summaryrefslogtreecommitdiff
path: root/lib/gcstar/GCPlugins/GCbooks/GCInternetBokHandeln.pm
diff options
context:
space:
mode:
authorJörg Frings-Fürst <jff@merkur>2014-07-06 15:20:38 +0200
committerJörg Frings-Fürst <jff@merkur>2014-07-06 15:20:38 +0200
commit126bb8cb6b93240bb4d3a2b816b74c286c3d422b (patch)
treee66e1dfe77d53a52539489765c88d23e4423ae27 /lib/gcstar/GCPlugins/GCbooks/GCInternetBokHandeln.pm
Imported Upstream version 1.7.0upstream/1.7.0
Diffstat (limited to 'lib/gcstar/GCPlugins/GCbooks/GCInternetBokHandeln.pm')
-rw-r--r--lib/gcstar/GCPlugins/GCbooks/GCInternetBokHandeln.pm464
1 files changed, 464 insertions, 0 deletions
diff --git a/lib/gcstar/GCPlugins/GCbooks/GCInternetBokHandeln.pm b/lib/gcstar/GCPlugins/GCbooks/GCInternetBokHandeln.pm
new file mode 100644
index 0000000..3b553e9
--- /dev/null
+++ b/lib/gcstar/GCPlugins/GCbooks/GCInternetBokHandeln.pm
@@ -0,0 +1,464 @@
+package GCPlugins::GCbooks::GCbooksInternetBokHandeln;
+
+###################################################
+#
+# Copyright 2005-2010 Christian Jodar
+#
+# This file is part of GCstar.
+#
+# GCstar is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or
+# (at your option) any later version.
+#
+# GCstar is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with GCstar; if not, write to the Free Software
+# Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
+#
+###################################################
+
+use strict;
+
+use GCPlugins::GCbooks::GCbooksCommon;
+
+{
+ package GCPlugins::GCbooks::GCPluginInternetBokHandeln;
+
+ use base qw(GCPlugins::GCbooks::GCbooksPluginsBase);
+ use URI::Escape;
+
+ sub start
+ {
+ my ($self, $tagname, $attr, $attrseq, $origtext) = @_;
+
+ $self->{inside}->{$tagname}++;
+
+ if ($self->{parsingList})
+ {
+
+ if (($tagname eq 'span') && ($attr->{class} eq 'title1'))
+ {
+ $self->{isFound} = 1 ;
+ $self->{itemIdx}++;
+ $self->{itemsList}[$self->{itemIdx}]->{url} = $self->{loadedUrl};
+ }
+ elsif (($tagname eq 'td') && ($attr->{rowspan} eq '4') && ($self->{isBook} eq '0') && ($self->{isFound} eq 0))
+ {
+ # En fait la sequence est un peu tordue. Je cherche le deuxieme passage dans la sequence
+ # rowspan/a
+ $self->{isBook} = 1 ;
+ $self->{isUrl} = 1 ;
+ }
+ elsif (($tagname eq 'img') && ($self->{isBook} eq '1') && ($self->{isUrl}))
+ {
+ $self->{isBook} = 2 ;
+ }
+ elsif (($tagname eq 'a') && ($self->{isBook} eq '2') && ($self->{isUrl}))
+ {
+ $self->{itemIdx}++;
+ $self->{itemsList}[$self->{itemIdx}]->{url} = "http://www.internetbokhandeln.se" . $attr->{href};
+ $self->{isUrl} = 0 ;
+ $self->{isTitle} = 1 ;
+ }
+ elsif (($tagname eq 'span') && ($attr->{class} eq 'author') && ($self->{isFound} eq 0))
+ {
+ $self->{isAuthor} = 1 ;
+ $self->{isBook} = 0 ;
+ }
+ elsif (($tagname eq 'span') && ($attr->{class} eq 'shaded') && ($self->{isFound} eq 0))
+ {
+ $self->{isEditor_Publication_Format_Lang} = 1 ;
+ $self->{isBook} = 0 ;
+ }
+ }
+ else
+ {
+ if ($self->{isAuthor} eq 1)
+ {
+ $self->{isAuthor} = 2 ;
+ }
+ elsif ($self->{isPublisher} eq 1)
+ {
+ $self->{isPublisher} = 2 ;
+ }
+ elsif ($self->{isISBN} eq 1)
+ {
+ $self->{isISBN} = 2 ;
+ }
+ elsif ($self->{isFormat} eq 1)
+ {
+ $self->{isFormat} = 2 ;
+ }
+ elsif ($self->{isEdition} eq 1)
+ {
+ $self->{isEdition} = 2 ;
+ }
+ elsif ($self->{isPage} eq 1)
+ {
+ $self->{isPage} = 2 ;
+ }
+ elsif ($self->{isLanguage} eq 1)
+ {
+ $self->{isLanguage} = 2 ;
+ }
+ elsif ($self->{isPublication} eq 1)
+ {
+ $self->{isPublication} = 2 ;
+ }
+ elsif ($self->{isSerie} eq 1)
+ {
+ $self->{isSerie} = 2 ;
+ }
+ elsif (($tagname eq 'span') && ($attr->{class} eq 'title1'))
+ {
+ $self->{isTitle} = 1 ;
+ # On initialise la variable ( sinon d une fiche sur l autre est n est pas reinitialisee )
+ $self->{isDescription} = 0;
+ }
+ elsif (($tagname eq 'span') && ($attr->{class} eq 'font5'))
+ {
+ $self->{isAnalyse} = 1 ;
+ }
+ elsif (($tagname eq 'p') && ($self->{curInfo}->{isbn} ne '') && ($self->{curInfo}->{description} eq '') && ($self->{isDescription} ne 2))
+ {
+ $self->{isDescription} = 1 ;
+ }
+ elsif (($tagname eq 'div') && ($attr->{id} eq 'largebook'))
+ {
+ # Pour etre sur s il n y a pas de commentaire de ne pas prendre n importe quoi
+ $self->{isDescription} = 2 ;
+ }
+ elsif (($tagname eq 'td') && ($attr->{class} eq 'pricecolumn'))
+ {
+ $self->{isCover} = 1 ;
+ }
+ elsif (($tagname eq 'img') && ($self->{isCover} eq 1))
+ {
+ # le but est de determiner s il y a une couverture ou non, et s il y en a une, on recuperera
+ # la version grand format qui est bien plus tard
+ if ($attr->{onclick} eq 'return showBig();')
+ {
+ $self->{isCover} = 2 ;
+ }
+ else
+ {
+ if ($attr->{src} eq '/i/dummy.gif')
+ {
+ # Il n y a pas d image
+ $self->{isCover} = 3 ;
+ }
+ else
+ {
+ $self->{curInfo}->{cover} = $attr->{src} ;
+ $self->{isCover} = 3 ;
+ }
+ }
+ }
+ elsif (($tagname eq 'img') && ($attr->{onclick} eq 'return hideBig();') && ($self->{isCover} eq 2))
+ {
+ $self->{curInfo}->{cover} = $attr->{src} ;
+ $self->{isCover} = 3 ;
+ }
+
+ }
+ }
+
+ sub end
+ {
+ my ($self, $tagname) = @_;
+
+ $self->{isFound} = 0;
+ $self->{inside}->{$tagname}--;
+ }
+
+ sub text
+ {
+ my ($self, $origtext) = @_;
+
+ if ($self->{parsingList})
+ {
+ if ($self->{isTitle})
+ {
+ $self->{itemsList}[$self->{itemIdx}]->{title} = $origtext;
+ $self->{isTitle} = 0 ;
+ }
+ elsif ($self->{isAuthor})
+ {
+ my @array = split(/;/,$origtext);
+ my $element;
+ foreach $element (@array)
+ {
+ my @nom_prenom = split(/,/,$element);
+ # Enleve les blancs en debut de chaine
+ $nom_prenom[0] =~ s/^\s//;
+ $nom_prenom[1] =~ s/^\s//;
+ # Enleve les blancs en fin de chaine
+ $nom_prenom[0] =~ s/\s$//;
+ $nom_prenom[1] =~ s/\s$//;
+ if ($self->{itemsList}[$self->{itemIdx}]->{authors} eq '')
+ {
+ if ($nom_prenom[1] ne '')
+ {
+ $self->{itemsList}[$self->{itemIdx}]->{authors} = $nom_prenom[1] ." " . $nom_prenom[0];
+ }
+ else
+ {
+ $self->{itemsList}[$self->{itemIdx}]->{authors} = $nom_prenom[0];
+ }
+ }
+ else
+ {
+ if ($nom_prenom[1] ne '')
+ {
+ $self->{itemsList}[$self->{itemIdx}]->{authors} .= ", " . $nom_prenom[1] ." " . $nom_prenom[0];
+ }
+ else
+ {
+ $self->{itemsList}[$self->{itemIdx}]->{authors} .= ", " . $nom_prenom[0];
+ }
+ }
+ }
+
+ $self->{isAuthor} = 0 ;
+ }
+ elsif ($self->{isEditor_Publication_Format_Lang})
+ {
+ my @Editor_Publication_Format_Lang = split(/\|/,$origtext);
+
+ $self->{itemsList}[$self->{itemIdx}]->{publication} = $Editor_Publication_Format_Lang[1];
+ $self->{itemsList}[$self->{itemIdx}]->{publication} =~ s/^\s+//;
+ $self->{itemsList}[$self->{itemIdx}]->{publication} =~ s/\s$+//;
+
+ $self->{itemsList}[$self->{itemIdx}]->{format} = $Editor_Publication_Format_Lang[2];
+ $self->{itemsList}[$self->{itemIdx}]->{format} =~ s/^\s+//;
+ $self->{itemsList}[$self->{itemIdx}]->{format} =~ s/\s$+//;
+
+ $self->{isEditor_Publication_Format_Lang} = 0 ;
+ }
+ }
+ else
+ {
+ # Enleve les blancs en debut de chaine
+ $origtext =~ s/^\s+//;
+ if ($self->{isTitle})
+ {
+ $self->{curInfo}->{title} = $origtext;
+ $self->{isTitle} = 0 ;
+ }
+ elsif ($self->{isAnalyse})
+ {
+ $self->{isAuthor} = 1 if ($origtext =~ m/F.*rfattare/i);
+ $self->{isISBN} = 1 if ($origtext =~ m/ISBN/i);
+ $self->{isPublisher} = 1 if ($origtext =~ m/F.*rlag/i);
+ $self->{isFormat} = 1 if ($origtext =~ m/Band/i);
+ $self->{isEdition} = 1 if ($origtext =~ m/Upplagenr/i);
+ $self->{isPage} = 1 if ($origtext =~ m/Sidor/i);
+ $self->{isLanguage} = 1 if ($origtext =~ m/Spr.*k/i);
+ $self->{isPublication} = 1 if ($origtext =~ m/Utgivning/i);
+ $self->{isSerie} = 1 if ($origtext =~ m/Serie/i);
+
+ $self->{isAnalyse} = 0 ;
+ }
+ elsif ($self->{isAuthor} eq 2)
+ {
+ my @array = split(/;/,$origtext);
+ my $element;
+ foreach $element (@array)
+ {
+ my @nom_prenom = split(/,/,$element);
+ # Enleve les blancs en debut de chaine
+ $nom_prenom[0] =~ s/^\s//;
+ $nom_prenom[1] =~ s/^\s//;
+ # Enleve les blancs en fin de chaine
+ $nom_prenom[0] =~ s/\s$//;
+ $nom_prenom[1] =~ s/\s$//;
+ if ($nom_prenom[1] ne '')
+ {
+ $self->{curInfo}->{authors} .= $nom_prenom[1] ." " . $nom_prenom[0];
+ }
+ else
+ {
+ $self->{curInfo}->{authors} .= $nom_prenom[0];
+ }
+ $self->{curInfo}->{authors} .= ",";
+ }
+
+ $self->{isAuthor} = 0 ;
+ }
+ elsif ($self->{isISBN} eq 2)
+ {
+ # Il y a 2 ISBN sur le site, seul le premier m interesse
+ if ($self->{curInfo}->{isbn} eq '')
+ {
+ $self->{curInfo}->{isbn} = $origtext;
+ }
+ $self->{isISBN} = 0 ;
+ }
+ elsif ($self->{isPublisher} eq 2)
+ {
+ $self->{curInfo}->{publisher} = $origtext;
+ $self->{isPublisher} = 0 ;
+ }
+ elsif ($self->{isFormat} eq 2)
+ {
+ my @array = split(/\n/,$origtext);
+
+ $self->{curInfo}->{format} = $array[0];
+ $self->{isFormat} = 0 ;
+ }
+ elsif ($self->{isEdition} eq 2)
+ {
+ # There is some trouble on the site with this field : it is not accurrate. For example for ISBN 9113014528
+ # this field is set to 7000. So for instance this field isn't taken.
+# $self->{curInfo}->{edition} = $origtext;
+ $self->{isEdition} = 0 ;
+ }
+ elsif ($self->{isPage} eq 2)
+ {
+ $self->{curInfo}->{pages} = $origtext;
+ $self->{isPage} = 0 ;
+ }
+ elsif ($self->{isLanguage} eq 2)
+ {
+ $self->{curInfo}->{language} = $origtext;
+ $self->{isLanguage} = 0 ;
+ }
+ elsif ($self->{isPublication} eq 2)
+ {
+ $self->{curInfo}->{publication} = $origtext;
+ $self->{curInfo}->{publication} =~ s|([0-9]*) ([A-Za-z]*) ([0-9]*)|$1.'/'.$self->{monthNumber}->{$2}.'/'.$3|e;
+ $self->{curInfo}->{publication} =~ s|([A-Za-z]*) ([0-9]*)|$self->{monthNumber}->{$1}.'/'.$2|e;
+ $self->{isPublication} = 0 ;
+ }
+ elsif ($self->{isSerie} eq 2)
+ {
+ $self->{curInfo}->{serie} = $origtext;
+ $self->{isSerie} = 0 ;
+ }
+ elsif ($self->{isDescription} eq 1)
+ {
+ $self->{curInfo}->{description} = $origtext;
+ $self->{isDescription} = 2 ;
+ }
+
+ }
+ }
+
+ sub new
+ {
+ my $proto = shift;
+ my $class = ref($proto) || $proto;
+ my $self = $class->SUPER::new();
+ bless ($self, $class);
+
+ $self->{monthNumber} = {
+ Januari => '01',
+ Februari => '02',
+ Mars => '03',
+ April => '04',
+ Maj => '05',
+ Juni => '06',
+ Juli => '07',
+ Augusti => '08',
+ September => '09',
+ Oktober => '10',
+ November => '11',
+ December => '12'
+ };
+
+ $self->{hasField} = {
+ title => 1,
+ authors => 1,
+ publication => 1,
+ format => 1,
+ edition => 0,
+ };
+
+ $self->{isBook} = 0;
+ $self->{isUrl} = 0;
+ $self->{isEditor_Publication_Format_Lang} = 0 ;
+ $self->{isAnalyse} = 0;
+ $self->{isFound} = 0;
+ $self->{isTitle} = 0;
+ $self->{isAuthor} = 0;
+ $self->{isPublisher} = 0;
+ $self->{isISBN} = 0;
+ $self->{isFormat} = 0;
+ $self->{isEdition} = 0;
+ $self->{isPage} = 0;
+ $self->{isLanguage} = 0;
+ $self->{isPublication} = 0;
+ $self->{isSerie} = 0;
+ $self->{isDescription} = 0;
+ $self->{isCover} = 0;
+
+ return $self;
+ }
+
+ sub preProcess
+ {
+ my ($self, $html) = @_;
+
+ if ($self->{parsingList})
+ {
+ $html =~ s|<b>||gi;
+ $html =~ s|</b>||gi;
+ }
+ else
+ {
+ $html =~ s|<li>|\n* |gi;
+ $html =~ s|<br>|\n|gi;
+ $html =~ s|<br />|\n|gi;
+ $html =~ s|<b>||gi;
+ $html =~ s|</b>||gi;
+ $html =~ s|<i>||gi;
+ $html =~ s|</i>||gi;
+ }
+
+ return $html;
+ }
+
+ sub getSearchUrl
+ {
+ my ($self, $word) = @_;
+
+ return "http://www.internetbokhandeln.se/results.html?new_search=1&all_search=" . $word. "&search_media=all";
+
+ }
+
+ sub getItemUrl
+ {
+ my ($self, $url) = @_;
+
+ return $url if $url;
+ return 'http://www.internetbokhandeln.se/';
+ }
+
+ sub getName
+ {
+ return "InternetBokHandeln";
+ }
+
+ sub getAuthor
+ {
+ return 'TPF';
+ }
+
+ sub getLang
+ {
+ return 'SV';
+ }
+
+ sub getSearchFieldsArray
+ {
+ return ['isbn', 'title'];
+ }
+
+}
+
+1;