nested models: build should directly assign the parent

We were thinking that the fix could be something like...

in nested_attributes.rb, assign_nested_attributes_for_one_to_one_association (around line 250) and like methods to add the reference to the parent:

assn = send("build_#{association_name}", attributes.except(*UNASSIGNABLE_KEYS)) assn.send("#{parentname}=", self)

not sure how to get the 'parentname' ...

Here is the patch. It only adds tests.

For now, the temporary workaround until Rails 2.3.4 is released would be:

@@@ruby p = Person.new(...) p.save(false) # save without validation if p.valid?

#...

else

#...

end

Very nice Wolfram!

Would you mind moving the last test, which tests nested attributes, to the nested_attributes_test.rb file?

Thanks

Done. Thanks for the suggestion and sorry for the slow turn-around.

damn it ... did this just miss 2.3.3?

@Wolfram: Unfortunately I'm unable to see your patch atm. It seems there's an issue with LH. If you could try again or maybe add a pastie/gist, I can verify it on the master branch.

@Bodaniel: The problem is that the :inverse_of patch hasn't been applied to the 2-3 branch, and won't be in the near future as I understand. So until then, this will stay an issue. Feel free to ask on the mailing list if the :inverse_of patch can be applied on the 2-3 branch, or why it isn't being applied.

@Eloy: OK, here I'm trying the attachment again, and here is a pastie link: http://pastie.org/543837 Thanks, Wolf

As the author of the :inverse_of patch, I have to say I agree with it not slipping into a 2.3.x release when it didn't make 2.3.0. That said, all is not lost, you could instead use the plugin that I extracted the :inverse_of code from: http://github.com/h-lame/parental_control/.

@Myrray: Thanks for the pointer to the plugin; that'll solve our problem. I agree with you that given the fundamental nature of the background of :inverse_of with object mapping and identity this is probably not adequate for a dot release.

(from [73f2d37505025a446bb5314a090f412d0fceb8ca]) Add test to verify that the new :inverse_of association option will indeed fix the validation problem for a belongs_to relationship that validates_presence_of the parent, when both the parent and the child are new (in-memory) records. Also check that this works when the parents adds child via nested_attributes_for.

Lastly, add a require 'models/pet' to association_validation_test.rb, so that test can be run independently (was failing due to that missing dependency). [#2815 status:resolved]

Signed-off-by: José Valim jose.valim@gmail.com
http://github.com/rails/rails/commit/73f2d37505025a446bb5314a090f41...

The following workaround has been fine for me:

class Person < ActiveRecord::Base

  has_many :children
  accepts_nested_attributes_for :children
  before_validation_on_create :initialize_children

  def initialize_children
    children.each { |c| c.parent = self }
  end

end

class Child < ActiveRecord::Base

  belongs_to :person

end

Bug #3588 is related to this - :autosave => true + accepts_nested_attributes_for = circular validation & blown stack.

+1 for that being documented. One point though here http://guides.rubyonrails.org/activerecord_validations_callbacks.ht... its said "If you want to be sure that an association is present, you’ll need to test whether the foreign key used to map the association is present, and not the associated object itself." which contradicts your "validates_presence_of :member"

This bug is not resolved in the 2.x branch (at least as of 2.3.10).