How to vary sort's block?

ruby-talk
1

When I have an array of objects (Thing's, say) with two properties (say
`name' and `amount') I want to be able to sort that array by either of
these properties, depending on circumstances (like the value of the
variable `order').

The script below is an example, and it works, but I don't like it.
I there a better way to do it, without an if..else construction?

For example, would it be possible to put several {|a,b|...} blocks for
the sort in a hash with keys telling what I want to sort on?

#!/usr/bin/ruby

class Thing
  attr_reader :name,:amount

  def initialize(name,amount)
    @name,@amount = name,amount
  end

  def list
    puts ['',@name,@amount].join("\t")
  end
end

class Array
  def byamount
    self.sort { |a,b| a.amount <=> b.amount }
  end
  def byname
    self.sort { |a,b| a.name <=> b.name }
  end
end

arr = [
  Thing.new('John',10),
  Thing.new('Anny',20)
]

order = :byname

puts "sorted #{order}:"
if order == :byname
  arr.sort { |a,b| a.name <=> b.name }.each { |a|
      a.list
      # much more code may occur here...
    }
elsif order == :byamount
  arr.sort { |a,b| a.amount <=> b.amount }.each { |a|
      a.list
      # much more code may occur here...
    }
else
  raise "illegal order"
end

···

--
Wybo

2

The easiest is probably to use sort_by.

Thing = Struct.new(:name, :amount)
arr = [
Thing.new('John',10),
Thing.new('Anny',20),
]

p arr.sort_by {|x| x.name}
p arr.sort_by {|x| x.amount}

Btw, I would *not* put byamount and byname into Array. If you want to do something general, then I'd do this:

module Enumerable
  def sort_field(*fields)
    sort_by {|x| fields.map {|f| x.send f}}
  end
end

Then you can do

p arr.sort_field :name
p arr.sort_field :amount

It might even be reasonable to replace the current implementation of sort_by with one that accepts either a list of symbols as arguments or a block. I think this might have been proposed already. Matz, any comment on this? Does this sound reasonable?

Kind regards

    robert

···

Wybo Dekker <wybo@servalys.nl> wrote:

When I have an array of objects (Thing's, say) with two properties
(say `name' and `amount') I want to be able to sort that array by
either of these properties, depending on circumstances (like the
value of the variable `order').

The script below is an example, and it works, but I don't like it.
I there a better way to do it, without an if..else construction?

For example, would it be possible to put several {|a,b|...} blocks for
the sort in a hash with keys telling what I want to sort on?

#!/usr/bin/ruby

class Thing
attr_reader :name,:amount

def initialize(name,amount)
   @name,@amount = name,amount
end

def list
   puts ['',@name,@amount].join("\t")
end
end

class Array
def byamount
   self.sort { |a,b| a.amount <=> b.amount }
end
def byname
   self.sort { |a,b| a.name <=> b.name }
end
end

arr = [
Thing.new('John',10),
Thing.new('Anny',20)
]

order = :byname

puts "sorted #{order}:"
if order == :byname
arr.sort { |a,b| a.name <=> b.name }.each { |a|
     a.list
     # much more code may occur here...
   }
elsif order == :byamount
arr.sort { |a,b| a.amount <=> b.amount }.each { |a|
     a.list
     # much more code may occur here...
   }
else
raise "illegal order"
end

3

When I have an array of objects (Thing's, say) with two properties (say
`name' and `amount') I want to be able to sort that array by either of
these properties, depending on circumstances (like the value of the
variable `order').

The script below is an example, and it works, but I don't like it.
I there a better way to do it, without an if..else construction?

For example, would it be possible to put several {|a,b|...} blocks for
the sort in a hash with keys telling what I want to sort on?

Yes, see below:

#!/usr/bin/ruby

class Thing
  attr_reader :name,:amount

  def initialize(name,amount)
    @name,@amount = name,amount
  end

  def list
    puts ['',@name,@amount].join("\t")
  end
end

class Array
  def byamount
    self.sort { |a,b| a.amount <=> b.amount }
  end
  def byname
    self.sort { |a,b| a.name <=> b.name }
  end
end

arr = [
  Thing.new('John',10),
  Thing.new('Anny',20)
]

# Replace rest of the code with:
orders = {
           :byname => lambda { |a,b| a.name <=> b.name },
           :byamount => lambda { |a,b| a.amount <=> b.amount }
         }

order = :byname

if blk = orders[order]
  puts "sorted #{order}:"
  arr.sort(&blk).each { |a| a.list }
else
  raise "illegal order"
end

__END__

This would be my first implementation I think, I like dispatch table
type stuff. Be careful what you enclose with all those blocks, though.

An alternative idea might be (replacing the code I show above):

  require 'curry' # Have to download this, see bottom

  sort_proc = lambda { |msg,a,b| a.send(msg) <=> b.send(msg) }

  by_name = sort_proc.curry(:name)
  by_amount = sort_proc.curry(:amount)

  puts "sorted name:"
  arr.sort(&by_name).each { |a| a.list }

# => sorted name:
# => Anny 20
# => John 10

  puts "sorted amount:"
  arr.sort(&by_amount).each { |a| a.list }

# => sorted amount:
# => John 10
# => Anny 20

(You could curry on the fly of course depending on the sort criteria,
it's reasonably low-overhead).

Does it show I've been searching for an opportunity to use currying
since I did that quiz entry? ;D

(Curry: http://www.ruby-talk.org/cgi-bin/scat.rb/ruby/ruby-talk/177343 )

···

On Sat, 2006-02-04 at 19:40 +0900, Wybo Dekker wrote:

--
Ross Bamford - rosco@roscopeco.REMOVE.co.uk

4

I would use send():

#!/usr/bin/ruby

class Thing
   attr_reader :name,:amount

   def initialize(name,amount)
     @name,@amount = name,amount
   end

   def list
     puts ['',@name,@amount].join("\t")
   end
end

arr = [
   Thing.new('John',10),
   Thing.new('Anny',20)
]

order = :name

puts "sorted by #{order}:"
arr.sort_by { |thing| thing.send(order) }.each { |a|
     a.list
     # much more code may occur here...
} rescue puts("illegal order")

__END__

Hope that helps.

James Edward Gray II

···

On Feb 4, 2006, at 4:40 AM, Wybo Dekker wrote:

When I have an array of objects (Thing's, say) with two properties (say
`name' and `amount') I want to be able to sort that array by either of
these properties, depending on circumstances (like the value of the
variable `order').

5

The easiest is probably to use sort_by.

Thing = Struct.new(:name, :amount)
arr = [
Thing.new('John',10),
Thing.new('Anny',20),
]

p arr.sort_by {|x| x.name}
p arr.sort_by {|x| x.amount}

this does not allow for more complex sorts, like on name *and* amount,
does it?

Btw, I would *not* put byamount and byname into Array. If you want to do
something general, then I'd do this:

module Enumerable
def sort_field(*fields)
sort_by {|x| fields.map {|f| x.send f}}
end
end

Then you can do

p arr.sort_field :name
p arr.sort_field :amount

but this is really great, I can now even say:
   p arr.sort_field(:name,:amount)
Thanks!

It might even be reasonable to replace the current implementation of sort_by
with one that accepts either a list of symbols as arguments or a block. I
think this might have been proposed already. Matz, any comment on this? Does
this sound reasonable?

I saw that it is also possible to give sort an argument instead of a
block:

require 'pp'

Thing = Struct.new(:name, :amount)
arr = [
  Thing.new('John',80),
  Thing.new('John',10),
  Thing.new('John',30),
  Thing.new('Anny',20),
  Thing.new('Anny',10),
  Thing.new('Anny',30),
  Thing.new('Anny',15)
]

def byname(a,b) a.name <=> b.name end
def byamount(a,b) a.amount <=> b.amount end
def by_name_amount(a,b)
  (a.name <=> b.name)*4 + (a.amount <=> b.amount)
end

[:byname,:byamount,:by_name_amount].each do |order|
  puts "sort #{order}:"
  pp arr.sort(&method(order))
end

But ri does not tell me that:

-------------------------------------------------------- Enumerable#sort
     enum.sort => array
     enum.sort {| a, b | block } => array

···

On Sat, 4 Feb 2006, Robert Klemme wrote:
------------------------------------------------------------------------
     Returns an array containing the items in _enum_ sorted, either
     according to their own +<=>+ method, or by using the results of the
     supplied block. The block should return -1, 0, or +1 depending on
     the comparison between _a_ and _b_. As of Ruby 1.8, the method
     +Enumerable#sort_by+ implements a built-in Schwartzian Transform,
     useful when key computation or comparison is expensive..

        %w(rhea kea flea).sort #=> ["flea", "kea", "rhea"]
        (1..10).sort {|a,b| b <=> a} #=> [10, 9, 8, 7, 6, 5, 4, 3, 2, 1]

--
Wybo

6

Thanks all for the very valuable explanations and suggestions.
I finally came to the conclusion that, for my problem (to do a sort,
depending on the value of some variable `order'), sort_by is the most
versatile way. I saved an example for future reference, which may be
useful for others on the list:

#!/usr/bin/ruby

class Thing
  attr_reader :name,:amount
  def initialize(name,amount)
    @name,@amount = name,amount
  end
  def list
    puts ['',@name,@amount].join("\t")
  end
end

arr = [
  Thing.new('Johnny',20), Thing.new('Anny',20),
  Thing.new('John',10), Thing.new('Anny',10),
  Thing.new('John',30), Thing.new('Anny',30),
]

for order in [:name, :amount]
  puts "by #{order}:"
  arr.sort_by{|x| x.send(order)}.each {|a| a.list}
end

order = :name
puts "by #{order}_length, #{order}, amount:"
arr.sort_by{|x| [ x.send(order).length,
                  x.send(order),
                  x.amount
                ]
}.each {|a| a.list}

PS. Good to see in the definition of arr that the extra comma after the
last element is not an error anymore.

···

--
Wybo

7

"Wybo Dekker" <wybo@servalys.nl> schrieb im Newsbeitrag news:Pine.LNX.4.61.0602041523170.10213@servalys.nl...

The easiest is probably to use sort_by.

Thing = Struct.new(:name, :amount)
arr = [
Thing.new('John',10),
Thing.new('Anny',20),
]

p arr.sort_by {|x| x.name}
p arr.sort_by {|x| x.amount}

this does not allow for more complex sorts, like on name *and* amount,
does it?

It does: arr.sort_by {|x| [x.name, x.amount]} (depending on the precedence you want). I believe, if you use Struct and want to order according to field order then you might be able to just use sort() without arguments (alternatively sort_by {|x| x.to_a} - to_a is defined by Struct).

I saw that it is also possible to give sort an argument instead of a
block:

require 'pp'

Thing = Struct.new(:name, :amount)
arr = [
Thing.new('John',80),
Thing.new('John',10),
Thing.new('John',30),
Thing.new('Anny',20),
Thing.new('Anny',10),
Thing.new('Anny',30),
Thing.new('Anny',15)
]

def byname(a,b) a.name <=> b.name end
def byamount(a,b) a.amount <=> b.amount end
def by_name_amount(a,b)
(a.name <=> b.name)*4 + (a.amount <=> b.amount)
end

[:byname,:byamount,:by_name_amount].each do |order|
puts "sort #{order}:"
pp arr.sort(&method(order))
end

But ri does not tell me that:

-------------------------------------------------------- Enumerable#sort
    enum.sort => array
    enum.sort {| a, b | block } => array
------------------------------------------------------------------------
    Returns an array containing the items in _enum_ sorted, either
    according to their own +<=>+ method, or by using the results of the
    supplied block. The block should return -1, 0, or +1 depending on
    the comparison between _a_ and _b_. As of Ruby 1.8, the method
    +Enumerable#sort_by+ implements a built-in Schwartzian Transform,
    useful when key computation or comparison is expensive..

       %w(rhea kea flea).sort #=> ["flea", "kea", "rhea"]
       (1..10).sort {|a,b| b <=> a} #=> [10, 9, 8, 7, 6, 5, 4, 3, 2, 1]

Actually ri tells you that - because it's the block form. Your methods are implicitely converted to blocks. In that case I'd rather not use a method definition but rather do the more direct lambda / proc:

Thing::BYNAME = lambda {|a,b| a.name <=> b.name}
Thing::BYAMOUNT = lambda {|a,b| a.amount <=> b.amount}
....
arr.sort(&Thing::BYNAME)

You can as well define those constants in global scope but I felt they are more appropriately placed in Thing's scope because they refer exactly to Things.

Kind regards

    robert

···

On Sat, 4 Feb 2006, Robert Klemme wrote:

8

"Wybo Dekker" <wybo@servalys.nl> wrote in message
news:Pine.LNX.4.61.0602051338380.10213@servalys.nl...

PS. Good to see in the definition of arr that the extra comma after the
last element is not an error anymore.

Agreed. C# is the same.