hi there,

i have a hopefully easy question to answer for you.
given the following code and mappings. how do i get hibernate
executing without throwing an exception.

now the classes being used


and then of course the mappings


and here finally the table definitions


here first the testcode.


this code will throw an exception as it tries to persist the class
Employee, which of course does not work as only the subclasses
have a persistent mapping.

any ideas how to solve this?

tia

robertj

I've done a one-to-one mapping and you need a <hibernate-mapping> for your Employee class. Otherwise, hibernate won't know how to save it to the database.

Regards,
David

well thats exactly the point. i dont want to save employees
to the database - only programmers or salespersons.

so i dont want/need a mapping for employee.

ciao robertj

You wrote you don't want to save employees to the database, so why did you set your one-to-one mapping to an employee record? If you want to save the employee as a type programmer or sales_person, why don't you use subclasses with a discriminator value so they get saved to the same table since they have the same property names? The discriminator should differentiate the account type. I've tried it before but I don't have any examples handy - the ref pdf should have one though.


Why all the tables when they have the same fields?

Regards,
David

hi david,



this was meant as a simplified example.
but let me rephrase my question:

if i have a given set of classes
(eg. Job, Employee, Programmer, SalesPerson) AND
a given set of tables
(eg.JOBS, PROGRAMMERS, SALES_PERSON)
WHERE neither class-structure nor table-structure
is changable,

how do i create a mapping so that for a polymorphic pair
of getter/setters the correct tables will be filled when cascading
is set to true?

ciao robertj

From your included examples, I can't figure out how you determine if the employee is a programmer or a sales person. Is it one depending on which table (programmer or sales_person) you find their id number in?

A discriminator (number, character, string, anything) would be helpful to have so you could use joined-subclasses (parts of the record in different tables) based on the discriminator. My only suggestion (and since no one else has posted ideas) is to use one-to-one mappings.

If you create an employee object which has a one-to-one to the Programmer and a one-to-one to the SalesPerson, you can leave one null and one assigned. If employee.getProgrammer() != null, then you have a programmer whereas if employee.getSalesPerson() != null, then you have a Sales Person. Your account add/change logic will have to make sure one of those types is set.

I'm not skilled enough to give you an example which will assign keys for you so you could just add a "Programmer" or "SalesPerson" to your "Job" and save just the "job" but someone else probably can if you ask it in another thread.

My working example shows if you save both the Job and Programmer/SalesPerson objects individually. First, here is a Java code snippet:
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
try {
tx = session.beginTransaction();
Programmer programmer = new Programmer();
Job job = new Job();
job.setDepartment("Programming");
programmer.setId(job.getId());
programmer.setName("David");
programmer.setPrimaryLanguage("Java");

session.save(job);
programmer.setId(job.getId());

session.save(programmer);
tx.commit();
session.close();
} catch (HibernateException he) {
System.out.println("Exception: " + he.getMessage() );
}
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

It came from the mapping file I made (used hbm2java and SchemaUpdate with it to create the classes and tables automatically, Hey, I'm lazy!).... Oh, I used "eg" as the package name so it won't interfere with your code if you try to test it.

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
<class name="eg.Job" table="JOBS" >
<id name="id" column="id" type="string">
<generator class="uuid.hex"/>
</id>
<property name="department" type="string"/>
<one-to-one
name="programmer"
class="eg.Programmer" />
<one-to-one
name="salesperson"
class="eg.SalesPerson" />
</class>

<class name="eg.Programmer" table="PROGRAMMER">
<id name="id" unsaved-value="any" type="string">
<generator class="foreign">
<param name="property">parent</param>
</generator>
</id>
<property name="name" type="string"/>
<property name="primaryLanguage" column="PRIM_LANG" type="string"/>
<one-to-one name="parent" class="eg.Job" />

</class>

<class name="eg.SalesPerson" table="SALES_PERSON">
<id name="id" unsaved-value="any" type="string">
<generator class="foreign">
<param name="property">parent</param>
</generator>
</id>
<property name="name" type="string"/>
<property name="primaryCustomer" column="PRIM_CUST" type="string"/>
<one-to-one name="parent" class="eg.Job" />
</class>

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

I'll play around with the assigned or foreign keys for a bit and post a follow-up if I can figure them out. I'd like to know for myself but I'm just not grasping the auto-assigning cascading id's on a parent object's save.

Regards,
David

Have you checked out the <any> mapping? This is how you map a poymorphic association to a "table-per-concrete-class" mapping strategy in Hibernate.

Gavin,

Would the 'any' mapping work if the record has no discriminator and could be in either of two tables like Robert described?

Regards,
David

Essentially, what the <any> mapping does is put the discriminator on the table that owns the association, instead of on the table(s) of the associated classes.

So, no, it won't map perfectly to his current schema - but it might, anyway, be what hes looking for.

hi,

ok maybe i am still not clear enough regarding my request.

all of the above is/was meant as an example.

the only real need is this:
i have a parent class that has an accessor which returns
a child class. during runtime the returned instances of
the child class will be of subclasses of this childclass but
never of the base child class.

once again an even more simplified example:



the following rules do apply
1) the semantics of get/setEmployee must not change
2) the semantics of job to employee is a "to one" - which
does not necessarily imply that the hibernate mapping
must be that way. i quite open to suggestions.
3) cascading inserts/updates/deletes must be supported by
the mapping.
4) the class structure can be extended (not changed in their
semantics) if need to be (but i am very hesistant to do so).
5) the table structures can be extended if need to be as long
as the basic structure (3 tables, 2 for subclasses of employee)
remains.

@gavin:
i have looked into the any-mapping, but to be honest the documentation is quite shallow at the moment regarding this
mapping element. i would rather need concrete example of
the usage of any (if any is the solution).

as a side note:
the selection of the word "any" is not such a good idea imo.
i quite understand the "technical" reasoning for it and it somehow
captures the intended semantics quite well but googling for "any" is obviously useless and for looking up examples this is quite the
only way at the moment.

btw. i will have a very simmilar problem soon when using collections.
i will define something like the following (this again is only an
example).

in the mapping i would have to do something like this

only that during runtime there never will be any Employee instances
but instances of subclasses.


if somebody can show me the concrete mappings for my example
i gladly write a small document which ties it all together.

thanks

ciao robertj

Search for the string "<any" in the net.sf.hibernate.test package. It is what you need.

oh, and for collections, there is the very-badly-named <many-to-any> element.