R – Filtering NHibernate Hierarchy using link tables

You are getting the first error because, when you remove the items from the collection, NHibernate's default mode of operation is to simply break the association. In the database, NHibernate tries to set the foreign key column on the child row to null. Since you do not allow nulls in that column, SQL Server raises the error. Clearing the collection will not necessarily delete the child object, but one way to do so is to set cascade=all-delete-orphan. This informs NHibernate that it should delete the newly orphaned rows instead of setting the foreign key column.

You are getting the second error because when you call SaveOrUpdate NHibernate first deletes all of the child objects. Then, because neither relationship is marked as inverse, NHibernate also tries to set the foreign key column in your child table to null. Since the rows have already been deleted, you receive the second error. You need to set inverse=true on one side of your relationship to fix this. This is usually done on the one (primary key or parent) side. If you do not do this, NHibernate will make the appropriate updates for each side of the relationship. Unfortunately, running two updates is not the appropriate thing to do.

You should always mark one side of your relationships as the inverse side. Depending on how you code, you may or may not need to use cascading. If you want to take advantage of one shot deletes as you are trying to do using Clear(), you need to define your cascade.

To get a bidirectional association with a not-null foreign key column in the Details table you can add the suggested Owner property, a References(...).CanNotBeNull() mapping in the DetailsMap class, and make the Summary end inverse.

To avoid having two different foreign key columns for the two association directions, you can either specify the column names manually or name the properties in a way that gives the same column name for both directions. In this case you I suggest renaming the Details.Owner property to Details.Summary.

I made the Summary id generated by increment to avoid problems when inserting into the table since Summary currenty has no columns besides id.

Domain:

public class Detail
{
    public int id { get; protected set; }
    public string ItemName { get; set; }

    // Renamed to use same column name as specified in the mapping of Summary.Details
    public Summary Summary {get; set;} 
}

public class Summary
{
    public Summary()
    {
        Details = new List<Detail>();
    }

    public int id { get; protected set; }
    public IList<Detail> Details { get; protected set; }
}

Mapping:

public class DetailMap : ClassMap<Detail>
{
    public DetailMap()
    {
        Id(x => x.id)
            .GeneratedBy.Native();

        Map(x => x.ItemName)
            .CanNotBeNull();

        References<Summary>(x => x.Summary)
            // If you don't want to rename the property in Summary,
            // you can do this instead:
            // .TheColumnNameIs("Summary_id")
            .CanNotBeNull();
    }
}

public class SummaryMap : ClassMap<Summary>
{
    public SummaryMap()
    {
        Id(x => x.id)
            .GeneratedBy.Increment();

        HasMany<Detail>(x => x.Details)
            .IsInverse()
            .AsBag(); // Use bag instead of list to avoid index updating issues
    }
}