Re: trim() magic

[Aristotle]

I tried to see if there were any knobs to twiddle on this one, using dragonchild’s benchmark cases.

First thing I tried: using a recursive call to alias $_. This lets you get rid of the ternary in the for list.

sub trim {
    return trim( $_ ) if not @_;
    @_ = @_ if defined wantarray;
    for ( @_ ) { s/^\s+//, s/\s+$// }
    return wantarray ? @_ : $_[ 0 ] if defined wantarray;
}

On my setup this is about 15% slower for the “inplace replacement of implicit $_” case, but ekes out a few percentage points on the other cases. But it let me proceed to switch from duplicate defined wantarray tests to a duplicate inner loop:

sub trim2 {
    return trim2( $_ ) if not @_;
    return map { local $_ = $_; s/^\s+//, s/\s+$//; $_ } @_
        if defined wantarray;
    for ( @_ ) { s/^\s+//, s/\s+$// }
}

This gets back most of the lost speed in the “inplace replacement of implicit $_” case, has roughly the same performance in other void contexts, but is also about 50% faster in many other cases, including the IMHO most important one – passing a scalar and assigning to one.

#!/usr/bin/perl
use strict;
use warnings;

use Benchmark qw( cmpthese timethese );

sub trim1 {
    @_ = $_ if not @_ and defined wantarray;
    @_ = @_ if defined wantarray;
    for ( @_ ? @_ : $_ ) { s/^\s+//, s/\s+$// }
    return wantarray ? @_ : $_[ 0 ] if defined wantarray;
}

sub trim2 {
    return trim2( $_ ) if not @_;
    return map { local $_ = $_; s/^\s+//, s/\s+$//; $_ } @_
        if defined wantarray;
    for ( @_ ) { s/^\s+//, s/\s+$// }
}

my $cpu = -1;

cmpthese timethese $cpu => {
    undef_default_1 => sub { $_ = '  asdf  '; trim1(); },
    undef_default_2 => sub { $_ = '  asdf  '; trim2(); },
};

cmpthese timethese $cpu => {
    scalar_default_1 => sub { $_ = '  asdf  '; my $n = trim1(); },
    scalar_default_2 => sub { $_ = '  asdf  '; my $n = trim2(); },
};

cmpthese timethese $cpu => {
    scalar_passed_1 => sub { my $x = '  asdf  '; my $n = trim1( $x ); 
+},
    scalar_passed_2 => sub { my $x = '  asdf  '; my $n = trim2( $x ); 
+},
};

cmpthese timethese $cpu => {
    list_default_1 => sub { $_ = '  asdf  '; my @n = trim1(); },
    list_default_2 => sub { $_ = '  asdf  '; my @n = trim2(); },
};

cmpthese timethese $cpu => {
    list_passed_1 => sub { my @l = ( '  asdf  ', '  asdf  ' ); my @n =
+ trim1( @l ); },
    list_passed_2 => sub { my @l = ( '  asdf  ', '  asdf  ' ); my @n =
+ trim2( @l ); },
};

cmpthese timethese $cpu => {
    undef_passed_1 => sub { my @l = ( '  asdf  ', '  asdf  ' ); trim1(
+ @l ); },
    undef_passed_2 => sub { my @l = ( '  asdf  ', '  asdf  ' ); trim2(
+ @l ); },
};

Sample run:

Benchmark: running undef_default_1, undef_default_2 for at least 1 CPU seconds...
undef_default_1:  0 wallclock secs ( 1.05 usr +  0.01 sys =  1.06 CPU) @ 190933.96/s (n=202390)
undef_default_2:  1 wallclock secs ( 1.04 usr +  0.00 sys =  1.04 CPU) @ 174121.15/s (n=181086)
                    Rate undef_default_2 undef_default_1
undef_default_2 174121/s              --             -9%
undef_default_1 190934/s             10%              --
Benchmark: running scalar_default_1, scalar_default_2 for at least 1 CPU seconds...
scalar_default_1:  1 wallclock secs ( 1.11 usr +  0.00 sys =  1.11 CPU) @ 103321.62/s (n=114687)
scalar_default_2:  1 wallclock secs ( 1.11 usr +  0.00 sys =  1.11 CPU) @ 154982.88/s (n=172031)
                     Rate scalar_default_1 scalar_default_2
scalar_default_1 103322/s               --             -33%
scalar_default_2 154983/s              50%               --
Benchmark: running scalar_passed_1, scalar_passed_2 for at least 1 CPU seconds...
scalar_passed_1:  1 wallclock secs ( 1.09 usr +  0.00 sys =  1.09 CPU) @ 121405.50/s (n=132332)
scalar_passed_2:  2 wallclock secs ( 1.04 usr +  0.00 sys =  1.04 CPU) @ 183794.23/s (n=191146)
                    Rate scalar_passed_1 scalar_passed_2
scalar_passed_1 121406/s              --            -34%
scalar_passed_2 183794/s             51%              --
Benchmark: running list_default_1, list_default_2 for at least 1 CPU seconds...
list_default_1:  2 wallclock secs ( 1.09 usr +  0.00 sys =  1.09 CPU) @ 98641.28/s (n=107519)
list_default_2:  2 wallclock secs ( 1.04 usr +  0.00 sys =  1.04 CPU) @ 150376.92/s (n=156392)
                   Rate list_default_1 list_default_2
list_default_1  98641/s             --           -34%
list_default_2 150377/s            52%             --
Benchmark: running list_passed_1, list_passed_2 for at least 1 CPU seconds...
list_passed_1:  1 wallclock secs ( 1.07 usr +  0.00 sys =  1.07 CPU) @ 80736.45/s (n=86388)
list_passed_2:  1 wallclock secs ( 1.08 usr +  0.00 sys =  1.08 CPU) @ 99554.63/s (n=107519)
                 Rate list_passed_1 list_passed_2
list_passed_1 80736/s            --          -19%
list_passed_2 99555/s           23%            --
Benchmark: running undef_passed_1, undef_passed_2 for at least 1 CPU seconds...
undef_passed_1:  1 wallclock secs ( 1.07 usr +  0.00 sys =  1.07 CPU) @ 133980.37/s (n=143359)
undef_passed_2:  1 wallclock secs ( 1.13 usr +  0.00 sys =  1.13 CPU) @ 138400.00/s (n=156392)
                   Rate undef_passed_1 undef_passed_2
undef_passed_1 133980/s             --            -3%
undef_passed_2 138400/s             3%             --

[Aristotle]

What I’d do differently now: at least this:

sub trim2 {
    return trim2 $_ if not @_;
    return map { s/^\s+//; s/\s+$//; $_ } my @c = @_
        if defined wantarray;
    for ( @_ ) { s/^\s+//, s/\s+$// }
}

And someone who can be bothered should benchmark the following variants:

sub trim3a {
    return trim3a $_ if not @_;

    s/^\s+//, s/\s+$// for
                wantarray ? my @c = @_     :
        defined wantarray ? my $c = $_[-1] :
                                    @_;

    @c ? @c : $c ? $c : ();
}

sub trim3b {
    return trim3b $_ if not @_;

    my $w = wantarray;
    s/^\s+//, s/\s+$// for
                $w ? my @c = @_     :
        defined $w ? my $c = $_[-1] :
                             @_;

    @c ? @c : $c ? $c : ();
}