[hibernate-dev] Implement UserCollectionType for non-Collection types such as Guava's Multimap

[Steve Ebersole]

On Mon, May 1, 2017 at 4:26 AM Jan-Willem Gmelig Meyling <
jan-willem at youngmediaexperts.nl> wrote:

> Hi Steve,
>
> Thanks for your response to my questions! Good to see that my intuition
> was right that this was currently not so trivial if even possible to do.
> I’ll read up on the 6.0 code base as I am definitely willing to join the
> ride for Hibernate 6. What is the current status of the Hibernate 6.0 type
> redesign?
>

It is still very much fluid, but getting pretty close.  The "runtime model"
is the piece you are interested in, and I have been working on that for the
past few weeks.  Specifically you would be looking at CollectionPersister
and friends.  You'd also be interested in CollectionType, but Type take on
a slightly different role in 6.0 - the important bits move to
CollectionPersister and the attribute definition.  CollectionPersister I
have not spent a lot of time on yet.

The code for 6.0 is still just part of my fork, and i branch named
`wip/6.0`: https://github.com/sebersole/hibernate-core/tree/wip/6.0



> What I was wondering though is whether the support of JDBC array types
> would pave the way for array-based subselect fetching of to-many relations.
> I am thinking of for example Postgresql’s array_agg function. In some of
> our more complex domain entities I’ve been struggling with eagerly fetching
> multiple to-many relations or deeper to-many relations without getting a
> relatively large cross product. I currently do use subselect fetching for
> this but I think having the ability to inline rows (perhaps even
> recursively) in arrays would decrease query time, number of queries and
> result set size significantly.
>

Possibly.  Hard to say yet.  First thing would be to analyze whether that
approach actually is more performant and if so how much better the
performance is (it has to justify the cost).