http://qs1969.pair.com?node_id=11134740


in reply to Using 'keys' on a list

Further down in this thread, I got trolled by an Anon Monk.

Instead of complicated syntax, I would recommend something easier.
Performance-wise it is always better to return fewer things from the sub() that more things.
That should be obvious!

The fastest way (performance-wise) and often the easiest way for a sub() to return a hash to the caller is via a reference to the hash.
Don't worry at all about taking a couple of lines of code vs one line. Sometimes 3 lines are faster and easier to understand than a single line.

Here are some benchmarks for your perusal.
Number (3) below is a "go to" winner!

use strict; use warnings; use Benchmark; use Data::Dumper; # Note: You can increment a string in Perl! WOW! # As long as you don't use the string in a numeric context! # $string++ is quite different than $string +=1 # # Below, this is used to make a bunch of unique hash keys # I don't think that the fact that they are sequential in # an alphabetic sense makes much difference in the generated # hash table because of the way that the Perl hash algorithm # works. sub return_hash { # create a large hash and return that entire hash as list my $string = "ABCDEFGHIJ"; my %hash = map{$string++ => 1}(1..100000); return %hash; } sub return_hash_ref { # create a large hash and return a ref to that hash my $string = "ABCDEFGHIJ"; my %hash = map{$string++ => 1}(1..100000); return \%hash; } sub return_just_keys { # create a large hash and return just the keys of that hash my $string = "ABCDEFGHIJ"; my %hash = map{$string++ => 1}(1..100000); return keys %hash; } timethese(1000, { '1)Keys of Hash via list' => 'my @keys = keys %{{return_hash()}}', '2)Keys of Local Hash copy' => 'my %hash2 = return_hash(); my @key +s = keys %hash2;', '3)Keys of local Hash Ref' => 'my $href = return_hash_ref(); my @k +eys = keys %$href;', '4)Just the returned keys' => 'my @keys = return_just_keys()', }); __END__ Benchmark: timing 1000 iterations of 1)Keys of Hash via list, 2)Keys of Local Hash copy, 3)Keys of local Hash Ref, 4)Just the returned keys... 1)Keys of Hash via list: 182 wallclock secs (179.20 usr + 1.97 sys = +181.17 CPU) @ 5.52/s (n=1000) 2)Keys of Local Hash copy: 184 wallclock secs (182.69 usr + 0.42 sys += 183.11 CPU) @ 5.46/s (n=1000) 3)Keys of local Hash Ref: 130 wallclock secs (129.56 usr + 0.72 sys = + 130.28 CPU) @ 7.68/s (n=1000) 4)Just the returned keys: 125 wallclock secs (125.20 usr + 0.52 sys = + 125.72 CPU) @ 7.95/s (n=1000)
Added: When looking at benchmarks, don't worry so much about 182 vs 184. That could be an artifact of a particular run. Most of the CPU MIP's are being consumed by creating and re-creating the hash. Look at ~180 vs ~130. The relative performance difference between the 2 methods of passing the result back is actually much, more than that because generating the hash "takes a lot of effort".