Don’t be a language slave

Last week I had an intriguing discussion with some Python guru (he got a nickname “ reticulatus ” a long ago.) I complained to him that the last year I was oblidged to code in PHP and now I am anxious about my IQ being decreased significally. That language virtually forces one to write an enormously nasty code. On the other hand, I finally feigned that PHP code might be written in functional manner as well. My friend archly grinned and suggested me to run benchmarks on functional-ity.

Well, I did.

The result is embarassing. I can’t even surmise how much scorn is to be felt to produce the interpreter that acts in the following way. Let’s take a look at the benchmarks. NB I wrote the tiny wrapper for calls to microtime . It does nothing but calculates times and memory usage between subsequent calls. Consider it a handy transparent utility.

$count = 1000000; // an amount of iterations

// test input, it will be mapped and reduced
$array = array();
for ($i = 0; $i < $count; $i++) {
  $array[] = rand(1, $count);
}

// the wrapper to benchmark easily
$ys = new \Mudasobwa\Screwdrivers\YardStick(true);

// Mapping
$ys->milestone('MAP#Start');
$countew_array1 = array();
foreach ($array as $a) { $countew_array1[] = $a / 2; }
$ys->milestone('MAP#FOREACH');
$countew_array2 = array();
for ($i = 0; $i < $count; $i++) { $countew_array2[] = $array[$i] / 2; }
$ys->milestone('MAP#FOR');
$countew_array3 = array_map(function($a) { return $a / 2; }, $array);
$ys->milestone('MAP#ARRAY_MAP');
$ys->report('MAP.+'); // report measures for milestones `MAP*`

// Reducing
$ys->milestone('REDUCE#Start');
$average1 = 0;
foreach ($array as $a) { $average1 += $a / $count; }
$ys->milestone('REDUCE#FOREACH');
$average2 = 0;
for ($i = 0; $i < $count; $i++) { $average2 += $array[$i] / $count; }
$ys->milestone('REDUCE#FOR');
$average3 = array_reduce($array, function($memo, $a) use($count) {
  return $memo + $a / $count;
}, 0);
$ys->milestone('REDUCE#ARRAY_REDUCE');
$ys->report('REDUCE.+'); // report measures for milestones `REDUCE*`

The results are shown below:

==== Diff for tags: [MAP#Start :: MAP#FOREACH]
--   ⌚ Time:    ⇒ 0.223850 sec
--   ⌛ Memory:  ⇒ 141,009.3 KB
==== Diff for tags: [MAP#FOREACH :: MAP#FOR]
--   ⌚ Time:    ⇒ 0.240624 sec
--   ⌛ Memory:  ⇒ 141,009.2 KB
==== Diff for tags: [MAP#FOR :: MAP#ARRAY_MAP]
--   ⌚ Time:    ⇒ 0.374310 sec
--   ⌛ Memory:  ⇒ 141,009.2 KB
——————————————————————————————————————
==== Diff for tags: [REDUCE#Start :: REDUCE#FOREACH]
--   ⌚ Time:    ⇒ 0.072303 sec
--   ⌛ Memory:  ⇒ 4.6 KB
==== Diff for tags: [REDUCE#FOREACH :: REDUCE#FOR]
--   ⌚ Time:    ⇒ 0.086437 sec
--   ⌛ Memory:  ⇒ 4.6 KB
==== Diff for tags: [REDUCE#FOR :: REDUCE#ARRAY_REDUCE]
--   ⌚ Time:    ⇒ 0.276434 sec
--   ⌛ Memory:  ⇒ 4.6 KB

Memory consumption is the same, but array_map execution time is almost twice as much as foreach , and array_reduce is almost four times slower! WUT? Let’s turn back to ruby (I’m not as reptile as my friend, I like jewels more than snakes.)

require 'benchmark'

n = 1_000_000
prng = Random.new
array = 1.upto(n).map { |_| 1.0 * (prng.rand n) }
array_new1 = []
array_new2 = []
array_new3 = []

Benchmark.bm do |x|
  x.report { for i in 1...n; array_new1 << array[i] / 2; end }
  x.report { array.each { |a| array_new2 << a / 2 } }
  x.report { array_new3 = array.map { |a| a / 2 } }
end

reduce1 = reduce2 = reduce3 = 0
Benchmark.bm do |x|
  x.report { for i in 1...n; reduce1 += array[i] / n; end }
  x.report { array.each { |a| reduce2 += a / n } }
  x.report { reduce3 = array.reduce(&:+) / n }
end

       user     system      total        real
   0.130000   0.000000   0.130000 (  0.130952)
   0.110000   0.000000   0.110000 (  0.104156)
   0.090000   0.010000   0.100000 (  0.097225)
       user     system      total        real
   0.120000   0.000000   0.120000 (  0.115846)
   0.150000   0.000000   0.150000 (  0.155771)
   0.090000   0.000000   0.090000 (  0.090643)

Ooh! Here we got what we expected from the modern language: mapping and reducing might be optimized comparing to dumb iteration and they occasionally were optimized. Both map and reduce are predictably faster than iterations. QED.

The difference in ruby is not significant, but the language is likely nudging us to use functional approach. Besides this code is a way more readable, it is just faster . While PHP... Well, thanks God my PHP diving season is over.