Hello everybody,

i have a problem or perhaps just a missunderstanding of session "cache" and the session-per-conversation pattern.

First the facts:

Hibernate version:
1.2.0 GA

Mapping documents:
<?xml version="1.0" encoding="utf-8" ?>

<hibernate-mapping xmlns="urn:nhibernate-mapping-2.2"
assembly="Benutzerverwaltung"
namespace="Benutzerverwaltung.Domain" >

<class name="Mandant">
<id name="Id">
<generator class="hilo"/>
</id>
<version name="Version"/>

<property name="Kennung" not-null="true" unique="true"/>
<property name="Beschreibung"/>
<property name="Praefix" not-null="true" unique="true"/>


<bag name="funktionsrechte" inverse="true" cascade="all-delete-orphan" access="field">
<key column="Mandant_Id"/>
<one-to-many class="Funktionsrecht"/>
</bag>

<bag name="benutzergruppen" inverse="true" cascade="all-delete-orphan" access="field">
<key column="Mandant_Id"/>
<one-to-many class="Benutzergruppe"/>
</bag>
</class>

</hibernate-mapping>


Test Code:
[Test]
public void BusinessObjectJustSavedShouldBeFound() {
ISession session = pHelper.OpenSession();
session.FlushMode = FlushMode.Never;

Mandant m = new Mandant("key", "text", "text");
session.Save(m);

IQuery query = session.CreateQuery("FROM Mandant m WHERE m.Kennung= 'key'");
IList<Mandant> list = query.List<Mandant>();
Assert.AreEqual(1, list.Count);
Mandant mLoaded = list[0];

Assert.AreEqual(m, mLoaded);
}


Name and version of the database you are using:
Microsoft SQL Server Express Edition (Version 9.00.1399.06)


As you can see in the NUnit-Test I expect that the just saved business object of the type Mandant would be found by the query. But the test fails.
(Well, I know that when i call session.Flush() after the session.Save it would be possible, but I don´t want to call it -> see the reason described in the story.)

The story:
I have a GUI where the user can change the values of business objects, delete them or create new ones. The user can also use a cancel button or a save button to "rollback" or "commit" changes.
In my first idea I realised this with transactions, but I found out that it is not a good idea to keep transactions open for a long time (user think time) because they might lock complete db-tables.

So I decided to use the Session-per-Conversation-Pattern.
Now it is possible to write changes to the session and when the user clicks "save" session.Flush() is called. When the user wants to cancel I call session.Clear(). (So you can see now why I can´t call Flush() before the user clicks "save" because otherwise a rollback wouldn´t be possible any more.)
So far it´s ok.

My problem now is, that any changes I write to the "unflushed" session are not reachable by queries. As in the test case above, I would expect that NHibernate knows that I deleted the object before in this session context and would not return it by a query called in the same session context.

I now have to manage differences between changes written to the session and the actual state of the database on my own. Well, I always thought that such a management is one of the key features of Hibernate.

Is there any possibility to reach the features I described with on-board features of Hibernate an not to implement them on my own?

I realy like NHibernate :-) - please help me that I don´t have to give up using it.

Thank you very much for your help!!!

Greetings
hiaso

It is, but you can't query into those changes without first flushing to the database. NH only translates HQL and Criteria into DB-specific SQL, it does not also query its own internal state.

As for having to "give up" on NH, it seems that you could build your own conversation abstraction on top of using NH as your mapping and unit-of-work provider.