Feature #6483: parametric map

• Status: Rejected
• Assignee: matz (Yukihiro Matsumoto)
• Target version: 2.6

[prijutme4ty (Ilya Vorontsov)]

I found very common use-case for map: map with parameters and &:meth syntax. For example:
matrix =[[1,2,3],[4,5,6]]
matrix.pmap(' ',&:join).join("\n") # => "1 2 3\n4 5 6
[1,2,3,4,5].pmap(2, &:to_s) # ['1', '10', '11', '100', '101']

[1,2,3,4,5].pmap(&:to_s) # ['1', '2, '3', '4', '5'] # empty parameter list behaves as usual map

Isn't it much better than ugly and verbose code:
matrix.map{|line| line.join(' ')}.join("\n")

I can write simple implementation
class Proc
def curry_except_self(*args)
Proc.new{|slf| curry[slf,*args] }
end
end

module Enumerable
def pmap!(*args,&block)
map! &block.curry_except_self(*args)
end
def pmap(*args,&block)
dup.pmap!(*args, &block)
end
end

Use-cases can be rewritten as tests (I can send my own unit-test if needed)

Also I've two related things to discuss.

1. First is &-syntax. Is it possible to change ruby-interpeter in such a way that &:proc could be at any place. matrix.pmap(&:join,' ') is much prettier than matrix.join(' ',&:join) What is the reason behind this restriction? And if one can remove this restriction, we'd have new nice syntax.
2. I'm not very experience in functional programming with curry etc, but it seems to me that currying proc without supplying first argument(self) can be common task when &:meth syntax is in play. If so, may be my curry_except_self(*args) also should be included in ruby.

[prijutme4ty (Ilya Vorontsov)]

Also I propose similar approach for tap method.

class Object
  def ptap(*args,&block)
    tap &block.curry_except_self(*args)
  end
end

So instead of

lines = File.read('filename.txt');
lines.delete('')

or

lines = File.read('filename.txt').tap{|x|x.delete ''}

I can use

lines = File.read('filename.txt').ptap('', &:delete)

[nobu (Nobuyoshi Nakada)]

I proposed a different approach, symbol with arguments syntax.

(1..5).map(&:to_s(2)) #=> ['1', '10', '11', '100', '101']

And another proposed Symbol#call in the same thread, instead.

(1..5).map(&:to_s.(2)) #=> ['1', '10', '11', '100', '101']

[prijutme4ty (Ilya Vorontsov)]

nobu (Nobuyoshi Nakada) wrote:

I proposed a different approach, symbol with arguments syntax.

(1..5).map(&:to_s(2)) #=> ['1', '10', '11', '100', '101']

And another proposed Symbol#call in the same thread, instead.

(1..5).map(&:to_s.(2)) #=> ['1', '10', '11', '100', '101']

Really cool syntax! I didn't even thought that about such a way. Second version I wrote like that:

class Symbol
  def call(*args)
    obj=Object.new.instance_exec(self,args){|sym,params| @sym=sym; @args = params; self}
    obj.define_singleton_method :to_proc do
      @sym.to_proc.curry_except_self(*@args)
    end
    obj
  end
end

Unfortunately symbol isn't cloneable, so I used auxiliary object
In such a way one shouldn't define both tap and map and pleorth of other methods!

But first approach you suggested cannot be implemented in ruby yet. Hope sometimes ruby'll supply such a syntax! I propose it shouldn't even have parentheses for args

[Eregon (Benoit Daloze)]

nobu (Nobuyoshi Nakada) wrote:

I proposed a different approach, symbol with arguments syntax.

(1..5).map(&:to_s(2)) #=> ['1', '10', '11', '100', '101']

And another proposed (({Symbol#call})) in the same thread, instead.

(1..5).map(&:to_s.(2)) #=> ['1', '10', '11', '100', '101']

For info, this is http://bugs.ruby-lang.org/issues/show/4146 (I had a hard time finding it back).

Another syntax is proposed by Koichi (from what I can understand):

p %w[12 45 32].map(&PM.to_i(9)).map(&PM * 2)
p %w[abc def ghi].map(&PM[1])

Which might be simplified to:

p %w[12 45 32].map(&.to_i(9)).map(& * 2)
p %w[abc def ghi].map(&[1])

I really like that one.

[trans (Thomas Sawyer)]

p %w[12 45 32].map(&.to_i(9)).map(& * 2)

That's pretty neat. I wonder about its implementation, so basically & becomes a special object that returns a proc when method is called on it?

class Ampersand < BasicObject
  def method_missing(s, *a, &b)
    ::Proc.new{ |x| x.public_send(s, *a, &b) }
  end
end

& = Ampersand.new

This is also interesting in that it has an appearance similar to an anaphora (default block argument):

p %w[12 45 32].map{it.to_i(9)}.map{it * 2}

But despite appearances they are very different in nature.

[trans (Thomas Sawyer)]

FYI https://github.com/raganwald/homoiconic/blob/master/2009-09-22/anaphora.md

[prijutme4ty (Ilya Vorontsov)]

Eregon (Benoit Daloze) wrote:

Another syntax is proposed by Koichi (from what I can understand):

p %w[12 45 32].map(&PM.to_i(9)).map(&PM * 2)
p %w[abc def ghi].map(&PM[1])

Which might be simplified to:

p %w[12 45 32].map(&.to_i(9)).map(& * 2)
p %w[abc def ghi].map(&[1])

I really like that one.

It's fine, but in version with PM it look like a hack. However if it'll be implemented in second version - I'd take my words back.

[mame (Yusuke Endoh)]

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

Assigning to matz.

Eregon (Benoit Daloze) wrote:

For info, this is http://bugs.ruby-lang.org/issues/show/4146 (I had a hard time finding it back).

Thanks Benoit! I added it as a related ticket.

--
Yusuke Endoh [e-mail]

[prijutme4ty (Ilya Vorontsov)]

I've made a simple extension that allows one to use nested symbolic-procs like this:

[[1,2,3],[4,5,6]].map(&:map.(&:to_s.(2))) # => [['1','10','11'],['100','101','110']]

Not to duplicate code - look at http://bugs.ruby-lang.org/issues/4146#change-26991

[nobu (Nobuyoshi Nakada)]

In golf_prelude.rb:

class Symbol
  def call(*args, &block)
    proc do |recv|
      recv.__send__(self, *args, &block)
    end
  end
end

It might use public_send, indeed.

[prijutme4ty (Ilya Vorontsov)]

Thank you for much more elegant code. It works with all of my specs except one:

['abc','cdef','xy','z','wwww'].select(&:size.() == 4)    # ==> ['cdef', 'wwww']

It looks that one cannot make this work without proxy object.

[mame (Yusuke Endoh)]

• Target version set to 2.6

[matz (Yukihiro Matsumoto)]

• Status changed from Assigned to Rejected

I reject the original idea of #pmap which might be read as 'parallel map' or 'parametric map' or something else.
The idea of adding parameter to block with argument specified e.g. a.map(&:to_s.(2)) is interesting,
but it must be proposed in separated issue.

Matz.