From: Robert Klemme <shortcutter@...>
Date: 2006-08-15T16:10:09+09:00
Subject: Re: hash.keys and hash.values

--------------030804050006010507080807
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

On 14.08.2006 21:01, Erik Veenstra wrote:
>> Your test doesn't prove higher memory usage for the inject
>> version. The problem with the transpose approach is that you
>> need an additional copy of the complete hash in mem which is
>> not needed for inject. However, inject version is likely to
>> cause more GC because of all the two element arrays created.
>> You could retest this:
>>
>> keys, vals = [], [] hash.each {|k,v| keys << k; vals << v}
>>
>> This should be the most efficient approach - memory and
>> performance wise.
> 
> Hihi... ;]
> 
> Remember, both to_a and transpose are implemented in C.

Yes, right.  Still some remarks

  - by not using GC you will see the memory used in total but this does 
not give a realistic result because it omits GC times and it does make a 
difference whether you only allocate small chunks and release them again 
or whether you need a copy of the whole data structure in memory, so

  - memory wise approach "each" is the most efficient

  - "traditional" is not semantically identical to the other approaches 
because it's not guaranteed that both are in the same order

Some more figures:

08:58:59 [Temp]: ./test2.rb
Rehearsal -----------------------------------------------
traditional   1.375000   0.000000   1.375000 (  1.375000)
transpose    11.203000   0.000000  11.203000 ( 11.203000)
inject       45.188000   0.000000  45.188000 ( 45.187000)
each         20.140000   0.000000  20.140000 ( 20.157000)
------------------------------------- total: 77.906000sec

                   user     system      total        real
traditional   1.391000   0.000000   1.391000 (  1.390000)
transpose    11.281000   0.000000  11.281000 ( 11.282000)
inject       45.219000   0.000000  45.219000 ( 45.234000)
each         19.813000   0.000000  19.813000 ( 19.812000)

Cheers

	robert


--------------030804050006010507080807
Content-Type: text/plain;
 name="test2.rb"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline;
 filename="test2.rb"

#!/usr/bin/env ruby

require 'benchmark'

REPEAT = 1_00_000

REP = 10_000

hash    = {}

1000.times do |n|
  hash[n] = n*n
end


Benchmark.bmbm do |bm|
  bm.report "traditional" do
    REP.times do
      keys      = hash.keys
      values    = hash.values
    end
  end

  bm.report "transpose" do
    REP.times do
      keys, values = hash.to_a.transpose
    end
  end

  bm.report "inject" do
    REP.times do
      keys, values = hash.inject([[],[]]){|(ks,vs),(k,v)| [ks << k, vs << v]}
    end
  end

  bm.report "each" do
    REP.times do
      keys, vals = [], []
      hash.each {|k,v| keys << k; vals << v}
    end
  end
end

--------------030804050006010507080807--