Require belongs_to by default.

[simi]

Deprecate required option in favor of optional for belongs_to.

closes #18233

[simi]

/cc @dhh @al2o3cr

[simi]

I wonder what to do with failing test.

rails generate scaffold LineItems product:references cart:belongs_to
bundle exec rake db:migrate test`

This code fails now, since belongs_to relation is required by default. Should I change generated test for belongs_to to make it pass or should I remove this test since it is not valid with this change?

[simi]

Better info will be needed here. Any ideas?

[dhh]

'required: true is now the default for belongs_to and can be removed.' – don't even need to talk about optional here. What they wanted to do is now being done for them.

[simi]

What about when required: false is specified explicitly? Maybe throw deprecation warning only for false value?

[dhh]

That was the old default, which means no client code will have that code, since it doesn’t make any sense.

On Feb 15, 2015, at 12:54 PM, Josef Šimánek notifications@github.com wrote:

In activerecord/lib/active_record/associations/builder/belongs_to.rb #18937 (comment):

@@ -110,5 +110,24 @@ def self.add_destroy_callbacks(model, reflection)
name = reflection.name
model.after_destroy lambda { |o| o.association(name).handle_dependency }
end
+

• def self.define_validations(model, reflection)

•  if reflection.options.key?(:required)
  

•    ActiveSupport::Deprecation.warn('`required` option for belongs_to relation is deprecated and replaced with `optional` option.')
  

  What about when required: false is specified explicitly? Maybe throw deprecation warning only for false value?

—
Reply to this email directly or view it on GitHub https://github.com/rails/rails/pull/18937/files#r24725940.

[simi]

OK, makes sense.

[dhh]

You can kill that test and replace it with a test that tests optional: true.

On Feb 14, 2015, at 3:24 AM, Josef Šimánek notifications@github.com wrote:

I wonder what to do with failing test.

rails generate scaffold LineItems product:references cart:belongs_to
bundle exec rake db:migrate test`
This code fails now, since belongs_to relation is required by default. Should I change generated test for belongs_to to make it pass or should I remove this test since it is not valid with this change?

—
Reply to this email directly or view it on GitHub #18937 (comment).

[simi]

Should I add also this to generator? rails g scaffold post author:belongs_to:optional ?

[dhh]

Nah, don’t think that’s necessary. Will be a very uncommon option to use, I think.

On Feb 18, 2015, at 14:30, Josef Šimánek notifications@github.com wrote:

Should I add also this to generator? rails g scaffold post author:belongs_to:optional ?

—
Reply to this email directly or view it on GitHub.

[simi]

OK I think my work is almost done here. I'll add also CHANGELOG note soon.

[dhh]

Awesome. Ping when ready to merge.

On Feb 18, 2015, at 15:35, Josef Šimánek notifications@github.com wrote:

OK I think my work is almost done here. I'll add also CHANGELOG note soon.

—
Reply to this email directly or view it on GitHub.

[simi]

I added CHANGELOG note, but it is a little fuzzy to me. Feel free to ping me with better one.

[dhh]

Let's try with:

belongs_to will now trigger a validation error by default if the association is not present. You can turn this off on a per-association basis with optional: true. (Note this new default only applies to new Rails apps that will be generated with config.active_record.require_belongs_to = true)

[simi]

Actually the final note is not true right now. config.active_record.require_belongs_to = true is set directly in railtie, not in application config. Should I move it to new generated application.rb and let it nil for existing applications? In that case your note will be true.

[dhh]

Let's update the config option to be a bit clearer: belongs_to_required_by_default = true

[dhh]

This is also missing the config/initializers/active_record_belongs_to_required_by_default.rb file that's going to set the belongs_to_required_by_default = true setting by default for new apps.

[simi]

@dhh I mentioned that https://github.com/rails/rails/pull/18937/files#r24953865 already.

[dhh]

Ah, we crossed streams. Didn't note that one. Yes, please do go ahead. Thanks!

[jeremy]

I'm finding the control flow here a little difficult to follow. Consider condensing the conditionals and using the :optional option to determine whether to declare the presence validation:

# Disallow conflicting options
if reflection.options.key?(:required) && reflection.options.key?(:optional)
  raise ArgumentError, "complain if we have conflicting options"
# Deprecate :required
elsif reflection.options.key?(:required)
  reflection.options[:optional] = !reflection.options[:required]
  ActiveSupport::Deprecation.warn "switch to `optional: #{reflection.options[:optional]}`"
# Set default :optional option
elsif !reflection.options.key?(:optional)
  reflection.options[:optional] = !model.require_belongs_to
end

super

if !reflection.options[:optional]
  model.validates_presence_of …
end

[simi]

@jeremy I try to rethink this flow, but deprecation was marked as unnecessary by @dhh here.

[jeremy]

Nice work @simi. Pardon the repeated remarks, looks like I missed concurrent feedback too. Thanks!

[simi]

Nevermind @jeremy. Thanks for review.

I have just one more question. I made change where config.active_record.require_belongs_to = true is generated to application.rb for new apps. But it is only one uncommented (unless :skip_active_record) configuration in that file. Should I left it there or should it be better to move into custom initializer?

[simi]

I bet it will be better as initializer after inspecting currently generated initializers.

[simi]

@jeremy @dhh I pushed new commit (I can squash later, just to highlight new code). config.active_record.belongs_to_required_by_default = true is now part of config/initializers/active_record_belongs_to_required_by_default.rb.

This initializer is generated only when ActiveRecord is not skipped and is not added for rake rails:update. I think this is compatible enough to skip deprecation.

[Silex]

I feel this change was a bit of a mistake at the design level... validations and relations are different things, and now we get validations done when defining relations.

I understand the "it's convenient" argument tho.

[Silex]

Typo: "then associated object will NOT be validated for presence"

[simi]

optional: By default, this is false: associated objects will be validated for presence.

[Silex]

I'm talking about the first part. Look more closely, the sentence says the same thing twice:

If you set the `:optional` option to `true`, then associated object 
will be validated for presence. By default, this is `false`: associated objects 
will be validated for presence.

I took the original sentence and just modified the alignement so it's visually clear they are the same.

[Silex]

Ok, this was fixed in 803ef74

[lekansogunle]

We are only testing for validation of the presence of the association ID, what if the given ID is invalid. And app behaviour is not to validate the existence of such association?
https://robots.thoughtbot.com/referential-integrity-with-foreign-keys