nested_attributes: validates_uniqueness_of fails

hmm.. this only happens with "@product.categorizations.build" in the products_controller.

I noticed something else. If I create/edit a product, choose a category and, lets say, type in an invalid product title, i get an "missing product_id" error for the chosen category.

All these bugs were not in 2.3-RC1.. that's really annoying. :/

Any news on this?

This is really a problem with a core rails philosophy. It assumes that, no matter what, if "valid?" returns true, then it will save successfully. That is a false presumption.

I think rails needs to completely rethink validation and incorporate database constraint detection. And it should go on the assumption that it might return true for "valid?", but not for "save". This is obviously beyond the scope of this ticket, but this is a pretty big problem.

I have the same problems. Shortly said: validations are not taken into account when accept_nested_attributes_for is used. I will give my problem, since I use a dirty trick on the column. I think it is not the cause, but good to add anyway: My 'short' situation:

class Employee < ActiveRecord::Base
require 'digest/sha1' attr_protected :id, :salt attr_accessor :password, :password_confirmation

belongs_to :address accepts_nested_attributes_for :address # Always have an address as employee alias :old_address :address def address

end

def validate_on_create

end ... end

When a hash containing :address_attributes => {.....} is given to Employee.new, and no password or password_confirmation, it will save => true. Without the :address_attri butes the validations work.

@Andrew Thanks for your comment. I'm still thinking about if and how we could fix this as simple as possible, but it might also be handy for people if you could document your workaround in either the NestedAttributes or validates_uniqueness_of docs.

Care to write such a doc patch and do you think it's a valid approach?

Hi Eloy, I think my workaround is a bit of a kludge:

class Author
  has_many :books

  # Could easily be made a validation-style class method of course
  validate :validate_unique_books

  def validate_unique_books
    validate_uniqueness_of_in_memory(
      books, [:title, :isbn], 'Duplicate book.')
  end
end

module ActiveRecord
  class Base
    # Validate that the the objects in +collection+ are unique
    # when compared against all their non-blank +attrs+. If not
    # add +message+ to the base errors.
    def validate_uniqueness_of_in_memory(collection, attrs, message)
      hashes = collection.inject({}) do |hash, record|
        key = attrs.map {|a| record.send(a).to_s }.join
        if key.blank? || record.marked_for_destruction?
          key = record.object_id
        end
        hash[key] = record unless hash[key]
        hash
      end
      if collection.length > hashes.length
        self.errors.add_to_base(message)
      end
    end
  end
end

If this is useful to people I'd be happy to tidy it up and document it somewhere, not really worth making a plugin out of it, I just put it in my initializers. It moves the validation responsibility to the parent object, which is probably bad too, but I don't think there's any way for a new instance to find out about its other unsaved instances.

Best of luck finding an elegant solution though, its got me stumped!

OMG Lighthouse is so shit, sorry about the mess.

I've fixed the formatting there.

Same problem on rails (2.3.2), validates_uniqueness_of is bypassed.

Yes! People are still running into problem. Why would the status be "invalid" if a validation is not working as it is supposed to be?

Why would the status be "invalid" if a validation is not working as it is supposed to be?

How about reading the comments? (that's rhetorical btw.)

It works as it currently should:

“the problem is simply that validates_uniqueness_of cannot work with nested attributes because the validation only checks against saved records”

If you have a good solution/patch to your own problem, please share it with us. But your message as is, doesn't add much, sorry.

I run into similar BUT not SAME problem. Fully explained here: http://railsforum.com/viewtopic.php?id=38343 , I attached there quick fix, but I think it is not elegant. Basing on Eloy summary I assume that is not already any "built in" solution, to use validates_uniqueness_of at model which is used like "dictionary" in context of accepts_nested_attributes_for? I think that filter could somehow discover from what context is it fired up and then check given data against DB and decide is @new_record or not.

Ran into exactly the same problem. Had a uniqueness constraint on a date of a ProductRate, scoped for a Product, but it wasn't validated. Andrew's patch works like a charm.