Hello everyone,

We're trying to determine by HQL when a lazy collection will be join-loaded. We're using fetch join but we get what it seems a strange result.

This HQL query works well:
select distinct d from DescripcioAssignatura d join d.fontsInformacio
(in fact, kinda "select distinct p from parent p join p.children")
It results in 4 rows, but it doesn't load the lazy collection children. That's ok.

This HQL query...:
select distinct d from DescripcioAssignatura d join fetch d.fontsInformacio
(in fact, kinda "select distinct p from parent p join fetch p.children")
It results in 20 rows, where each original row is repeated as many times as children it has.

Is this a bug or am I missing something here?

P.S. The collection fontsInformacio (childre) is mapped as follows:

<set name="fontsInformacio" lazy="true" inverse="true" >
<key column="descripcioAssignaturaId"/>
<one-to-many class="es.castinfo.ub.domain.FontInformacio"/>
</set>

Thanks in advance,

Jordi Pradel & Jos

This is a known issue - HIB-123

As I recall, the last recommended work around was to insert your resultset into a map.

Also, turn on your "show_sql" attribute. Then you can see what Hibernate is doing. In the first scenario, Hibernate is just returning your result set from your main table "DescripcioAssignatura". It is creating proxies for each of your collection associations. When you access your collections, you will see Hibernate generate the appropriate sql to "lazy" load them at that time.

In the second scenario, the "fetch" option is telling Hibernate to "left outer join" your data all at once, no lazy loading. Look at the generated sql. Hibernate is basically passing back to you what it is getting from jdbc. jdbc will return 20 rows due to the joins. Hibernate still loads your collections for you.

I havn't yet decided whether Hibernate should distinctify the query result set automatically. Believe it or not, this can have a significant performance impact, so I lean toward leaving things the way they are and letting the application handle this, if required.

Ok, no problem. The application can insert the obtained List in a Set or do any other handling of duplicates.

The only disadvantage then is the difference in semantics between join and fetch join.

P.S. Great forums! Thanks to evaala, gavin and everyone participating in Hibernate forums.

Jordi Pradel