Re^5: Concerning hash operations (appending, concatenating)

[japhy]

Here's my benchmark:

use Benchmark 'cmpthese';
use strict;

my %hash = 1 .. 10_000;
my %add = 9_001 .. 11_000;

cmpthese(-5, {
  BASE => sub {
    my %copy = %hash;
  },
  JAPHY => sub {
    my %copy = %hash;
    my @new = grep !exists $copy{$_}, keys %add;
    @copy{@new} = @add{@new};
  },
  ZAXO => sub {
    my %copy = %hash;
    %copy = (%add, %copy);
  },
  NULL => sub {
    my %copy = %hash;
    %copy = (%copy);
  },
});

Here are the results I get:

        Rate  ZAXO  NULL JAPHY  BASE
ZAXO  15.1/s    --  -22%  -61%  -76%
NULL  19.3/s   28%    --  -50%  -69%
JAPHY 38.9/s  157%  101%    --  -38%
BASE  62.4/s  312%  223%   60%    --

The two cases 'BASE' and 'NULL' show that Perl does not optimize %hash = %hash at all, since 'NULL' is three times slower than 'BASE'. 'JAPHY' is 2.5 times faster than 'ZAXO' in this run, with 500 old keys and 500 new keys. When I change %add to be (5_001 .. 12_000), the run times are:

        Rate  ZAXO  NULL JAPHY  BASE
ZAXO  11.4/s    --  -39%  -58%  -81%
NULL  18.6/s   63%    --  -31%  -69%
JAPHY 26.9/s  136%   45%    --  -56%
BASE  60.5/s  431%  225%  125%    --

The point is that %a = (%b, %a) is slow because it is O(N+M) where N is the size of %a and M is the size of %b. My code is O(M) -- specifically, O(M(1+x)) where x is the fraction of keys in %b that aren't in %a -- because it doesn't iterate over %a at all, only %b.