conditional validates_associated

[Rafael C. de Almeida]

Hello,

Have you guys noticed that conditional validation with validates_associated does not work well when you are creating a new record?

Consider this gist: https://gist.github.com/aflag/4780225 

The Lawyer class has validates_associated on address conditioned on whether the Lawyer data comes from a known source or not. So, if lawyer.source equals to some string, then lawyer.address must not be validated. However, if I'm trying to create a new record, it is validated regardless of whether source is nil or not. Is it expected to behave like that? Is that a bug?

Cheers,

Rafael

[Matt Jones]

The behavior is intentional, though perhaps not documented as well as it should be. It has to do with the autosave behavior for has_many on new records - the part that enables you to build a new record, add some unsaved child records (phones, etc in your example) and then save the parent object and get all the others saved as well.

You should be able to turn off this validation by passing 'validate: false' to your has_many associations as needed.

--Matt Jones 

[Rafael C. de Almeida]

On Wednesday, February 13, 2013 12:59:11 PM UTC-2, Matt Jones wrote:

On Tuesday, 12 February 2013 20:29:49 UTC-5, Rafael C. de Almeida wrote:

Hello,

Have you guys noticed that conditional validation with validates_associated does not work well when you are creating a new record?

Consider this gist: https://gist.github.com/aflag/4780225 

The Lawyer class has validates_associated on address conditioned on whether the Lawyer data comes from a known source or not. So, if lawyer.source equals to some string, then lawyer.address must not be validated. However, if I'm trying to create a new record, it is validated regardless of whether source is nil or not. Is it expected to behave like that? Is that a bug?

The behavior is intentional, though perhaps not documented as well as it should be. It has to do with the autosave behavior for has_many on new records - the part that enables you to build a new record, add some unsaved child records (phones, etc in your example) and then save the parent object and get all the others saved as well.

I see. Thank you for the clarification. However, isn't that behavior counter-intuitive? Why should "valid?" return a different thing depending on whether a nested model is saved or not? If you want to skip the validation of one nested model, but still validate other things, then your only option is doing something like:

    object.save validate:false

    object.reload

    object.valid? || object.delete

That also has the potential problem of leaving some of the nested model data behind in the database.

Maybe there's some gem that changes that behavior. Anyhow, at least now I know what's going on, I can come up with a workaround.

Cheers,

Rafael

[Matt Jones]

On Wednesday, 13 February 2013 15:41:49 UTC-5, Rafael C. de Almeida wrote:

On Wednesday, February 13, 2013 12:59:11 PM UTC-2, Matt Jones wrote:

On Tuesday, 12 February 2013 20:29:49 UTC-5, Rafael C. de Almeida wrote:

Hello,

Have you guys noticed that conditional validation with validates_associated does not work well when you are creating a new record?

Consider this gist: https://gist.github.com/aflag/4780225 

The Lawyer class has validates_associated on address conditioned on whether the Lawyer data comes from a known source or not. So, if lawyer.source equals to some string, then lawyer.address must not be validated. However, if I'm trying to create a new record, it is validated regardless of whether source is nil or not. Is it expected to behave like that? Is that a bug?

The behavior is intentional, though perhaps not documented as well as it should be. It has to do with the autosave behavior for has_many on new records - the part that enables you to build a new record, add some unsaved child records (phones, etc in your example) and then save the parent object and get all the others saved as well.

I see. Thank you for the clarification. However, isn't that behavior counter-intuitive? Why should "valid?" return a different thing depending on whether a nested model is saved or not? If you want to skip the validation of one nested model, but still validate other things, then your only option is doing something like:

Better option: don't save records to the DB that are invalid. That's typically considered a bad thing, so Rails makes it hard to do.

I'd recommend extending the "extracted?" method into your related models:

class Lawyer < ActiveRecord::Base

  has_many :phones, :inverse_of => :lawyer

...

end

class Phone < ActiveRecord::Base

  belongs_to :lawyer, :inverse_of => :phones

  validates_something_of :foo, :if => :extracted?

  def extracted?

    lawyer.extracted?

  end

end

This has the advantage of making the validations you want to skip *explicit* (they get the :if => :extracted? condition) instead of just skipping them entirely when needed.

 

--Matt Jones