Feature #15483: Proc or Method combination with Symbol

• Status: Rejected
• Assignee: -
• Target version: -

[aycabta (aycabta .)]

In [Feature #6284], Matz said

We need more discussion if we would add combination methods to the Symbol class.

Right, let's get started to discuss.

For your information, recent a few months I'm discussing this with @osyo .

This is a discussion of "design"

I understand that all features of this issue have both merits and demerits, but I guess that language design is most important. All features of this issue related to each other.

Abstract

At present, you can use Proc#>> or Proc#<< with Symbol#to_proc.

%w{72 101 108 108 111}.map(&(:to_i.to_proc >> :chr.to_proc))
# => ["H", "e", "l", "l", "o"]

This is convenient but methods that take block can take a proc with & syntax sugar instead of #to_proc by right, like [1, 2, 3].map(&:to_s). So Symbol#to_proc looks like too long for Proc#>> or Proc#<<. Therefore, you need new syntax sugar.

Receiver

Symbol#>> and Symbol#<<

Symbol#>> and Symbol#<< will be considered, but this means that Symbol is treated as Proc partially. The [1, 2, 3].map(&:to_s) treats Symbol as Proc partially too, but it's with pre-positioned &.

%w{72 101 108 108 111}.map(&(:to_i >> :chr.to_proc))
# => ["H", "e", "l", "l", "o"]

I can't come up with other ideas for the Symbol receiver.

New &:symbol_name syntax sugar for :symbol_name.to_proc

%w{72 101 108 108 111}.map(&(&:to_i >> :chr.to_proc)))
# => ["H", "e", "l", "l", "o"]

Argument

Calls #to_proc by Proc#>> or Proc#<< internally as a duck typing

%w{72 101 108 108 111}.map(&(:to_i.to_proc >> :chr))
# => ["H", "e", "l", "l", "o"]

In this case, Proc#>>(:to_i.to_proc >>) calls Symbol#to_proc(for :chr) inside.

This is useful to use with Hash#to_proc:

h = { Alice: 30, Bob: 60, Cris: 90 }
%w{Alice Bob Cris}.map(&(:to_sym.to_proc >> h))
# => [30, 60, 90]

Proc#>> and Proc#<< take block as an argument

%w{72 101 108 108 111}.map(&(:to_i.to_proc >> &:chr))

Combination of receiver and argument

Symbol#>> and calling #to_proc internally:

%w{72 101 108 108 111}.map(&(:to_i >> :chr))
# => ["H", "e", "l", "l", "o"]

&:symbol_name syntax sugar for :symbol_name.to_proc and Symbol#>> and taking block:

%w{72 101 108 108 111}.map(&(&:to_i >> &:chr))
# => ["H", "e", "l", "l", "o"]

[aycabta (aycabta .)]

• Tracker changed from Bug to Feature
• Backport deleted (2.4: UNKNOWN, 2.5: UNKNOWN, 2.6: UNKNOWN)

[aycabta (aycabta .)]

• Description updated (diff)

[shevegen (Robert A. Heiler)]

I am biased so I do not want to digress from this thread too much while explaining my bias. However had,
I still want to state a few things:

• In regards to Symbol, this is a language design decision, how Symbols are to be used. I think we can
  have valid arguments for both main variants, e. g. to keep Symbols simple, or to allow more flexibility.
  Personally I'd rather prefer them simple, largely because I don't feel most proposals for change make
  them better and most definitely not prettier; but I have no real problem either way here.

Still, in regards to proposals allowing for more flexibility of Symbols, this leads me to:

• Syntax consideration. To me personally the proposed syntax is not very elegant.

In particular:

.map(&(&:to_i >> &:chr))

Is really not pretty. We use '& three' times there; and the new >>. It does not really feel consistent
with other parts of ruby in my opinion, syntax-wise alone. I have less of a problem with a single & but
I also dislike that I have to look carefully, e. g to distinguish between a** .map(&:)** versus a .map(&)
variant. Do we really want to have to look for & now carefully and a : or no :, on top of it? The second
variant also packs a lot more information into the method-call, which makes it a bit hard to see what
is going on to me, e. g. .map(&(&:to_i >> :chr.to_proc))). And the >> which I am also not a big fan of,
but as said in the beginning, I am biased already, so my comments will be biased as well.

• Another issue I have, and this is more general, that I do not really see the massive benefit. This is not
  solely confined to the proposal here, and is obviously subject to personal opinion/evaluation and how
  you use ruby ("more than one way to use ruby", too), but more generally about some other related
  proposals too, where I am not really sure if the change is needed or provides a lot of really useful
  things that we need.

I understand it if the goal is more flexibility in what we can do; for example, I think I also stated before
that I am in agreement with proposals to allow arguments to methods given rather than solely be able
to use e. g. .map(&:method SOME WAY FOR ARGUMENTS HERE). The major problem I have with most
of these proposals I have seen so far is syntax-wise. We do not have that many characters while staying
in ASCII land, but the core of ruby is very elegant and quite simple, syntax-wise (for me). Several of the
proposals in the last ~3 years or so, are, to me, syntax-wise, not really elegant. Syntax is not everything
but if I have to stare at code a lot then I'd rather look at good syntax than bad one.

Anyway, I'll close my comment here.

[osyo (manga osyo)]

I am thinking like this.

NOTE: Here we define it as follows.

• functional object
  • defined #call (and #<< #>>) object
  • e.g. Proc Method
• blockable object
  • defined #to_proc object
  • e.g. Symbol Hash

Current

• Proc#<< and Proc#>> arguments is functional object
  call #call.
• Proc#<< and Proc#>> is not call #to_proc
• Proc#<< and Proc#>> is not accept block argument

Composite function in Ruby

• Composite function is functional object and functional object
• functional object >> functional object # => OK
• functional object >> other object # => NG
• other object >> functional object # => NG

Symbol is functional object

• Symbol is blockable object
• Symbol is not functional object
• Handling Symbol with compositing functions is incorrect
• What about other blockable objects?
  • e.g. Hash
  • Hash is functional object?

Proc#<< is call #to_proc ?

• It should be explicitly converted to Proc (functional object) with # to_proc
  • proc << :hoge => NG: :hoge is not Proc
  • proc << :hoge.to_proc => OK : Explicitly convert :hoge to Proc
• Same as not handling "42" as an Integer
  • 1 + "42" => NG : "42" is not an Integer
  • 1 + "42".to_i => OK : Explicitly convert "42" to a Proc

Proposal1 : Symbol to functional object

• define Symbol#>> Symbol#<< Symbol#call
• What about other blockable objects?
  • Hash is functional object?
• Is it really necessary for Symbol ?
• Is Symbol really a "functinal object" ?

# Symbol to functional object
class Symbol
	def call(*args, &block)
		to_proc.call(*args, &block)
	end

	def <<(other)
		to_proc << other
	end

	def >>(other)
		to_proc >> other
	end
end

p %w{72 101 108 108 111}.map(&(:to_i >> :chr))
# => ["H", "e", "l", "l", "o"]

Proposal2 : Symbol to functional object

• Proc#<<(other) to Proc#<<(other, &block)
• Prioritize other ?

class Proc
	prepend Module.new {
		def <<(other = nil, &block)
			# other or block?
			super(other || block)
		end

		def >>(other = nil, &block)
			# other or block?
			super(other || block)
		end
	}
end

# :to_i convert to Proc
# must be `.>>`
p %w{72 101 108 108 111}.map(&(:to_i.to_proc.>> &:chr))
# => ["H", "e", "l", "l", "o"]

Proposal3 : Define syntax sugar for #to_proc

• For example, define #to_proc to ~@.
  • or other Unary operator
  • +@ -@ ! & ?
• Do not change current specifications
• I think this is good

# Add ~@
class Object
	# ~ is to_proc
	# ~ or other unary operator?
	def ~@
		to_proc
	end
end

# Use Symbol#to_proc
p %w{72 101 108 108 111}.map(&(:to_i.to_proc >> :chr.to_proc))

# alias ~ is to_proc
p %w{72 101 108 108 111}.map(&~:to_i >> ~:chr)

Thank you :)

Japanese

[nobu (Nobuyoshi Nakada)]

Why not using refinements?

# symbol/functionalized.rb
module Symbol::Functionalized
  refine(Symbol) do
    def call(*args, &block)
      to_proc.call(*args, &block)
    end

    def <<(other = (b = true), &block)
      to_proc << (b ? block : other.to_proc)
    end

    def >>(other = (b = true), &block)
      to_proc >> (b ? block : other.to_proc)
    end
  end
end

require 'symbol/functionalized'
using Symbol::Functionalized

p %w{72 101 108 108 111}.map(&:to_i >> :chr) #=> ["H", "e", "l", "l", "o"]

[osyo (manga osyo)]

hi, nobu :)

引用
Why not using refinements?

It is example code.
Also, Symbol#call is not called in Proc#<<.

# Error: undefined method `call' for :chr:Symbol (NoMethodError)
p %w{72 101 108 108 111}.map(&proc { |s| s.to_i } >> :chr) #=> ["H", "e", "l", "l", "o"]

[nobu (Nobuyoshi Nakada)]

# symbol/functionalized.rb
module Symbol::Functionalized
  refine(Symbol) do
    def call(*args, &block)
      to_proc.call(*args, &block)
    end

    def <<(other = (b = true), &block)
      to_proc << (b ? block : other.to_proc)
    end

    def >>(other = (b = true), &block)
      to_proc >> (b ? block : other.to_proc)
    end
  end

  refine(Proc) do
    def <<(other)
      super(other.to_proc)
    end

    def >>(other)
      super(other.to_proc)
    end
  end
end

[nobu (Nobuyoshi Nakada)]

I made function-composite gem, as a PoC.

[osyo (manga osyo)]

I think it will not work in the following cases.

# NG: Error undefined method `call' for :chr:Symbol (NoMethodError)
p (30.method(:+) >> :chr).call 42


h = { Alice: 30, Bob: 60, Cris: 90 }

# OK
p (:to_sym >> h).call "Alice"
# => 30

# NG
p (h << :to_sym).call "Bob"

Would you like to add Method#>> and Hash#>>, or other object #>> definitions?
I do not think that is good.
I think it is necessary to clearly separate "functional object"(e.g. Proc, Method) and "blockable object"(e.g. Symbol, Hash).
I think that it should handle only functional object in the composite function.

[mrkn (Kenta Murata)]

• Related to Bug #15428: Refactor Proc#>> and #<< added

[matz (Yukihiro Matsumoto)]

• Status changed from Open to Rejected

I feel the expression ary.map(&(:to_i << :chr)) is far less readable than ary.map{|x|x.to_i.chr}.
And the latter is faster and can take arguments NOW e.g. ary.map{|x|x.to_i(16).chr}.

Given these superiorities, this proposal does not sound attractive.

Matz.

p.s.
And this can lead to the default block parameter like it.