Add option for class_attribute default #29270

dhh · 2017-05-29T14:58:32Z

It's extremely common, as evidenced by the commit in this PR of Rails conversions, to set a default value for a class_attribute. So let's provide that directly.

matthewd · 2017-05-29T15:03:38Z

actionview/lib/action_view/layouts.rb

-      class_attribute :_layout, :_layout_conditions, instance_accessor: false
-      self._layout = nil
-      self._layout_conditions = {}
+      class_attribute :_layout


Lost instance_accessor: false

matthewd · 2017-05-29T15:08:08Z

activesupport/lib/active_support/core_ext/class/attribute.rb

@@ -123,6 +124,10 @@ def class_attribute(*attrs)
        remove_possible_method "#{name}="
        attr_writer name
      end
+
+      if default_value


unless default_value.nil?

.. or just change the nil on line 79/80

dhh · 2017-05-29T15:19:29Z

actionpack/lib/action_controller/metal/helpers.rb

@@ -53,9 +53,8 @@ class << self; attr_accessor :helpers_path; end
    include AbstractController::Helpers

    included do
-      class_attribute :helpers_path, :include_all_helpers
-      self.helpers_path ||= []


Probably shouldn't replace this with a default:, right? Given that it checks the parent class first.

I hesitated at that, but ultimately convinced myself the ||= is a lie: class_attribute will always define a method on the current class, with the value set to nil... so AFAICS, this is no different from if it were a straight assignment.

Ah, right. Yup, good 👍

dhh · 2017-05-29T15:19:43Z

activesupport/lib/active_support/callbacks.rb

@@ -62,8 +62,7 @@ module Callbacks

    included do
      extend ActiveSupport::DescendantsTracker
-      class_attribute :__callbacks, instance_writer: false
-      self.__callbacks ||= {}


Same issue here: "Probably shouldn't replace this with a default:, right? Given that it checks the parent class first."

matthewd · 2017-05-29T15:33:48Z

activesupport/lib/active_support/core_ext/class/attribute.rb

    instance_predicate = options.fetch(:instance_predicate, true)
+    default_value      = options.fetch(:default, nil)


This (really, the fact it's in the options hash at all) does mean the methods below will capture the default value in their closure -- so even if the value is changed on the root class, the original default won't be GCed.

I don't think we particularly need to care about a couple of extra empty hashes and arrays floating around, but I figured I'd at least mention it.

Would be nice to convert all these to kwargs. Should probably do a big sweep for Rails 6.

We can do it right now. We only support Ruby versions that work with kwargs

simi · 2017-05-29T15:46:47Z

activesupport/lib/active_support/core_ext/class/attribute.rb

@@ -70,9 +70,10 @@ class Class
  # To opt out of both instance methods, pass <tt>instance_accessor: false</tt>.
  def class_attribute(*attrs)
    options = attrs.extract_options!
-    instance_reader = options.fetch(:instance_accessor, true) && options.fetch(:instance_reader, true)
-    instance_writer = options.fetch(:instance_accessor, true) && options.fetch(:instance_writer, true)
+    instance_reader    = options.fetch(:instance_accessor, true) && options.fetch(:instance_reader, true)


Sorry for a little off-topic here: I have spotted this formatting changes in few recent PRs. Is this new standard? Wasn't this considered "cosmetic" change before?

dhh · 2017-05-29T15:56:32Z

Not a general new standard. Just a dash of BDFL privilege 😂

…

On Mon, May 29, 2017 at 5:47 PM, Josef Šimánek ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In activesupport/lib/active_support/core_ext/class/attribute.rb <#29270 (comment)>: > @@ -70,9 +70,10 @@ class Class # To opt out of both instance methods, pass <tt>instance_accessor: false</tt>. def class_attribute(*attrs) options = attrs.extract_options! - instance_reader = options.fetch(:instance_accessor, true) && options.fetch(:instance_reader, true) - instance_writer = options.fetch(:instance_accessor, true) && options.fetch(:instance_writer, true) + instance_reader = options.fetch(:instance_accessor, true) && options.fetch(:instance_reader, true) *Sorry for a little off-topic here: I have spotted this formatting changes in few recent PRs. Is this new standard? Wasn't this considered "cosmetic" change before?* — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#29270 (review)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAAKtUkXZF1vQlGV1xuQ1Of4NX4ShM9-ks5r-uiKgaJpZM4NpZ0L> .

gsamokovarov · 2017-05-29T17:37:59Z

David, have you tried defaulting with a block?

class_attribute :periodic_timers, instance_reader: false, default: []

vs

class_attribute(:periodic_timers, instance_reader: false) { [] }

The reason I'm bringing it up is because mattr_accessor works like 👆 and does not have the default option. Should we pair the functionality so it's symmetrical across all the attribute extensions?

dhh · 2017-05-29T18:48:38Z

That strikes me as less revealing, but the upside is that it won't get evaluated until you access it, which is a nice perk. Default does have symmetry with belongs_to/default, though.

gsamokovarov · 2017-05-29T19:11:04Z

Add a default to mattr_accessor as well then? Dunno if it will make the interface too big though.

dhh · 2017-05-29T20:44:21Z

Other aspect is that class_attribute already has other named options. mattr_accessor doesn't and follows the rest of the Ruby methods by same name.

…

On May 29, 2017, at 21:11, Genadi Samokovarov ***@***.***> wrote: Add a default to mattr_accessor as well then? Dunno if it will make the interface too big though. — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or mute the thread.

dhh · 2017-05-30T08:08:34Z

Let's make it so!

…

On May 30, 2017, at 03:43, Rafael França ***@***.***> wrote: @rafaelfranca commented on this pull request. In activesupport/lib/active_support/core_ext/class/attribute.rb: > instance_predicate = options.fetch(:instance_predicate, true) + default_value = options.fetch(:default, nil) We can do it right now. We only support Ruby versions that work with kwargs — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or mute the thread.

gsamokovarov · 2017-05-30T08:20:10Z

mattr_accessor family actually accept options as well, they let you choose whether or not to have the instance methods. The interfaces between class_attributes and cattr_accessor aren't that far off.

kaspth · 2017-05-30T08:51:50Z

Let's make it so!

@dhh GitHub didn't connect your email reply: this was in reference to kwargs across the board right?

dhh · 2017-05-30T11:11:14Z

Kasper, yes.

…

On Tue, May 30, 2017 at 10:52 AM, Kasper Timm Hansen < ***@***.***> wrote: Let's make it so! @dhh <https://github.com/dhh> GitHub didn't connect your email reply: this was in reference to kwargs across the board right? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#29270 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAAKtT_2B2SEt8D8gPlrBs-vmHpVBUL-ks5r-9jKgaJpZM4NpZ0L> .

dhh · 2017-05-30T11:11:59Z

True. I'd rather make mattr_accessor use default:, then, than going with {}. For lazy resolution, we could always have default take a lambda.

…

On Tue, May 30, 2017 at 10:20 AM, Genadi Samokovarov < ***@***.***> wrote: mattr_accessor family actually accept options as well <https://github.com/rails/rails/blob/6847877a30fd8d578f72db0cf40674f71a9b6286/activesupport/lib/active_support/core_ext/module/attribute_accessors.rb#L67>, they let you choose whether or not to have the instance methods. The interfaces between class_attributes and cattr_accessor aren't actually that far off. — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#29270 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAAKtV3g5Nes_o1Dp59zdfMyw3kYvJysks5r-9FggaJpZM4NpZ0L> .

kaspth · 2017-05-30T11:20:44Z

Looks like mattr_* just yields to the block immediately, so there's no lazy loading at all:

rails/activesupport/lib/active_support/core_ext/module/attribute_accessors.rb

Line 74 in 3fbe657

class_variable_set("@@#{sym}", yield) if block_given?

Then there doesn't seem to be a need for default: to take lambdas — besides all our class_attribute didn't need deferred evaluation, and I haven't seen the need for it in my app's usage of mattr_* either.

I'd be 👍 on deprecating the block to mattr_* and cattr_* and replacing it with a default: arg.

dhh · 2017-05-30T11:32:27Z

Ah, yeah, then I definitely don't think it's worth it. A kwarg seems clearly better to me.

…

On Tue, May 30, 2017 at 1:21 PM, Kasper Timm Hansen < ***@***.***> wrote: Looks like mattr_* just yields to the block immediately, so there's no lazy loading at all: https://github.com/rails/rails/blob/ 3fbe657/activesupport/lib/ active_support/core_ext/module/attribute_accessors.rb#L74 Then there doesn't seem to be a need for default: to take lambdas — besides all our class_attribute didn't need deferred evaluation, and I haven't seen the need for it in my app's usage of mattr_* either. I'd be 👍 on deprecating the block to mattr_* and cattr_* and replacing it with a default: arg. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#29270 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAAKtUEcs1Osnx8aP9vavCQhZAVbZm_dks5r-_u2gaJpZM4NpZ0L> .

gsamokovarov · 2017-05-30T11:51:29Z

Cool, I can take stab at that.

PR rails#29270 changed the number of arguments that gets passed to Procs defined in ActionMail::Base.default. This fixes it and introduce a deprecation so the hack can be removed (the argument is useless since it has the same value as `self`).

PR rails#29270 changed the number of arguments that gets passed to Procs defined in ActionMail::Base.default. With this changeset, Procs can now have 1 or 0 arguments Also adds test coverage for AM::Base.default Proc arity.

As default value for class_attribute introduced only in Rails 5 rails/rails#29270

* Allow a default value to be declared for class_attribute * Convert to using class_attribute default rather than explicit setter * Removed instance_accessor option by mistake * False is a valid default value * Documentation

dhh added 2 commits May 29, 2017 16:56

Allow a default value to be declared for class_attribute

3662870

Convert to using class_attribute default rather than explicit setter

18ab5c3

matthewd reviewed May 29, 2017

View reviewed changes

dhh added 2 commits May 29, 2017 17:18

Removed instance_accessor option by mistake

ecc1a83

False is a valid default value

96dabf2

dhh commented May 29, 2017

View reviewed changes

matthewd reviewed May 29, 2017

View reviewed changes

simi reviewed May 29, 2017

View reviewed changes

Documentation

4eb9b55

dhh merged commit 1c275d8 into master May 29, 2017

dhh deleted the class-attribute-default branch May 29, 2017 16:01

maclover7 added the activesupport label May 29, 2017

teeceepee mentioned this pull request Jul 11, 2017

Incompatible with Rails 5.1.2 rails/activestorage#39

Closed

georgeclaghorn mentioned this pull request Jul 11, 2017

Add something like default_value for initializing class_attribute #29754

Closed

glaucocustodio mentioned this pull request Jul 11, 2017

Add documentation for class_attribute default option #29758

Merged

kennyadsl mentioned this pull request Aug 1, 2017

Allow dev mode code reloading of configured classes solidusio/solidus#2126

Merged

jbourassa mentioned this pull request Aug 24, 2017

Fix actionmailer lambda default #30391

Merged

antonzaytsev added a commit to sdtechdev/activestorage-rails4 that referenced this pull request Dec 19, 2017

Set default value of service.url_expires_in after initialization

255b9b0

As default value for class_attribute introduced only in Rails 5 rails/rails#29270

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add option for class_attribute default #29270

Add option for class_attribute default #29270

dhh commented May 29, 2017

matthewd May 29, 2017

matthewd May 29, 2017

matthewd May 29, 2017

dhh May 29, 2017

matthewd May 29, 2017

dhh May 29, 2017

dhh May 29, 2017

matthewd May 29, 2017

dhh May 29, 2017

rafaelfranca May 30, 2017

simi May 29, 2017

dhh commented May 29, 2017 via email

gsamokovarov commented May 29, 2017

dhh commented May 29, 2017

gsamokovarov commented May 29, 2017

dhh commented May 29, 2017 via email

dhh commented May 30, 2017 via email

gsamokovarov commented May 30, 2017 •

edited

kaspth commented May 30, 2017

dhh commented May 30, 2017 via email

dhh commented May 30, 2017 via email

kaspth commented May 30, 2017

dhh commented May 30, 2017 via email

gsamokovarov commented May 30, 2017

		instance_predicate = options.fetch(:instance_predicate, true)
		default_value = options.fetch(:default, nil)

Add option for class_attribute default #29270

Add option for class_attribute default #29270

Conversation

dhh commented May 29, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dhh commented May 29, 2017 via email

gsamokovarov commented May 29, 2017

dhh commented May 29, 2017

gsamokovarov commented May 29, 2017

dhh commented May 29, 2017 via email

dhh commented May 30, 2017 via email

gsamokovarov commented May 30, 2017 • edited

kaspth commented May 30, 2017

dhh commented May 30, 2017 via email

dhh commented May 30, 2017 via email

kaspth commented May 30, 2017

dhh commented May 30, 2017 via email

gsamokovarov commented May 30, 2017

gsamokovarov commented May 30, 2017 •

edited