This tutorial explains a setup of NHibernate 5.0.0 within a Microsoft environment. The tools used in this tutorial are:

First, we have to create a new Web project. We use the name QuickStart. In the project, add a NuGet reference to NHibernate. Visual Studio will automatically copy the library and its dependencies to the project output directory. If you are using a database other than SQL Server, add a reference to its driver assembly to your project.

We now set up the database connection information for NHibernate. To do this, open the file Web.config automatically generated for your project and add configuration elements according to the listing below:

<?xml version="1.0" encoding="utf-8" ?>
<configuration>
  <!-- Add this element -->
  <configSections>
    <section
        name="hibernate-configuration"
        type="NHibernate.Cfg.ConfigurationSectionHandler, NHibernate" />
  </configSections>

  <!-- Add this element -->
  <hibernate-configuration xmlns="urn:nhibernate-configuration-2.2">
    <session-factory>
      <property name="dialect">NHibernate.Dialect.MsSql2012Dialect</property>
      <property name="connection.connection_string">
        Server=localhost\SQLEXPRESS;initial catalog=quickstart;Integrated Security=True
      </property>

      <mapping assembly="QuickStart" />
    </session-factory>
  </hibernate-configuration>

  <!-- Leave the other sections unchanged -->
  <system.web>
    ...
  </system.web>
</configuration>

The <configSections> element contains definitions of sections that follow and handlers to use to process their content. We declare the handler for the configuration section here. The <hibernate-configuration> section contains the configuration itself, telling NHibernate that we will use a Microsoft SQL Server 2012 database and connect to it through the specified connection string. The dialect is a required setting, databases differ in their interpretation of the SQL "standard". NHibernate will take care of the differences and comes bundled with dialects for several major commercial and open source databases.

An ISessionFactory is NHibernate's concept of a single datastore, multiple databases can be used by creating multiple XML configuration files and creating multiple Configuration and ISessionFactory objects in your application.

The last element of the <hibernate-configuration> section declares QuickStart as the name of an assembly containing class declarations and mapping files. The mapping files contain the metadata for the mapping of the POCO class to a database table (or multiple tables). We'll come back to mapping files soon. Let's write the POCO class first and then declare the mapping metadata for it.

NHibernate works best with the Plain Old CLR Objects (POCOs, sometimes called Plain Ordinary CLR Objects) programming model for persistent classes. A POCO has its data accessible through the standard .NET property mechanisms, shielding the internal representation from the publicly visible interface:

namespace QuickStart
{
    public class Cat
    {
        public virtual string Id { get; set; }

        public virtual string Name { get; set; }

        public virtual char Sex { get; set; }

        public virtual float Weight { get; set; }
    }
}

NHibernate is not restricted in its usage of property types, all .NET types and primitives (like string, char and DateTime) can be mapped, including classes from the System.Collections.Generic namespace. You can map them as values, collections of values, or associations to other entities. The Id is a special property that represents the database identifier (primary key) of that class, it is highly recommended for entities like a Cat. NHibernate can use identifiers only internally, without having to declare them on the class, but we would lose some of the flexibility in our application architecture.

No special interface has to be implemented for persistent classes nor do we have to subclass from a special root persistent class. NHibernate also doesn't use any build time processing, such as IL manipulation, it relies solely on .NET reflection and runtime class enhancement. So, without any dependency in the POCO class on NHibernate, we can map it to a database table.

For the above mentioned runtime class enhancement to work, NHibernate requires that all public properties of an entity class are declared as virtual. It also requires a parameter-less constructor: if you add a constructor having parameters, make sure to add a parameter-less constructor too.

The Cat.hbm.xml mapping file contains the metadata required for the object/relational mapping. The metadata includes declaration of persistent classes and the mapping of properties (to columns and foreign key relationships to other entities) to database tables.

Please note that the Cat.hbm.xml file should be set to an embedded resource.

<?xml version="1.0" encoding="utf-8" ?>
<hibernate-mapping xmlns="urn:nhibernate-mapping-2.2"
    namespace="QuickStart" assembly="QuickStart">

    <class name="Cat" table="Cat">

        <!-- A 32 hex character is our surrogate key. It's automatically
            generated by NHibernate with the UUID pattern. -->
        <id name="Id">
            <column name="CatId" sql-type="char(32)" not-null="true"/>
            <generator class="uuid.hex" />
        </id>

        <!-- A cat has to have a name, but it shouldn't be too long. -->
        <property name="Name">
            <column name="Name" length="16" not-null="true" />
        </property>
        <property name="Sex" />
        <property name="Weight" />
    </class>

</hibernate-mapping>

Every persistent class should have an identifier attribute (actually, only classes representing entities, not dependent value objects, which are mapped as components of an entity). This property is used to distinguish persistent objects: Two cats are equal if catA.Id.Equals(catB.Id) is true, this concept is called database identity. NHibernate comes bundled with various identifier generators for different scenarios (including native generators for database sequences, hi/lo identifier tables, and application assigned identifiers). We use the UUID generator (only recommended for testing, as integer surrogate keys generated by the database should be preferred) and also specify the column CatId of the table Cat for the NHibernate generated identifier value (as a primary key of the table).

All other properties of Cat are mapped to the same table. In the case of the Name property, we mapped it with an explicit database column declaration. This is especially useful when the database schema is automatically generated (as SQL DDL statements) from the mapping declaration with NHibernate's SchemaExport tool. All other properties are mapped using NHibernate's default settings, which is what you need most of the time. Here the specification of the table name with the attribute table is redundant, it default to the class name when not specified. The table Cat in the database looks like this:

 Column |     Type     | Modifiers
--------+--------------+----------------------
 CatId  | char(32)     | not null, primary key
 Name   | nvarchar(16) | not null
 Sex    | nchar(1)     |
 Weight | real         |

You should now create the database and this table manually, and later read Chapter 21, Toolset Guide if you want to automate this step with the SchemaExport tool. This tool can create a full SQL DDL, including table definition, custom column type constraints, unique constraints and indexes. If you are using SQL Server, you should also make sure the ASPNET user has permissions to use the database.

We're now ready to start NHibernate's ISession. It is the persistence manager interface, we use it to store and retrieve Cats to and from the database. But first, we've to get an ISession (NHibernate's unit-of-work) from the ISessionFactory:

ISessionFactory sessionFactory =
            new Configuration().Configure().BuildSessionFactory();

An ISessionFactory is responsible for one database and may only use one XML configuration file (Web.config or hibernate.cfg.xml). You can set other properties (and even change the mapping metadata) by accessing the Configuration before you build the ISessionFactory (it is immutable). Where do we create the ISessionFactory and how can we access it in our application?

An ISessionFactory is usually only built once, e.g. at start-up inside Application_Start event handler. This also means you should not keep it in an instance variable in your ASP.NET pages or MVC controllers, but in some other location. Furthermore, we need some kind of Singleton, so we can access the ISessionFactory easily in application code. The approach shown next solves both problems: configuration and easy access to a ISessionFactory.

We implement a NHibernateHelper helper class:

using System;
using System.Web;
using NHibernate;
using NHibernate.Cfg;

namespace QuickStart
{
    public sealed class NHibernateHelper
    {
        private const string CurrentSessionKey = "nhibernate.current_session";
        private static readonly ISessionFactory _sessionFactory;

        static NHibernateHelper()
        {
            _sessionFactory = new Configuration().Configure().BuildSessionFactory();
        }

        public static ISession GetCurrentSession()
        {
            var context = HttpContext.Current;
            var currentSession = context.Items[CurrentSessionKey] as ISession;

            if (currentSession == null)
            {
                currentSession = _sessionFactory.OpenSession();
                context.Items[CurrentSessionKey] = currentSession;
            }

            return currentSession;
        }

        public static void CloseSession()
        {
            var context = HttpContext.Current;
            var currentSession = context.Items[CurrentSessionKey] as ISession;

            if (currentSession == null)
            {
                // No current session
                return;
            }

            currentSession.Close();
            context.Items.Remove(CurrentSessionKey);
        }

        public static void CloseSessionFactory()
        {
            if (_sessionFactory != null)
            {
                _sessionFactory.Close();
            }
        }
    }
}

This class does not only take care of the ISessionFactory with its static attribute, but also has code to remember the ISession for the current HTTP request.

An ISessionFactory is threadsafe, many threads can access it concurrently and request ISessions. An ISession is a non-threadsafe object that represents a single unit-of-work with the database. ISessions are opened by an ISessionFactory and are closed when all work is completed:

ISession session = NHibernateHelper.GetCurrentSession();
try
{
    using (ITransaction tx = session.BeginTransaction())
    {
        var princess = new Cat
        {
            Name = "Princess",
            Sex = 'F',
            Weight = 7.4f
        };

        session.Save(princess);
        tx.Commit();
    }
}
finally
{
    NHibernateHelper.CloseSession();
}

In an ISession, every database operation occurs inside a transaction that isolates the database operations (even read-only operations). We use NHibernate's ITransaction API to abstract from the underlying transaction strategy (in our case, ADO.NET transactions). Please note that the example above does not handle any exceptions.

Also note that you may call NHibernateHelper.GetCurrentSession(); as many times as you like, you will always get the current ISession of this HTTP request. You have to make sure the ISession is closed after your unit-of-work completes, either in Application_EndRequest event handler in your application class, or with a MVC action filter, or in a HttpModule before the HTTP response is sent. The nice side effect of the latter is easy lazy initialization: the ISession is still open when the view is rendered, so NHibernate can load uninitialized objects while you navigate the graph.

NHibernate has various methods that can be used to retrieve objects from the database. Nowadays the most standard way is using Linq:

using(var tx = session.BeginTransaction())
{
    var females = session
        .Query<Cat>()
        .Where(c => c.Sex == 'F')
        .ToList();
    foreach (var cat in females)
    {
        Console.Out.WriteLine("Female Cat: " + cat.Name);
    }

    tx.Commit();
}

If you use an older NHibernate, you may have to import the NHibernate.Linq namespace.

NHibernate also offers an object-oriented query by criteria API that can be used to formulate type-safe queries, the Hibernate Query Language (HQL), which is an easy to learn and powerful object-oriented extension to SQL, as well as a strongly-typed LINQ API which translates internally to HQL. NHibernate of course uses DbCommands and parameter binding for all SQL communication with the database. You may also use NHibernate's direct SQL query feature or get a plain ADO.NET connection from an ISession in rare cases.

Since NHibernate 5.0, the session and its queries IO bound methods have async counterparts. Each call to an async method must be awaited before further interacting with the session or its queries.

We only scratched the surface of NHibernate in this small tutorial. Please note that we don't include any ASP.NET specific code in our examples. You have to create an ASP.NET page yourself and insert the NHibernate code as you see fit.

Keep in mind that NHibernate, as a data access layer, is tightly integrated into your application. Usually, all other layers depend on the persistence mechanism. Make sure you understand the implications of this design.

A (very) high-level view of the NHibernate architecture:

This diagram shows NHibernate using the database and configuration data to provide persistence services (and persistent objects) to the application.

We would like to show a more detailed view of the runtime architecture. Unfortunately, NHibernate is flexible and supports several approaches. We will show the two extremes. The "lite" architecture has the application provide its own ADO.NET connections and manage its own transactions. This approach uses a minimal subset of NHibernate's APIs:

The "full cream" architecture abstracts the application away from the underlying ADO.NET APIs and lets NHibernate take care of the details.

Here are some definitions of the objects in the diagrams:

Given a "lite" architecture, the application bypasses the ITransaction/ITransactionFactory and/or IConnectionProvider APIs to talk to ADO.NET directly.

An instance of a persistent classes may be in one of three different states, which are defined with respect to a persistence context. The NHibernate ISession object is the persistence context:

Most applications using NHibernate need some form of "contextual" sessions, where a given session is in effect throughout the scope of a given context. However, across applications the definition of what constitutes a context is typically different; and different contexts define different scopes to the notion of current.

Starting with version 1.2, NHibernate added the ISessionFactory.GetCurrentSession() method. The processing behind ISessionFactory.GetCurrentSession() is pluggable. An extension interface (NHibernate.Context.ICurrentSessionContext) and a new configuration parameter (current_session_context_class) have been added to allow pluggability of the scope and context of defining current sessions.

See the API documentation for the NHibernate.Context.ICurrentSessionContext interface for a detailed discussion of its contract. It defines a single method, CurrentSession(), by which the implementation is responsible for tracking the current contextual session. Out-of-the-box, NHibernate comes with several implementations of this interface:

The current_session_context_class configuration parameter defines which NHibernate.Context.ICurrentSessionContext implementation should be used. Typically, the value of this parameter would just name the implementation class to use (including the assembly name); for the out-of-the-box implementations, however, there are corresponding short names: async_local, call, thread_static, web and wcf_operation, respectively.

An instance of NHibernate.Cfg.Configuration represents an entire set of mappings of an application's .NET types to a SQL database. The Configuration is used to build an (immutable) ISessionFactory. The mappings are compiled from various XML mapping files.

You may obtain a Configuration instance by instantiating it directly. Here is an example of setting up a datastore from mappings defined in two XML configuration files:

Configuration cfg = new Configuration()
    .AddFile("Item.hbm.xml")
    .AddFile("Bid.hbm.xml");

An alternative (sometimes better) way is to let NHibernate load a mapping file from an embedded resource:

Configuration cfg = new Configuration()
    .AddClass(typeof(NHibernate.Auction.Item))
    .AddClass(typeof(NHibernate.Auction.Bid));

Then NHibernate will look for mapping files named NHibernate.Auction.Item.hbm.xml and NHibernate.Auction.Bid.hbm.xml embedded as resources in the assembly that the types are contained in. This approach eliminates any hardcoded filenames.

Another alternative (probably the best) way is to let NHibernate load all of the mapping files contained in an Assembly:

Configuration cfg = new Configuration()
    .AddAssembly( "NHibernate.Auction" );

Then NHibernate will look through the assembly for any resources that end with .hbm.xml. This approach eliminates any hardcoded filenames and ensures the mapping files in the assembly get added.

If a tool like Visual Studio .NET or NAnt is used to build the assembly, then make sure that the .hbm.xml files are compiled into the assembly as Embedded Resources.

A Configuration also specifies various optional properties:

var props = new Dictionary<string, string>();
...
Configuration cfg = new Configuration()
    .AddClass(typeof(NHibernate.Auction.Item))
    .AddClass(typeof(NHibernate.Auction.Bind))
    .SetProperties(props);

A Configuration is intended as a configuration-time object, to be discarded once an ISessionFactory is built.

When all mappings have been parsed by the Configuration, the application must obtain a factory for ISession instances. This factory is intended to be shared by all application threads:

ISessionFactory sessions = cfg.BuildSessionFactory();

However, NHibernate does allow your application to instantiate more than one ISessionFactory. This is useful if you are using more than one database.

An ISessionFactory may open an ISession on a user-provided ADO.NET connection. This design choice frees the application to obtain ADO.NET connections wherever it pleases:

var conn = myApp.GetOpenConnection();
var session = sessions.OpenSession(conn);

// do some data access work

The application must be careful not to open two concurrent ISessions on the same ADO.NET connection!

Alternatively, you can have the ISessionFactory open connections for you. The ISessionFactory must be provided with ADO.NET connection properties in one of the following ways:

If you take this approach, opening an ISession is as simple as:

ISession session = sessions.OpenSession(); // open a new Session
// do some data access work, an ADO.NET connection will be used on demand

All NHibernate property names and semantics are defined on the class NHibernate.Cfg.Environment. We will now describe the most important settings for ADO.NET connection configuration.

NHibernate will obtain (and pool) connections using an ADO.NET data provider if you set the following properties:

This is an example of how to specify the database connection properties inside a web.config:

<?xml version="1.0" encoding="utf-8" ?>
<configuration>
  <configSections>
    <section name="hibernate-configuration"
        type="NHibernate.Cfg.ConfigurationSectionHandler, NHibernate" />
  </configSections>

  <hibernate-configuration xmlns="urn:nhibernate-configuration-2.2">
    <session-factory>
      <property name="dialect">NHibernate.Dialect.MsSql2012Dialect</property>
      <property name="connection.connection_string">
        Server=(local);initial catalog=theDb;Integrated Security=SSPI
      </property>
      <property name="connection.isolation">ReadCommitted</property>
    </session-factory>
  </hibernate-configuration>

  <!-- other app specific config follows -->
</configuration>

NHibernate relies on the ADO.NET data provider implementation of connection pooling.

You may define your own plug-in strategy for obtaining ADO.NET connections by implementing the interface NHibernate.Connection.IConnectionProvider. You may select a custom implementation by setting connection.provider.

There are a number of other properties that control the behaviour of NHibernate at runtime. All are optional and have reasonable default values.

Some properties are system-level properties. They can only be set manually by setting static properties of NHibernate.Cfg.Environment class or be defined in the <hibernate-configuration> section of the application configuration file. These properties cannot be set using Configuration.SetProperties or the hibernate.cfg.xml configuration file.

System.Data.Odbc.OdbcException :
ERROR [22008]
[Microsoft][SQL Server Native Client 11.0]
Datetime field overflow. Fractional second
precision exceeds the scale specified
in the parameter binding.

query.substitutions true=1, false=0

query.substitutions toLowercase=LOWER

NHibernate logs various events using Apache log4net.

You may download log4net from https://logging.apache.org/log4net/, or install it with NuGet. To use log4net you will need a log4net configuration section in the application configuration file. An example of the configuration section is distributed with NHibernate in the src/NHibernate.Test project.

We strongly recommend that you familiarize yourself with NHibernate's log messages. A lot of work has been put into making the NHibernate log as detailed as possible, without making it unreadable. It is an essential troubleshooting device. Also don't forget to enable SQL logging as described above (show_sql), it is your first step when looking for performance problems.

The interface NHibernate.Cfg.INamingStrategy allows you to specify a "naming standard" for database objects and schema elements.

You may provide rules for automatically generating database identifiers from .NET identifiers or for processing "logical" column and table names given in the mapping file into "physical" table and column names. This feature helps reduce the verbosity of the mapping document, eliminating repetitive noise (TBL_ prefixes, for example). The default strategy used by NHibernate is quite minimal.

You may specify a different strategy by calling Configuration.SetNamingStrategy() before adding mappings:

ISessionFactory sf = new Configuration()
    .SetNamingStrategy(ImprovedNamingStrategy.Instance)
    .AddFile("Item.hbm.xml")
    .AddFile("Bid.hbm.xml")
    .BuildSessionFactory();

NHibernate.Cfg.ImprovedNamingStrategy is a built-in strategy that might be a useful starting point for some applications.

An alternative approach is to specify a full configuration in a file named hibernate.cfg.xml. This file can be used as a replacement for the <hibernate-configuration> sections of the application configuration file.

The XML configuration file is by default expected to be in your application directory. Here is an example:

<?xml version='1.0' encoding='utf-8'?>
<hibernate-configuration xmlns="urn:nhibernate-configuration-2.2">

  <!-- an ISessionFactory instance -->
  <session-factory>

    <!-- properties -->
    <property name="connection.connection_string">
      Server=localhost;initial catalog=nhibernate;User Id=;Password=
    </property>
    <property name="show_sql">false</property>
    <property name="dialect">NHibernate.Dialect.MsSql2012Dialect</property>

    <!-- mapping files -->
    <mapping resource="NHibernate.Auction.Item.hbm.xml" assembly="NHibernate.Auction" />
    <mapping resource="NHibernate.Auction.Bid.hbm.xml" assembly="NHibernate.Auction" />

  </session-factory>

</hibernate-configuration>

Configuring NHibernate is then as simple as

ISessionFactory sf = new Configuration().Configure().BuildSessionFactory();

You can pick a different XML configuration file using

ISessionFactory sf = new Configuration()
    .Configure("/path/to/config.cfg.xml")
    .BuildSessionFactory();

Most .NET applications require a persistent class representing felines.

using System;
using System.Collections.Generic;

namespace Eg
{
    public class Cat
    {
        long _id;
        // identifier

        public virtual long Id
        {
            get { return _id; }
            protected set { _id = value; }
        }

        public virtual string Name { get; set; }
        public virtual Cat Mate { get; set; }
        public virtual DateTime Birthdate { get; set; }
        public virtual float Weight { get; set; }
        public virtual Color Color { get; set; }
        public virtual ISet<Cat> Kittens { get; set; }
        public virtual char Sex { get; set; }

        // AddKitten not needed by NHibernate
        public virtual void AddKitten(Cat kitten)
        {
            kittens.Add(kitten);
        }
    }
}

There are four main rules to follow here:

A subclass must also observe the first and second rules. It inherits its identifier property from Cat.

using System;
namespace Eg
{
    public class DomesticCat : Cat
    {
        public virtual string Name { get; set; }
    }
}

You have to override the Equals() and GetHashCode() methods if you intend to mix objects of persistent classes (e.g. in an ISet).

This only applies if these objects are loaded in two different ISessions, as NHibernate only guarantees identity ( a == b , the default implementation of Equals()) inside a single ISession!

Even if both objects a and b are the same database row (they have the same primary key value as their identifier), we can't guarantee that they are the same object instance outside of a particular ISession context.

The most obvious way is to implement Equals()/GetHashCode() by comparing the identifier value of both objects. If the value is the same, both must be the same database row, they are therefore equal (if both are added to an ISet, we will only have one element in the ISet). Unfortunately, we can't use that approach. NHibernate will only assign identifier values to objects that are persistent, a newly created instance will not have any identifier value! We recommend implementing Equals() and GetHashCode() using Business key equality.

Business key equality means that the Equals() method compares only the properties that form the business key, a key that would identify our instance in the real world (a natural candidate key):

public class Cat
{

    ...
    public override bool Equals(object other)
    {
        if (this == other) return true;
        
        Cat cat = other as Cat;
        if (cat == null) return false; // null or not a cat

        if (Name != cat.Name) return false;
        if (!Birthday.Equals(cat.Birthday)) return false;

        return true;
    }

    public override int GetHashCode()
    {
        unchecked
        {
            int result;
            result = Name.GetHashCode();
            result = 29 * result + Birthday.GetHashCode();
            return result;
        }
    }

}

Keep in mind that our candidate key (in this case a composite of name and birthday) has to be only valid for a particular comparison operation (maybe even only in a single use case). We don't need the stability criteria we usually apply to a real primary key!

Note that the following features are currently considered experimental and may change in the near future.

Persistent entities don't necessarily have to be represented as POCO classes at runtime. NHibernate also supports dynamic models (using Dictionaries of Dictionarys at runtime) . With this approach, you don't write persistent classes, only mapping files.

The following examples demonstrates the representation using Maps (Dictionary). First, in the mapping file, an entity-name has to be declared instead of a class name:

<hibernate-mapping>

    <class entity-name="Customer">

        <id name="id"
            type="long"
            column="ID">
            <generator class="sequence"/>
        </id>

        <property name="name"
            column="NAME"
            type="string"/>

        <property name="address"
            column="ADDRESS"
            type="string"/>

        <many-to-one name="organization"
            column="ORGANIZATION_ID"
            class="Organization"/>

        <bag name="orders"
            inverse="true"
            lazy="false"
            cascade="all">
            <key column="CUSTOMER_ID"/>
            <one-to-many class="Order"/>
        </bag>

    </class>
    
</hibernate-mapping>

Note that even though associations are declared using target class names, the target type of an associations may also be a dynamic entity instead of a POCO.

At runtime we can work with Dictionaries of Dictionaries:

using(ISession s = OpenSession())
using(ITransaction tx = s.BeginTransaction())
{
    // Create a customer
    var frank = new Dictionary<string, object>();
    frank["name"] = "Frank";

    // Create an organization
    var foobar = new Dictionary<string, object>();
    foobar["name"] = "Foobar Inc.";

    // Link both
    frank["organization"] =  foobar;

    // Save both
    s.Save("Customer", frank);
    s.Save("Organization", foobar);

    tx.Commit();
}

The advantages of a dynamic mapping are quick turnaround time for prototyping without the need for entity class implementation. However, you lose compile-time type checking and will very likely deal with many exceptions at runtime. Thanks to the NHibernate mapping, the database schema can easily be normalized and sound, allowing to add a proper domain model implementation on top later on.

NHibernate.Tuple.Tuplizer, and its sub-interfaces, are responsible for managing a particular representation of a piece of data, given that representation's NHibernate.EntityMode. If a given piece of data is thought of as a data structure, then a tuplizer is the thing which knows how to create such a data structure and how to extract values from and inject values into such a data structure. For example, for the POCO entity mode, the corresponding tuplizer knows how create the POCO through its constructor and how to access the POCO properties using the defined property accessors. There are two high-level types of Tuplizers, represented by the NHibernate.Tuple.Entity.IEntityTuplizer and NHibernate.Tuple.Component.IComponentTuplizer interfaces. IEntityTuplizers are responsible for managing the above mentioned contracts in regards to entities, while IComponentTuplizers do the same for components.

Users may also plug in their own tuplizers. Perhaps you require that a System.Collections.IDictionary implementation other than System.Collections.Hashtable be used while in the dynamic-map entity-mode; or perhaps you need to define a different proxy generation strategy than the one used by default. Both would be achieved by defining a custom tuplizer implementation. Tuplizers definitions are attached to the entity or component mapping they are meant to manage. Going back to the example of our customer entity:

<hibernate-mapping>
    <class entity-name="Customer">
        <!--
            Override the dynamic-map entity-mode
            tuplizer for the customer entity
        -->
        <tuplizer entity-mode="dynamic-map"
                class="CustomMapTuplizerImpl"/>

        <id name="id" type="long" column="ID">
            <generator class="sequence"/>
        </id>

        <!-- other properties -->
        ...
    </class>
</hibernate-mapping>


public class CustomMapTuplizerImpl : NHibernate.Tuple.Entity.DynamicMapEntityTuplizer
{
    // override the BuildInstantiator() method to plug in our custom map...
    protected override IInstantiator BuildInstantiator(
        NHibernate.Mapping.PersistentClass mappingInfo)
    {
        return new CustomMapInstantiator(mappingInfo);
    }

    private sealed class CustomMapInstantiator : NHibernate.Tuple.DynamicMapInstantiator
    {
        // override the generateMap() method to return our custom map...
        protected override IDictionary GenerateMap()
        {
            return new CustomMap();
        }
    }
}

Optionally, a persistent class might implement the interface ILifecycle which provides some callbacks that allow the persistent object to perform necessary initialization/cleanup after save or load and before deletion or update.

The NHibernate IInterceptor offers a less intrusive alternative, however.

public interface ILifecycle
{                                                                    (1)
        LifecycleVeto OnSave(ISession s);                            (2)
        LifecycleVeto OnUpdate(ISession s);                          (3)
        LifecycleVeto OnDelete(ISession s);                          (4)
        void OnLoad(ISession s, object id);
}

OnSave(), OnDelete() and OnUpdate() may be used to cascade saves and deletions of dependent objects. This is an alternative to declaring cascaded operations in the mapping file. OnLoad() may be used to initialize transient properties of the object from its persistent state. It may not be used to load dependent objects since the ISession interface may not be invoked from inside this method. A further intended usage of OnLoad(), OnSave() and OnUpdate() is to store a reference to the current ISession for later use.

Note that OnUpdate() is not called every time the object's persistent state is updated. It is called only when a transient object is passed to ISession.Update().

If OnSave(), OnUpdate() or OnDelete() return LifecycleVeto.Veto, the operation is silently vetoed. If a CallbackException is thrown, the operation is vetoed and the exception is passed back to the application.

Note that OnSave() is called after an identifier is assigned to the object, except when native key generation is used.

If the persistent class needs to check invariants before its state is persisted, it may implement the following interface:

public interface IValidatable
{
        void Validate();
}

The object should throw a ValidationFailure if an invariant was violated. An instance of Validatable should not change its state from inside Validate().

Unlike the callback methods of the ILifecycle interface, Validate() might be called at unpredictable times. The application should not rely upon calls to Validate() for business functionality.

Object/relational mappings are defined in an XML document. The mapping document is designed to be readable and hand-editable. The mapping language is object-centric, meaning that mappings are constructed around persistent class declarations, not table declarations.

Note that, even though many NHibernate users choose to define XML mappings by hand, a number of tools exist to generate the mapping document, including NHibernate.Mapping.Attributes library and various template-based code generators (CodeSmith, MyGeneration). You may also use NHibernate.Mapping.ByCode available since NHibernate 3.2, or Fluent NHibernate.

Let's kick off with an example mapping:

<?xml version="1.0"?>
<hibernate-mapping xmlns="urn:nhibernate-mapping-2.2" assembly="Eg"
    namespace="Eg">

        <class name="Cat" table="CATS" discriminator-value="C">
                <id name="Id" column="uid" type="Int64">
                        <generator class="hilo"/>
                </id>
                <discriminator column="subclass" type="Char"/>
                <property name="BirthDate" type="Date"/>
                <property name="Color" not-null="true"/>
                <property name="Sex" not-null="true" update="false"/>
                <property name="Weight"/>
                <many-to-one name="Mate" column="mate_id"/>
                <set name="Kittens">
                        <key column="mother_id"/>
                        <one-to-many class="Cat"/>
                </set>
                <subclass name="DomesticCat" discriminator-value="D">
                        <property name="Name" type="String"/>
                </subclass>
        </class>

        <class name="Dog">
                <!-- mapping for Dog could go here -->
        </class>

</hibernate-mapping>

We will now discuss the content of the mapping document. We will only describe the document elements and attributes that are used by NHibernate at runtime. The mapping document also contains some extra optional attributes and elements that affect the database schemas exported by the schema export tool. (For example the not-null attribute.)

<hibernate-mapping
         schema="schemaName"                          (1)
         default-cascade="none|save-update"           (2)
         auto-import="true|false"                     (3)
         assembly="Eg"                                (4)
         namespace="Eg"                               (5)
         default-access="field|property|field.camecase(6)..."
         default-lazy="true|false"                    (7)
 />

<class
        name="ClassName"                              (1)
        table="tableName"                             (2)
        discriminator-value="discriminator_value"     (3)
        mutable="true|false"                          (4)
        schema="owner"                                (5)
        proxy="ProxyInterface"                        (6)
        dynamic-update="true|false"                   (7)
        dynamic-insert="true|false"                   (8)
        select-before-update="true|false"             (9)
        polymorphism="implicit|explicit"              (10)
        where="arbitrary sql where condition"         (11)
        persister="PersisterClass"                    (12)
        batch-size="N"                                (13)
        optimistic-lock="none|version|dirty|all"      (14)
        lazy="true|false"                             (15)
        abstract="true|false"                         (16)
/>

<subselect>
    <![CDATA[
    SELECT cat.ID, cat.NAME, cat.SEX, cat.MATE FROM cat
    ]]>
</subselect>

<subselect>
    <![CDATA[
    SELECT cat.ID, cat.NAME, cat.SEX, cat.MATE FROM cat
    ]]>
</subselect>
<syncronize table="cat"/>

<id
        name="PropertyName"                      (1)
        type="typename"                          (2)
        column="column_name"                     (3)
        unsaved-value="any|none|null|id_value"   (4)
        access="field|property|nosetter|ClassName(5)">

        <generator class="generatorClass"/>
</id>

<id name="Id" type="Int64" column="uid" unsaved-value="0">
        <generator class="NHibernate.Id.TableHiLoGenerator">
                <param name="table">uid_table</param>
                <param name="column">next_hi_value_column</param>
        </generator>
</id>

<id name="Id" type="Int64" column="uid" unsaved-value="0" generator="native" />

<id name="Id" type="Int64" column="cat_id">
        <generator class="hilo">
                <param name="table">hi_value</param>
                <param name="column">next_value</param>
                <param name="max_lo">100</param>
        </generator>
</id>

<id name="Id" type="Int64" column="cat_id">
        <generator class="seqhilo">
                <param name="sequence">hi_value</param>
                <param name="max_lo">100</param>
        </generator>
</id>

<id name="Id" type="String" column="cat_id">
        <generator class="uuid.hex">
            <param name="format">format_value</param>
            <param name="separator">separator_value</param>
        </generator>
</id>

<id name="Id" type="Int64" column="uid">
        <generator class="sequence">
                <param name="sequence">uid_sequence</param>
        </generator>
</id>

<id name="Id" type="Int64" column="uid" unsaved-value="0">
        <generator class="identity"/>
</id>

<composite-id
        name="PropertyName"
        class="ClassName"
        unsaved-value="any|none"
        access="field|property|nosetter|ClassName">

        <key-property name="PropertyName" type="typename" column="column_name"/>
        <key-many-to-one name="PropertyName class="ClassName" column="column_name"/>
        ......
</composite-id>

<composite-id>
        <key-property name="MedicareNumber"/>
        <key-property name="Dependent"/>
</composite-id>

<discriminator
        column="discriminator_column"  (1)
        type="discriminator_type"      (2)
        force="true|false"             (3)
        insert="true|false"            (4)
        formula="arbitrary SQL expressi(5)on"
/>

<discriminator
    formula="case when CLASS_TYPE in ('a', 'b', 'c') then 0 else 1 end"
    type="Int32"/>

<version
        column="version_column"                            (1)
        name="PropertyName"                                (2)
        type="typename"                                    (3)
        access="field|property|nosetter|ClassName"         (4)
        unsaved-value="null|negative|undefined|value"      (5)
        generated="never|always"                           (6)
/>

<timestamp
        column="timestamp_column"           (1)
        name="PropertyName"                 (2)
        access="field|property|nosetter|Clas(3)sName"
        unsaved-value="null|undefined|value"(4)
        generated="never|always"            (5)
/>

<property
        name="propertyName"                 (1)
        column="column_name"                (2)
        type="typename"                     (3)
        update="true|false"                 (4)
        insert="true|false"                 (4)
        formula="arbitrary SQL expression"  (5)
        access="field|property|ClassName"   (6)
        optimistic-lock="true|false"        (7)
        generated="never|insert|always"     (8)
        lazy="true|false"                   (9)
/>

<many-to-one
        name="PropertyName"                                (1)
        column="column_name"                               (2)
        class="ClassName"                                  (3)
        cascade="all|none|save-update|delete|delete-orphan|(4)all-delete-orphan"
        fetch="join|select"                                (5)
        update="true|false"                                (6)
        insert="true|false"                                (6)
        property-ref="PropertyNameFromAssociatedClass"     (7)
        access="field|property|nosetter|ClassName"         (8)
        unique="true|false"                                (9)
        optimistic-lock="true|false"                       (10)
        not-found="ignore|exception"                       (11)
/>

<many-to-one name="product" class="Product" column="PRODUCT_ID"/>

<property name="serialNumber" unique="true" type="string" column="SERIAL_NUMBER"/>

<many-to-one name="product" property-ref="serialNumber" column="PRODUCT_SERIAL_NUMBER"/>

<one-to-one
        name="PropertyName"                                (1)
        class="ClassName"                                  (2)
        cascade="all|none|save-update|delete|delete-orphan|(3)all-delete-orphan"
        constrained="true|false"                           (4)
        fetch="join|select"                                (5)
        property-ref="PropertyNameFromAssociatedClass"     (6)
        access="field|property|nosetter|ClassName"         (7)
/>

<one-to-one name="Person" class="Person"/>

<one-to-one name="Employee" class="Employee" constrained="true"/>

<class name="Person" table="PERSON">
    <id name="Id" column="PERSON_ID">
        <generator class="foreign">
            <param name="property">Employee</param>
        </generator>
    </id>
    ...
    <one-to-one name="Employee"
        class="Employee"
        constrained="true"/>
</class>

<many-to-one name="Person" class="Person" column="PERSON_ID" unique="true"/>

<one-to-one name="Employee" class="Employee" property-ref="Person"/>

<natural-id mutable="true|false"/>
        <property ... />
        <many-to-one ... />
        ......
</natural-id>

<component 
        name="PropertyName"                                (1)
        class="ClassName"                                  (2)
        insert="true|false"                                (3)
        upate="true|false"                                 (4)
        access="field|property|nosetter|ClassName"         (5)
        optimistic-lock="true|false">                      (6)

        <property ...../>
        <many-to-one .... />
        ........
</component>

<properties
      name="logicalName"                                   (1)
      insert="true|false"                                  (2)
      update="true|false"                                  (3)
      optimistic-lock="true|false"                         (4)
      unique="true|false">                                 (5)

      <property .../>
      <many-to-one .../>
      ........
</properties>

<class name="Person">
      <id name="personNumber" />
      <properties name="name" unique="true" update="false">
          <property name="firstName" />
          <property name="lastName" />
          <property name="initial" />
      </properties>
</class>

<many-to-one name="owner" class="Person" property-ref="name">
        <column name="firstName" />
        <column name="lastName" />
        <column name="initial" />
</many-to-one>

<subclass
        name="ClassName"                              (1)
        discriminator-value="discriminator_value"     (2)
        proxy="ProxyInterface"                        (3)
        lazy="true|false"                             (4)
        dynamic-update="true|false"
        dynamic-insert="true|false">

        <property .... />
        <properties .... />
        .....
</subclass>

<joined-subclass
        name="ClassName"                    (1)
        proxy="ProxyInterface"              (2)
        lazy="true|false"                   (3)
        dynamic-update="true|false"
        dynamic-insert="true|false">

        <key .... >

        <property .... />
        <properties .... />
        .....
</joined-subclass>

<?xml version="1.0"?>
<hibernate-mapping xmlns="urn:nhibernate-mapping-2.2" assembly="Eg"
    namespace="Eg">

        <class name="Cat" table="CATS">
                <id name="Id" column="uid" type="Int64">
                        <generator class="hilo"/>
                </id>
                <property name="BirthDate" type="Date"/>
                <property name="Color" not-null="true"/>
                <property name="Sex" not-null="true"/>
                <property name="Weight"/>
                <many-to-one name="Mate"/>
                <set name="Kittens">
                        <key column="MOTHER"/>
                        <one-to-many class="Cat"/>
                </set>
                <joined-subclass name="DomesticCat" table="DOMESTIC_CATS">
                    <key column="CAT"/>
                        <property name="Name" type="String"/>
                </joined-subclass>
        </class>

        <class name="Dog">
                <!-- mapping for Dog could go here -->
        </class>

</hibernate-mapping>

<union-subclass
        name="ClassName"                    (1)
        table="tablename"                   (2)
        proxy="ProxyInterface"              (3)
        lazy="true|false"                   (4)
        dynamic-update="true|false"
        dynamic-insert="true|false"
        schema="schema"
        catalog="catalog"
        extends="SuperclassName"
        abstract="true|false"
        persister="ClassName"
        subselect="SQL expression"
        entity-name="EntityName"
        node="element-name">

        <property .... />
        <properties .... />
        .....
</union-subclass>

<join
        table="tablename"                        (1)
        schema="owner"                           (2)
        fetch="join|select"                      (3)
        inverse="true|false"                     (4)
        optional="true|false">                   (5)

        <key ... />

        <property ... />
        ...
</join>

<class name="Person"
    table="PERSON">

    <id name="id" column="PERSON_ID">...</id>

    <join table="ADDRESS">
        <key column="ADDRESS_ID"/>
        <property name="address"/>
        <property name="zip"/>
        <property name="country"/>
    </join>
    ...

<import class="System.Object" rename="Universe"/>

<import
        class="ClassName"              (1)
        rename="ShortName"             (2)
/>

You may force NHibernate to quote an identifier in the generated SQL by enclosing the table or column name in back-ticks in the mapping document. NHibernate will use the correct quotation style for the SQL Dialect (usually double quotes, but brackets for SQL Server and back-ticks for MySQL).

<class name="LineItem" table="`Line Item`">
    <id name="Id" column="`Item Id`"/><generator class="assigned"/></id>
    <property name="ItemNumber" column="`Item #`"/>
    ...
</class>

Quoting column identifiers is required if a table contains two columns differing only by case. Ensure you use consistent casing when quoting identifiers.

It is possible to define subclass and joined-subclass mappings in separate mapping documents, directly beneath hibernate-mapping. This allows you to extend a class hierarchy just by adding a new mapping file. You must specify an extends attribute in the subclass mapping, naming a previously mapped superclass. Use of this feature makes the ordering of the mapping documents important!

<hibernate-mapping>
        <subclass name="Eg.Subclass.DomesticCat, Eg"
            extends="Eg.Cat, Eg" discriminator-value="D">
             <property name="name" type="string"/>
        </subclass>
</hibernate-mapping>

Generated properties are properties which have their values generated by the database. Typically, NHibernate applications needed to Refresh objects which contain any properties for which the database was generating values. Marking properties as generated, however, lets the application delegate this responsibility to NHibernate. Essentially, whenever NHibernate issues an SQL INSERT or UPDATE for an entity which has defined generated properties, it immediately issues a select afterwards to retrieve the generated values.

Properties marked as generated must additionally be non-insertable and non-updatable. Only Section 5.1.8, “version (optional)”, Section 5.1.9, “timestamp (optional)”, and Section 5.1.10, “property” can be marked as generated.

never (the default) - means that the given property value is not generated within the database.

insert - states that the given property value is generated on insert, but is not regenerated on subsequent updates. Things like created-date would fall into this category. Note that even though Section 5.1.8, “version (optional)” and Section 5.1.9, “timestamp (optional)” properties can be marked as generated, this option is not available there...

always - states that the property value is generated both on insert and on update.

Allows CREATE and DROP of arbitrary database objects, in conjunction with NHibernate's schema evolution tools, to provide the ability to fully define a user schema within the NHibernate mapping files. Although designed specifically for creating and dropping things like triggers or stored procedures, really any SQL command that can be run via a DbCommand.ExecuteNonQuery() method is valid here (ALTERs, INSERTS, etc). There are essentially two modes for defining auxiliary database objects.

The first mode is to explicitly list the CREATE and DROP commands out in the mapping file:

<nhibernate-mapping>
    ...
    <database-object>
        <create>CREATE TRIGGER my_trigger ...</create>
        <drop>DROP TRIGGER my_trigger</drop>
    </database-object>
</nhibernate-mapping>

The second mode is to supply a custom class which knows how to construct the CREATE and DROP commands. This custom class must implement the NHibernate.Mapping.IAuxiliaryDatabaseObject interface.

<hibernate-mapping>
    ...
    <database-object>
        <definition class="MyTriggerDefinition, MyAssembly"/>
    </database-object>
</hibernate-mapping>

You may also specify parameters to be passed to the database object:

<hibernate-mapping>
    ...
    <database-object>
        <definition class="MyTriggerDefinition, MyAssembly">
            <param name="parameterName">parameterValue</param>
        </definition>
    </database-object>
</hibernate-mapping>

NHibernate will call IAuxiliaryDatabaseObject.SetParameterValues passing it a dictionary of parameter names and values.

Additionally, these database objects can be optionally scoped such that they only apply when certain dialects are used.

<hibernate-mapping>
    ...
    <database-object>
        <definition class="MyTriggerDefinition"/>
        <dialect-scope name="NHibernate.Dialect.Oracle9iDialect"/>
        <dialect-scope name="NHibernate.Dialect.Oracle8iDialect"/>
    </database-object>
</hibernate-mapping>

NHibernate requires that persistent collection-valued fields be declared as a generic interface type, for example:

public class Product
{
    public ISet<Part> Parts { get; set; } = new HashSet<Part>();

    public string SerialNumber { get; set; }
}

The actual interface might be System.Collections.Generic.ICollection<T>, System.Collections.Generic.IList<T>, System.Collections.Generic.IDictionary<K, V>, System.Collections.Generic.ISet<T> or ... anything you like! (Where "anything you like" means you will have to write an implementation of NHibernate.UserType.IUserCollectionType.)

Notice how we initialized the instance variable with an instance of HashSet<T>. This is the best way to initialize collection valued properties of newly instantiated (non-persistent) instances. When you make the instance persistent - by calling Save(), for example - NHibernate will actually replace the HashSet<T> with an instance of NHibernate's own implementation of ISet<T>. Watch out for errors like this:

Cat cat = new DomesticCat();
Cat kitten = new DomesticCat();
....
ISet<Cat> kittens = new HashSet<Cat>();
kittens.Add(kitten);
cat.Kittens = kittens;
session.Save(cat);
kittens = cat.Kittens; //Okay, kittens collection is an ISet
HashSet<Cat> hs = (HashSet<Cat>) cat.Kittens; //Error!

Collection instances have the usual behavior of value types. They are automatically persisted when referenced by a persistent object and automatically deleted when unreferenced. If a collection is passed from one persistent object to another, its elements might be moved from one table to another. Two entities may not share a reference to the same collection instance. Due to the underlying relational model, collection-valued properties do not support null value semantics; NHibernate does not distinguish between a null collection reference and an empty collection.

You shouldn't have to worry much about any of this. Just use NHibernate's collections the same way you use ordinary .NET collections, but make sure you understand the semantics of bidirectional associations (discussed later) before using them.

Collection instances are distinguished in the database by a foreign key to the owning entity. This foreign key is referred to as the collection key . The collection key is mapped by the <key> element.

Collections may contain almost any other NHibernate type, including all basic types, custom types, entity types and components. This is an important definition: An object in a collection can either be handled with "pass by value" semantics (it therefore fully depends on the collection owner) or it can be a reference to another entity with an own lifecycle. Collections may not contain other collections. The contained type is referred to as the collection element type. Collection elements are mapped by <element>, <composite-element>, <one-to-many>, <many-to-many> or <many-to-any>. The first two map elements with value semantics, the other three are used to map entity associations.

All collection types except ISet and bag have an index column - a column that maps to an array or IList index or IDictionary key. The index of an IDictionary may be of any basic type, an entity type or even a composite type (it may not be a collection). The index of an array or list is always of type Int32. Indexes are mapped using <index>, <index-many-to-many>, <composite-index> or <index-many-to-any>.

There are quite a range of mappings that can be generated for collections, covering many common relational models. We suggest you experiment with the schema generation tool to get a feeling for how various mapping declarations translate to database tables.

Collections are declared by the <set>, <list>, <map>, <bag>, <array> and <primitive-array> elements. <map> is representative:

<map
    name="propertyName"                                         (1)
    table="table_name"                                          (2)
    schema="schema_name"                                        (3)
    lazy="true|false|extra"                                     (4)
    inverse="true|false"                                        (5)
    cascade="all|none|save-update|delete|all-delete-orphan"     (6)
    sort="unsorted|natural|comparatorClass"                     (7)
    order-by="column_name asc|desc"                             (8)
    where="arbitrary sql where condition"                       (9)
    fetch="select|join"                                         (10)
    batch-size="N"                                              (11)
    access="field|property|ClassName"                           (12)
    optimistic-lock="true|false"                                (13)
    generic="true|false"                                        (14)
>

    <key .... />
    <index .... />
    <element .... />
</map>

The mapping of an IList or array requires a separate table column holding the array or list index (the i in foo[i]). If your relational model doesn't have an index column, e.g. if you're working with legacy data, use an unordered ISet instead. This seems to put people off who assume that IList should just be a more convenient way of accessing an unordered collection. NHibernate collections strictly obey the actual semantics attached to the ISet, IList and IDictionary interfaces. IList elements don't just spontaneously rearrange themselves!

On the other hand, people who planned to use the IList to emulate bag semantics have a legitimate grievance here. A bag is an unordered, unindexed collection which may contain the same element multiple times. The .NET collections framework lacks an IBag interface, hence you have to emulate it with an IList. NHibernate lets you map properties of type IList or ICollection with the <bag> element. Note that bag semantics are not really part of the ICollection contract and they actually conflict with the semantics of the IList contract (however, you can sort the bag arbitrarily, discussed later in this chapter).

Note: Large NHibernate bags mapped with inverse="false" are inefficient and should be avoided; NHibernate can't create, delete or update rows individually, because there is no key that may be used to identify an individual row.

A collection table is required for any collection of values and any collection of references to other entities mapped as a many-to-many association (the natural semantics for a .NET collection). The table requires (foreign) key column(s), element column(s) and possibly index column(s).

The foreign key from the collection table to the table of the owning class is declared using a <key> element.

<key column="column_name"/>

For indexed collections like maps and lists, we require an <index> element. For lists, this column contains sequential integers numbered from zero. Make sure that your index really starts from zero if you have to deal with legacy data. For maps, the column may contain any values of any NHibernate type.

<index
        column="column_name"                (1)
        type="typename"                     (2)
/>

Alternatively, a map may be indexed by objects of entity type. We use the <index-many-to-many> element.

<index-many-to-many
        column="column_name"                (1)
        class="ClassName"                   (2)
/>

For a collection of values, we use the <element> tag.

<element
        column="column_name"                (1)
        type="typename"                     (2)
/>

A collection of entities with its own table corresponds to the relational notion of many-to-many association. A many to many association is the most natural mapping of a .NET collection but is not usually the best relational model.

<many-to-many
        column="column_name"                               (1)
        class="ClassName"                                  (2)
        fetch="join|select"                                (3)
        not-found="ignore|exception"                       (4)
    />

Some examples, first, a set of strings:

<set name="Names" table="NAMES">
    <key column="GROUPID"/>
    <element column="NAME" type="String"/>
</set>

A bag containing integers (with an iteration order determined by the order-by attribute):

<bag name="Sizes" table="SIZES" order-by="SIZE ASC">
    <key column="OWNER"/>
    <element column="SIZE" type="Int32"/>
</bag>

An array of entities - in this case, a many to many association (note that the entities are lifecycle objects, cascade="all"):

<array name="Foos" table="BAR_FOOS" cascade="all">
    <key column="BAR_ID"/>
    <index column="I"/>
    <many-to-many column="FOO_ID" class="Eg.Foo, Eg"/>
</array>

A map from string indices to dates:

<map name="Holidays" table="holidays" schema="dbo" order-by="hol_name asc">
    <key column="id"/>
    <index column="hol_name" type="String"/>
    <element column="hol_date" type="Date"/>
</map>

A list of components (discussed in the next chapter):

<list name="CarComponents" table="car_components">
    <key column="car_id"/>
    <index column="posn"/>
    <composite-element class="Eg.Car.CarComponent">
            <property name="Price" type="float"/>
            <property name="Type" type="Eg.Car.ComponentType, Eg"/>
            <property name="SerialNumber" column="serial_no" type="String"/>
    </composite-element>
</list>

A one to many association links the tables of two classes directly, with no intervening collection table. (This implements a one-to-many relational model.) This relational model loses some of the semantics of .NET collections:

An association from Foo to Bar requires the addition of a key column and possibly an index column to the table of the contained entity class, Bar. These columns are mapped using the <key> and <index> elements described above.

The <one-to-many> tag indicates a one to many association.

<one-to-many
        class="ClassName"                                  (1)
        not-found="ignore|exception"                       (2)
    />

Example:

<set name="Bars">
    <key column="foo_id"/>
    <one-to-many class="Eg.Bar, Eg"/>
</set>

Notice that the <one-to-many> element does not need to declare any columns. Nor is it necessary to specify the table name anywhere.

Very Important Note: If the <key> column of a <one-to-many> association is declared NOT NULL, NHibernate may cause constraint violations when it creates or updates the association. To prevent this problem, you must use a bidirectional association with the many valued end (the set or bag) marked as inverse="true". See the discussion of bidirectional associations later in this chapter.

Collections (other than arrays) may be lazily initialized, meaning they load their state from the database only when the application needs to access it. Initialization happens transparently to the user so the application would not normally need to worry about this (in fact, transparent lazy initialization is the main reason why NHibernate needs its own collection implementations). However, if the application tries something like this:

IDictionary<string, int> permissions;
using (s = sessions.OpenSession())
using (ITransaction tx = sessions.BeginTransaction())
{
    var u = s.Load<User>(userId);
    permissions = u.Permissions;
    tx.Commit();
}

int accessLevel = permissions["accounts"];  // Error!

It could be in for a nasty surprise. Since the permissions collection was not initialized when the ISession was committed, the collection will never be able to load its state. The fix is to move the line that reads from the collection to just before the commit. (There are other more advanced ways to solve this problem, however.)

Alternatively, use a non-lazy collection. Since lazy initialization can lead to bugs like that above, non-laziness is the default. However, it is intended that lazy initialization be used for almost all collections, especially for collections of entities (for reasons of efficiency).

Exceptions that occur while lazily initializing a collection are wrapped in a LazyInitializationException.

Declare a lazy collection using the optional lazy attribute:

<set name="Names" table="NAMES" lazy="true">
    <key column="group_id"/>
    <element column="NAME" type="String"/>
</set>

In some application architectures, particularly where the code that accesses data using NHibernate, and the code that uses it are in different application layers, it can be a problem to ensure that the ISession is open when a collection is initialized. There are two basic ways to deal with this issue:

You can use the CreateFilter() method of the NHibernate ISession API to get the size of a collection without initializing it:

var count = s
    .CreateFilter(collection, "select count(*)")
    .UniqueResult<long>();

CreateFilter() is also used to efficiently retrieve subsets of a collection without needing to initialize the whole collection.

NHibernate supports collections implemented by System.Collections.Generic.SortedList<T> and System.Collections.Generic.SortedSet<T>. You must specify a comparer in the mapping file:

<set name="Aliases" table="person_aliases" sort="natural">
    <key column="person"/>
    <element column="name" type="String"/>
</set>

<map name="Holidays" sort="My.Custom.HolidayComparer, MyAssembly" lazy="true">
    <key column="year_id"/>
    <index column="hol_name" type="String"/>
    <element column="hol_date" type="Date"/>
</map>

Allowed values of the sort attribute are unsorted, natural and the name of a class implementing System.Collections.Generic.IComparer<T>.

If you want the database itself to order the collection elements use the order-by attribute of set, bag or map mappings. This performs the ordering in the SQL query, not in memory.

Setting the order-by attribute tells NHibernate to use Iesi.Collections.Generic.LinkedHashSet class internally for sets, maintaining the order of the elements. It is not supported on maps.

<set name="Aliases" table="person_aliases" order-by="name asc">
    <key column="person"/>
    <element column="name" type="String"/>
</set>

<map name="Holidays" order-by="hol_date, hol_name" lazy="true">
    <key column="year_id"/>
    <index column="hol_name" type="String"/>
    <element column="hol_date type="Date"/>
</map>

Note that the value of the order-by attribute is an SQL ordering, not a HQL ordering!

Associations may even be sorted by some arbitrary criteria at runtime using a CreateFilter().

sortedUsers = s
    .CreateFilter(group.Users, "order by this.Name")
    .List<User>();

If you've fully embraced our view that composite keys are a bad thing and that entities should have synthetic identifiers (surrogate keys), then you might find it a bit odd that the many to many associations and collections of values that we've shown so far all map to tables with composite keys! Now, this point is quite arguable; a pure association table doesn't seem to benefit much from a surrogate key (though a collection of composite values might). Nevertheless, NHibernate provides a feature that allows you to map many to many associations and collections of values to a table with a surrogate key.

The <idbag> element lets you map a List (or Collection) with bag semantics.

<idbag name="Lovers" table="LOVERS" lazy="true">
    <collection-id column="ID" type="Int64">
        <generator class="hilo"/>
    </collection-id>
    <key column="PERSON1"/>
    <many-to-many column="PERSON2" class="Eg.Person" fetch="join"/>
</idbag>

As you can see, an <idbag> has a synthetic id generator, just like an entity class! A different surrogate key is assigned to each collection row. NHibernate does not provide any mechanism to discover the surrogate key value of a particular row, however.

Note that the update performance of an <idbag> is much better than a regular <bag>! NHibernate can locate individual rows efficiently and update or delete them individually, just like a list, map or set.

As of version 2.0, the native identifier generation strategy is supported for <idbag> collection identifiers.

A bidirectional association allows navigation from both "ends" of the association. Two kinds of bidirectional association are supported:

You may specify a bidirectional many-to-many association simply by mapping two many-to-many associations to the same database table and declaring one end as inverse (which one is your choice). Here's an example of a bidirectional many-to-many association from a class back to itself (each category can have many items and each item can be in many categories):

<class name="NHibernate.Auction.Category, NHibernate.Auction">
  <id name="Id" column="ID"/>
  ...
  <bag name="Items" table="CATEGORY_ITEM" lazy="true">
    <key column="CATEGORY_ID"/>
    <many-to-many class="NHibernate.Auction.Item, NHibernate.Auction" column="ITEM_ID"/>
  </bag>
</class>

<class name="NHibernate.Auction.Item, NHibernate.Auction">
  <id name="id" column="ID"/>
  ...

  <!-- inverse end -->
  <bag name="categories" table="CATEGORY_ITEM" inverse="true" lazy="true">
    <key column="ITEM_ID"/>
    <many-to-many class="NHibernate.Auction.Category, NHibernate.Auction"
        column="CATEGORY_ID"/>
  </bag>
</class>

Changes made only to the inverse end of the association are not persisted. This means that NHibernate has two representations in memory for every bidirectional association, one link from A to B and another link from B to A. This is easier to understand if you think about the .NET object model and how we create a many-to-many relationship in C#:

category.Items.Add(item);          // The category now "knows" about the relationship
item.Categories.Add(category);     // The item now "knows" about the relationship

session.Update(item);                     // No effect, nothing will be saved!
session.Update(category);                 // The relationship will be saved

The non-inverse side is used to save the in-memory representation to the database. We would get an unnecessary INSERT/UPDATE and probably even a foreign key violation if both would trigger changes! The same is of course also true for bidirectional one-to-many associations.

You may map a bidirectional one-to-many association by mapping a one-to-many association to the same table column(s) as a many-to-one association and declaring the many-valued end inverse="true".

<class name="Eg.Parent, Eg">
    <id name="Id" column="id"/>
    ....
    <set name="Children" inverse="true" lazy="true">
        <key column="parent_id"/>
        <one-to-many class="Eg.Child, Eg"/>
    </set>
</class>

<class name="Eg.Child, Eg">
    <id name="Id" column="id"/>
    ....
    <many-to-one name="Parent" class="Eg.Parent, Eg" column="parent_id"/>
</class>

Mapping one end of an association with inverse="true" doesn't affect the operation of cascades, both are different concepts!

There are some additional considerations for bidirectional mappings with indexed collections (where one end is represented as a <list> or <map>) when using NHibernate mapping files. If there is a property of the child class that maps to the index column you can use inverse="true" on the collection mapping:

<class name="Parent">
    <id name="Id" column="parent_id"/>
    ....
    <map name="Children" inverse="true">
        <key column="parent_id"/>
        <map-key column="name"
            type="string"/>
        <one-to-many class="Child"/>
    </map>
</class>

<class name="Child">
    <id name="Id" column="child_id"/>
    ....
    <property name="Name" column="name"
        not-null="true"/>
    <many-to-one name="Parent"
        class="Parent"
        column="parent_id"
        not-null="true"/>
</class>

If there is no such property on the child class, the association cannot be considered truly bidirectional. That is, there is information available at one end of the association that is not available at the other end. In this case, you cannot map the collection inverse="true". Instead, you could use the following mapping:

<class name="Parent">
    <id name="Id" column="parent_id"/>
    ....
    <map name="Children">
        <key column="parent_id"
            not-null="true"/>
        <map-key column="name"
            type="string"/>
        <one-to-many class="Child"/>
    </map>
</class>

<class name="Child">
    <id name="Id" column="child_id"/>
    ....
    <many-to-one name="Parent"
        class="Parent"
        column="parent_id"
        insert="false"
        update="false"
        not-null="true"/>
</class>

Note that in this mapping, the collection-valued end of the association is responsible for updates to the foreign key.

There are two possible approaches to mapping a ternary association. One approach is to use composite elements (discussed below). Another is to use an IDictionary with an association as its index:

<map name="Contracts" lazy="true">
    <key column="employer_id"/>
    <index-many-to-many column="employee_id" class="Employee"/>
    <one-to-many class="Contract"/>
</map>

<map name="Connections" lazy="true">
    <key column="node1_id"/>
    <index-many-to-many column="node2_id" class="Node"/>
    <many-to-many column="connection_id" class="Connection"/>
</map>

The <many-to-any> and <index-many-to-any> elements provide for true heterogeneous associations. These mapping elements work in the same way as the <any> element - and should also be used rarely, if ever.

The previous sections are pretty confusing. So lets look at an example. This class:

using System;
using System.Collections.Generic;

namespace Eg

    public class Parent
    {
        public long Id { get; set; }

        private ISet<Child> Children { get; set; }

        ....
        ....
    }
}

has a collection of Eg.Child instances. If each child has at most one parent, the most natural mapping is a one-to-many association:

<hibernate-mapping xmlns="urn:nhibernate-mapping-2.2"
    assembly="Eg" namespace="Eg">

    <class name="Parent">
        <id name="Id">
            <generator class="sequence"/>
        </id>
        <set name="Children" lazy="true">
            <key column="parent_id"/>
            <one-to-many class="Child"/>
        </set>
    </class>

    <class name="Child">
        <id name="Id">
            <generator class="sequence"/>
        </id>
        <property name="Name"/>
    </class>

</hibernate-mapping>

This maps to the following table definitions:

create table parent (Id bigint not null primary key)
create table child (Id bigint not null primary key, Name varchar(255), parent_id bigint)
alter table child add constraint childfk0 (parent_id) references parent

If the parent is required, use a bidirectional one-to-many association:

<hibernate-mapping xmlns="urn:nhibernate-mapping-2.2"
    assembly="Eg" namespace="Eg">

    <class name="Parent">
        <id name="Id">
            <generator class="sequence"/>
        </id>
        <set name="Children" inverse="true" lazy="true">
            <key column="parent_id"/>
            <one-to-many class="Child"/>
        </set>
    </class>

    <class name="Child">
        <id name="Id">
            <generator class="sequence"/>
        </id>
        <property name="Name"/>
        <many-to-one name="parent" class="Parent" column="parent_id" not-null="true"/>
    </class>

</hibernate-mapping>

Notice the NOT NULL constraint:

create table parent ( Id bigint not null primary key )
create table child ( Id bigint not null
                     primary key,
                     Name varchar(255),
                     parent_id bigint not null )
alter table child add constraint childfk0 (parent_id) references parent

On the other hand, if a child might have multiple parents, a many-to-many association is appropriate:

<hibernate-mapping xmlns="urn:nhibernate-mapping-2.2"
    assembly="Eg" namespace="Eg">

    <class name="Parent">
        <id name="Id">
            <generator class="sequence"/>
        </id>
        <set name="Children" lazy="true" table="childset">
            <key column="parent_id"/>
            <many-to-many class="Child" column="child_id"/>
        </set>
    </class>

    <class name="eg.Child">
        <id name="Id">
            <generator class="sequence"/>
        </id>
        <property name="Name"/>
    </class>

</hibernate-mapping>

Table definitions:

create table parent ( Id bigint not null primary key )
create table child ( Id bigint not null primary key, name varchar(255) )
create table childset ( parent_id bigint not null,
                        child_id bigint not null,
                        primary key ( parent_id, child_id ) )
alter table childset add constraint childsetfk0 (parent_id) references parent
alter table childset add constraint childsetfk1 (child_id) references child

See also Chapter 22, Example: Parent/Child.

A component is a contained object that is persisted as a value type, not an entity. The term "component" refers to the object-oriented notion of composition (not to architecture-level components). For example, you might model a person like this:

public class Person
{
    public string Key { get; set; }

    public DateTime Birthday { get; set; }

    public Name Name { get; set; }

    ......
    ......
}

public class Name
{
    public string First { get; set; }

    public string Last { get; set; }

    public char Initial { get; set; }
}

Now Name may be persisted as a component of Person. Notice that Name defines getter and setter methods for its persistent properties, but doesn't need to declare any interfaces or identifier properties.

Our NHibernate mapping would look like:

<class name="Eg.Person, Eg" table="person">
    <id name="Key" column="pid" type="string">
        <generator class="uuid.hex"/>
    </id>
    <property name="Birthday" type="date"/>
    <component name="Name" class="Eg.Name, Eg"> <!-- class attribute optional -->
        <property name="Initial"/>
        <property name="First"/>
        <property name="Last"/>
    </component>
</class>

The person table would have the columns pid, Birthday, Initial, First and Last.

Like all value types, components do not support shared references. The null value semantics of a component are ad hoc. When reloading the containing object, NHibernate will assume that if all component columns are null, then the entire component is null. This should be okay for most purposes.

The properties of a component may be of any NHibernate type (collections, many-to-one associations, other components, etc). Nested components should not be considered an exotic usage. NHibernate is intended to support a very fine-grained object model.

The <component> element allows a <parent> sub-element that maps a property of the component class as a reference back to the containing entity.

<class name="Eg.Person, Eg" table="person">
    <id name="Key" column="pid" type="string">
        <generator class="uuid.hex"/>
    </id>
    <property name="Birthday" type="date"/>
    <component name="Name" class="Eg.Name, Eg">
        <parent name="NamedPerson"/> <!-- reference back to the Person -->
        <property name="Initial"/>
        <property name="First"/>
        <property name="Last"/>
    </component>
</class>

Collections of components are supported (eg. an array of type Name). Declare your component collection by replacing the <element> tag with a <composite-element> tag.

<set name="SomeNames" table="some_names" lazy="true">
    <key column="id"/>
    <composite-element class="Eg.Name, Eg"> <!-- class attribute required -->
        <property name="Initial"/>
        <property name="First"/>
        <property name="Last"/>
    </composite-element>
</set>

Note: if you define an ISet of composite elements, it is very important to implement Equals() and GetHashCode() correctly.

Composite elements may contain components but not collections. If your composite element itself contains components, use the <nested-composite-element> tag. This is a pretty exotic case - a collection of components which themselves have components. By this stage you should be asking yourself if a one-to-many association is more appropriate. Try remodelling the composite element as an entity - but note that even though the object model is the same, the relational model and persistence semantics are still slightly different.

Please note that a composite element mapping doesn't support null-able properties if you're using a <set>. NHibernate has to use each columns value to identify a record when deleting objects (there is no separate primary key column in the composite element table), which is not possible with null values. You have to either use only not-null properties in a composite-element or choose a <list>, <map>, <bag> or <idbag>.

A special case of a composite element is a composite element with a nested <many-to-one> element. A mapping like this allows you to map extra columns of a many-to-many association table to the composite element class. The following is a many-to-many association from Order to Item where PurchaseDate, Price and Quantity are properties of the association:

<class name="Order" .... >
    ....
    <set name="PurchasedItems" table="purchase_items" lazy="true">
        <key column="order_id">
        <composite-element class="Purchase">
            <property name="PurchaseDate"/>
            <property name="Price"/>
            <property name="Quantity"/>
            <many-to-one name="Item" class="Item"/> <!-- class attribute is optional -->
        </composite-element>
    </set>
</class>

Even ternary (or quaternary, etc) associations are possible:

<class name="Order" .... >
    ....
    <set name="PurchasedItems" table="purchase_items" lazy="true">
        <key column="order_id">
        <composite-element class="OrderLine">
            <many-to-one name="PurchaseDetails class="Purchase"/>
            <many-to-one name="Item" class="Item"/>
        </composite-element>
    </set>
</class>

Composite elements may appear in queries using the same syntax as associations to other entities.

The <composite-index> element lets you map a component class as the key of an IDictionary. Make sure you override GetHashCode() and Equals() correctly on the component class.

You may use a component as an identifier of an entity class. Your component class must satisfy certain requirements:

You can't use an IIdentifierGenerator to generate composite keys. Instead the application must assign its own identifiers.

Since a composite identifier must be assigned to the object before saving it, we can't use unsaved-value of the identifier to distinguish between newly instantiated instances and instances saved in a previous session.

You may instead implement IInterceptor.IsTransient() if you wish to use SaveOrUpdate() or cascading save / update. As an alternative, you may also set the unsaved-value attribute on a <version> (or <timestamp>) element to specify a value that indicates a new transient instance. In this case, the version of the entity is used instead of the (assigned) identifier and you don't have to implement IInterceptor.IsTransient() yourself.

Use the <composite-id> tag (same attributes and elements as <component>) in place of <id> for the declaration of a composite identifier class:

<class name="Foo" table="FOOS">
    <composite-id name="CompId" class="FooCompositeID">
        <key-property name="String"/>
        <key-property name="Short"/>
        <key-property name="Date" column="date_" type="Date"/>
    </composite-id>
    <property name="Name"/>
    ....
</class>

Now, any foreign keys into the table FOOS are also composite. You must declare this in your mappings for other classes. An association to Foo would be declared like this:

<many-to-one name="Foo" class="Foo">
<!-- the "class" attribute is optional, as usual -->
    <column name="foo_string"/>
    <column name="foo_short"/>
    <column name="foo_date"/>
</many-to-one>

This new <column> tag is also used by multi-column custom types. Actually it is an alternative to the column attribute everywhere. A collection with elements of type Foo would use:

<set name="Foos">
    <key column="owner_id"/>
    <many-to-many class="Foo">
        <column name="foo_string"/>
        <column name="foo_short"/>
        <column name="foo_date"/>
    </many-to-many>
</set>

On the other hand, <one-to-many>, as usual, declares no columns.

If Foo itself contains collections, they will also need a composite foreign key.

<class name="Foo">
    ....
    ....
    <set name="Dates" lazy="true">
        <key>   <!-- a collection inherits the composite key type -->
            <column name="foo_string"/>
            <column name="foo_short"/>
            <column name="foo_date"/>
        </key>
        <element column="foo_date" type="Date"/>
    </set>
</class>

You may even map a property of type IDictionary:

<dynamic-component name="UserAttributes">
    <property name="Foo" column="FOO"/>
    <property name="Bar" column="BAR"/>
    <many-to-one name="Baz" class="Baz" column="BAZ"/>
</dynamic-component>

The semantics of a <dynamic-component> mapping are identical to <component>. The advantage of this kind of mapping is the ability to determine the actual properties of the component at deployment time, just by editing the mapping document. (Runtime manipulation of the mapping document is also possible, using a DOM parser.)

NHibernate supports the three basic inheritance mapping strategies.

In addition, NHibernate supports a fourth, slightly different kind of polymorphism:

It is possible to use different mapping strategies for different branches of the same inheritance hierarchy, and then make use of implicit polymorphism to achieve polymorphism across the whole hierarchy. However, NHibernate does not support mixing <subclass>, and <joined-subclass> and <union-subclass> mappings under the same root <class> element. It is possible to mix together the table per hierarchy and table per subclass strategies, under the the same <class> element, by combining the <subclass> and <join> elements (see below).

It is possible to define subclass, union-subclass, and joined-subclass mappings in separate mapping documents, directly beneath hibernate-mapping. This allows you to extend a class hierarchy just by adding a new mapping file. You must specify an extends attribute in the subclass mapping, naming a previously mapped superclass.

 <hibernate-mapping>
     <subclass name="DomesticCat" extends="Cat" discriminator-value="D">
          <property name="name" type="string"/>
     </subclass>
 </hibernate-mapping>

<class name="IPayment" table="PAYMENT">
    <id name="Id" type="Int64" column="PAYMENT_ID">
        <generator class="native"/>
    </id>
    <discriminator column="PAYMENT_TYPE" type="String"/>
    <property name="Amount" column="AMOUNT"/>
    ...
    <subclass name="CreditCardPayment" discriminator-value="CREDIT">
        ...
    </subclass>
    <subclass name="CashPayment" discriminator-value="CASH">
        ...
    </subclass>
    <subclass name="ChequePayment" discriminator-value="CHEQUE">
        ...
    </subclass>
</class>

<class name="IPayment" table="PAYMENT">
    <id name="Id" type="Int64" column="PAYMENT_ID">
        <generator class="native"/>
    </id>
    <property name="Amount" column="AMOUNT"/>
    ...
    <joined-subclass name="CreditCardPayment" table="CREDIT_PAYMENT">
        <key column="PAYMENT_ID"/>
        ...
    </joined-subclass>
    <joined-subclass name="CashPayment" table="CASH_PAYMENT">
        <key column="PAYMENT_ID"/>
        ...
    </joined-subclass>
    <joined-subclass name="ChequePayment" table="CHEQUE_PAYMENT">
        <key column="PAYMENT_ID"/>
        ...
    </joined-subclass>
</class>

<class name="Payment" table="PAYMENT">
    <id name="Id" type="Int64" column="PAYMENT_ID">
        <generator class="native"/>
    </id>
    <discriminator column="PAYMENT_TYPE" type="string"/>
    <property name="Amount" column="AMOUNT"/>
    ...
    <subclass name="CreditCardPayment" discriminator-value="CREDIT">
        <join table="CREDIT_PAYMENT">
            <key column="PAYMENT_ID"/>
            <property name="CreditCardType" column="CCTYPE"/>
            ...
        </join>
    </subclass>
    <subclass name="CashPayment" discriminator-value="CASH">
        <join table="CASH_PAYMENT">
            <key column="PAYMENT_ID"/>
            ...
        </join>
    </subclass>
    <subclass name="ChequePayment" discriminator-value="CHEQUE">
        <join table="CHEQUE_PAYMENT" fetch="select">
            <key column="PAYMENT_ID"/>
            ...
        </join>
    </subclass>
</class>

<class name="Payment" table="PAYMENT">
    <id name="Id" type="Int64" column="PAYMENT_ID">
        <generator class="native"/>
    </id>
    <discriminator column="PAYMENT_TYPE" type="string"/>
    <property name="Amount" column="AMOUNT"/>
    ...
    <subclass name="CreditCardPayment" discriminator-value="CREDIT">
        <join table="CREDIT_PAYMENT">
            <property name="CreditCardType" column="CCTYPE"/>
            ...
        </join>
    </subclass>
    <subclass name="CashPayment" discriminator-value="CASH">
        ...
    </subclass>
    <subclass name="ChequePayment" discriminator-value="CHEQUE">
        ...
    </subclass>
</class>

<many-to-one name="Payment" column="PAYMENT" class="IPayment"/>

<class name="Payment">
    <id name="Id" type="Int64" column="PAYMENT_ID">
        <generator class="sequence"/>
    </id>
    <property name="Amount" column="AMOUNT"/>
    ...
    <union-subclass name="CreditCardPayment" table="CREDIT_PAYMENT">
        <property name="CreditCardType" column="CCTYPE"/>
        ...
    </union-subclass>
    <union-subclass name="CashPayment" table="CASH_PAYMENT">
        ...
    </union-subclass>
    <union-subclass name="ChequePayment" table="CHEQUE_PAYMENT">
        ...
    </union-subclass>
</class>

<class name="CreditCardPayment" table="CREDIT_PAYMENT">
    <id name="Id" type="Int64" column="CREDIT_PAYMENT_ID">
        <generator class="native"/>
    </id>
    <property name="Amount" column="CREDIT_AMOUNT"/>
    ...
</class>

<class name="CashPayment" table="CASH_PAYMENT">
    <id name="Id" type="Int64" column="CASH_PAYMENT_ID">
        <generator class="native"/>
    </id>
    <property name="Amount" column="CASH_AMOUNT"/>
    ...
</class>

<class name="ChequePayment" table="CHEQUE_PAYMENT">
    <id name="Id" type="Int64" column="CHEQUE_PAYMENT_ID">
        <generator class="native"/>
    </id>
    <property name="Amount" column="CHEQUE_AMOUNT"/>
    ...
</class>

<any name="Payment" meta-type="string" id-type="Int64">
    <meta-value value="CREDIT" class="CreditCardPayment"/>
    <meta-value value="CASH" class="CashPayment"/>
    <meta-value value="CHEQUE" class="ChequePayment"/>
    <column name="PAYMENT_CLASS"/>
    <column name="PAYMENT_ID"/>
</any>

<class name="CreditCardPayment" table="CREDIT_PAYMENT">
    <id name="Id" type="Int64" column="CREDIT_PAYMENT_ID">
        <generator class="native"/>
    </id>
    <discriminator column="CREDIT_CARD" type="String"/>
    <property name="Amount" column="CREDIT_AMOUNT"/>
    ...
    <subclass name="MasterCardPayment" discriminator-value="MDC"/>
    <subclass name="VisaPayment" discriminator-value="VISA"/>
</class>

<class name="NonelectronicTransaction" table="NONELECTRONIC_TXN">
    <id name="Id" type="Int64" column="TXN_ID">
        <generator class="native"/>
    </id>
    ...
    <joined-subclass name="CashPayment" table="CASH_PAYMENT">
        <key column="PAYMENT_ID"/>
        <property name="Amount" column="CASH_AMOUNT"/>
        ...
    </joined-subclass>
    <joined-subclass name="ChequePayment" table="CHEQUE_PAYMENT">
        <key column="PAYMENT_ID"/>
        <property name="Amount" column="CHEQUE_AMOUNT"/>
        ...
    </joined-subclass>
</class>

There are certain limitations to the "implicit polymorphism" approach to the table per concrete-class mapping strategy. There are somewhat less restrictive limitations to <union-subclass> mappings.

The following table shows the limitations of table per concrete-class mappings, and of implicit polymorphism, in NHibernate.

An object (entity instance) is either transient or persistent with respect to a particular ISession. Newly instantiated objects are, of course, transient. The session offers services for saving (ie. persisting) transient instances:

DomesticCat fritz = new DomesticCat();
fritz.Color = Color.Ginger;
fritz.Sex = 'M';
fritz.Name = "Fritz";
long generatedId = (long) sess.Save(fritz);

DomesticCat pk = new DomesticCat();
pk.Color = Color.Tabby;
pk.Sex = 'F';
pk.Name = "PK";
pk.Kittens = new HashSet<Cat>();
pk.AddKitten(fritz);
sess.Save( pk, 1234L );

The single-argument Save() generates and assigns a unique identifier to fritz. The two-argument form attempts to persist pk using the given identifier. We generally discourage the use of the two-argument form since it may be used to create primary keys with business meaning.

Associated objects may be made persistent in any order you like unless you have a NOT NULL constraint upon a foreign key column. There is never a risk of violating foreign key constraints. However, you might violate a NOT NULL constraint if you Save() the objects in the wrong order.

The Load() methods of ISession give you a way to retrieve a persistent instance if you already know its identifier. One version takes a class object and will load the state into a newly instantiated object. The second version allows you to supply an instance into which the state will be loaded. The form which takes an instance is only useful in special circumstances (DIY instance pooling etc.)

Cat fritz = sess.Load<Cat>(generatedId);

long pkId = 1234;
DomesticCat pk = sess.Load<DomesticCat>(pkId);

Cat cat = new DomesticCat();
// load pk's state into cat
sess.Load( cat, pkId );
var kittens = cat.Kittens;

Note that Load() will throw an unrecoverable exception if there is no matching database row. If the class is mapped with a proxy, Load() returns an object that is an uninitialized proxy and does not actually hit the database until you invoke a method of the object. This behaviour is very useful if you wish to create an association to an object without actually loading it from the database.

If you are not certain that a matching row exists, you should use the Get() method, which hits the database immediately and returns null if there is no matching row.

Cat cat = sess.Get<Cat>(id);
if (cat==null) {
    cat = new Cat();
    sess.Save(cat, id);
}
return cat;

You may also load an objects using an SQL SELECT ... FOR UPDATE. See the next section for a discussion of NHibernate LockModes.

Cat cat = sess.Get<Cat>(id, LockMode.Upgrade);

Note that any associated instances or contained collections are not selected FOR UPDATE.

It is possible to re-load an object and all its collections at any time, using the Refresh() method. This is useful when database triggers are used to initialize some of the properties of the object.

sess.Save(cat);
sess.Flush(); //force the SQL INSERT
sess.Refresh(cat); //re-read the state (after the trigger executes)

An important question usually appears at this point: How much does NHibernate load from the database and how many SQL SELECTs will it use? This depends on the fetching strategy and is explained in Section 20.1, “Fetching strategies”.

If you don't know the identifier(s) of the object(s) you are looking for, use the CreateQuery() method of ISession. NHibernate supports a simple but powerful object oriented query language.

IList<Cat> cats = sess
    .CreateQuery("from Cat as cat where cat.Birthdate = ?")
    .SetDateTime(0, date)
    .List<Cat>();

var mates = sess
    .CreateQuery("select mate from Cat as cat join cat.Mate as mate " +
        "where cat.name = ?")
    .SetString(0, name)
    .List<Cat>();

var cats = sess
    .CreateQuery("from Cat as cat where cat.Mate.Birthdate is null")
    .List<Cat>();

var moreCats = sess
    .CreateQuery("from Cat as cat where " +
        "cat.Name = 'Fritz' or cat.id = ? or cat.id = ?")
    .SetInt64(0, id1)
    .SetParameter(1, id2, NHibernateUtil.Int64)
    .List<Cat>();

var mates = sess
    .CreateQuery("from Cat as cat where cat.Mate = ?")
    .SetEntity(0, izi)
    .List<Cat>();
);

var problems = sess
    .CreateQuery("from GoldFish as fish " +
        "where fish.Birthday > fish.Deceased or fish.Birthday is null")
    .List<GoldFish>();

These given Set parameters are used to bind the given values to the ? query placeholders (which map to input parameters of an ADO.NET DbCommand). Just as in ADO.NET, you should use this binding mechanism in preference to string manipulation.

The NHibernateUtil class defines a number of static methods and constants, providing access to most of the built-in types, as instances of NHibernate.Type.IType.

If you expect your query to return a very large number of objects, but you don't expect to use them all, you might get better performance from the Enumerable() method, which return a IEnumerable. The iterator will load objects on demand, using the identifiers returned by an initial SQL query (n+1 selects total).

// fetch ids
IEnumerable<Qux> en = sess
    .CreateQuery("from eg.Qux q order by q.Likeliness")
    .Enumerable<Qux>();
foreach (Qux qux in en)
{
    // something we couldnt express in the query
    if ( qux.CalculateComplicatedAlgorithm() ) {
        // dont need to process the rest
        break;
    }
}

The Enumerable() method also performs better if you expect that many of the objects are already loaded and cached by the session, or if the query results contain the same objects many times. (When no data is cached or repeated, CreateQuery() is almost always faster.) Here is an example of a query that should be called using Enumerable():

var en = sess
    .CreateQuery(
        "select customer, product " +
        "from Customer customer, " +
        "Product product " +
        "join customer.Purchases purchase " +
        "where product = purchase.Product")
    .Enumerable<object[]>();

Calling the previous query using CreateQuery() would return a very large ADO.NET result set containing the same data many times.

NHibernate queries sometimes return tuples of objects, in which case each tuple is returned as an array:

var foosAndBars = sess
    .CreateQuery(
        "select foo, bar from Foo foo, Bar bar " +
        "where bar.Date = foo.Date")
    .Enumerable<object[]>();
foreach (object[] tuple in foosAndBars)
{
    Foo foo = tuple[0]; Bar bar = tuple[1];
    ....
}

var results = sess
    .CreateQuery(
        "select cat.Color, min(cat.Birthdate), count(cat) from Cat cat " +
        "group by cat.Color")
    .Enumerable<object[]>();
foreach (object[] row in results)
{
    Color type = (Color) row[0];
    DateTime oldest = (DateTime) row[1];
    int count = (int) row[2];
    .....
}

var en = sess
    .CreateQuery(
        "select cat.Type, cat.Birthdate, cat.Name from DomesticCat cat")
    .Enumerable<object[]>();

IList<object[]> list = sess
    .CreateQuery("select cat, cat.Mate.Name from DomesticCat cat")
    .List<object[]>();

IQuery q = sess.CreateQuery("from DomesticCat cat");
q.SetFirstResult(20);
q.SetMaxResults(10);
var cats = q.List<Cat>();

<query name="Eg.DomesticCat.by.name.and.minimum.weight"><![CDATA[
    from Eg.DomesticCat as cat
        where cat.Name = ?
        and cat.Weight > ?
] ]></query>

IQuery q = sess.GetNamedQuery("Eg.DomesticCat.by.name.and.minimum.weight");
q.SetString(0, name);
q.SetInt32(1, minWeight);
var cats = q.List<Cat>();

//named parameter (preferred)
IQuery q = sess.CreateQuery("from DomesticCat cat where cat.Name = :name");
q.SetString("name", "Fritz");
var cats = q.Enumerable<DomesticCat>();

//positional parameter
IQuery q = sess.CreateQuery("from DomesticCat cat where cat.Name = ?");
q.SetString(0, "Izi");
var cats = q.Enumerable<DomesticCat>();

//named parameter list
var names = new List<string>();
names.Add("Izi");
names.Add("Fritz");
IQuery q = sess.CreateQuery("from DomesticCat cat where cat.Name in (:namesList)");
q.SetParameterList("namesList", names);
var cats = q.List<DomesticCat>();

var blackKittens = session
    .CreateFilter(pk.Kittens, "where this.Color = ?")
    .SetEnum(0, Color.Black)
    .List<Cat>();

var blackKittenMates = session
    .CreateFilter(pk.Kittens,
        "select this.Mate where this.Color = Eg.Color.Black")
    .List<Cat>();

ICriteria crit = session.CreateCriteria<Cat>();
crit.Add(Expression.Eq("color", Eg.Color.Black));
crit.SetMaxResults(10);
var cats = crit.List<Cat>();

var cats = session
    .CreateSQLQuery("SELECT {cat.*} FROM CAT {cat} WHERE ROWNUM<10")
    .AddEntity("cat", typeof(Cat))
    .List<Cat>();

var cats = session
    .CreateSQLQuery(
        "SELECT {cat}.ID AS {cat.Id}, {cat}.SEX AS {cat.Sex}, " +
           "{cat}.MATE AS {cat.Mate}, {cat}.SUBCLASS AS {cat.class}, ... " +
        "FROM CAT {cat} WHERE ROWNUM<10")
    .AddEntity("cat", typeof(Cat))
    .List<Cat>()

ISession.Delete() will remove an object's state from the database. Of course, your application might still hold a reference to it. So it's best to think of Delete() as making a persistent instance transient.

sess.Delete(cat);

You may also delete many objects at once by passing a NHibernate query string to Delete().

sess.Delete("from Cat");

You may now delete objects in any order you like, without risk of foreign key constraint violations. Of course, it is still possible to violate a NOT NULL constraint on a foreign key column by deleting objects in the wrong order.

From time to time the ISession will execute the SQL statements needed to synchronize the ADO.NET connection's state with the state of objects held in memory. This process, flush, occurs by default at the following points

The SQL statements are issued in the following order

(An exception is that objects using native ID generation are inserted when they are saved.)

Except when you explicitly Flush(), there are absolutely no guarantees about when the Session executes the ADO.NET calls, only the order in which they are executed. However, NHibernate does guarantee that the queries methods will never return stale data; nor will they return the wrong data.

It is possible to change the default behavior so that flush occurs less frequently. The FlushMode class defines three different modes: only flush at commit time (and only when the NHibernate ITransaction API is used, or inside a transaction scope), flush automatically using the explained routine (will only work inside an explicit NHibernate ITransaction or inside a transaction scope), or never flush unless Flush() is called explicitly. The last mode is useful for long running units of work, where an ISession is kept open and disconnected for a long time (see Section 11.4, “Optimistic concurrency control”).

sess = sf.OpenSession();
using (ITransaction tx = sess.BeginTransaction())
{
    // allow queries to return stale state
    sess.FlushMode = FlushMode.Commit;
    Cat izi = sess.Load<Cat>(id);
    izi.Name = "iznizi";
    // execute some queries....
    sess.CreateQuery("from Cat as cat left outer join cat.Kittens kitten")
        .List<object[]>();
    // change to izi is not flushed!
    ...
    tx.Commit(); // flush occurs
}

Ending a session involves four distinct phases:

tx.Commit(); // flush the session and commit the transaction

sess.Flush();
currentTransaction.Commit();

tx.Rollback();  // rollback the transaction

currentTransaction.Rollback();

tx.Commit();
sess.Close();

sess.Flush();
currentTransaction.Commit();
sess.Close();

NHibernate use might lead to exceptions, usually HibernateException. This exception can have a nested inner exception (the root cause), use the InnerException property to access it.

If the ISession throws an exception you should immediately rollback the transaction, call ISession.Close() and discard the ISession instance. Certain methods of ISession will not leave the session in a consistent state.

For exceptions thrown by the data provider while interacting with the database, NHibernate will wrap the error in an instance of ADOException. The underlying exception is accessible by calling ADOException.InnerException. NHibernate converts the DbException into an appropriate ADOException subclass using the ISQLExceptionConverter attached to the SessionFactory. By default, the ISQLExceptionConverter is defined by the configured dialect; however, it is also possible to plug in a custom implementation (see the api-docs for the ISQLExceptionConverter class for details).

The following exception handling idiom shows the typical case in NHibernate applications:

using (ISession sess = factory.OpenSession())
using (ITransaction tx = sess.BeginTransaction())
{
    // do some work
    ...
    tx.Commit();
}

Or, when manually managing ADO.NET transactions:

ISession sess = factory.openSession();
try
{
    // do some work
    ...
    sess.Flush();
    currentTransaction.Commit();
}
catch (Exception e)
{
    currentTransaction.Rollback();
    throw;
}
finally
{
    sess.Close();
}

To save or update all objects in a graph of associated objects, you must either

Likewise, to delete all objects in a graph, either

Recommendation:

Mapping an association (many-to-one, one-to-one or collection) with cascade="all" marks the association as a parent/child style relationship where save/update/deletion of the parent results in save/update/deletion of the child(ren). Furthermore, a mere reference to a child from a persistent parent will result in save / update of the child. The metaphor is incomplete, however. A child which becomes unreferenced by its parent is not automatically deleted, except in the cases of <one-to-many> and <one-to-one> associations that have been mapped with cascade="all-delete-orphan" or cascade="delete-orphan". The precise semantics of cascading operations are as follows:

NHibernate does not fully implement "persistence by reachability", which would imply (inefficient) persistent garbage collection. However, due to popular demand, NHibernate does support the notion of entities becoming persistent when referenced by another persistent object. Associations marked cascade="save-update" behave in this way. If you wish to use this approach throughout your application, it's easier to specify the default-cascade attribute of the <hibernate-mapping> element.

The IInterceptor interface provides callbacks from the session to the application allowing the application to inspect and / or manipulate properties of a persistent object before it is saved, updated, deleted or loaded. One possible use for this is to track auditing information. For example, the following IInterceptor automatically sets the CreateTimestamp when an IAuditable is created and updates the LastUpdateTimestamp property when an IAuditable is updated.

using System;
using NHibernate.Type;

namespace NHibernate.Test
{
    [Serializable]
    public class AuditInterceptor : IInterceptor
    {
    
        private int updates;
        private int creates;
    
        public void OnDelete(object entity,
                             object id,
                             object[] state,
                             string[] propertyNames,
                             IType[] types)
        {
            // do nothing
        }
    
        public boolean OnFlushDirty(object entity, 
                                    object id, 
                                    object[] currentState,
                                    object[] previousState,
                                    string[] propertyNames,
                                    IType[] types) {
    
            if ( entity is IAuditable )
            {
                updates++;
                for ( int i=0; i < propertyNames.Length; i++ )
                {
                    if ( "LastUpdateTimestamp" == propertyNames[i] )
                    {
                        currentState[i] = DateTime.Now;
                        return true;
                    }
                }
            }
            return false;
        }
    
        public boolean OnLoad(object entity, 
                              object id,
                              object[] state,
                              string[] propertyNames,
                              IType[] types)
        {
            return false;
        }
    
        public boolean OnSave(object entity,
                              object id,
                              object[] state,
                              string[] propertyNames,
                              IType[] types)
        {
            if ( entity is IAuditable )
            {
                creates++;
                for ( int i=0; i<propertyNames.Length; i++ )
                {
                    if ( "CreateTimestamp" == propertyNames[i] )
                    {
                        state[i] = DateTime.Now;
                        return true;
                    }
                }
            }
            return false;
        }
    
        public void PostFlush(ICollection entities)
        {
            Console.Out.WriteLine("Creations: {0}, Updates: {1}", creates, updates);
        }
    
        public void PreFlush(ICollection entities) {
            updates=0;
            creates=0;
        }
        
        ......
        ......
    }
}

The interceptor would be specified when a session is created.

ISession session = sf.OpenSession( new AuditInterceptor() );

You may also set an interceptor on a global level, using the Configuration:

new Configuration().SetInterceptor( new AuditInterceptor() );

NHibernate requires a very rich meta-level model of all entity and value types. From time to time, this model is very useful to the application itself. For example, the application might use NHibernate's metadata to implement a "smart" deep-copy algorithm that understands which objects should be copied (eg. mutable value types) and which should not (eg. immutable value types and, possibly, associated entities).

NHibernate exposes metadata via the IClassMetadata and ICollectionMetadata interfaces and the IType hierarchy. Instances of the metadata interfaces may be obtained from the ISessionFactory.

Cat fritz = ......;
IClassMetadata catMeta = sessionfactory.GetClassMetadata(typeof(Cat));
long id = (long) catMeta.GetIdentifier(fritz);
object[] propertyValues = catMeta.GetPropertyValues(fritz);
string[] propertyNames = catMeta.PropertyNames;
IType[] propertyTypes = catMeta.PropertyTypes;

// get an dictionary of all properties which are not collections or associations
// TODO: what about components?

var namedValues = new Dictionary<string, object>();
for (int i = 0; i < propertyNames.Length; i++)
{
    if (!propertyTypes[i].IsEntityType && !propertyTypes[i].IsCollectionType)
    {
        namedValues[propertyNames[i]] = propertyValues[i];
    }
}

An ISessionFactory is an expensive-to-create, threadsafe object intended to be shared by all application threads. An ISession is an inexpensive, non-threadsafe object that should be used once, for a single business process, and then discarded. For example, when using NHibernate in an ASP.NET application, pages could obtain an ISessionFactory using:

ISessionFactory sf = Global.SessionFactory;

Each call to a service method could create a new ISession, Flush() it, Commit() its transaction, Close() it and finally discard it. (The ISessionFactory may also be kept in a static Singleton helper variable.)

We use the NHibernate ITransaction API as discussed previously, a single Commit() of a NHibernate ITransaction flushes the state and commits any underlying database connection. Transaction scopes can be used instead, which allows the transaction to be distributed. See Section 11.8, “Transaction scopes (System.Transactions)”.

NHibernate may be used without managing transactions neither with the ITransaction API nor with transaction scopes. We advise against such a pattern, because this causes each single interaction of the session with the database to be a transaction on its own. This causes overhead, breaks the unit of work in case of errors during flushing, and causes some features to be disabled like auto-flush and second level cache. If the application encounters locks contention or deadlocks, investigate lowering the isolation level (like enabling read committed snapshot with SQL Server) or shortening your transactions rather than removing them.

Ensure you understand the semantics of Flush(). Flushing synchronizes the persistent store with in-memory changes but not vice-versa. Note that for all NHibernate ADO.NET connections/transactions, the transaction isolation level for that connection applies to all operations executed by NHibernate!

The next few sections will discuss alternative approaches that utilize versioning to ensure transaction atomicity. These are considered "advanced" approaches to be used with care.

You should observe the following practices when creating NHibernate Sessions:

The application may concurrently access the same persistent state in two different units-of-work. However, an instance of a persistent class is never shared between two ISession instances. Hence there are two different notions of identity:

Then for objects attached to a particular Session, the two notions are equivalent. However, while the application might concurrently access the "same" (persistent identity) business object in two different sessions, the two instances will actually be "different" (CLR identity).

This approach leaves NHibernate and the database to worry about concurrency. The application never needs to synchronize on any business object, as long as it sticks to a single thread per ISession or object identity (within an ISession the application may safely use == to compare objects).

Many business processes require a whole series of interactions with the user interleaved with database accesses. In web and enterprise applications it is not acceptable for a database transaction to span a user interaction.

Maintaining isolation of business processes becomes the partial responsibility of the application tier, hence we call this process a long running application transaction. A single application transaction usually spans several database transactions. It will be atomic if only one of these database transactions (the last one) stores the updated data, all others simply read data.

The only approach that is consistent with high concurrency and high scalability is optimistic concurrency control with versioning. NHibernate provides for three possible approaches to writing application code that uses optimistic concurrency.

// foo is an instance loaded earlier by the Session
session.Reconnect();
using (var transaction = session.BeginTransaction())
{
    foo.Property = "bar";
    session.Flush();
    transaction.Commit();
}
session.Disconnect();

// foo is an instance loaded by a previous Session
foo.Property = "bar";
using (var session = factory.OpenSession())
using (var transaction = session.BeginTransaction())
{
    session.SaveOrUpdate(foo);
    session.Flush();
    transaction.Commit();
}

// foo is an instance loaded by a previous Session
using (var session = factory.OpenSession())
using (var transaction = session.BeginTransaction())
{
    int oldVersion = foo.Version;
    session.Load( foo, foo.Key );
    if ( oldVersion != foo.Version ) throw new StaleObjectStateException();
    foo.Property = "bar";
    session.Flush();
    transaction.Commit();
}

The first approach described above is to maintain a single ISession for a whole business process that spans user think time. (For example, a servlet might keep an ISession in the user's HttpSession.) For performance reasons you should

before waiting for user activity. The method ISession.Disconnect() will disconnect the session from the ADO.NET connection and return the connection to the pool (unless you provided the connection).

ISession.Reconnect() obtains a new connection (or you may supply one) and restarts the session. After reconnection, to force a version check on data you aren't updating, you may call ISession.Lock() on any objects that might have been updated by another transaction. You don't need to lock any data that you are updating.

Here is an example:

ISessionFactory sessions;
IList<Foo> fooList;
Bar bar;
....
ISession s = sessions.OpenSession();
ITransaction tx = null;

try
{
    tx = s.BeginTransaction())

    fooList = s
        .CreateQuery(
            "select foo from Eg.Foo foo where foo.Date = current date"
            // uses db2 date function
        ).List<Foo>();

    bar = new Bar();
    s.Save(bar);

    tx.Commit();
}
catch (Exception)
{
    if (tx != null) tx.Rollback();
    s.Close();
    throw;
}
s.Disconnect();

Later on:

s.Reconnect();

try
{
    tx = s.BeginTransaction();

    bar.FooTable = new HashMap();
    foreach (Foo foo in fooList)
    {
        s.Lock(foo, LockMode.Read);    //check that foo isn't stale
        bar.FooTable.Put( foo.Name, foo );
    }

    tx.Commit();
}
catch (Exception)
{
    if (tx != null) tx.Rollback();
    throw;
}
finally
{
    s.Close();
}

You can see from this how the relationship between ITransactions and ISessions is many-to-one, An ISession represents a conversation between the application and the database. The ITransaction breaks that conversation up into atomic units of work at the database level.

It is not intended that users spend much time worrying about locking strategies. It's usually enough to specify an isolation level for the ADO.NET connections and then simply let the database do all the work. However, advanced users may sometimes wish to obtain exclusive pessimistic locks, or re-obtain locks at the start of a new transaction.

NHibernate will always use the locking mechanism of the database, never lock objects in memory!

The LockMode class defines the different lock levels that may be acquired by NHibernate. A lock is obtained by the following mechanisms:

The "explicit user request" is expressed in one of the following ways:

If ISession.Load() is called with Upgrade or UpgradeNoWait, and the requested object was not yet loaded by the session, the object is loaded using SELECT ... FOR UPDATE. If Load() is called for an object that is already loaded with a less restrictive lock than the one requested, NHibernate calls Lock() for that object.

ISession.Lock() performs a version number check if the specified lock mode is Read, Upgrade or UpgradeNoWait. (In the case of Upgrade or UpgradeNoWait, SELECT ... FOR UPDATE is used.)

If the database does not support the requested lock mode, NHibernate will use an appropriate alternate mode (instead of throwing an exception). This ensures that applications will be portable.

The legacy (1.0.x) behavior of NHibernate in regards to ADO.NET connection management was that a ISession would obtain a connection when it was first needed and then hold unto that connection until the session was closed. NHibernate introduced the notion of connection release modes to tell a session how to handle its ADO.NET connections. Note that the following discussion is pertinent only to connections provided through a configured IConnectionProvider; user-supplied connections are outside the breadth of this discussion. The different release modes are identified by the enumerated values of NHibernate.ConnectionReleaseMode:

The configuration parameter connection.release_mode is used to specify which release mode to use. The possible values:

If your application manages transactions through .NET APIs such as System.Transactions library while not using a compatible transaction factory (see transaction.factory_class in Section 3.5, “Optional configuration properties”), ConnectionReleaseMode.AfterTransaction may cause NHibernate to open and close several connections during one transaction, leading to unnecessary overhead and transaction promotion from local to distributed. Specifying ConnectionReleaseMode.OnClose will revert to the legacy behavior and prevent this problem from occurring.

Instead of using NHibernate ITransaction, TransactionScope can be used. Please do not use both simultaneously. Using TransactionScope requires using a compatible transaction factory (see transaction.factory_class in Section 3.5, “Optional configuration properties”). The default transaction factory supports scopes.

When using TransactionScope with NHibernate, you need to be aware of following points:

The IInterceptor interface provides callbacks from the session to the application allowing the application to inspect and/or manipulate properties of a persistent object before it is saved, updated, deleted or loaded. One possible use for this is to track auditing information. For example, the following IInterceptor automatically sets the createTimestamp when an IAuditable is created and updates the lastUpdateTimestamp property when an IAuditable is updated.

You may either implement IInterceptor directly or (better) extend EmptyInterceptor.

using System;
	
using NHibernate;
using NHibernate.Type;

public class AuditInterceptor : EmptyInterceptor {

    private int updates;
    private int creates;
    private int loads;

    public override void OnDelete(object entity,
                                  object id,
                                  object[] state,
                                  string[] propertyNames,
                                  IType[] types)
    {
        // do nothing
    }

    public override bool OnFlushDirty(object entity, 
                                      object id, 
                                      object[] currentState,
                                      object[] previousState,
                                      string[] propertyNames,
                                      IType[] types)
    {
        if ( entity is IAuditable ) {
            updates++;
            for ( int i=0; i < propertyNames.Length; i++ ) {
                if ( "lastUpdateTimestamp".Equals( propertyNames[i] ) ) {
                    currentState[i] = new DateTime();
                    return true;
                }
            }
        }
        return false;
    }

    public override bool OnLoad(object entity, 
                                object id, 
                                object[] state,
                                string[] propertyNames,
                                IType[] types)
    {
        if ( entity is IAuditable ) {
            loads++;
        }
        return false;
    }

    public override bool OnSave(object entity, 
                                object id, 
                                object[] state,
                                string[] propertyNames,
                                IType[] types)
    {
        if ( entity is IAuditable ) {
            creates++;
            for ( int i=0; i<propertyNames.Length; i++ ) {
                if ( "createTimestamp".Equals( propertyNames[i] ) ) {
                    state[i] = new DateTime();
                    return true;
                }
            }
        }
        return false;
    }

    public override void AfterTransactionCompletion(ITransaction tx)
    {
        if ( tx.WasCommitted ) {
            System.Console.WriteLine(
                "Creations: " + creates +
                ", Updates: " + updates +
                ", Loads: " + loads);
        }
        updates=0;
        creates=0;
        loads=0;
    }

}

Interceptors come in two flavors: ISession-scoped and ISessionFactory-scoped.

An ISession-scoped interceptor is specified when a session is opened using one of the overloaded ISessionFactory.OpenSession() methods accepting an IInterceptor.

ISession session = sf.OpenSession( new AuditInterceptor() );

An ISessionFactory-scoped interceptor is registered with the Configuration object prior to building the ISessionFactory. In this case, the supplied interceptor will be applied to all sessions opened from that ISessionFactory; this is true unless a session is opened explicitly specifying the interceptor to use. ISessionFactory-scoped interceptors must be thread safe, taking care to not store session-specific state since multiple sessions will use this interceptor (potentially) concurrently.

new Configuration().SetInterceptor( new AuditInterceptor() );

If you have to react to particular events in your persistence layer, you may also use the NHibernate2 event architecture. The event system can be used in addition or as a replacement for interceptors.

Essentially all of the methods of the ISession interface correlate to an event. You have a LoadEvent, a FlushEvent, etc (consult the XML configuration-file XSD or the NHibernate.Event namespace for the full list of defined event types). When a request is made of one of these methods, the ISession generates an appropriate event and passes it to the configured event listeners for that type. Out-of-the-box, these listeners implement the same processing in which those methods always resulted. However, you are free to implement a customization of one of the listener interfaces (i.e., the LoadEvent is processed by the registered implementation of the ILoadEventListener interface), in which case their implementation would be responsible for processing any Load() requests made of the ISession.

The listeners should be considered effectively singletons; meaning, they are shared between requests, and thus should not save any state as instance variables.

A custom listener should implement the appropriate interface for the event it wants to process and/or extend one of the convenience base classes (or even the default event listeners used by NHibernate out-of-the-box as their methods are declared virtual for this purpose). Custom listeners can either be registered programmatically through the Configuration object, or specified in the NHibernate configuration XML. Here's an example of a custom load event listener:

public class MyLoadListener : ILoadEventListener 
{
    // this is the single method defined by the LoadEventListener interface
    public void OnLoad(LoadEvent theEvent, LoadType loadType)
    {
        if ( !MySecurity.IsAuthorized( theEvent.EntityClassName, theEvent.EntityId ) ) {
            throw new MySecurityException("Unauthorized access");
        }
    }
}

You also need a configuration entry telling NHibernate to use the listener in addition to the default listener:

<hibernate-configuration>
    <session-factory>
        ...
        <event type="load">
            <listener class="MyLoadListener"/>
            <listener class="NHibernate.Event.Default.DefaultLoadEventListener"/>
        </event>
    </session-factory>
</hibernate-configuration>

Instead, you may register it programmatically:

Configuration cfg = new Configuration();
ILoadEventListener[] stack =
    new ILoadEventListener[] { new MyLoadListener(), new DefaultLoadEventListener() };
cfg.EventListeners.LoadEventListeners = stack;

Listeners registered declaratively cannot share instances. If the same class name is used in multiple <listener/> elements, each reference will result in a separate instance of that class. If you need the capability to share listener instances between listener types you must use the programmatic registration approach.

Why implement an interface and define the specific type during configuration? Well, a listener implementation could implement multiple event listener interfaces. Having the type additionally defined during registration makes it easier to turn custom listeners on or off during configuration.

When making new objects persistent, you must Flush() and then Clear() the session regularly, to control the size of the first-level cache.

using (ISession session = sessionFactory.openSession())
using (ITransaction tx = session.BeginTransaction())
{
    for (int i = 0; i < 100000; i++)
    {
        Customer customer = new Customer(.....);
        session.Save(customer);
        // 20, same as the ADO batch size
        if (i % 20 == 0)
        {
            // flush a batch of inserts and release memory:
            session.Flush();
            session.Clear();
        }
    }

    tx.Commit();
}

Alternatively, NHibernate provides a command-oriented API that may be used for streaming data to and from the database in the form of detached objects. A IStatelessSession has no persistence context associated with it and does not provide many of the higher-level life cycle semantics. In particular, a stateless session does not implement a first-level cache nor interact with any second-level or query cache. It does not implement transactional write-behind or automatic dirty checking. Operations performed using a stateless session do not ever cascade to associated instances. Collections are ignored by a stateless session. Operations performed via a stateless session bypass NHibernate's event model and interceptors. Stateless sessions are vulnerable to data aliasing effects, due to the lack of a first-level cache. A stateless session is a lower-level abstraction, much closer to the underlying ADO.

using (IStatelessSession session = sessionFactory.OpenStatelessSession())
using (ITransaction tx = session.BeginTransaction())
{
    var customers = session.GetNamedQuery("GetCustomers")
        .Enumerable<Customer>();
    while (customers.MoveNext())
    {
        Customer customer = customers.Current;
        customer.updateStuff(...);
        session.Update(customer);
    }

    tx.Commit();
}

Note that in this code example, the Customer instances returned by the query are immediately detached. They are never associated with any persistence context.

The insert(), update() and delete() operations defined by the StatelessSession interface are considered to be direct database row-level operations, which result in immediate execution of a SQL INSERT, UPDATE or DELETE respectively. Thus, they have very different semantics to the Save(), SaveOrUpdate() and Delete() operations defined by the ISession interface.

As already discussed, automatic and transparent object/relational mapping is concerned with the management of object state. This implies that the object state is available in memory, hence manipulating (using the SQL Data Manipulation Language (DML) statements: INSERT, UPDATE, DELETE) data directly in the database will not affect in-memory state. However, NHibernate provides methods for bulk SQL-style DML statement execution which are performed through the Hibernate Query Language (HQL). A Linq implementation is available too.

The pseudo-syntax for UPDATE and DELETE statements is: ( UPDATE | DELETE ) FROM? EntityName (WHERE where_conditions)?. Some points to note:

As an example, to execute an HQL UPDATE, use the IQuery.ExecuteUpdate() method:

using (ISession session = sessionFactory.OpenSession())
using (ITransaction tx = session.BeginTransaction())
{
    string hqlUpdate = "update Customer c set c.name = :newName where c.name = :oldName";
    // or string hqlUpdate = "update Customer set name = :newName where name = :oldName";
    int updatedEntities = s.CreateQuery(hqlUpdate)
        .SetString("newName", newName)
        .SetString("oldName", oldName)
        .ExecuteUpdate();
    tx.Commit();
}

HQL UPDATE statements, by default do not effect the version or the timestamp property values for the affected entities. However, you can force NHibernate to properly reset the version or timestamp property values through the use of a versioned update. This is achieved by adding the VERSIONED keyword after the UPDATE keyword.

using (ISession session = sessionFactory.OpenSession())
using (ITransaction tx = session.BeginTransaction())
{
    string hqlVersionedUpdate =
        "update versioned Customer set name = :newName where name = :oldName";
    int updatedEntities = s.CreateQuery(hqlUpdate)
        .SetString("newName", newName)
        .SetString("oldName", oldName)
        .ExecuteUpdate();
    tx.Commit();
}

Note that custom version types (NHibernate.Usertype.IUserVersionType) are not allowed in conjunction with a update versioned statement.

To execute an HQL DELETE, use the same IQuery.ExecuteUpdate() method:

using (ISession session = sessionFactory.OpenSession())
using (ITransaction tx = session.BeginTransaction())
{
    string hqlDelete = "delete Customer c where c.name = :oldName";
    // or String hqlDelete = "delete Customer where name = :oldName";
    int deletedEntities = s.CreateQuery(hqlDelete)
        .SetString("oldName", oldName)
        .ExecuteUpdate();
    tx.Commit();
}

The int value returned by the IQuery.ExecuteUpdate() method indicate the number of entities effected by the operation. Consider this may or may not correlate to the number of rows effected in the database. An HQL bulk operation might result in multiple actual SQL statements being executed, for joined-subclass, for example. The returned number indicates the number of actual entities affected by the statement. Going back to the example of joined-subclass, a delete against one of the subclasses may actually result in deletes against not just the table to which that subclass is mapped, but also the "root" table and potentially joined-subclass tables further down the inheritance hierarchy.

The pseudo-syntax for INSERT statements is: INSERT INTO EntityName properties_list select_statement. Some points to note:

An example HQL INSERT statement execution:

using (ISession session = sessionFactory.OpenSession())
using (ITransaction tx = session.BeginTransaction())
{
    var hqlInsert =
        "insert into DelinquentAccount (id, name) " +
        "select c.id, c.name from Customer c where ...";
    int createdEntities = s.CreateQuery(hqlInsert)
        .ExecuteUpdate();
    tx.Commit();
}

Queries are case-insensitive, except for names of .NET classes and properties. So SeLeCT is the same as sELEct is the same as SELECT but Eg.FOO is not Eg.Foo and foo.barSet is not foo.BARSET.

This manual uses lowercase HQL keywords. Some users find queries with uppercase keywords more readable, but we find this convention ugly when embedded in C# code.

The simplest possible NHibernate query is of the form:

from Eg.Cat

which simply returns all instances of the class Eg.Cat.

Most of the time, you will need to assign an alias, since you will want to refer to the Cat in other parts of the query.

from Eg.Cat as cat

This query assigns the alias cat to Cat instances, so we could use that alias later in the query. The as keyword is optional; we could also write:

from Eg.Cat cat

Multiple classes may appear, resulting in a cartesian product or "cross" join.

from Formula, Parameter

from Formula as form, Parameter as param

It is considered good practice to name query aliases using an initial lowercase, consistent with naming standards for local variables (eg. domesticCat).

We may also assign aliases to associated entities, or even to elements of a collection of values, using a join.

from Eg.Cat as cat 
    inner join cat.Mate as mate
    left outer join cat.Kittens as kitten

from Eg.Cat as cat left join cat.Mate.Kittens as kittens

from Formula form full join form.Parameter param

The supported join types are borrowed from ANSI SQL

The inner join, left outer join and right outer join constructs may be abbreviated.

from Eg.Cat as cat 
    join cat.Mate as mate
    left join cat.Kittens as kitten

In addition, a "fetch" join allows associations or collections of values to be initialized along with their parent objects, using a single select. This is particularly useful in the case of a collection. It effectively overrides the outer join and lazy declarations of the mapping file for associations and collections. See Section 20.1, “Fetching strategies” for more information.

from Eg.Cat as cat 
    inner join fetch cat.Mate
    left join fetch cat.Kittens

The associated objects are not returned directly in the query results. Instead, they may be accessed via the parent object.

It is possible to create a cartesian product by join fetching more than one collection in a query, so take care in this case. Join fetching multiple collection roles is also disabled for bag mappings. Note also that the fetch construct may not be used in queries called using Enumerable(). Finally, note that full join fetch and right join fetch are not meaningful.

The select clause picks which objects and properties to return in the query result set. Consider:

select mate 
from Eg.Cat as cat 
    inner join cat.Mate as mate

The query will select Mates of other Cats. Actually, you may express this query more compactly as:

select cat.Mate from Eg.Cat cat

You may even select collection elements, using the special elements function. The following query returns all kittens of any cat.

select elements(cat.Kittens) from Eg.Cat cat

Queries may return properties of any value type including properties of component type:

select cat.Name from Eg.DomesticCat cat
where cat.Name like 'fri%'

select cust.Name.FirstName from Customer as cust

Queries may return multiple objects and/or properties as an array of type object[]

select mother, offspr, mate.Name 
from Eg.DomesticCat as mother
    inner join mother.Mate as mate
    left outer join mother.Kittens as offspr

or as an actual type-safe object

select new Family(mother, mate, offspr)
from Eg.DomesticCat as mother
    join mother.Mate as mate
    left join mother.Kittens as offspr

assuming that the class Family has an appropriate constructor.

HQL queries may even return the results of aggregate functions on properties:

select avg(cat.Weight), sum(cat.Weight), max(cat.Weight), count(cat)
from Eg.Cat cat

Collections may also appear inside aggregate functions in the select clause.

select cat, count( elements(cat.Kittens) ) 
from Eg.Cat cat group by cat.Id, cat.Weight, ...

The supported aggregate functions are

The distinct and all keywords may be used and have the same semantics as in SQL.

select distinct cat.Name from Eg.Cat cat

select count(distinct cat.Name), count(cat) from Eg.Cat cat

A query like:

from Eg.Cat as cat

returns instances not only of Cat, but also of subclasses like DomesticCat. NHibernate queries may name any .NET class or interface in the from clause. The query will return instances of all persistent classes that extend that class or implement the interface. The following query would return all persistent objects:

from System.Object o

The interface INamed might be implemented by various persistent classes:

from Eg.Named n, Eg.Named m where n.Name = m.Name

Note that these last two queries will require more than one SQL SELECT. This means that the order by clause does not correctly order the whole result set.

In order to use non-mapped base classes or interfaces in HQL queries, they have to be imported. See Section 5.1.21, “import” for more information.

The where clause allows you to narrow the list of instances returned.

from Eg.Cat as cat where cat.Name='Fritz'

returns instances of Cat named 'Fritz'.

select foo 
from Eg.Foo foo, Eg.Bar bar
where foo.StartDate = bar.Date

will return all instances of Foo for which there exists an instance of Bar with a Date property equal to the StartDate property of the Foo. Compound path expressions make the where clause extremely powerful. Consider:

from Eg.Cat cat where cat.Mate.Name is not null

This query translates to an SQL query with a table (inner) join. If you were to write something like

from Eg.Foo foo  
where foo.Bar.Baz.Customer.Address.City is not null

you would end up with a query that would require four table joins in SQL.

The = operator may be used to compare not only properties, but also instances:

from Eg.Cat cat, Eg.Cat rival where cat.Mate = rival.Mate

select cat, mate 
from Eg.Cat cat, Eg.Cat mate
where cat.Mate = mate

The special property (lowercase) id may be used to reference the unique identifier of an object. (You may also use its property name.)

from Eg.Cat as cat where cat.id = 123

from Eg.Cat as cat where cat.Mate.id = 69

The second query is efficient. No table join is required!

Properties of composite identifiers may also be used. Suppose Person has a composite identifier consisting of Country and MedicareNumber.

from Bank.Person person
where person.id.Country = 'AU' 
    and person.id.MedicareNumber = 123456

from Bank.Account account
where account.Owner.id.Country = 'AU' 
    and account.Owner.id.MedicareNumber = 123456

Once again, the second query requires no table join.

Likewise, the special property class accesses the discriminator value of an instance in the case of polymorphic persistence. A .Net class name embedded in the where clause will be translated to its discriminator value.

from Eg.Cat cat where cat.class = Eg.DomesticCat

You may also specify properties of components or composite user types (and of components of components, etc). Never try to use a path-expression that ends in a property of component type (as opposed to a property of a component). For example, if store.Owner is an entity with a component Address

store.Owner.Address.City    // okay
store.Owner.Address         // error!

An "any" type has the special properties id and class, allowing us to express a join in the following way (where AuditLog.Item is a property mapped with <any>).

from Eg.AuditLog log, Eg.Payment payment 
where log.Item.class = 'Eg.Payment, Eg, Version=...' and log.Item.id = payment.id

Notice that log.Item.class and payment.class would refer to the values of completely different database columns in the above query.

Expressions allowed in the where clause include most of the kind of things you could write in SQL:

in and between may be used as follows:

from Eg.DomesticCat cat where cat.Name between 'A' and 'B'

from Eg.DomesticCat cat where cat.Name in ( 'Foo', 'Bar', 'Baz' )

and the negated forms may be written

from Eg.DomesticCat cat where cat.Name not between 'A' and 'B'

from Eg.DomesticCat cat where cat.Name not in ( 'Foo', 'Bar', 'Baz' )

Likewise, is null and is not null may be used to test for null values.

Booleans may be easily used in expressions by declaring HQL query substitutions in NHibernate configuration:

<property name="query.substitutions">true 1, false 0</property>

This will replace the keywords true and false with the literals 1 and 0 in the translated SQL from this HQL:

from Eg.Cat cat where cat.Alive = true

You may test the size of a collection with the special property size, or the special size() function.

from Eg.Cat cat where cat.Kittens.size > 0

from Eg.Cat cat where size(cat.Kittens) > 0

For indexed collections, you may refer to the minimum and maximum indices using minIndex and maxIndex. Similarly, you may refer to the minimum and maximum elements of a collection of basic type using minElement and maxElement.

from Calendar cal where cal.Holidays.maxElement > current date

There are also functional forms (which, unlike the constructs above, are not case sensitive):

from Order order where maxindex(order.Items) > 100

from Order order where minelement(order.Items) > 10000

The SQL functions any, some, all, exists, in are supported when passed the element or index set of a collection (elements and indices functions) or the result of a sub-query (see below).

select mother from Eg.Cat as mother, Eg.Cat as kit
where kit in elements(mother.Kittens)

select p from Eg.NameList list, Eg.Person p
where p.Name = some elements(list.Names)

from Eg.Cat cat where exists elements(cat.Kittens)

from Eg.Player p where 3 > all elements(p.Scores)

from Eg.Show show where 'fizard' in indices(show.Acts)

Note that these constructs - size, elements, indices, minIndex, maxIndex, minElement, maxElement - have certain usage restrictions: