Project

General

Profile

Actions

Feature #12115

open

Add Symbol#call to allow to_proc shorthand with arguments

Added by felixbuenemann (Felix Bünemann) over 8 years ago. Updated almost 6 years ago.

Status:
Open
Assignee:
-
Target version:
-
[ruby-core:74011]

Description

I am a great fan of the Symbol#to_proc shorthand when mapping or reducing collections:

[1,2,16].map(&:to_s)
=> ["1", "2", "16"]
[1,2,16].reduce(&:*)
=> 32

I often wish it would be possible to pass an argument to the method when doing this, which currently requires a block and is more verbose:

[1,2,16].map { |n| n.to_s(16) }
=> ["1", "2", "10"]
# active_support example
{id: 1, parent_id: nil}.as_json.transform_keys { |k| k.camelize :lower }.to_json
=> '{"id":1,"parentId":null}'

It would be much shorter, if ruby allowed this:

[1,2,16].map(&:to_s.(16))
=> ["1", "2", "10"]
# active_support example
{id: 1, parent_id: nil}.as_json.transform_keys(&:camelize.(:lower)).to_json
=> '{"id":1,"parentId":null}'

This can be implemented easily, by adding the Symbol#call method:

class Symbol
  def call(*args, &block)
    ->(caller, *rest) { caller.send(self, *rest, *args, &block) }
  end
end

Source: stackoverflow: Can you supply arguments to the map(&:method) syntax in Ruby?

I think this is a rather common use case, so I propose to add Symbol#call to the Ruby standard library.


Related issues 2 (0 open2 closed)

Related to Ruby master - Feature #4146: Improvement of Symbol and ProcRejectednobu (Nobuyoshi Nakada)Actions
Has duplicate Ruby master - Feature #15301: Symbol#call, returning method bound with argumentsClosedActions

Updated by felixbuenemann (Felix Bünemann) over 8 years ago

  • Description updated (diff)

Updated by felixbuenemann (Felix Bünemann) over 8 years ago

Edited to remove **kwargs argument I added, which would require checking if the called method supports them.

Actions #3

Updated by nobu (Nobuyoshi Nakada) over 8 years ago

Updated by shelvacu (Shel vacu) over 8 years ago

I agree that there should be some syntax for doing this, but I don't think this is the proper way to do it.

Personally, the syntax is confusing to me. I would prefer something like:

[1,2,16].map(&:to_s(16))

This way, the way my brain parses it is that &: is the operator for turning a symbol into a block that calls the method on the argument. However, this would require changes in the parser instead of stdlib.

My other concern is that Symbol#call returning a proc feels wrong. It leads to code like this:

a = :to_s
a.call(16).call(15)

While such code may never be written even if this is implemented, I hope it conveys how odd it feels to have a method named "call" always return a proc which is then meant to be called, instead of calling anything.

Would the change from &(:meth_name_as_symbol) to special operator &: followed by method name and optionally arguments (ie. the syntax I used above) break any existing code?

Updated by shevegen (Robert A. Heiler) over 8 years ago

I think there have been many other similar proposals. Nobu linked to other
discussions.

From what I have seen, I think the major problem is coming up with a nice
syntax proposal.

.map(&:foo)

is ok because it is short.

Adding implicit arguments to it is harder.

[1,2,16].map(&:to_s(16))

Is probably ok. But I am not sure if the parser is happy with it.

[1,2,16].map(&:to_s.(16))

Is not good IMHO, the . there is very confusing for me.

There is a slight alternative to .call() which is the []

I like [] a lot, but I think it looks a bit weird too
inside of ().

Perhaps we do not have a syntax that will be better if
we require arguments for &: ?

Updated by felixbuenemann (Felix Bünemann) over 8 years ago

Although I don't understand the Japanese, the linked issue, with a similar syntax to what Shel vacu proposed above, was rejected by Matz. So probably not too much hope on getting this into core…

Updated by nobu (Nobuyoshi Nakada) over 8 years ago

Yes, &:to_s(16) is exactly my (rejected) proposal.

Updated by sawa (Tsuyoshi Sawada) over 8 years ago

For a similar proposal, please cf. #10394.

Updated by knu (Akinori MUSHA) about 7 years ago

Wouldn't Array#to_proc make sense?

class Array
  def to_proc
    proc { |x| x.__send__(*self) }
  end
end

[100, 200, 300].map(&[:to_s, 16])
# => ["64", "c8", "12c"]

Updated by zverok (Victor Shepelev) about 7 years ago

Wouldn't Array#to_proc make sense?

For me, it looks confusing. Array is (usually) a set of homogenous objects, so my first guess for Array#to_proc would be this:

[100,200,300].map(&%i[to_s reverse to_i])

(chain of calls on argument)

Actions #11

Updated by jwmittag (Jörg W Mittag) about 7 years ago

knu (Akinori MUSHA) wrote:

Wouldn't Array#to_proc make sense?

class Array
  def to_proc
    proc { |x| x.__send__(*self) }
  end
end

[100, 200, 300].map(&[:to_s, 16])
# => ["64", "c8", "12c"]

I disagree: responding to to_proc in Ruby more or less means "I am a function-like thing". And arrays are function-like things, they are basically functions from their indices to their values. Having to_proc mean something different than that would be a big mistake, IMO. It would also be inconsistent with Hash#to_proc.

See #11653 (Hash#to_proc) for what I mean, and #11262 for a more comprehensive argument.

Actions #12

Updated by nobu (Nobuyoshi Nakada) about 7 years ago

jwmittag (Jörg W Mittag) wrote:

I disagree: responding to to_proc in Ruby more or less means "I am a function-like thing". And arrays are function-like things, they are basically functions from their indices to their values. Having to_proc mean something different than that would be a big mistake, IMO. It would also be inconsistent with Hash#to_proc.

Even if arrays were function-like things, the elements are not arguments.

Updated by duerst (Martin Dürst) about 7 years ago

nobu (Nobuyoshi Nakada) wrote:

jwmittag (Jörg W Mittag) wrote:

I disagree: responding to to_proc in Ruby more or less means "I am a function-like thing". And arrays are function-like things, they are basically functions from their indices to their values. Having to_proc mean something different than that would be a big mistake, IMO. It would also be inconsistent with Hash#to_proc.

Even if arrays were function-like things, the elements are not arguments.

In Jörg's proposals, array elements are indeed not arguments, they are return values. Index values are arguments.

Updated by knu (Akinori MUSHA) about 7 years ago

I think &[symbol, *args] can be a natural extension to &symbol, as they are both a shorthand for { |_| _.__send__(*object) }.

For Array to provide #to_proc would be just a little bit weird convention for greater convenience, just as Symbol#to_proc is. Symbol had been in no way a function-like entity, but once Symbol#to_proc was added we almost instantly grew used to it and now we all take it for granted because being able to map(&:to_s) is so handy and useful.

The original proposal that is to introduce a new syntax is a bit too costly for one of the most frequently wanted features like this because it takes years of time before everyone can start using it. On the other hand, adding Array#to_proc is easily backportable and you can start using it today. FWIW, Matz once said he was not in favor of extending the syntax just for this: http://blade.nagaokaut.ac.jp/cgi-bin/vframe.rb/ruby/ruby-dev/45404?45384-45592 (written in Japanese, try Google translate)

I've done some more googling and it turned out that Array#to_proc was not a new idea at all.

So, these people independently have reached the same idea! Isn't that a good sign?

Actions #15

Updated by nobu (Nobuyoshi Nakada) almost 6 years ago

  • Has duplicate Feature #15301: Symbol#call, returning method bound with arguments added

Updated by RichOrElse (Ritchie Buitre) almost 6 years ago

I have a related proposal #15302 with a different implementation and interface.

Actions

Also available in: Atom PDF

Like0
Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0