Make AR guess the inverse of an association if not specified

+1 for sure. This would be great - DataMapper and Sequel already support correct inverse associations, so it's really something that AR should support as well.

After discussion with Eloy in IRC it seems it's useful to put a note about some of the gotchas here.

With this patch, AR will try to guess the association on a loaded model that is the inverse of the association it was loaded through, but this It has the exact same limitations as the explicit :inverse_of feature. Namely that it doesn't work for :through associations, and it won't work for belongs_to associations that have a has_many association as the inverse.

The reason for the has_many limitation is partly implementation, in that has_many associations don't have the set_XXX_target that other associations do, and that's what the current :inverse_of implementation uses to set inverses. The other reason for the limitation is that it's not hugely clear what exactly we should do. Given:

  class Car < ActiveRecord::Base
    has_many :wheels
  end
  
  class Wheel < ActiveRecord::Base
    belongs_to :car
  end

When we do:

  car = Car.first
  wheels = car.wheels

It's obvious what should be set as an inverse, each instance of Wheel should have the Car we loaded first set as the target of their car association.

If we do:

  wheel = Wheel.first
  car = wheel.car
  wheels = car.wheels

It's not particularly obvious what should be in this wheels array. Should we have all the wheels associated to the car loaded from the DB, apart from the one we started with? Should we just have the wheel we started with? Clearly the first option is correct, it's just hard to see the benefit of using the pre-loaded wheel here. We can't avoid a DB query to find out which other wheels we need and then inserting the wheel we already have into the set of objects returned from that query is extra work.

It seems more pragmatic to ignore the belongs_to -> has_many case for now until we have a clear solution or full identity map.

As discussed with Murray on IRC, I'm gonna take a look later this week if we can also solve the problem with autosave as outlined.

Also inviting Jeremy to the ticket.