hild1.no

Kagee's random rantings

Perl Regexp Oneliners and UTF-8

For my project to find as many .no domains as possible, I needed a regexp for extracting valid domains. This task is made more fun by the inclusion of Norwegian and Sami characters in the set of valid characters.

In addition to [a-z0-9\-], valid dot-no domains can contain the Norwegian æ (ae), ø (o with stroke) and å (a with ring above) (Stargate, anyone?) and a number of Sami characters. ŧ (t with stroke), ç (c with cedilla) and ŋ (simply called “eng”) are some of my favourites.

The following code will print only the first match per line, and uses ŧ directly in the regexp.

1
2
echo "fooŧ.no baŧ.no" | perl -ne 'if(/([a-zŧ]{2,63}\.no)/ig) { print $1,"\n"; }'
fooŧ.no

If we replace if with while we will print any match found in the whole line.

1
2
3
echo "fooŧ.no baŧ.no" | perl -ne 'while(/([a-zŧ]{2,63}\.no)/ig) { print $1,"\n"; }'
fooŧ.no
baŧ.no

Because I’m afraid the regexp (specifically the non-ASCII characters) may be mangled by being saved and moved between systems, I want to write the Norwegian and Sami characters using their Unicode code points. Perl has support for this using \x{<number>} (see perl unicode)

1
2
3
echo "fooŧ.no baŧ.no" | perl -CSD -ne 'while(/([a-z\x{167}]{2,63}\.no)/ig) { print $1,"\n"; }'
fooŧ.no
baŧ.no

When using code points, I have to specify -CSD for the matching to work. I am not really sure why this is required. If you can explain, please comment or tell my by other means. As you can read in perlrun, -CSD specifies that STDIN, STDOUT, STDERR and all input and output streams should be treated as being UTF-8.

Another problem is that if this last solution is is fed invalid UTF-8, it will die fatally and stop processing input.

1
Malformed UTF-8 character (fatal) at -e line 1, <> line X.

To prevent this happening I currently sanitize my dirty input using iconv -f utf-8 -t utf-8 -c. If you have a better solution for this, Perl or otherwise, please tell me!.

A simple regexp would match the valid characters for a length between 2 and 63 followed by .no. However, I wanted only and all “domains under .no” as counted by Norid in their statistics. Norids definition of “domains under .no” are all the domains directly under .no, but also domains under category domains i.e. ohv.oslo.no and ola.priv.no. To get comparable results, I have to collect both *.no and *.<category domain>.no domains when scraping data.

The resulting “oneliner” I use is this…. It once was a oneliner, but with more than 10k characters in the regexp it was hard to manage. The resulting script builds up a regexp that is valid for all Norwegian domains using a list of valid category domains, all valid characters and other rules for .no domains.

Comments