### Re: Dictionary-style sort a la Tcl?

by Juerd (Abbot)
 on Apr 18, 2002 at 11:43 UTC ( #160175=note: print w/replies, xml ) Need Help??

in reply to Dictionary-style sort a la Tcl?

```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;
That's my try, and it seems to work perfectly with the given example, and some of my own tries.

@unsorted
```qw(
bigbang
x10y
x9y
bigboy
bigBoy
x11y
)
@sorted
```qw(
bigbang
bigBoy
bigboy
x9y
x10y
x11y
)
Update - integer-detection altered per argarthin's reply.

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

Replies are listed 'Best First'.
Re: Re: Dictionary-style sort a la Tcl?
by argathin (Acolyte) on Apr 18, 2002 at 12:56 UTC
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.
Yes, you are right, of course - it just looked strange on first glance. However, after I've reviewed my requirements, I don't think it'll cause any problems either way, so the solutions offered here are perfect. :-)

Cheerio,
argathin

Re^2: Dictionary-style sort a la Tcl?
by Rohaq (Initiate) on Feb 02, 2012 at 15:09 UTC

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
}
\$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;
}

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
Sorry, I guess I could have worded that better - it doesn't 'strip' leading zeroes, it just ignores them as it's doing a numerical comparison. In any case, the change I added seems to make the sort identical to the Tcl dictionary sort. I'm not sure about yours.

Create A New User
Node Status?
node history
Node Type: note [id://160175]
help
Chatterbox?
and all is quiet...

How do I use this? | Other CB clients
Other Users?
Others making s'mores by the fire in the courtyard of the Monastery: (5)
As of 2017-11-19 22:40 GMT
Sections?
Information?
Find Nodes?
Leftovers?
Voting Booth?
In order to be able to say "I know Perl", you must have:

Results (282 votes). Check out past polls.

Notices?