lookup_name" />
</join>
</subclass>
<subclass name="EntityUser" discriminator-value="Entity">
<join table="entity" >
<key column="entity_id" property-ref="login_cust_id" />
<property name="CompanyName"/>
</join>
</subclass>
</class>
Unfortunately at this time this feature is supported in Hibernate but not in NHibernate. Please see here and here for the outstanding tickets. Some work has gone towards adding this feature which can be seen on this fork on github.
Solution 2: Discriminator based mapping with Many-to-One
Another option is to still use the discriminator based mapping, but use a many-to-one mapping within each of the sub-classes, which would allow you to join on the foreign key using a property-ref. This has the disadvantage of requiring separate classes for all of the properties in your customer and entity tables but is a workable solution.
<class name="IUser" abstract="true" table="login">
<id name="Id" column="login_sk">
<generator class="identity"/>
</id>
<discriminator column="login_holder_type" not-null="true" type="System.String"/>
<subclass name="CustomerUser" discriminator-value="Customer">
<many-to-one name="CustomerProps" property-ref="login_cust_id" />
</subclass>
<subclass name="EntityUser" discriminator-value="entity">
<many-to-one name="EntityProps" property-ref="login_cust_id" />
</subclass>
</class>
<class name="CustomerProps" Table="customer" >
<id name="Id" column="cust_id">
<generator class="assigned"/>
</id>
<property name="DisplayName" column="cust_mail_name"/>
<property name="RecordChangeName" column="cust_lookup_name" />
</class>
<class name="EntityProps" Table="entity" >
<id name="Id" column="entity_id">
<generator class="assigned"/>
</id>
<property name="CompanyName"/>
</class>
Solution 3: Discriminator based mapping with Joins to Updatable Views
The final option is to create an Updatable View in the DB for the customer and entity tables which contains the login_sk field. You can then use Join within each sub-class as you wouldn't require the property-ref.
<class name="IUser" abstract="true" table="login">
<id name="Id" column="login_sk">
<generator class="identity"/>
</id>
<discriminator column="login_holder_type" not-null="true" type="System.String"/>
<subclass name="CustomerUser" discriminator-value="Customer">
<join table="customerView" >
<key column="login_sk" />
<property name="DisplayName" column="cust_mail_name"/>
<property name="RecordChangeName" column="cust_lookup_name" />
</join>
</subclass>
<subclass name="EntityUser" discriminator-value="Entity">
<join table="entityView" >
<key column="login_sk" />
<property name="CompanyName"/>
</join>
</subclass>
</class>
I think that the main issue is that the subclass deletes all constraints related to the base mapped collections.
Using the table per concrete class strategy, I have found that the parent collections are not asociated with the subclasses also in another (maybe related) problem, the associations between Basetypes and ChildTypes are not created either.
I have a schema similar to this:
public class Parent{
public virtual Int64 Id{get; set;}
public virtual IList<Foo> foos{get; set;}
public virtual IList<ParentType> _pts{get; set;}
}
public class child: Parent{
public virtual int chilInt{get; set;}
}
public class BaseType{
public virtual Int64 Id{get; set;}
public virtual Parent ParentReference{get; set;}
}
public class ChildType: BaseType{
public virtual string childBacon{get; set;}
}
Mapping Files
<class name="Parent" abstract="true">
<id name="Id" type="Int64" column="Id" unsaved-value="0">
<generator class="native"/>
</id>
<set name="foos" inverse="false" >
<key column="Id"/>
<one-to-many class="Foo" />
</set>
<set name="pts" inverse="false" >
<key column="Id"/>
<one-to-many class="ParentType" />
</set>
</class>
<union-subclass name="Child" table="Child" extends="Parent">
<property name="childInt" type="int" />
</union-subclass>
<class name="ParentType" abstract="true">
<id name="Id" type="Int64" column="Id" unsaved-value="0">
<generator class="native"/>
</id>
<many-to-one name="ParentReference" class="Parent"/>
</class>
<union-subclass name="ChildType" table="ChildType" extends="ParentType">
<property name="childBacon" type="string" />
</union-subclass>
The result that the child table don't have any relation with foo table.
If you use the <union-subclass> mapping it's clear that there is no direct relation of the foo entries to your child table because the child table only contains the additional things declared in the child class.
When instantiating a child instance with union-subclass mapping you get a row in both, the parent and child tables. And if your child instance contains entries in the foo set, you get some rows in the foo table with relation to the parent table.
Using table per concrete class mapping does not make sense with associations pointing to the parent class (as the foo class not part of your code example seems to do) because then the different derived classes of parent all inherit the foo set but the foo table cannot have foreign keys to all those tables.
Well, there are three common approaches for ORM and inheritance (table per class hierarchy, table per subclass, table per concrete class). <union-subclass>, one you use, is used in table per concrete class and it should be embedded in parent <class>. Read about it here (8.1.5).
Maybe it won't resolve all your issues, but at least it should help with establishing mapping for inheritance.
How do I map relationship, where child endpoint is exposed via Id property and not via whole Parent object?
Here is the example:
class Parent {
public Guid Id { get; set; }
public List<Child> Chlidren { get; set; }
}
class Child {
public Guid Id { get; set; }
public Guid ParentId { get; set; }
}
Here are the equivalent mappings I'm using:
<?xml version="1.0" encoding="utf-8" ?>
<hibernate-mapping xmlns="urn:nhibernate-mapping-2.2"
assembly="Blabla"
namespace="Blabla"
auto-import="false">
<typedef name="ChildrenList" class="Blabla" />
<class name="Parent" table="Parent" lazy="false">
<id name="Id" column="ID" type="Guid">
<generator class="guid" />
</id>
<bag name="Children" table="Child"
cascade="save-update"
collection-type="ChildrenList"
lazy="false">
<key column="ParentID" not-null="true" />
<one-to-many class="Child" />
</bag>
</class>
<class name="Child" table="Child" lazy="false">
<id name="Id" column="ID" type="Guid">
<generator class="guid" />
</id>
<!-- How to map ParentID here? -->
</class>
</hibernate-mapping>
When I create a parent, add some children to Children collection and then save the parent, everything is fine. But if save a parent object first, then create a child, setting its ParentID property to ID of the parent, then I get
NHibernate.PropertyValueException:
not-null property references a null or transient value Child._Parent.ChildrenBackref
All attempts to map many-to-one relationship resulted in different exceptions while creating NHibernate configuration. Mostly about object type mismatch.
I'm sure NHibernate is capable to handle this scenario. There must something fairly basic that I miss.
EDIT:
I think it make sense to the example test, which fails with above exception:
var child = new Child(Create.Saved<Parent>().Id); // this sets the ParentId property
this.Repository.Save(child); // here I get the exception
My thoughts why NHibernate is raising this: Children property of Parent class mapped in a way that says that a child cannot exist without a parent (<key column="ParentID" not-null="true" />). When I try to persist a child, NHibernate tries to resolve this relationship (to find a parent this child relates to) and fails, since being given no child endpoint (which otherwise would be ParentId property) in the mapping, it check for its own Child._Parent.ChildrenBackref endpoint, whatever it is.
This looks like a desired solution: Mapping ParentId property as child endpoint of the relationship. This would force NHibernate to resolve a parent by using value of ParentId property as parent's primary key.
The thing is I don't know if it's possible.
The one-to-many / many-to-one relationships you have in NHibernate always needs to have a dominant side (i.e. the side that manages the "saving").
<bag name="Children" table="Child"
cascade="save-update"
collection-type="ChildrenList"
lazy="false">
<key column="ParentID" not-null="true" />
<one-to-many class="Child" />
</bag>
The above is a one-to-many relationship where the dominant side is the parent. That means, you save the parent ... and that will save the parent first, then, the children (with the ParentId being null), then a subsequent update will be issued to set the child.ParentId.
Note:
The child is inserted first with ParentId=null ... if you have a db or mapping restriction to say ParentId cannot be null, this action will fail.
<bag name="Children" table="Child"
cascade="save-update"
collection-type="ChildrenList"
lazy="false"
inverse=true>
<key column="ParentID" not-null="true" />
<one-to-many class="Child" />
</bag>
Note the inverse=true attribute. This means the child object is dominant in the relationship, meaning the child object is in charge. The parent will be inserted, then the Id will be assiged to the child.ParentId, and then the child will be inserted with the ParentId already set.
In many cases, of course, you want to go either way. The easiest way to do this is to manage the relationship on both ends (unfortunately, you have to do this yourself).
On the Parent, you have a method:
public void AddChild(Child child)
{
Children.Add(child);
child.ParentId = Id;
}
public void RemoveChild(Child child)
{
Children.Remove(child);
child.ParentId = null;
}
On the Child, you have a method:
public void SetParent(Parent parent)
{
ParentId = parent.Id;
parent.Children.Add(this);
}
Using these methods to Add/Remove/Set, both sides are consistent after the action is performed. It, then, wouldn't matter whether you set inverse=true on the bag or not.
see http://www.nhforge.org/doc/nh/en/index.html#collections-example
class FooBase{...}
class FooDerived : FooBase {...}
class BaseContainer
{
public virtual FooBase Foo {get;set;}
}
class DerivedContainer : BaseContainer
{
public virtual new FooDerived Foo {get;set;}
}
Hibernate mapping options
Option 1 below
Fails to persist on a/c of NHibernate generating additional member declaration in the xml (index out of range error)
<class name="BaseContainer" discriminator-value="0">
<discriminator column="ContainerType" type="int" />
<many-to-one name="Foo"
foreign-key="..."
class="FooBase"
column="FooId"
unique="true"/>
<subclass name="DerivedContainer" discriminator-value="1">
<many-to-one name="Foo"
foreign-key="..."
class="FooDerived"
column="FooId"
unique="true"/>
</subclass>
</class>
Option 2 independent mappings !
Fetch operation erroneous, does not discriminate the types
<class name="BaseContainer" discriminator-value="0">
<discriminator column="ContainerType" type="int" />
<many-to-one name="Foo"
foreign-key="..."
class="FooBase"
column="FooId"
unique="true"/>
</class>
<class name="DerivedContainer" discriminator-value="1">
<many-to-one name="Foo"
foreign-key="..."
class="FooDerived"
column="FooId"
unique="true"/>
</class>
Stuck, would be grateful for any pointers, although I understand this can easily achieved if done via table per subclass, is there any way above can be achieved via table per class hierarchy
I'm having one self-referencing class. A child has a reference to its parent and a parent has a list of children. Since the list of children is ordered, I'm trying to map the relation using NHibernate's .
This is my mapping:
<class name="MyClass">
<id name="Id">
<generator class="native"/>
</id>
<list name="Children" cascade="delete" inverse="true">
<key column="ParentId"/>
<index column="ListOrder"/>
<one-to-many class="MyClass"/>
</list>
<many-to-one name="Parent" class="MyClass" column="ParentId"/>
</class>
The problem I'm having is when having a bi-directional mapping child<->parent, the list index (ListOrder) isn't updated in the database when I do my CRUD dance. This means that when I e.g. remove a child, I get holes in the children list after saving to the database and fetching the parent again. If I remove the bidirectionality, by not having a many-to-one from the children to the parent (and no inverse=true), the ListOrder is updated correctly.
Have any of you seen this before? Is there any simple solution?
Yes, it's because of inverse=true, an alternate solution would be to use a set or bag instead of list with order="ListOrder", add the ListOrder column as a property to the MyClass class with an empty setter and a getter that always returns it's index from it's parent's child collection. Like this:
<class name="MyClass">
<id name="Id">
<generator class="native"/>
</id>
<bag name="Children" cascade="delete" inverse="true" order-by="ListOrder">
<key column="ParentId"/>
<one-to-many class="MyClass"/>
</bag>
<property name="ListOrder" column="ListOrder"/>
<many-to-one name="Parent" class="MyClass" column="ParentId"/>
</class>
and the class
public class MyClass
{
public virtual int ID { get; set; }
public virtual IList<MyClass> Children { get; set; }
public virtual MyClass Parent { get; set; }
public virtual int ListOrder
{
get
{
if (Parent == null || !Parent.Children.Contains(this)) return -1;
return Parent.Children.IndexOf(this);
}
set { }
}
}