Allow to override the full_message error format

[Larochelle]

Summary

The goal of this PR is to make it easier for an app to transition from a #{attribute} #{message} to a #{message}, full_message error format.

full_message formats error messages with a #{attribute} #{message} format which generates error messages as “name cannot be nil”.

It is possible to override the format with :"errors.format" but only language wide. https://github.com/rails/rails/blob/master/activemodel/lib/active_model/errors.rb#L371 allows a language to define a custom format, but changing it forces to extract all messages, including the implicit ones as :blank

The #{attribute} #{message} format prevents languages to move the attribute name to somewhere else within the message, for example, in some cases it can be preferred to have error messages as “The person's name cannot be blank”, so changing the format to #{message} and including the attribute name in the message itself can be preferable.

To make such transition easier, this PR allows to override errors.format at the attribute or model level, so an app can use either:

en:
  errors:
    format: '%{message}'

en:
  activemodel:
    errors:
      models:
        person:
          format: '%{message}'

en:
  activemodel:
    errors:
      models:
        person:
          attributes:
            name:
              format: '%{message}'

How

This is based on active_model/translation.rb#human_attribute_name
https://github.com/rails/rails/blob/master/activemodel/lib/active_model/translation.rb#L44

And active_model/errors.rb#generate_message
https://github.com/rails/rails/blob/master/activemodel/lib/active_model/errors.rb#L401

• The @base.class.respond_to?(:i18n_scope) is needed for cases where ActiveModel::Errors is used without including the Translation module

Benchmark

require "active_model"

module ActiveModel
  class Errors
    def full_message_i18n(attribute, message)
      return message if attribute == :base

      if @base.class.respond_to?(:i18n_scope)
        parts = attribute.to_s.split(".")
        attribute_name = parts.pop
        namespace = parts.join("/") unless parts.empty?
        attributes_scope = "#{@base.class.i18n_scope}.errors.models"

        if namespace
          defaults = @base.class.lookup_ancestors.map do |klass|
            [
              :"#{attributes_scope}.#{klass.model_name.i18n_key}/#{namespace}.attributes.#{attribute_name}.format",
              :"#{attributes_scope}.#{klass.model_name.i18n_key}/#{namespace}.format",
            ]
          end
        else
          defaults = @base.class.lookup_ancestors.map do |klass|
            [
              :"#{attributes_scope}.#{klass.model_name.i18n_key}.attributes.#{attribute_name}.format",
              :"#{attributes_scope}.#{klass.model_name.i18n_key}.format",
            ]
          end
        end

        defaults.flatten!
      else
        defaults = []
      end

      defaults << :"errors.format"
      defaults << "%{attribute} %{message}"

      attr_name = attribute.to_s.tr(".", "_").humanize
      attr_name = @base.class.human_attribute_name(attribute, default: attr_name)
      I18n.t(defaults.shift,
        default:  defaults,
        attribute: attr_name,
        message:   message)
    end
  end
end

class Person
  include ActiveModel::Validations
  extend  ActiveModel::Translation

  attr_accessor :title, :karma, :salary, :gender

  def condition_is_true
    true
  end

  def condition_is_false
    false
  end
end

person = Person.new

# Enumerate some representative scenarios here.
#
# It is very easy to make an optimization that improves performance for a
# specific scenario you care about but regresses on other common cases.
# Therefore, you should test your change against a list of representative
# scenarios. Ideally, they should be based on real-world scenarios extracted
# from production applications.
SCENARIOS = {
  "Deeply nested model attributes" => {
    attribute: :'contacts/addresses.street',
    translations: { errors: { format: "%{attribute} %{message}" } }
  },
  "Attribute with base format" => {
    attribute: :name,
    translations: { errors: { format: "%{attribute} %{message}" } }
  },
  "Attribute with attribute format" => {
    attribute: :name,
    translations: { activemodel: { errors: { models: { person: { attributes: { name: { format: "%{message}" } } } } } } }
  }
}

SCENARIOS.each_pair do |name, value|
  attribute = value[:attribute]
  translations = value[:translations]

  I18n.load_path.clear
  I18n.backend = I18n::Backend::Simple.new
  I18n.backend.store_translations("en", translations)

  puts
  puts " #{name} ".center(80, "=")
  puts

  Benchmark.ips do |x|
    x.report("full_message_i18n") { person.errors.full_message_i18n(attribute, "cannot be blank") }
    x.report("full_message") { person.errors.full_message(attribute, "cannot be blank") }
    x.compare!
  end
end

Results

The result with the variance between runs

======================== Deeply nested model attributes ========================

Warming up --------------------------------------
   full_message_i18n     1.011k i/100ms
        full_message     1.423k i/100ms
Calculating -------------------------------------
   full_message_i18n     10.145k (± 3.9%) i/s -     51.561k in   5.090447s
        full_message     14.456k (± 3.9%) i/s -     72.573k in   5.028127s

Comparison:
        full_message:    14456.1 i/s
   full_message_i18n:    10144.8 i/s - 1.38x - 1.44x  slower


========================== Attribute with base format ==========================

Warming up --------------------------------------
   full_message_i18n     1.203k i/100ms
        full_message     1.834k i/100ms
Calculating -------------------------------------
   full_message_i18n     12.047k (± 4.0%) i/s -     60.150k in   5.000757s
        full_message     17.978k (± 1.9%) i/s -     89.866k in   5.000616s

Comparison:
        full_message:    17977.9 i/s
   full_message_i18n:    12047.3 i/s - 1.49x - 1.65x  slower


======================= Attribute with attribute format ========================

Warming up --------------------------------------
   full_message_i18n     1.553k i/100ms
        full_message     1.772k i/100ms
Calculating -------------------------------------
   full_message_i18n     16.465k (± 3.9%) i/s -     82.309k in   5.006884s
        full_message     17.041k (± 2.7%) i/s -     86.828k in   5.099021s

Comparison:
        full_message:    17040.9 i/s
   full_message_i18n:    16464.7 i/s - same-ish: difference falls within error to  1.20x  slower

[sikachu]

I think this is a good patch. I personally found myself having to find way to workaround this a few times. It'd be nice to be able to do this in i18n config.

wdyt @rafaelfranca @matthewd ?

[rafaelfranca]

Weird. Why did you need to add this? Active Model is not a ORM.

[Larochelle]

Removed.
Looking the active_record and http://api.rubyonrails.org/classes/Rails/Railtie.html as examples, I though it was needed.

[rafaelfranca]

Let's document this option in the configuring guide https://github.com/rails/rails/blob/master/guides/source/configuring.md.

[Larochelle]

Added

[bogdanvlviv]

Is it added intentionally?

It seems extra since we set default value of i18n_full_message in activemodel/lib/active_model/railtie.rb.

diff --git a/activemodel/lib/active_model/errors.rb b/activemodel/lib/active_model/errors.rb
index 56404a036c..7400bb5a6c 100644
--- a/activemodel/lib/active_model/errors.rb
+++ b/activemodel/lib/active_model/errors.rb
@@ -65,7 +65,6 @@ class Errors
     class << self
       attr_accessor :i18n_full_message # :nodoc:
     end
-    self.i18n_full_message = false

     attr_reader :messages, :details

diff --git a/activemodel/lib/active_model/railtie.rb b/activemodel/lib/active_model/railtie.rb
index 0ed70bd473..ed70d65c8e 100644
--- a/activemodel/lib/active_model/railtie.rb
+++ b/activemodel/lib/active_model/railtie.rb
@@ -8,13 +8,14 @@ class Railtie < Rails::Railtie # :nodoc:
     config.eager_load_namespaces << ActiveModel

     config.active_model = ActiveSupport::OrderedOptions.new
+    config.active_model.i18n_full_message = false

     initializer "active_model.secure_password" do
       ActiveModel::SecurePassword.min_cost = Rails.env.test?
     end

     initializer "active_model.i18n_full_message" do
-      ActiveModel::Errors.i18n_full_message = config.active_model.delete(:i18n_full_message) || false
+      ActiveModel::Errors.i18n_full_message = config.active_model.delete(:i18n_full_message)
     end
   end
 end

[rafaelfranca]

Yes, it is intentional. We want a default value if the railtie is not loaded, for example if you use Active Model without Rails.

[bogdanvlviv]

It makes sense then. Thanks for explanation.
What do you think about making ActiveModel::Errors.i18n_full_message as public API in order to allow users use this new feature if use Active Model without Rails?