[Dbix-class] PATCH: Support for arbitrary SQL in relationship definition

[Rob Kinyon]

On Tue, Jun 30, 2009 at 11:14, David Ihnen<davidi at norchemlab.com> wrote:
> But the question seemed to be 'I don't understand why you would need this'
> and 'is there a need for this feature' and I must vociferously respond "IT
> IS USEFULE" and "YES THERE IS".  As for 'is there a need to do it this way'
> to which I'd agree with you, probably not this way, its awfully cludgy to
> use the refs.
>
> BTW, I am okay with the join terms *not* being query time dynamic/bindable
> stuff - in my application I want to join based on two separate relations
> between the two tables, not just one, and not using a query-time parameter
> that affects the relationship spec.

I don't think anyone has said "This is a bad idea." For my part, I've
been pushing back to try and get some use-cases that cannot be written
as methods or rewritten with subqueries.

My point in all this is that relationships should be created when you
want to use that join in a query later or via X_related(). Otherwise,
it's nothing more than an auto-generated method that returns a
resultset against another table with a where condition set for you. I
think that a lot of people are trying to create relationships where a
relationship may not be appropriate. Just chain resultsets!

Now, if someone can provide a solid use-case, then I'll be onboard.
But, adding features for the sake of adding features is how you end up
with a crappy API.

Rob