I have a inheritance hierarchy which I have mapped in NHibernate using Table-per-class. My mappping file looks like the one below (lots of properties omitted).
To query this hierarchy, I am building a dynamic DetachedCriteria for Message based on filter input from the user. Messages (of any type in the hierarchy) should be returned to the user in one list.
I would like to build a criteria based on the type of message, ie. the user could specify to get all messages of type SMSMessage or EmailMessage with a ReceivedDate > '2009-01-01'. How would I go about to do that?
In the same query, the user could specify that if the Message is an InternalMessage, it should have Priority = 2. How would I specify such type-specific predicates?
All this is possible to do in LINQ, so I am hoping I can do it in NHibernate as well.
<class name="Message" table="Message" abstract="true" discriminator-value="null">
<id name="MessageId">
<generator class="identity" />
</id>
<discriminator column="Type" type="byte" />
<property name="ParentId" />
<property name="ReceivedDate" />
...
<subclass name="SMSMessage" discriminator-value="0">
<property name="Text" column="Text" />
...
</subclass>
<subclass name="MMSMessage" discriminator-value="1">
<property name="Subject" />
...
</subclass>
<subclass name="EmailMessage" discriminator-value="2">
<property name="BodyPlainText" />
...
</subclass>
<subclass name="InternalMessage" discriminator-value="4">
<property name="Priority" />
...
</subclass>
</class>
I kind of figured this out myself, but in the end I ended up reverting to pure SQL since I hit too many roadblocks with HQL/Criterias. Anyways, I can share how I did this.
Maybe not pretty, but I solved it by adding the discriminator column as a regular property to the top level class in the hierarchy (Message) and employed restrictions against that column.
It turns out that you can specify restrictions against properties for subclasses even in the top-level query, so this was easier than I thought. It was just a matter of specifying the restrictions.
Related
Having trouble with implementing second level cache in Nhibernate. I have a class mapped as follows:
<?xml version="1.0" encoding="utf-8" ?>
<hibernate-mapping xmlns="urn:nhibernate-mapping-2.2" assembly="Data" namespace="Data">
<class name="Account" table="Accounts" lazy="false">
<cache region="Standard" usage="read-write" include="all"/>
<id column="ID" name="ID">
<generator class="assigned" />
</id>
<version name="VersionStamp" column="VersionStamp" type="integer" unsaved-value="0" />
<property name="Name" not-null="true" />
<property name="Type" not-null="true" />
<property name="ClientID" not-null="true" insert="false" update="false" />
<property name="DateCreated" not-null="true" type="UtcDateTime" />
<property name="LastUpdatedDate" not-null="true" type="UtcDateTime" />
<property name="IsActive" not-null="true" />
<many-to-one name="Client" class="Client" column="ClientID" not-found="exception" not-null="true" />
</class>
</hibernate-mapping>
The property "ClientID" is a foreign key into the Clients table and the Client many-to-one property uses it to look up the associated client object.
When I add a new Account, I look up the Client object from the database with a Session.Get and assign it to my Account object's Client property. Behind the scenes, this also automatically populates the ClientID property when the object is written to the database, and the ID is correctly stored in the database. When I retrieve the Account object from the database by ID using Session.Get, all the fields are populated correctly when the object is retrieved.
However, when I implement the second level cache using the settings shown above, the ClientID property is NOT populated when the Account object is retrieved using Session.Get, but the Client property is populated correctly. Is there some reason why this will not work with second level cache? Or have I done something wrong in my mapping/configuration?
For now I am just using SysCache as my caching provider, and both query and second level cache are turned on. The Client class mapping contains a corresponding one-to-many property for the Accounts.
I like the convenience of having the ClientID property on my Account class, so that I can read it without using the Client property, and it seems to work fine without caching.
Thanks for any help.
Rich
I tried to reproduce your situation locally. With your mapping (used the same as the snippet above) I was able to get incorrect behaviour only on UPDATE. In that case, the ClientID was cached, and while the Client reference was changed, the ClientID remained unchanged. In other cases caching was working as expected.
The solution is to change the mapping. The below suggested mapping is the most suitable for read-only properties like ClientID. (I am using that approach as well).
<property name="ClientID" formula="[ClientId]"
not-null="true" insert="false" update="false" />
So the trick is in the formula mapping instead of Column (the default when none is provided)
I have an entity class that I use to represent the results of an sql-query. The mapping for the class shown below. Yet as far as I can tell nhiberate treats the mapping as if there is a real database table - when in fact there is not. In this case there is nothing in the database that represents this entity at all. I am using this to map a query through, but the same would be true of a view. Is there no way to indicate to nhibernate that there isn't a table represented by the mapping?
<class name="Models.UserTransaction"> <!-- Defaults table name same as Entity even though table doesn’t exist -->
<id name="Id">
<column name="Id" not-null="true" unique="true" />
<generator class="native" />
</id>
<property name="TransType" />
<property name="Date" />
<property name="Amount" />
<property name="Balance" />
</class>
This is the query I am mapping, which uses a user defined table. I couldn't get it working without having a mapping even though the example I copied appeared to.
<sql-query name="UserTransactions">
<query-param name="userId" type="string" />
<return class="Models.UserTransaction" alias="userTx">
<return-property name="TransType" column="TransType" />
<return-property name="Id" column="Id" />
<return-property name="Date" column="TransDate" />
<return-property name="Amount" column="Amount" />
<return-property name="Balance" column="Balance" />
</return>
<![CDATA[
SELECT userTx.[Type] as TransType, userTx.[Id] as Id, userTx.[Date] as TransDate, userTx.[Amount] as Amount, userTx.[Balance] as Balance
FROM dbo.User_AccountStatement(:userId) userTx
]]>
</sql-query>
If you have a db view, you can use nhibernate to map to that, but if all you are doing is storing the projection fields of the query there doesn't need to be a map at all.
How are you querying this data?
if you are using the criteria API, you can use the resultstransformer to map the returned object array to your class. the types have to match between your class that the projection.
if you are using the linq provider you can project directly into your class. so you'd have something like this
from s in Session.Query
where s.some-property== "some-value"
select new your-type
{
some-property-on-your-type = s.some-property,
some-other-property-on-your-type = s.some-other-property
}
There is no need to write a mapping to the database since you aren't mapping to an object in the database.
I guess you should at least specify a view as the tablename of your mapping.
The view should have the same resulting columns as your query (and hopefully return any row that your query could return)
Then you will be able to :
map your properties to your view/SQL query result columns
set your class as mutable="false"
declare your query as a named query (see : http://ayende.com/blog/3948/nhibernate-mapping-named-queries-query-and-sql-query )
I have two classes: Family and Address.
A family has a physical address and a mailing address.
The mapping file for Family looks like:
....
<id name="Id" column="Id" type="Int32" unsaved-value="0">
<generator class="native"></generator>
</id>
<many-to-one name="PhysicalAddress" class="Address" column="PhysicalAddressId" cascade="all" unique="true" />
<many-to-one name="MailingAddress" class="Address" column="MailingAddressId" cascade="all" unique="true" />
...
The mapping file for Address looks like:
...
<id name="Id" column="Id" type="Int32" unsaved-value="0">
<generator class="native"></generator>
</id>
<property name="StreetAddress1" column="StreetAddress1" />
<property name="StreetAddress2" column="StreetAddress2"/>
<property name="City" column="City" />
<property name="State" column="State" />
<property name="ZipCode" column="ZipCode" />
...
(Note that Family-PhysicalAddress and Family-MailingAddress are one-to-one relationships.)
What I would like to happen is that when I execute
aFamily.MailingAddress = null;
session.Save(aFamily);
session.Flush();
I expect NHibernate to automatically delete the mailing address record from SQL Server for me.
BUT, that doesn't happen. NHibernate does not delete the address record from SQL Server.
Is there any way I can make it work?
Thank you!
This behaviour isn't supported by NHibernate. Of course the problem is that you probably don't have access to the NHibernate session in your domain logic where the change is made.
One possible -- though admittedly not ideal solution -- is to simply run another process to clean up orphaned entities.
Here is a discussion of this scenario:
http://colinjack.blogspot.com/2008/03/nhibernate-gotchas-orphans-and-one-to.html
And a link to a ticket on the issue:
https://nhibernate.jira.com/browse/NH-1262
Unfortunately NHibernate currently does not support automatic deletions of orhphans for many-to-one (Hibernate v3 in Java does support it). It is only supported on lists (cascade="all-delete-orphan").
What you can try to do is to use component mapping. Maybe it is possible to embed many-to-one into a component.
But I think it would better to explicitly delete the related object.
I have a base class for content items in a CMS I'm building. It's currently marked abstract because I only want derived classes to be instantiated. Derived classes like BlogPost, Article, Photo, etc. are set up as a joined subclass to my ContentBase class in nHibernate.
I'm trying to set up a many-to-many mapping between this class and a Tag class. I want to have a collection of Tags on the ContentBase class, and a collection of ContentBase items on the tag class.
Will nHibernate allow me to map the abstract ContentBase class as a collection on the Tag class? I'm assuming not since it wouldn't be able to instantiate any instances of this class when reconstituting a Tag entity from the db. I really don't want to have to have to use a collection of content items per type (e.g. TaggedBlogPosts, TaggedArticles, etc.) on the Tag class.
The whole reason I'm doing this is because logically, a content item can have many tags, and 1 tag can belong to multiple content items. in order for nHibernate to manage the relationships for me in a mapping table, I believe I have to set up a many-to-many association and add the Tag to the ContentBase.Tags collection and then the content item to the Tags.TaggedContentItems collection before the mapping table entry is created in nHibernate.
Here are my mappings for reference:
<class name="CMS.Core.Model.Tag,CMS.Core" table="bp_Tags">
<id column="TagName" name="TagName" type="String" unsaved-value="">
<generator class="assigned" />
</id>
<bag name="_taggedContentList" table="bp_Tags_Mappings" inverse="true" cascade="save-update" lazy="true">
<key column="TagName" />
<many-to-many class="CMS.Core.Model.ContentBase,CMS.Core" column="Target_Id" />
</bag>
</class>
<class name="CMS.Core.Model.ContentBase,CMS.Core" table="bp_Content">
<id name="Id" column="Id" type="Int32" unsaved-value="0">
<generator class="native"></generator>
</id>
<property name="SubmittedBy" column="SubmittedBy" type="string" length="256" not-null="true" />
<property name="SubmittedDate" column="SubmittedDate" type="datetime" not-null="true" />
<property name="PublishDate" column="PublishDate" type="datetime" not-null="true" />
<property name="State" column="State" type="CMS.Core.Model.ContentStates,CMS.Core" not-null="true" />
<property name="ContentType" column="ContentType" type="CMS.Core.Model.ContentTypes,CMS.Core" not-null="true" />
<bag name="_tagsList" table="bp_Tags_Mappings" lazy="false" cascade="save-update">
<key column="Target_Id" />
<many-to-many class="CMS.Core.Model.Tag,CMS.Core" column="TagName" lazy="false" />
</bag>
...
<joined-subclass name="CMS.Core.Model.BlogPost,CMS.Core" table="bp_Content_BlogPosts" >
<key column="Id" />
<property name="Body" type="string" column="Body" />
<property name="Title" type="string" column="Title" />
</joined-subclass>
...
I would also assume that Hibernate would need to instantiate the base class, which makes sense, as there is data directly linked to it in the DB.
It looks like your data entities are in a separate assembly from your main application. The point of not instantiating the base is from a business point of view, so if you make the constructor internal to the Core assembly, does that accomplish what you want? If not, it may be helpful to ask yourself: who am I protecting this functionality from?
NHibernate should allow you to map the abstract class as a collection as long as you map the abstract class itself (it looks like you have).
An alternative is to change your strategy to the Table-per-class-hierachy approach. This puts all your content into a single table with a discriminator column to define the type of content. NHibernate knows how to materialize each content type based on the discriminator. The downside is that:
Each unique property of the concrete classes has to be of type nullable.
This table can get unwieldy if there are a lot of properties specific to each concrete class.
The table-per-class-hierachy is the preferred default mapping strategy (per NHibernate in Action). I would start with the approach and modify it to table-per-subclass when the need is certain.
A whole separate solution would be keep the mapping as is and rely on query to get the content by tag without worrying about the tags keeping a reference to a collection of content.
I'm using nHibernate to map an object very similar to .NET's System.Web.SiteMapNode. In order to keep my object similar to this .NET object I would like to have it contain a ParentNode, PreviousSibling, NextSibling, and ChildNodes complex properties.
The table looks somewhat like this and is open to be changed:
ID (int)
Title (string)
Description (string)
Key (string)
ParentNodeId (int)
OrdinalPosition (int)
ReadOnly (bool)
Url (string)
I may have some other properties that are not needed to mimic the .NET SiteMapNode object (like an isExternal bool), but I think those are inconsequential to this question.
My current mapping looks like this:
<hibernate-mapping xmlns="urn:nhibernate-mapping-2.2" namespace="AthletesCafe.Core.Domain.System.SiteMap" assembly="AthletesCafe.Core">
<class name="SiteMapNode" table="SiteMapNode" lazy="true" >
<id name="ID" type="Int32" unsaved-value="0">
<column name="ID" not-null="true" unique="true" index="PK_SiteMapNode"/>
<generator class="identity" />
</id>
<property name="Title" column="Title" type="String" length="255" not-null="true" />
<property name="Description" column="Description" type="String" not-null="false" />
<property name="Url" column="Description" type="String" not-null="true" />
<property name="SiteMapKey" column="SiteMapKey" type="String" not-null="true" length="255" />
<property name="OrdinalPosition" column="OrdinalPosition" type="Int32" not-null="true" />
<property name="ReadOnly" column="ReadOnly" not-null="true" type="System.Boolean" />
<property name="IsExternal" column="IsExternal" not-null="true" type="System.Boolean" />
<many-to-one name="ParentNode" column="ParentNodeId" class="AthletesCafe.Core.Domain.System.SiteMap.SiteMapNode, AthletesCafe.Core"
access="field.pascalcase-underscore" not-null="false" />
<many-to-one name="PreviousNode" column="ParentNodeId" class="EatMyTrainer.Core.Domain.SiteMap.SiteMapNode, EatMyTrainer.Core" not-null="false" /></hibernate-mapping>
The ParentNode mapping is easy as it should be just a simple many-to-one mapping. This is the code I have for it (untested, but I believe it to be correct):
<many-to-one name="ParentNode" column="ParentNodeId" class="AthletesCafe.Core.Domain.System.SiteMap.SiteMapNode, AthletesCafe.Core"
access="field.pascalcase-underscore" not-null="false" />
The mapping for the child nodes should just be a simple bag which will bring back all SiteMapNode objects that have the ParentNodeId equal to the current ID. I haven't written this bag yet, but I believe it to be not such a big deal.
The issue that I cannot seem to resolve is how to do the Next/Previous Sibling properties. This objects can be derived from the following formula for each node:
PreviousSibling: Has the same ParentNode (ParentNodeId) as the current object and its OrdinalPosition should be one less than the current object's OrdinalPosition.
NextSibling: Has the same ParentNode (ParentNodeId) as the current object and its OrdinalPosition should be one more than the current object's OrdinalPosition.
I think this is achievable through the formual attribute on a many-to-one mapping. Is this possible? I haven't found a good example of how this works.
I don't think what you're asking for is strictly possible (although I would be very interested to see the solution if it is). There would be a relatively simple workaround, but NHibernate does not support bidirectional one-to-many mappings with indexed collections on the many end.
The only thing that comes to mind is a bit ugly: have the parent object keep its own index map (keyed off the OrdinalPosition) to each child object. On the child do something like:
public SiteMapNode NextSibling()
{
return this.Parent.NextSibling(this);
}
I believe Stuart is correct in this situation. It is impossible to do for the many-to-one mapping. If NHibernate provided a way to do where clausing on this mapping then I may have a chance.
Another possible solution although inefficient is to create a bag that uses a field setter. The public property that would be Next/Previous setting would still return an object reference (as opposed to an enumerable). In the getter it would just reference the first position of the enumerable in the field. Lazy loading would be ideal because NHibernate wouldn't be able to load this object in one get with the initial load of the object. You would have a penalty for accessing this object every time.
I guess both solutions have a similar penalty.