See the test case for details, but basically, I was trying to create a HQL query that would return a parent object if either its sub-objects matched on a name, or it had no links to its sub-objects. I mistakenly used the identification variable in an is empty instead of the actual collection because I misunderstood that the identification variable is the single object in the collection, not an alias for the collection itself. Whilst this is not the bug, what I believe is the bug, is that hibernate went on to create this sql query:
{code sql }
select
car0_.id as id1_0_
from
car car0_
left outer join
car_parts_join_table parts1_
on car0_.id=parts1_.car_id
left outer join
part part2_
on parts1_.part_id=part2_.id
left outer join
part_subpart_join_table subparts3_
on part2_.id=subparts3_.part_id
left outer join
subpart subpart4_
on subparts3_.subpart_id=subpart4_.id
where
not (exists (select
part2_.id
from
part part2_))
or subpart4_.name='axle'
{code}
Instead of just erroring or at least throwing a warning saying that I was using invalid syntax. The problem with this query is that if you look closely at the first where fragment, it will always return true for exists (..), so it will always return false when used in the not(..), therefore will never return any Cars with no parts. This obviously does this logically not make any sense at all if you've not twigged you're using the hql wrong.
Maybe this will not be agreed with, but I feel like what was happening was very unexpected to the user, and this situation would be greatly improved if hibernate could detect whether someone was using a non collection or collection identification variable in an is [not] empty fragment (or indeed any other collection-related fragments). |
|
