Feature #5825

Sweet instance var assignment in the object initializer

Added by Gosha Arinich over 4 years ago. Updated about 2 years ago.



I'm very excited about this feature in CoffeeScript, and think it might be a nice-to-have thing in Ruby 2.0.

That's how I think it would look like:

class Me
  def initialize(@name, @age, @location); end

So we can declare @variables in the initializer method parameters definition to avoid assigning instance variables from method arguments by hand, like:

class Me
  def initialize(name, age, location)
    @name = name
    @age = age
    @location = location

Want to hear what do you guys think, does that feature worth being included in 2.0?

Related issues

Duplicated by Ruby trunk - Feature #8563: Instance variable arguments Assigned
Duplicated by Ruby trunk - Feature #12023: Allow ivars to be used as method arguments Open


#1 Updated by Vlad Lisosvky over 4 years ago

I like it. No stupid assigns.

#2 [ruby-core:41850] Updated by Eero Saynatkari over 4 years ago

Would be nice, and should be able to coexist with normal parameters:

def foo(bar, @baz, quux = @moomin)

And so on. What about splat- and block arguments? It gets a little ugly:

def foo(bar, *@baz, &@quux)

#3 [ruby-core:41851] Updated by Rodrigo Rosenfeld Rosas over 4 years ago

+1 - too common use case

#4 [ruby-core:41859] Updated by Shyouhei Urabe over 4 years ago

I liked this 1.8-specific trick:

define_method(:intialize){|@foo, @bar|}

but it was abondoned for any reason.

#5 Updated by Anuj Dutta over 4 years ago

+1. Common use case.

#6 [ruby-core:41866] Updated by Alexey Muranov over 4 years ago

+1, why only initialize?

#7 [ruby-core:41868] Updated by Benoit Daloze over 4 years ago

I think most of the time you need to parse or check your arguments, in which case this syntax would not be practical.
Otherwise, you could use Struct to avoid the duplication:

class Me <, :age, :location)

#8 Updated by Gosha Arinich over 4 years ago

Alexey Muranov wrote:

+1, why only initialize?

It's just too common case. It could work for other methods as well.

#9 Updated by Robert A. Heiler over 4 years ago

Is this even possible with the Parser?

It would eliminate a few lines of code.

Also, I am not sure if this violates the principle of least matz surprise.

If I initially see

def initialize(@name, @age, @location)

I wonder a bit, because it does not feel consistent.

Perhaps it would be different if some kind of attr* could be

attr_initialize :name, :age, :location

And then the above could work. Where the name would work just
similar to an attr_writer, but different in that it assumes
default values passed to initialize to automatically go towards
those instance variables (the order must be the same of course)

#10 [ruby-core:41914] Updated by Yui NARUSE over 4 years ago

  • Status changed from Open to Assigned
  • Assignee set to Yukihiro Matsumoto

#11 [ruby-core:42896] Updated by Koichi Sasada about 4 years ago

  • Target version set to 2.0.0

if my memory serves me right, matz dislikes such a style of arguments.
Matz, could you explain the reason again?

I know some people like this style.

#12 [ruby-core:42916] Updated by Thomas Sawyer about 4 years ago

I think it's really not such a good idea. Often you'll just end up having to redo it anyway when you finally decide to coerce and/or validate arguments to make your code more robust, e,g,

def initialize(@name, @age, @location); end


def initialize(name, age, location)
  @name = name.to_s
  @age = age.to_i
  @location =

Might as well write it out from the get-go in preparation.

#13 [ruby-core:48363] Updated by Koichi Sasada over 3 years ago

  • Target version changed from 2.0.0 to next minor

I changed target to next minor.
I think someone who want to introduce it need to persuade matz.

#14 [ruby-core:50704] Updated by Anonymous over 3 years ago

Well... I like the sweetness... But to have such a feature working syntactically from
inside of #initialize method, but not from other methods... I don't know.

It's not like this is my suggestion, but since it is up for discussion, let me try to
straighten this proposal according to my own thinking:

These "attributes on steroids" are a thing to be done not at the level of the #initialize
method, but at the level of the module, along with #attr_accessor and friends.

Imho, it would be necessary to update #attr_accessor & friends to accept :autoinit
named argument:

attr_reader :name, :age, autoinit: true

Now there are two flavors of this candy, one with ordered arguments, one with named.
So we could have to specify it:

attr_reader :name, autoinit: :ordered
attr_reader :age, autoinit: :named

I'm sure you know what I mean here. Default option could be eg. :named.

Also, #autoinit method, or rather, #autoinit_named, #autoinit_ordered,
would have to be added to the Module, for those times, when we want to
autoinit, but don't want the reader/writer/accessor:

autoinit_named :age
autoinit_ordered :name

Afaik, current #attr_accessor & friends work by defining instance methods on
the module. How #autoinit should work, is a question. Two possibilities come
to my mind:

  1. By patching #initialize method.

  2. By creating and including a mixin patching #new class method, that would
    set the appropriate instance variables right after creating a new instance,
    in effect something like this:

     module AgeNamedArgAutoinit
       def new *args, &block
         named_args = args.extract_options!
         age = named_args.delete :age
         modified_args = args + named_args.empty? ? [] : [named_args]
         new_instance = super *modified_args, &block
         new_instance.instance_variable_set :@age, age
     module NameOrderedArgAutoinit
       def new *args, &block
         name = args.shift
         new_instance = super *args, &block
         new_instance.instance_variable_set :@name, name
     class MyClass
       include AgeNamedArgAutoinit
       include NameOrderedArgAutoinit

Now "John Smith", :whatever, age: 35, other_stuff: :whatever ) should
behave in the expected way.

Again, I have not come up with this proposal, I do not give +1 or -1 to it,
I am only trying to iron it to be more consistent, leaving the decision to others.

#15 [ruby-core:50705] Updated by Matthew Kerwin over 3 years ago

alexeymuranov (Alexey Muranov) wrote:

+1, why only initialize?

I agree. Is there a reason not to specify something other than a local variable as the receiver for a method parameter?

For example:

# precondition: ???
# postcondition: updates @instance_var and $global_var
def some_thing(local_var, @instance_var, $global_var)
  # ...

I know it's not safe, in that it makes it quite easy to shoot one's self in the foot, but I don't know that it's necessarily a bad thing unless someone tells me it is.

#16 [ruby-core:52900] Updated by Prem Sichanugrist about 3 years ago

I think this is a good feature, so I'd like to support this (and possibly, provide a patch for this)

Reading from all the comments, I saw that someone has some concern about having this feature on another method definition (not just initialize) as well. I think we should implement it for any type of method definition if that's going to make the code cleaner, but the main focus here is for the initialize method.

As per comment 12, I think it's OK to start up with def initialize(@foo, @bar, @baz) and then refactor it if you need to perform any method on those arguments before store it to an instance variable.

I think adding support for only local and instance variables make sense, and I think we need to make sure that we're not supporting global variable like in comment 15.

Anyway, the benefit I'm seeing here is that we're not cluttering the initializer with all those obvious instance variable assignments. I also think it's a good syntactic sugar and it make sense after I saw all the usage from CoffeeScript. I hope I can help to make this happen in the next minor.

#17 [ruby-core:55597] Updated by Nobuyoshi Nakada almost 3 years ago

  • Description updated (diff)

#18 [ruby-core:55599] Updated by Nobuyoshi Nakada almost 3 years ago

  • Target version changed from next minor to Next Major
  • Category set to syntax

#19 [ruby-core:61475] Updated by Tyler Rick about 2 years ago

I would love to see this feature in Ruby. Assigning an argument to an instance variable in a constructor is something that we do in almost every constructor we write, so I think this should be made as easy and simple as possible, by adding a little syntactic sugar to the language.

I shouldn't have to repeat myself and type out each argument name 3 times in every constructor I write, for something as mundane as this:

def initialize(name₁, …)
  @name₂ = name₃

This constant repetition feels inelegant to me and goes against one of the Ruby community's most fundamental values (Don't Repeat Yourself).

This method could be simplified to simply this:

def initialize(@name, …)

I think CoffeeScript solved this problem quite nicely. Many constructors in CoffeeScript end up being beautiful, simple one-liners!

constructor: (@name) ->

And with the rising popularity of CoffeeScript, there are going to be more and more Rubyists not only wishing for this but also expecting this same feature to exist in Ruby as well. :)

Here are a few more "votes" for this feature:

Plus various attempts at removing the duplication from assign variables in constructors, using only pure Ruby:
* (def_init :arg1, :arg2)
* (initializer :arg1, :arg2)
* (assign(hash))
* (magic_initialize and magic_options(hash))
(But pure Ruby solutions can only go so far, and none of those solutions really solve the problem nicely enough...)

#20 [ruby-core:61483] Updated by Nobuyoshi Nakada about 2 years ago

  • Description updated (diff)

Tyler Rick wrote:

I think CoffeeScript solved this problem quite nicely. Many constructors in CoffeeScript end up being beautiful, simple one-liners!

constructor: (@name) ->

Sorry, it doesn look beautiful to me.

#21 [ruby-core:61485] Updated by Matthew Kerwin about 2 years ago

Does anyone have a link to discussions/logs that lead to the decision to remove instance/global variables from block parameters?

#22 Updated by Nobuyoshi Nakada 4 months ago

  • Duplicated by Feature #12023: Allow ivars to be used as method arguments added

Also available in: Atom PDF