I'm struggling to find the best way to model 1 : 0,1 relationships ("may have one" or "has at most one"). I believe this is called Z cardinality.
For example, suppose I have two classes Widget and WidgetTest. Not all Widgets are tested and the test is destructive so there can be at most one WidgetTest per Widget. Also assume that it's inappropriate to add the WidgetTest fields to Widget.
I would like my public interface to be:
Widget
WidgetTest { get; set; }
WidgetTest
Widget { get; }
Model 1: Widget has a WidgetTest property and in the database the Widget table has a uniquely constrained foreign key to WidgetTest. My DBA argues that this would allow a WidgetTest record to exist without a Widget.
WidgetTable
WidgetTestId (FK, UQ)
Model 2: Widget has a private collection of WidgetTest and enforces the 0,1 relationship by adding or removing a single object from the collection controlled by a public WidgetTest property. The database models this as 1:m with WidgetTest having a uniquely constrained foreign key to Widget. I argue that this means adopting the model to fit the database schema (i.e. more work for me).
WidgetTestTable
WidgetId (FK, UQ)
Which model is better? Which is easier to achieve with NHibernate? Or is there a third way?
Edit ... Here's what I ended up with:
public class Widget
{
// This is mapped in NH using a access strategy
private IList<WidgetTest> _widgetTests = new List<WidgetTest>(1);
public WidgetTest
{
get { return _widgetTests.FirstOrDefault(); }
set
{
_widgetTests.Clear();
if (value != null)
{
_widgetTests.Add(value);
}
}
}
}
My approach has been to model a one-to-many relationship in the mappings, but to constrain the "many" to a single item. This allows the optional one-to-one, and also guarantees that your WidgetTest instance is persisted when you save the Widget. For example:
public class Widget
{
/// <summary>
/// This property is ignored by the NHibernate mappings.
/// </summary>
public virtual WidgetTest WidgetTest { get; set; }
/// <summary>
/// For easier persistence with NHibernate, this property repackages the
/// WidgetTest property as a list containing a single item. If an
/// attempt is made to set this property to a list containing more than
/// one item, an exception will be thrown. But why bother? Just use the
/// WidgetTest property.
/// </summary>
public virtual IList<WidgetTest> WidgetTests
{
get
{
IList<WidgetTest> widgetTests = new List<WidgetTest>();
if (this.WidgetTest != null)
{
widgetTests.Add(this.WidgetTest);
}
return widgetTests;
}
set
{
if (value != null && value.Count > 1)
{
throw new Exception("The WidgetTests collection may not contain more than one item.");
}
else if (value != null && value.Count == 1)
{
this.WidgetTest = value[0];
}
else
{
this.WidgetTest = null;
}
}
}
}
When you say "assume that it's inappropriate to add the WidgetTest fields to Widget", do you mean in your domain objects or in the database. If you are happy for the fields to be in the same table in the database, how about mapping WidgetTest as a component of Widget? Have the NHibernate mapping file look like:
<class name="Widget" table="Widget">
...
<property name="WidgetProperty"/>
...
<component name="WidgetTest" class="WidgetTest">
<property name="WidgetTestProperty"/>
</component>
</class>
Giving the table structure:
WidgetTable
WidgetProperty
WidgetTestProperty
Which would still let you have the public interface you've specified, however, WidgetTest would become a value object which you may or may not want.
i have 2 other ideas here
Join table and map as Component
ignore Id of dependant class
The answer given by nw. can result in the exception "A collection with cascade=”all-delete-orphan” was no longer referenced by the owning entity instance".
You will find this to be the case if you're using inverse="true" and cascade="all-delete-orphan" in your mapping file.
This is because nw.'s answer creates a new list every time the get accessor is called and doesn't do anything with the list passed in through the set accessor. As such, NHibernate doesn't have the IList<WidgetTest> reference it originally passed in when creating the object and can't proceed with the cascade.
So in order to fix this, we need to do something with that IList<WidgetTest> reference and be careful not to de-reference it.
public class Widget
{
public Widget()
{
_widgetTests = new List<WidgetTest>();
}
/// <summary>
/// This property is ignored by the NHibernate mappings.
/// </summary>
public WidgetTest WidgetTest { get; set; }
/// <summary>
/// For easier persistence with NHibernate, this property repackages the
/// WidgetTest property as a list containing a single item. If an
/// attempt is made to set this property to a list containing more than
/// one item, an exception will be thrown. But why bother? Just use the
/// WidgetTest property.
/// </summary>
private IList<WidgetTest> _widgetTests;
protected virtual IList<WidgetTest> WidgetTests
{
get
{
if (_widgetTests.Count == 0 && WidgetTest != null)
{
_widgetTests.Add(WidgetTest);
}
else if (_widgetTests.Count > 0 && WidgetTest == null)
{
_widgetTests.Clear();
}
else if (_widgetTests.Count > 0 && WidgetTest != _widgetTests[0])
{
_widgetTests.Clear();
_widgetTests.Add(WidgetTest);
}
return _widgetTests;
}
set
{
if (value != null && value.Count > 1)
{
throw new Exception("The WidgetTest collection may not contain more than one item.");
}
if (value != null && value.Count == 1)
{
WidgetTest = value[0];
}
else
{
WidgetTest = null;
}
//Store the reference
_widgetTests = value;
}
}
}
Mapping:
<class name="Widget" table="widgets">
...
<id name="Id" type="Guid" column="widgetId">
...
</id>
...
<bag name="WidgetTests" inverse="true" cascade="all-delete-orphan" access="property">
...
<key column="widgetId" />
<one-to-many class="WidgetTest" />
</bag>
</class>
Inspiration for the enhancement:
http://www.onkarjoshi.com/blog/188/hibernateexception-a-collection-with-cascade-all-delete-orphan-was-no-longer-referenced-by-the-owning-entity-instance/comment-page-1/
Related
I have a model like this:
public class Order
{
public virtual int OrderType { get; set; }
}
(lots of other properties omitted of course) which maps directly to an int type in the DB.
The thing is, the numeric order type is meaningless to my application. There are single-letter codes that the user sees which denote the order type. So, I could do something like this:
public class Order
{
public virtual int OrderTypeIgnored { get; set; }
public virtual char OrderType
{
get
{
return translateForward(OrderTypeIgnored);
}
set(char val)
{
OrderTypeIgnored = translateBackward(val);
}
}
}
(lots of air code/pseudocode there, I'm relatively new to C#) and just map the OrderTypeIgnored property. But is there a cleaner way to do this? Perhaps somehow overriding the getter and setter on the mapped property itself?
A few notes: The values are static enough that embedding the translation in the code is not a problem. No, there's no LOV table, and no, I don't have control over the database structure.
Sorry if there are answers for this, but searching for things like "mapping" and "translation" don't really get me the results I'm looking for, obviously.
You could create a public char property that uses a private int field and only map the field.
Model:
public class Order
{
private int _orderType;
public virtual char OrderType
{
get
{
return TranslateForward(_orderType);
}
set
{
_orderType = TranslateBackward(value);
}
}
}
Mapping:
<property name="_orderType" access="field" />
If you don't want to map the field directly (because you use a compile-safe mapping) you can map the public property using the access strategy "field", a naming strategy like "camelcase-underscore" and explicitly specify the "Int32" type.
you can always use enums for this kind of situation.
You can define it like this:
namespace MyApp.Domain
{
using System.ComponentModel;
public enum OrderType : short
{
[Description("Order Suspended")]
Suspended = 1,
[Description("Order Delivered")]
Delivered = 2,
[Description("Order New")]
Inserted = 3
}
}
and map it this way:
<property name="Type" type="MyApp.Domain.OrderType, MyApp.Domain" >
<column name="Type" not-null="true"/>
</property>
so you can write your QueryOver in a simple way like this:
var orders = this.Session.QueryOver<MyApp.Domain.Orders>()
.Where(x => x.Type == MyApp.Domain.OrderType.Inserted)
.List();
I want to get ClassA.ClassBCollection property filtered and paged. I need to change filtering dynamically.
The default querying will result in something like:
select * from ClassA
left outer join ClassB
on id == FK_ClassB
Can I customize querying of nhibernate set somehow?
Mappings:
<class name="ClassA">
<property name="Name" />
<set name="ClassBCollection">
<key column="FK_ClassB" on-delete="cascade" />
<one-to-many class="ClassB" />
</set>
</class>
<class name="ClassB">
<property name="Something"/>
</class>
If I do understand your question...
Can I customize querying of nhibernate set somehow?
...correctly, the answer is NO.
I mean, if you think about getting the instance of ClassA and doing some paging and filtering over its <set> collection. That would be always done in memory. (What we can do with mapping I appended at the end).
we can change the approach
In this case, when you need a filter and paging over the collection items, I would strongly recommend to go the other way. Create Criteria (QueryOver, HQL) not over the ClassA but over the ClassB.
First of all we have to extend ClassB mapping:
<class name="ClassB">
<property name="Something" />
<many-to-one name="ClassA" column="FK_ClassB" fetch="join" />
</class>
And then create a Criteria like this
var criteria = NHSession.Current.CreateCriteria<ClassB>();
criteria
.Add(new InExpression("ClassA", new object[] {1})) // id of one or more ClassA
.AddOrder(new Order("Something", true)) // Order By
.SetFirstResult(2) // Skip
.SetMaxResults(10); // Take
var list = criteria.List<ClassB>();
Because we used mapping of ClassA fetch="join" the resulting SQL statement will be very similar to the first snippet in this question.
So this way, we can achieve the desired SQL Select, but we cannot use ClassA.ClassBCollection directly. We did it this way...
NOTE:
Filters / paging which we can influence on the <set> mapping are static filter in the where clause and style of fetching values.
Where clause will always be evaluated when loading ClassBCollection as a property of the ClassA. It could be like where="IsActive=true"
In case that ClassA can have a lot of items in ClassBCollection, we can manage how they will be loaed. Very effective way is attribute batch-size documented here
Can I customize querying of nhibernate set somehow?
I'm not entirely sure what this means. If you meant, can I query and use WHERE clauses over the collection, the answer is yes. Here's how:
[TestFixture]
public class StackOverflowQuestion13496270Tests
{
public ISession session;
[SetUp]
public void SetUp()
{
session = // Get the current NHibernate session
}
[Test]
public void Query_ClassA()
{
var results = session.Query<ClassA>()
.Where( x => x.ClassBCollection.Any( y => y.Name == "Bob" ) )
.Fetch( x => x.ClassBCollection )
.Skip( 0 )
.Take( 50 )
.ToList();
}
[Test]
public void Query_ClassB()
{
var results = session.Query<ClassB>()
.Where( x => x.Name == "Bob" )
.Fetch( x => x.ClassAParent )
.Skip( 0 )
.Take( 50 )
.ToList();
}
public class ClassA
{
public virtual int Id { get; set; }
public virtual string Name { get; set; }
public virtual IList<ClassB> ClassBCollection { get; set; }
}
public class ClassB
{
public virtual int Id { get; set; }
public virtual string Name { get; set; }
// Add this and the appropriate mapping modifications to be able to navigate back to the parent
public virtual ClassA ClassAParent { get; set; }
}
}
I have begun to test Fluent NHibernate in C#
I have a well normalized object structure with 20 related classes.
I currently use Fluent 1.3 with NHibernate 3.2.
So far I have managed to use the AutoMap feature which suits me fine,
Very convenient!
BUT ...
3 of the tables are "enum tables" that need to have their records set with specific Id value.
I tried to make manual mappings of these tables and let the rest be automapped.
But when the manual table is created it fails because it references a table that is automapped (and not available for manual mapper?)
Is it possible to use AutoMapping but for some very few classes override identity creation on primary key?
I tried to make a custom convention but without success.
public class OverrideIdentityGeneration : Attribute
{
}
public class ConventionIdentity : AttributePropertyConvention<OverrideIdentityGeneration>
{
protected override void Apply(OverrideIdentityGeneration attribute, IPropertyInstance instance)
{
instance.Generated.Never();
}
}
Is there some other way?
It would be sad to be forced back to use manual mapping for all classes ....
class MyIdConvention : IIdConvention
{
public void Apply(IIdentityInstance instance)
{
if (instance.EntityType == ...)
{
instance.GeneratedBy.Assigned();
}
}
}
Update:
for enum-like classes it's often easier to define an enum as id
class ConfigValue
{
public virtual Config Id { get; set; }
}
// the convention is easy
if (instance.EntityType.IsEnum)
{
instance.GeneratedBy.Assigned();
// to save as int and not string
instance.CustomType(typeof(Config));
}
// querying without magic int values
var configValue = Session.Get<ConfigValue>(Config.UIColor);
I used the idea given by Fifo and extended it to use a custom attribute instead.
To make code readable and avoid redundance when using similar idea in other conventions I added an extension method to check for custom attribute.
This is the code I ended up with:
/// <summary>
/// Convention to instruct FluentNHIbernate to NOT generate identity columns
/// when custom attribute is set.
/// </summary>
public class ConventionIdentity : IIdConvention
{
public void Apply(IIdentityInstance instance)
{
if(instance.CustomAttributeIsSet<NoIdentity>())
instance.GeneratedBy.Assigned();
}
}
/// <summary>
/// Custom attribute definition.
/// </summary>
public class NoIdentity : Attribute
{
}
/// <summary>
/// Example on how to set attribute.
/// </summary>
public class Category
{
[NoIdentity]
public int Id { get; set; }
public string Name { get; set; }
}
public static class IInspectorExtender
{
/// <summary>
/// Extender to make convention usage easier.
/// </summary>
public static T GetCustomAttribute<T>(this IInspector instance)
{
var memberInfos = instance.EntityType.GetMember(instance.StringIdentifierForModel);
if(memberInfos.Length > 0)
{
var customAttributes = memberInfos[0].GetCustomAttributes(false);
return customAttributes.OfType<T>().FirstOrDefault();
}
return default(T);
}
}
I have some existing asp.net membership and roles tables in a legacy db and I am mapping them to new entities with Fluent Nhibernate.
I also generate the schema directly from Fluent Nhibernate and I then manually tweak the generated sql script to exclude the existing tables.
Is it possible to say to Fluent Nhibernate to exclude from generation certain tables?
SchemaAction.None() in your ClassMap.
Another option would be to create an attribute, say
public class DoNotAutoPersistAttribute : Attribute
{
}
Then in AutoPersistenceModelGenerator you could check for this attribute in the Where clause of AddEntityAssembly.
I've managed this with an attribute + convention:
public enum SchemaAction
{
None
}
[Serializable]
[AttributeUsage(AttributeTargets.Class)]
public class SchemaActionAttribute : Attribute
{
private readonly SchemaAction schemaAction = SchemaAction.None;
public SchemaActionAttribute()
{
}
public SchemaActionAttribute(SchemaAction schemaAction)
{
this.schemaAction = schemaAction;
}
public SchemaAction GetSchemaAction()
{
return schemaAction;
}
}
/// <summary>
/// overrides the default action for entities when creating/updating the schema
/// based on the class having a Schema attribute (<see cref="SchemaActionAttribute" />)
/// </summary>
public class SchemaActionConvention : IClassConvention
{
public void Apply(IClassInstance instance)
{
object[] attributes = instance.EntityType.GetCustomAttributes(true);
foreach (object t in attributes)
{
if (t is SchemaActionAttribute)
{
var a = (SchemaActionAttribute) t;
switch(a.GetSchemaAction())
{
case SchemaAction.None:
instance.SchemaAction.None();
return;
default: throw new ApplicationException("That schema action:" + a.GetSchemaAction().ToString() + " is not currently implemented.");
}
}
}
}
}
...
[SchemaAction(SchemaAction.None)]
public class TextItem : Entity
...
I have two classes:
public class Parent
{
public virtual long? ID { get; set; } // native
public virtual IList<Child> Children { get; set; }
public virtual string Name { get; set; }
}
public class Child
{
public virtual long ID { get; set; } // assigned
public virtual string Name { get; set; }
}
Instantiating and saving parent and child:
child = new Child() { ID = 1, Name = "SomeName" };
parent = new Parent() { Children = new List() { child } };
session.Save(parent);
Which gives me:
NHibernate.StaleStateException: Unexpected row count: 0; expected: 1.
I think the problem is with the assigned id on the child. Since it has an id, NHibernate thinks it has previously saved before which is not the case.
The generated (trimmed & renamed) SQL is:
NHibernate: select child0_.ID as child1_1_, child0_.NAME as NAME1_, child0_.PARENT_ID as COMMAND7_1_, from CHILD child0_
NHibernate: select parent0_.PARENT_ID as parent1_10_
NHibernate: select parent0_.PARENT_ID as parent1_10_, parent0_.NAME as parent2_10_ from PARENT parent0_
NHibernate: UPDATE CHILD SET PARENT_ID = #p0 WHERE CHILD_ID = #p1;#p0 = 2, #p1 = 1
Mapping files:
<class name="MyNamespace.Child" table="CHILD">
<id name="ID" column="CHILD_ID" type="System.Int64">
<generator class="assigned"></generator>
</id>
<property name="Name" column="NAME"></property>
</class>
<class name="MyNamespace.Parent" table="PARENT">
<id name="ID" column="PARENT_ID" type="System.Int64">
<generator class="native"></generator>
</id>
<property name="Name" column="NAME"></property>
<bag name="Children">
<key column="PARENT_ID"></key>
<one-to-many class="MyNamespace.Child"></one-to-many>
</bag>
</class>
While searching google, I found about version tag which may be a solution but I do not have a persistent field to use as version. In this case, how can I save (insert) a child with assigned id and its parent?
When cascading from a parent to a child, NHibernate uses the SaveOrUpdate method. You are correct that NHibernate need some way to determine whether it should perform an insert or an update. It will look at three different fields for an unsaved value to determine if the entity is new.
Id
Version
Timestamp
With an assigned Id, you will need either a Version or Timestamp field in order to indicate that the entity is new.
An alternative would be to call Save() on the children explicitly.
I'm not 100% sure if this is the same problem you are having, but my database is 100% assigned id (ugh) and I had to create an Interceptor that kept track of whether a child is or isn't persisted for cascades to work.
The code is cut/paste (which is why it has dumb names... I didn't understand 100% at first!), and I initially got 90% of it from online documentation (which I can't find via google right now ... sorry):
Base class you put on object that has an assigned ID you want to cascade:
public class Persistent
{
private bool _saved = false;
public virtual void OnSave()
{
_saved = true;
}
public virtual void OnLoad()
{
_saved = true;
}
public virtual bool IsSaved
{
get { return _saved; }
}
}
The interceptor you add to session:
public class TrackingNumberInterceptor : EmptyInterceptor
{
public override bool? IsTransient(object entity)
{
if (entity is Persistent)
{
return !((Persistent)entity).IsSaved;
}
else
{
return null;
}
}
public override bool OnLoad(object entity, object id, object[] state, string[] propertyNames, IType[] types)
{
if (entity is Persistent) ((Persistent)entity).OnLoad();
return false;
}
public override bool OnSave(object entity, object id, object[] state, string[] propertyNames, IType[] types)
{
if (entity is Persistent) ((Persistent)entity).OnSave();
return false;
}
}
Basically the idea is that since NHibernate doesn't know if an assigned id entity is persisted or not, you keep track for it.
By default the object starts with persisted (_saved) at false. When the entity is either loaded or saved by NHibernate, the trigger sets the objects persisted (_saved) flag to true.
So for a fresh item that isn't persisted, it starts at false and stays false because NHibernate has never saved or loaded it. When NHibernate checks whether the child is transient, the trigger responds that it is transient, and a save happens which marks the child as persisted. Also now any future use will require a load which again marks it as persisted.
Calling session.SaveOrUpdate( childObject ) should solve the problem.