Ruby inject vs. tap vs. each_with_object. Inspired by http://phrogz.net/tap-vs-each_with_object

benchmark_dead_horse.rb

require 'benchmark'

N = 1000000
nums = N.times.map{ rand(N) }
enum = [1, 2, 'string', {}, [], false, true, nil]

def process(v)
  v
end

def i_want_this?(v)
  !v.nil?
end

Benchmark.bmbm do |x|
  x.report('inj') { enum.inject([]) { |arr, v|  arr << process(v) if i_want_this?(v); arr } }
  x.report('tap') { [].tap { |arr| enum.each { |v| arr << process(v) if i_want_this?(v) } } }
  x.report('ewo') { enum.each_with_object([]) { |v, arr|  arr << process(v) if i_want_this?(v) } }
  x.report('sel') { enum.select {|v| i_want_this?(v) }.map { |v| process(v) } }
  x.report('map') { enum.map { |v| process(v) if i_want_this?(v) }.reject { |v| !v } }
end


# Rehearsal ---------------------------------------
# inj   0.000000   0.000000   0.000000 (  0.000027)
# tap   0.000000   0.000000   0.000000 (  0.000017)
# ewo   0.000000   0.000000   0.000000 (  0.000023)
# sel   0.000000   0.000000   0.000000 (  0.000016)
# map   0.000000   0.000000   0.000000 (  0.000021)
# ------------------------------ total: 0.000000sec
# 
#           user     system      total        real
# inj   0.000000   0.000000   0.000000 (  0.000025)
# tap   0.000000   0.000000   0.000000 (  0.000024)
# ewo   0.000000   0.000000   0.000000 (  0.000026)
# sel   0.000000   0.000000   0.000000 (  0.000023)
# map   0.000000   0.000000   0.000000 (  0.000024)

For no raisin:

$ ruby -v
ruby 2.0.0p353 (2013-11-22 revision 43784) [x86_64-darwin12.5.0]
$ ruby deadhorse.rb 
Rehearsal ---------------------------------------
inj   0.000000   0.000000   0.000000 (  0.000015)
tap   0.000000   0.000000   0.000000 (  0.000008)
ewo   0.000000   0.000000   0.000000 (  0.000011)
sel   0.000000   0.000000   0.000000 (  0.000008)
map   0.000000   0.000000   0.000000 (  0.000015)
------------------------------ total: 0.000000sec

          user     system      total        real
inj   0.000000   0.000000   0.000000 (  0.000014)
tap   0.000000   0.000000   0.000000 (  0.000015)
ewo   0.000000   0.000000   0.000000 (  0.000017)
sel   0.000000   0.000000   0.000000 (  0.000016)
map   0.000000   0.000000   0.000000 (  0.000012)

And

$ jruby -v
jruby 1.7.4 (1.9.3p392) 2013-05-16 2390d3b on Java HotSpot(TM) 64-Bit Server VM 1.7.0_40-b43 +indy [darwin-x86_64]
oscar:~ ajconsta$ jruby deadhorse.rb 
Rehearsal ---------------------------------------
inj   0.030000   0.000000   0.030000 (  0.017000)
tap   0.040000   0.000000   0.040000 (  0.018000)
ewo   0.030000   0.000000   0.030000 (  0.010000)
sel   0.040000   0.000000   0.040000 (  0.015000)
map   0.050000   0.000000   0.050000 (  0.017000)
------------------------------ total: 0.190000sec

          user     system      total        real
inj   0.000000   0.000000   0.000000 (  0.001000)
tap   0.010000   0.000000   0.010000 (  0.001000)
ewo   0.000000   0.000000   0.000000 (  0.001000)
sel   0.000000   0.000000   0.000000 (  0.000000)
map   0.000000   0.000000   0.000000 (  0.001000)

Couldn't help myself; does JIT make the order matter under jruby ('sel' moved to top)?

$ jruby deadhorse.rb 
Rehearsal ---------------------------------------
sel   0.030000   0.000000   0.030000 (  0.020000)
inj   0.010000   0.000000   0.010000 (  0.012000)
tap   0.050000   0.000000   0.050000 (  0.015000)
ewo   0.030000   0.000000   0.030000 (  0.012000)
map   0.050000   0.000000   0.050000 (  0.017000)
------------------------------ total: 0.170000sec

          user     system      total        real
sel   0.000000   0.000000   0.000000 (  0.000000)
inj   0.000000   0.000000   0.000000 (  0.000000)
tap   0.010000   0.000000   0.010000 (  0.000000)
ewo   0.000000   0.000000   0.000000 (  0.000000)
map   0.000000   0.000000   0.000000 (  0.001000)

The takeaway, obviously, is that you can hardly measure the time of any of them with a million iterations so it doesn't really matter. I tried using an array of 800 elements, too, and got nothing any different.