Re: Choosing the right sort (Or: WUDsamatter?)

[tlm]

in reply to Choosing the right sort (Or: WUDsamatter?)

Following up on a sharp observation of tye's in CB, I decided to check whether some of your sorts were in fact no-ops. Here's what I benchmarked, and the results:

srand 0;
my @fixture = map 'foo' . int( rand 1000 ). '.tla', 1..5000;
use Benchmark 'cmpthese';
cmpthese( -1,
          {
           naive       => sub { ( naive( @fixture ) )[ 0 ] },
           orcish      => sub { ( orcish( @fixture ) )[ 0 ] },
           schwartzian => sub { ( schwartzian( @fixture ) )[ 0 ] },
           guttros     => sub { ( guttros( @fixture ) )[ 0 ] },
          }
        );
__END__
              Rate       naive      orcish schwartzian     guttros
naive       3.05/s          --        -65%        -76%        -87%
orcish      8.74/s        186%          --        -31%        -61%
schwartzian 12.7/s        317%         46%          --        -44%
guttros     22.6/s        642%        159%         78%          --

Note that the subs that are benchmarked return the first element of the sorted array.

[ikegami]

Just to clarify Pustular Postulant, the OP must have called the functions in a void or scalar context. sort is optimized to do nothing in that situation, so the functions naive and orcish return instantly. Read more for confirmation of Pustular Postulant's results and confirmation that running the sorts in scalar context give results similar to those of the OP.

Read more... (2 kB)

That's why it's important to show one's work :)

[tlm]

Yes, that's what gave me the idea of asking for the first element of sorted array, though in hindsight there are definitely clearer, more direct ways to force a list context than the one I chose.