I'm finally happy with the code I wrote to browse wikipedia offline.

The most tricky part was to keep the database small. So I made a database with blocks of 256 articles. Each block is frozen using Storable and then compressed with Bzip2. Doing so, the created database is only about 15% larger than the original xml.bz2

I also use XML::Parser to parse wikipedia's database dump.

Here is the most difficult part: converting the XML database (see into a usable one:

#!/usr/bin/perl -w use v5.14; use strict; use warnings; die "please provide a database name" unless $ARGV[0]; my $rootname = $ARGV[0] =~ s/\.xml\.bz2\E//r =~ s,.*/,,r; use Encode; use XML::Parser; use IO::Uncompress::Bunzip2; use IO::Compress::Bzip2 qw(bzip2 $Bzip2Error); use Digest::MD5 qw(md5); use Storable qw(freeze thaw); open my $db, "> $rootname.db"; END { close $db } open my $t, "> $rootname.titles"; END { close $t } my ($title, @block, $char); my %debug; use DB_File; tie my %index, 'DB_File', "$rootname.index"; END { untie %index } $SIG{INT} = sub { die "caught INT signal" }; END { printf "%d entries made\n", scalar keys %index } sub store { my $freeze = freeze shift; bzip2 \($freeze, my $z); my $start = tell $db; print $db pack('L', length $z), $z; printf "block %d -> %d, compressed ratio is %2.2f%%\n", $start, tell($db), 100*length($z)/length($freeze), ; } my $parser = new XML::Parser Handlers => { Char => sub { shift; $char .= shift }, Start => sub { undef $char }, End => sub { shift; given( $_[0] ) { when( 'title' ) { $title = encode 'utf8', $char; say $t $title + } when( 'text' ) { push @block, $char; $index{length($title) > 16 ? md5 $title : $title} = pack 'LC', tell($db), scalar(@block) - 1; if (@block == 256) { store \@block; undef @block; } } } }, }; $parser->parse( new IO::Uncompress::Bunzip2 $ARGV[0] ); END { store \@block if @block }

I think it works pretty well, even if the rendering of the Text::Mediawiki module is a bit ugly for some pages. I need to take care of the references for instance. Still, it does the job, and it's much faster than on-line browsing.

I posted everything (including the CGI script) on my wikipedia userpage, as it also concerns wikipedia users:

EDIT. I also set up a github repo: