Feature #5825: Sweet instance var assignment in the object initializer

• Status: Assigned
• Assignee: matz (Yukihiro Matsumoto)
• Target version: -

[goshakkk (Gosha Arinich)]

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
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
  end
end

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

[lisovskyvlad (Vlad Lisosvky)]

I like it. No stupid assigns.

[rue (Eero Saynatkari)]

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

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

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

def foo(bar, *@baz, &@quux)
  …
end

[rosenfeld (Rodrigo Rosenfeld Rosas)]

+1 - too common use case

[shyouhei (Shyouhei Urabe)]

I liked this 1.8-specific trick:

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

but it was abondoned for any reason.

[andhapp (Anuj Dutta)]

+1. Common use case.

[alexeymuranov (Alexey Muranov)]

+1, why only initialize?

[Eregon (Benoit Daloze)]

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 < Struct.new(:name, :age, :location)
end

[goshakkk (Gosha Arinich)]

Alexey Muranov wrote:

+1, why only initialize?

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

[shevegen (Robert A. Heiler)]

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
used.

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)

[naruse (Yui NARUSE)]

• Status changed from Open to Assigned
• Assignee set to matz (Yukihiro Matsumoto)

[ko1 (Koichi Sasada)]

• 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.

[trans (Thomas Sawyer)]

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

Becomes

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

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

[ko1 (Koichi Sasada)]

• Target version changed from 2.0.0 to 2.6

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

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
     end
   end
   
   module NameOrderedArgAutoinit
     def new *args, &block
       name = args.shift
       new_instance = super *args, &block
       new_instance.instance_variable_set :@name, name
     end
   end
   
   class MyClass
     include AgeNamedArgAutoinit
     include NameOrderedArgAutoinit
   end
   

Now MyClass.new( "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.

[phluid61 (Matthew Kerwin)]

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)
  # ...
end

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.

[sikachu (Prem Sichanugrist)]

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.

[nobu (Nobuyoshi Nakada)]

• Description updated (diff)

[nobu (Nobuyoshi Nakada)]

• Category set to syntax
• Target version changed from 2.6 to 3.0

[TylerRick (Tyler Rick)]

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
  …
end

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, …)
end

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:

• http://stackoverflow.com/questions/10856191/ruby-automatically-set-instance-variable-as-method-argument
• http://stackoverflow.com/questions/9597249/in-ruby-can-i-automatically-populate-instance-variables-somehow-in-the-initializ/10855962
• http://stackoverflow.com/questions/16072965/why-do-method-arguments-not-work-for-assignment

Plus various attempts at removing the duplication from assign variables in constructors, using only pure Ruby:

• http://redsquirrel.com/cgi-bin/dave/dynamic/def_init.html (def_init :arg1, :arg2)
• http://blog.jayfields.com/2007/04/ruby-assigning-instance-variables-in.html (initializer :arg1, :arg2)
• https://github.com/rubyworks/facets/blob/master/lib/core/facets/kernel/assign.rb (assign(hash))
• https://github.com/sheldonh/magic_options (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...)

[nobu (Nobuyoshi Nakada)]

• 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.

[phluid61 (Matthew Kerwin)]

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

[nobu (Nobuyoshi Nakada)]

• Has duplicate Feature #12023: Allow ivars to be used as method arguments added

[nobu (Nobuyoshi Nakada)]

• Has duplicate Feature #12578: Instance Variables Assigned In parameters ( ala Crystal? ) added

[nobu (Nobuyoshi Nakada)]

• Has duplicate Feature #12820: Shorter syntax for assigning a method argument to an instance variable added

[mame (Yusuke Endoh)]

• Related to Feature #15192: Introduce a new "shortcut assigning" syntax to convenient setup instance variables added

[nobu (Nobuyoshi Nakada)]

• Related to deleted (Feature #15192: Introduce a new "shortcut assigning" syntax to convenient setup instance variables)

[nobu (Nobuyoshi Nakada)]

• Has duplicate Feature #15192: Introduce a new "shortcut assigning" syntax to convenient setup instance variables added

[nobu (Nobuyoshi Nakada)]

• Has duplicate Feature #16095: 2 Features: remove (simplify) 'new' keyword and Property Shorthand added

[nobu (Nobuyoshi Nakada)]

• Has duplicate deleted (Feature #16095: 2 Features: remove (simplify) 'new' keyword and Property Shorthand)

[naruse (Yui NARUSE)]

• Target version deleted (3.0)

[mame (Yusuke Endoh)]

• Related to Feature #17942: Add a `initialize(public @a, private @b)` shortcut syntax for defining public/private accessors for instance vars as part of constructor added

[[e-mail] (Eike Send)]

I would really love to see native support for an initializer instance variable assignment shorthand without adding the problems when inheriting from Struct.

[nobu (Nobuyoshi Nakada)]

• Has duplicate Feature #19898: Special syntax for instance variable assignment added

[matheusrich (Matheus Richard)]

Maybe a different proposal, but if the def initialize(@a, @b) is ugly, how about a new method similar to attr_{reader,writter} like

class User
  init_with :name, :age

  # or maybe
  attributes :name, :age
end

User.new("Matz", 21)

We could even have an option like keyword_init: true for keyword args.

On the other hand, a simple alternative is using the new Data class, which already provides this nicety:

User = Data.define(:name, :age)

User.new("matz", 21)

# or

User.new(name: "matz", age: 21)