my @sorted =
    map $_->[0],
    sort {
        my $i = 0;
        {
            my $A = $a->[1][$i];
            my $B = $b->[1][$i];
            defined($A) || defined($B)       # Stop if both undef
            and (
                defined($A) <=> defined($B)  # Defined wins over undef
                or (
                    $A !~ /\d/ || $B !~ /\d/ # $A or $B is non-integer
                    ?    (lc $A cmp lc $B)   # ?? Stringy lowercase
                      || (   $A cmp    $B)   #    -> Tie breaker
                    : $A <=> $B              # :: $A and $B are intege
+rs
                )
                or ++$i && redo              # tie => next part
            );
        }
    }
    map [ $_, [ split /(\d+)/ ] ], @unsorted;
[download]

qw(
    bigbang
    x10y
    x9y
    bigboy
    bigBoy
    x11y
)
[download]

qw(
    bigbang
    bigBoy
    bigboy
    x9y
    x10y
    x11y
)
[download]

- Yes, I reinvent wheels.
- Spam: Visit eurotraQ.

First of all: A big THANKS to all who responded! The solutions offered by ariels and Juerd seem to come the closest to what I need (BTW: Thanks for pointing out that I need to rethink my requirements!). blakem's solution probably won't work, as I can't make the assumption mentioned by him about the data I get and moodster's solution doesn't seem to work for all cases (though I didn't quite find out why).
In any case, I learned a few more things, which is good... :-)

As for Juerd's code:
That's a Schwartzian Transform, right? (Didn't know it existed until today... :-})
However, there seems to be one problem: I get quite a few of Argument "" isn't numeric in numeric comparison (<=>) ... if any of the strings starts with a number. If I understand your code correctly, the problem lies in the : $A <=> $B part. When the string starts with a number, the split generates a first field that's defined, but empty. After replacing $A =~ /\D/ || $B =~ /\D/ with $A !~ /\d+/ || $B !~ /\d+/ (is there a better way?), it worked.

There's also a second problem, but that's partly due to my apparently incomplete requirements - I wasn't aware of that until I experimented a bit with the solutions offered here. Take the following data set:

Unsorted:x10y 1abc a10y x9y b1 abc DEF 123DEF bigboys big9boys 123DEFG 123def

Using your code or ariels' code, the result is:

Sorted:1abc 123DEF 123def 123DEFG a10y abc b1 big9boys bigboys DEF x9y x10y

Whether "big9boys" should appear before "bigboys" is probably debatable - I'm not 100% sure myself wich way round would be better in my case...
I am, however, wondering about "1abc" coming before "123DEF" (and the other "123..."), though I can't see a way of changing that without breaking the other requirements. I'll have to think about it.

Thanks again,
argathin

hat's a Schwartzian Transform, right?

It is. In short: by creating a datastructure that you store together with the original string, you don't have to re-build the original string and don't have to take apart the string at every iteration.

After replacing $A =~ /\D/ || $B =~ /\D/ with $A !~ /\d+/ || $B !~ /\d+/ (is there a better way?), it worked.

Your correction was correct. You can actually even drop the plusses. I'll update my code right away.

I am, however, wondering about "1abc" coming before "123DEF" (and the other "123...")

As I see it, that is correct. Integers should be used as such, according to your definition. That was the challenge :) First, the string is split to integer and non-integer parts, and then they are compared to eachother.

out of (1, abc) and (123, DEF), the first wins because 1 < 123.

- Yes, I reinvent wheels.
- Spam: Visit eurotraQ.

Actually, I think that would be correct. 1 would come before 123 in numeric sequence (according to your definition of the -dictionary sort). Also, I think big9boys would appear before bigboys, but just because in ascii order the digits come before letters.

Wow, I feel like the ultimate necroposter here.

I stumbled across this solution, and found it fell over when it met the likes of the following:

qw{
  1.01
  1.3
  1.02
  1.2
}
[download]

$A <=> $B does a numerical comparison, but strips any leading zeroes from the number, so how 1.02 and 1.2 are sorted will depend on their original order in the list - Tcl's dictionary sort doesn't seem to do this, and will place 1.2 above 1.02 in the sort.

So I extended your example to check for the string length of the compared numbers if they match numerically - If they match numerically, but have differing string lengths, then one must have leading zeroes. I also implemented the code into a subroutine:

sub dict_sort {
  my @unsorted = @_;
  my @sorted =
    map $_->[0],
    sort {
      my $i = 0;
      {
        my $A = $a->[1][$i];
        my $B = $b->[1][$i];
        defined($A) || defined($B)       # Stop if both undef
        and (
          defined($A) <=> defined($B)  # Defined wins over undef
          or (
            $A !~ /\d/ || $B !~ /\d/ # $A or $B is non-integer
            ?    (lc $A cmp lc $B)   # ?? Stringy lowercase
              || (   $A cmp    $B)   #    -> Tie breaker
            : $A <=> $B              # :: $A and $B are integers
              or (
                length($A) <=> length($B)  # If numeric comparison ret
+urns the same, check length to sort by leading zeroes
              )
          )
          or ++$i && redo              # tie => next part
        );
      }
    }
  map [ $_, [ split /(\d+)/ ] ], @unsorted;
  return @sorted;
}
[download]

I'm not sure that this thread will ever get any more posts, but hopefully this helps somebody out!

I stumbled across this solution, and found it fell over when it met the likes of the following: qw{ 1.01 1.3 1.02 1.2 };

$A <=> $B does a numerical comparison, but strips any leading zeroes from the number, so how 1.02 and 1.2 are sorted will depend on their original order in the list

No, the spaceship operator  <=> does no stripping of any kind, it does a numerical comparison, and numbers don't have leading zeros , 0002 is 02 is 2

The problem is with the code/regex, which turns decimals into an array of integers , and then does a numerical comparison on each portion -- not gonna work

I'm not too familiar with Tcl's sort order, but what works for me (esp for mp3s) is to lowercase the string, and pad all digits with zeroes (sometimes 6, sometimes 20)

#!/usr/bin/perl --
use strict; use warnings;
my @list = ( qw{
bigBoy bigbang bigboy x10y x9y x11y 
1.01   1.3   1.02   1.2
x1.1y x1.01y x2.3y x2.1y } );

print join "\n", 
  map { $$_[0] }
  sort { $$a[1] cmp $$b[1] }
  map {
    my $f = $_;
    $f =~ s/(\d+)/sprintf '%06d',$1/ge;
    [ $_, lc $f ]
  } @list;
__END__
1.01
1.02
1.2
1.3
bigbang
bigBoy
bigboy
x1.1y
x1.01y
x2.1y
x2.3y
x9y
x10y
x11y
[download]

#!/usr/local/bin/perl -w
use strict;

sub dictionary {
  dictionary_case(lc $a,lc $b) || dictionary_case($a,$b)
}

sub dictionary_case {
  my @a = split /(\d+)/, $_[0];
  my @b = split /(\d+)/, $_[1];
  my $lex = 1;
  for(;;) {
    my $x = shift @a;
    my $y = shift @b;
    if (! defined $x) {
      return -(defined $y);
    }
    elsif (! defined $y) {
      return +1;
    }
    my $c = $lex ? $x cmp $y : $x <=> $y;
    $lex = !$lex;
    return $c if $c;
  }
  return 0;
}

print join "\n",
  sort dictionary (qw(X10y x9y x bigboy y x11y bigbang bigBoy)),
  "\n";
[download]

#!/usr/bin/perl -wT
use strict;

{ # Create an interleaved 'AaBbCc' mapping
  my %map;
  $map{$_} = 2*(ord($_)-ord('A'))   for 'A'..'Z';
  $map{$_} = 2*(ord($_)-ord('a'))+1 for 'a'..'z';

  # use interleaved map to munge word into a string that will sort "co
+rrectly"
  sub dictize {
    my $word = shift;
    $word =~ s/(\d+|[A-Za-z])/chr($map{$1}||60+$1)/ge;
    return $word;
  }
}


my @words = qw(bigBoy bigbang bigboy
               x10y x9y x11y);

# Sort based on lowercased munged words,
# use the non-lowercased version as a tiebreaker
my @sorted = map  {$_->[0]}
             sort {$a->[1] cmp $b->[1] or
                   $a->[2] cmp $b->[2]}
             map  {[$_,dictize(lc),dictize($_)]} 
             @words;

print "$_\n" for @sorted;

__END__

bigbang
bigBoy
bigboy
x9y
x10y
x11y
[download]

-Blake

sub dictcmp
{
   my $ac = $a;
   my $bc = $b;
   while( 1 ) {
      my @a = $ac =~ /^\s*(([A-Za-z]+)|(\d+))(.*)/;
      my @b = $bc =~ /^\s*(([A-Za-z]+)|(\d+))(.*)/;
      return 0 if !defined $a[0] & !defined $b[0];
      return -1 if !defined $a[0];
      return 1 if !defined $b[0];
      my $res;
      if( $a[0] =~ /\d+/ && $b[0] =~ /\d+/) {
         $res = $a[0] <=> $b[0];
      }
      else {
         $res = $a[0] cmp $b[0];
      }
      return $res if $res;
      $ac = $a[3];
      $bc = $b[3];
   }
}
[download]

my @list = ( "x10 y", "1abc", "a10y", "x9y", " b1" );
print join "\n", sort dictcmp @list;
[download]

1abc
a10y
 b1
x9y
x10 y
[download]

Cheers,
--Moodster