Fellow boriz
Thank you very much for your suggestion, but I'm not sure if

perl -MData::Dumper -e "print Dumper @{$_[3]}{'a','b','c','d'}, $_[3];
+ "
$VAR1 = undef;
$VAR2 = undef;
$VAR3 = undef;
$VAR4 = undef;
$VAR1 = undef;
$VAR2 = undef;
$VAR3 = undef;
$VAR4 = undef;
$VAR1 = undef;
$VAR2 = undef;
$VAR3 = undef;
$VAR4 = undef;
$VAR5 = '1';
[download]

is the same thing as

perl -MData::Dumper -e 'print Dumper @$data{"a","b","c"}, $data'
$VAR1 = undef;
$VAR2 = undef;
$VAR3 = undef;
$VAR4 = {
          'c' => undef,
          'a' => undef,
          'b' => undef
        };
[download]

or I'm doing something really wrong here, or your data structure definition is not the same as my example. :-) Nice try, anyway.

-- monsieur_champs

Your $data and @_ are empty here, so this won't work the same because of autovivication. The number of $VAR variables is different, because you have a .. d in the first and a .. c in the second expample.

All of this should work if you actually fill $data and $_[3] with a hash reference that uses those keys.

"What should it profit a man, if he should win a flame war, yet lose his cool?"

My perl works this way, Im sure there is another mistake.

#!/usr/bin/perl

use Data::Dumper;

%xx = ( a => 1, b => 2 );

sub inputBox{
    @{$_[3]}{'a','b','c','d'};
}
sub inputBox2{
  my $data = $_[3];
    @$data{'a','b','c','d'};
}
print Dumper (inputBox( 1,2,3,\%xx));
print Dumper (inputBox2( 1,2,3,\%xx));
[download]

Boris

In fact, if you use $_[3] a lot in the same function, it might even be slower than copying it to a lexical variable, because of the array indexing that needs to be done. (haven't benchmarked this at all, so I might be wrong)

use strict;
use Benchmark qw(:all);

my %data = (
    Lore    =>'mipsumdo', ente    =>'squemoll',
    lors    =>'itametco', isod    =>'ioetnonu',
    nsec    =>'tetuerad', mmyf    =>'acilisis',
    ipis    =>'cingelit', augu    =>'eliberoi',
    Sedr    =>'honcusma', acul    =>'isodioat',
    ssai    =>'dmollisp', cons    =>'ectetuer',
    hare    =>'travelit', nisl    =>'nislquis',
    null    =>'amattise', feli    =>'sDonecma',
    nimq    =>'uisferme', gnar    =>'isusulla',
    ntum    =>'nequenul', mcor    =>'peridele',
    lase    =>'dduiDone', ifen    =>'dutfeugi',
    cbla    =>'nditmetu', atas    =>'emAliqua',
    svit    =>'aecondim', msed    =>'magnaado',
    entu    =>'mluctusa', lorn    =>'onummysa',
    ntem    =>'assaeuis', gitt    =>'isNuncne',
    moda    =>'nteaport', corc    =>'iMorbima',
    amii    =>'psumnonl', ttis    =>'blandits',
    eoAl    =>'iquamcon', emPh    =>'asellusq',
    dime    =>'ntumblan', uiso    =>'rciInfer',
    ditf    =>'elisPell', ment    =>'umturpis',
);

sub WithCopy {
    my $hashref = $_[0];

    my @dummy = @{$hashref}{'Lore','lors','nsec'};
}

sub WithoutCopy {
    my @dummy = @{$_[0]}{'Lore','lors','nsec'};
}

cmpthese(5000000, {
    'WithCopy'        => 'WithCopy(\%data);',
    'WithoutCopy'    => 'WithoutCopy(\%data);',
});

__END__

                Rate    WithCopy WithoutCopy
WithCopy    548908/s          --         -6%
WithoutCopy 581801/s          6%          --
[download]

On my machine that gives (with 500000 iterations)

                Rate    WithCopy WithoutCopy
WithCopy    450450/s          --         -9%
WithoutCopy 495050/s         10%          --
[download]

But if you access the hashref more than once in the functions - say about 10 times:

use strict;
use Benchmark qw(:all);
 
my %data = (
    Lore    =>'mipsumdo', ente    =>'squemoll',
    lors    =>'itametco', isod    =>'ioetnonu',
    nsec    =>'tetuerad', mmyf    =>'acilisis',
    ipis    =>'cingelit', augu    =>'eliberoi',
    Sedr    =>'honcusma', acul    =>'isodioat',
    ssai    =>'dmollisp', cons    =>'ectetuer',
    hare    =>'travelit', nisl    =>'nislquis',
    null    =>'amattise', feli    =>'sDonecma',
    nimq    =>'uisferme', gnar    =>'isusulla',
    ntum    =>'nequenul', mcor    =>'peridele',
    lase    =>'dduiDone', ifen    =>'dutfeugi',
    cbla    =>'nditmetu', atas    =>'emAliqua',
    svit    =>'aecondim', msed    =>'magnaado',
    entu    =>'mluctusa', lorn    =>'onummysa',
    ntem    =>'assaeuis', gitt    =>'isNuncne',
    moda    =>'nteaport', corc    =>'iMorbima',
    amii    =>'psumnonl', ttis    =>'blandits',
    eoAl    =>'iquamcon', emPh    =>'asellusq',
    dime    =>'ntumblan', uiso    =>'rciInfer',
    ditf    =>'elisPell', ment    =>'umturpis',
);
 
sub WithCopy {
    my $hashref = $_[0];
 
    my @dummy1 = @{$hashref}{'Lore','lors','nsec'};
    my @dummy2 = @{$hashref}{'Lore','lors','nsec'};
    my @dummy3 = @{$hashref}{'Lore','lors','nsec'};
    my @dummy4 = @{$hashref}{'Lore','lors','nsec'};
    my @dummy5 = @{$hashref}{'Lore','lors','nsec'};
    my @dummy6 = @{$hashref}{'Lore','lors','nsec'};
    my @dummy7 = @{$hashref}{'Lore','lors','nsec'};
    my @dummy8 = @{$hashref}{'Lore','lors','nsec'};
    my @dummy9 = @{$hashref}{'Lore','lors','nsec'};
    my @dummy0 = @{$hashref}{'Lore','lors','nsec'};
}
 
 
sub WithoutCopy {
   my @dummy1 = @{$_[0]}{'Lore','lors','nsec'};
   my @dummy2 = @{$_[0]}{'Lore','lors','nsec'};
   my @dummy3 = @{$_[0]}{'Lore','lors','nsec'};
   my @dummy4 = @{$_[0]}{'Lore','lors','nsec'};
   my @dummy5 = @{$_[0]}{'Lore','lors','nsec'};
   my @dummy6 = @{$_[0]}{'Lore','lors','nsec'};
   my @dummy7 = @{$_[0]}{'Lore','lors','nsec'};
   my @dummy8 = @{$_[0]}{'Lore','lors','nsec'};
   my @dummy9 = @{$_[0]}{'Lore','lors','nsec'};
   my @dummy0 = @{$_[0]}{'Lore','lors','nsec'};
}
 
 
cmpthese(500000, {
    'WithCopy'        => 'WithCopy(\%data);',
    'WithoutCopy'    => 'WithoutCopy(\%data);',
});
 
__END__
               Rate WithoutCopy    WithCopy
WithoutCopy 82645/s          --         -8%
WithCopy    89445/s          8%          --
[download]

Which seems to show that the performance increase is less when you index the @_ array a lot.

Of course it doesn't show that - it could also mean that the time needed to copy the keys of the hash reference 10 times overshadows the time needed to index the array and/or copy the hash reference, thereby decreasing the influence of doing either.

Benchmarking is hard.

"What should it profit a man, if he should win a flame war, yet lose his cool?"

Interesting point. I use $data 8 times. Half as a boolean test (no de-reference) and half as a hash reference (to recover one or more values). I guess (and maybe I'm wrong about this) that it doesn't worth the benchmark...

Here is the complete function code:

sub inputBox{
  # Don't uncomment, use @_ instead
  #   my $action_url = shift; # $_[0]
  #   my $msgid      = shift; # $_[1]
  #   my $action     = shift; # $_[2]
  my $data = $_[3]; # oh, I needed this one, I wasn't able to use a ha
+sh-slice from a hash-reference in one shot.
<readmore>
  return
qq{<form action="$_[0]" method="post">\n}.
($_[1]? qq{  <input type="hidden" name="message_id" value="$_[1]" />\n
+}:'').
qq{  <input type="hidden" name="action" value="$_[2]" />
  <input type="hidden" name="commit" value="1" />
  <table align="center" border="1" cellpadding="5" cellspacing="0">
    <tr valign="middle" align="center" bgcolor="$color{TABLE_HEAD}">
      <td colspan="2"> }
  .($_[2] eq 'edit'? "Edit Message #$_[1]" : 'Compose New Message' ).
  qq{ </td>
    </tr>
    <tr valign="middle" align="left" >
      <td> Author: </td>
      <td>
}.( $_[2] eq 'edit'? q{        <input type="text" name="author" size="
+40" }.
  ( $data? qq{value="$data->{author}"} : &TWiki::Func::getWikiUserName
+() )
 . qq{ />} : &TWiki::Func::getWikiUserName() ) . q{
      </td>
    </tr>
    <tr valign="middle" align="left">
      <td>Due Date: </td>
      <td>} . &gen_date_selector( 'due', ( $data? @$data{'day','month'
+,'year','hour','minute'} : (localtime)[5]+1900, (localtime)[4]+1, (lo
+caltime)[3], 23, 59 ) ) . qq{
      </td>
    </tr>
    <tr valign="middle" align="left">
      <td>Message: </td>
      <td>
        <textarea rows="5" name="msg" cols="50">}.($data? $data->{msg}
+ : '' ).qq{</textarea>
      </td>
    </tr>
    <tr valign="middle" align="left">
      <td>
        <input type="checkbox" name="dropped" value="Y" }.($data && $d
+ata->{dropped} eq 'Y'? 'CHECKED':'' ).q{>&nbsp;Dropped.
      </td>
      <td align="right">
        <input type="submit" name="change" value="Change Message">
      </td>
    </tr>
  </table>
</form>
};
}
[download]

-- monsieur_champs

Like Joost, I am a bit skeptical about the need for this "optimization." If the performance of your application hinges on the speed of accessing subroutine arguments, then I think Perl might not be the right choice. Have you done any profiling to see if this is the case, or are you simply stabbing in the dark? I suggest you do the former, rather than the latter, when it comes to expending time and effort to optimize your code.

Dear fellow revdiablo

In fact, you're right about this: I'm stabbling in the dark.

But I'm not alone: this is a standard (?), required (?) optimization for TWiki Plugins like this and this.

Maybe I shall alert the main project developers about this "optimization".

Thank you for the insight.

-- monsieur_champs