Koha/misc/cronjobs/cloud-kw.pl
Nick Clemens 906d5b1a85
Bug 31203: Alter other cronjobs that currenlty use cronlogaction
Added command line ooption logging and completion logging where
cronlogaction was already imported. We should probably standardize all
cronjobs, but this is a start

One cron didn't log on confirm, likely we need to update all crons to log
if confirm, and possibly not log if running in test mode? Another bug as well

Signed-off-by: David Nind <david@davidnind.com>

Signed-off-by: Marcel de Rooy <m.de.rooy@rijksmuseum.nl>
Signed-off-by: Tomas Cohen Arazi <tomascohen@theke.io>
2022-10-05 16:12:59 -03:00

413 lines
11 KiB
Perl
Executable file

#!/usr/bin/perl
#
# Copyright 2008 Tamil s.a.r.l.
#
# This file is part of Koha.
#
# Koha 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 3 of the License, or
# (at your option) any later version.
#
# Koha 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 Koha; if not, see <http://www.gnu.org/licenses>.
use strict;
use warnings;
use diagnostics;
use Carp qw( carp croak );
use YAML::XS;
use Pod::Usage qw( pod2usage );
use Getopt::Long qw( GetOptions );
use Koha::Script -cron;
use C4::Context;
use C4::Log qw( cronlogaction );
my $verbose = 0;
my $help = 0;
my $conf = '';
my $command_line_options = join(" ",@ARGV);
GetOptions(
'verbose' => \$verbose,
'help' => \$help,
'conf=s' => \$conf,
);
sub usage {
pod2usage( -verbose => 2 );
exit;
}
usage() if $help || !$conf;
cronlogaction({ info => $command_line_options });
my @clouds;
print "Reading configuration file: $conf\n" if $verbose;
eval {
@clouds = YAML::XS::LoadFile( $conf );
};
croak "Unable to read configuration file: $conf\n" if $@;
for my $cloud ( @clouds ) {
print "Create a cloud\n",
" Koha conf file: ", $cloud->{KohaConf} ? $cloud->{KohaConf} : "default", "\n",
" Zebra Index: ", $cloud->{ZebraIndex}, "\n",
" Koha Keyword: ", $cloud->{KohaIndex}, "\n",
" Count: ", $cloud->{Count}, "\n",
" Withcss: ", $cloud->{Withcss}, "\n",
" Output: ", $cloud->{Output}, "\n",
if $verbose;
# Set Koha context if KohaConf is present
my $set_new_context = 0;
if ( $cloud->{KohaConf} ) {
if ( -e $cloud->{KohaConf} ) {
my $context = C4::Context->new( $cloud->{KohaConf} );
$context->set_context();
$set_new_context = 1;
}
else {
carp "Koha conf file doesn't exist: ", $cloud->{KohaConf}, " ; use KOHA_CONF\n";
}
}
my $index = ZebraIndex->new( $cloud->{ZebraIndex} );
$index->scan( $cloud->{Count} );
open my $fh, ">", $cloud->{Output}
or croak "Unable to create file ", $cloud->{Output};
my $withcss = $cloud->{Withcss} =~ /^yes/i;
print $fh $index->html_cloud( $cloud->{KohaIndex}, $withcss );
close $fh;
$set_new_context && restore_context C4::Context;
}
cronlogaction({ action => 'End', info => "COMPLETED" });
package ZebraIndex;
use strict;
use warnings;
use diagnostics;
use Carp qw( carp croak );
sub new {
my $self = {};
my $class = shift;
$self->{ zebra_index } = shift;
$self->{ top_terms } = undef;
$self->{ levels_cloud } = 24;
bless $self, $class;
# Test Zebra index
my $zbiblio = C4::Context->Zconn( "biblioserver" );
eval {
my $ss = $zbiblio->scan_pqf(
'@attr 1=' . $self->{ zebra_index } . ' @attr 4=1 @attr 6=3 "a"'
);
};
croak "Invalid Zebra index: ", $self->{ zebra_index } if $@;
return $self;
}
#
# scan
# Scan zebra index and populate an array of top terms
#
# PARAMETERS:
# $max_terms Max number of top terms
#
# RETURN:
# A 4-dimensionnal array in $self->{top_terms}
# [0] term
# [1] term number of occurrences
# [2] term proportional relative weight in terms set E[0-1]
# [3] term logarithmic relative weight E [0-levels_cloud]
#
# This array is sorted alphabetically by terms ([0])
# It can be easily sorted by occurrences:
# @t = sort { $a[1] <=> $a[1] } @{$self->{top_terms}};
#
sub scan {
my $self = shift;
my $index_name = $self->{ zebra_index };
my $max_terms = shift;
my $MAX_OCCURENCE = 1000000000;
my $zbiblio = C4::Context->Zconn( "biblioserver" );
my $number_of_terms = 0;
my @terms; # 2 dimensions array
my $min_occurence_index = -1;
my $min_occurence;
my $from = '0';
while (1) {
my $ss;
eval {
print "$from\n" if $verbose;
$from =~ s/\"/\\\"/g;
my $query = '@attr 1=' . $index_name . ' @attr 4=1 @attr 6=3 "'
. $from . 'a"';
$ss = $zbiblio->scan_pqf( $query );
};
if ($@) {
chop $from;
next;
}
$ss->option( rpnCharset => 'UTF-8' );
last if $ss->size() == 0;
my $term = '';
my $occ = 0;
for my $index ( 0..$ss->size()-1 ) {
($term, $occ) = $ss->display_term($index);
#print "$term:$occ\n";
if ( $number_of_terms < $max_terms ) {
push( @terms, [ $term, $occ ] );
++$number_of_terms;
if ( $number_of_terms == $max_terms ) {
$min_occurence = $MAX_OCCURENCE;
for (0..$number_of_terms-1) {
my @term = @{ $terms[$_] };
if ( $term[1] <= $min_occurence ) {
$min_occurence = $term[1];
$min_occurence_index = $_;
}
}
}
}
else {
if ( $occ > $min_occurence) {
@{ $terms[$min_occurence_index] }[0] = $term;
@{ $terms[$min_occurence_index] }[1] = $occ;
$min_occurence = $MAX_OCCURENCE;
for (0..$max_terms-1) {
my @term = @{ $terms[$_] };
if ( $term[1] <= $min_occurence ) {
$min_occurence = $term[1];
$min_occurence_index = $_;
}
}
}
}
}
$from = $term;
}
# Sort array of array by terms weight
@terms = sort { @{$a}[1] <=> @{$b}[1] } @terms;
# A relatif weight to other set terms is added to each term
my $min = $terms[0][1];
my $log_min = log( $min );
my $max = $terms[$#terms][1];
my $log_max = log( $max );
my $delta = $max - $min;
$delta = 1 if $delta == 0; # Very unlikely
my $factor;
if ($log_max - $log_min == 0) {
$log_min = $log_min - $self->{levels_cloud};
$factor = 1;
}
else {
$factor = $self->{levels_cloud} / ($log_max - $log_min);
}
foreach (0..$#terms) {
my $count = @{ $terms[$_] }[1];
my $weight = ( $count - $min ) / $delta;
my $log_weight = int( (log($count) - $log_min) * $factor);
push( @{ $terms[$_] }, $weight );
push( @{ $terms[$_] }, $log_weight );
}
$self->{ top_terms } = \@terms;
# Sort array of array by terms alphabetical order
@terms = sort { @{$a}[0] cmp @{$b}[0] } @terms;
}
#
# Returns a HTML version of index top terms formatted
# as a 'tag cloud'.
#
sub html_cloud {
my $self = shift;
my $koha_index = shift;
my $withcss = shift;
my @terms = @{ $self->{top_terms} };
my $html = '';
if ( $withcss ) {
$html = <<EOS;
<style>
.subjectcloud {
text-align: center;
line-height: 16px;
margin: 20px;
background: #f0f0f0;
padding: 3%;
}
.subjectcloud a {
font-weight: lighter;
text-decoration: none;
}
span.tagcloud0 { font-size: 12px;}
span.tagcloud1 { font-size: 13px;}
span.tagcloud2 { font-size: 14px;}
span.tagcloud3 { font-size: 15px;}
span.tagcloud4 { font-size: 16px;}
span.tagcloud5 { font-size: 17px;}
span.tagcloud6 { font-size: 18px;}
span.tagcloud7 { font-size: 19px;}
span.tagcloud8 { font-size: 20px;}
span.tagcloud9 { font-size: 21px;}
span.tagcloud10 { font-size: 22px;}
span.tagcloud11 { font-size: 23px;}
span.tagcloud12 { font-size: 24px;}
span.tagcloud13 { font-size: 25px;}
span.tagcloud14 { font-size: 26px;}
span.tagcloud15 { font-size: 27px;}
span.tagcloud16 { font-size: 28px;}
span.tagcloud17 { font-size: 29px;}
span.tagcloud18 { font-size: 30px;}
span.tagcloud19 { font-size: 31px;}
span.tagcloud20 { font-size: 32px;}
span.tagcloud21 { font-size: 33px;}
span.tagcloud22 { font-size: 34px;}
span.tagcloud23 { font-size: 35px;}
span.tagcloud24 { font-size: 36px;}
</style>
<div class="subjectcloud">
EOS
}
for (0..$#terms) {
my @term = @{ $terms[$_] };
my $uri = $term[0];
$uri =~ s/\(//g;
#print " 0=", $term[0]," - 1=", $term[1], " - 2=", $term[2], " - 3=", $term[3],"\n";
$html = $html
. '<span class="tagcloud'
. $term[3]
. '">'
. '<a href="/cgi-bin/koha/opac-search.pl?q='
. $koha_index
. '%3A'
. $uri
. '">'
. $term[0]
. "</a></span>\n";
}
$html .= "</div>\n";
return $html;
}
=head1 NAME
cloud-kw.pl - Creates HTML keywords clouds from Koha Zebra Indexes
=head1 USAGE
=over
=item cloud-kw.pl [--verbose|--help] --conf=F<cloud.conf>
Creates multiple HTML files containing kewords cloud with top terms sorted
by their logarithmic weight.
F<cloud.conf> is a YAML configuration file driving cloud generation
process.
=back
=head1 PARAMETERS
=over
=item B<--conf=configuration file>
Specify configuration file name
=item B<--verbose|-v>
Enable script verbose mode.
=item B<--help|-h>
Print this help page.
=back
=head1 CONFIGURATION
Configuration file looks like that:
---
# Koha configuration file for a specific installation
# If not present, defaults to KOHA_CONF
KohaConf: /home/koha/mylibray/etc/koha-conf.xml
# Zebra index to scan
ZebraIndex: Author
# Koha index used to link found kewords with an opac search URL
KohaIndex: au
# Number of top keyword to use for the cloud
Count: 50
# Include CSS style directives with the cloud
# This could be used as a model and then CSS directives are
# put in the appropriate CSS file directly.
Withcss: Yes
# HTML file where to output the cloud
Output: /home/koha/mylibrary/koharoot/koha-tmpl/cloud-author.html
---
KohaConf: /home/koha/yourlibray/etc/koha-conf.xml
ZebraIndex: Subject
KohaIndex: su
Count: 200
Withcss: no
Output: /home/koha/yourlibrary/koharoot/koha-tmpl/cloud-subject.html
=head1 IMPROVEMENTS
Generated top terms have more informations than those outputted from
the time being. Some parameters could be easily added to improve
this script:
=over
=item B<WithCount>
In order to output terms with the number of occurrences they
have been found in Koha Catalogue by Zebra.
=item B<CloudLevels>
Number of levels in the cloud. Now 24 levels are hardcoded.
=item B<Weithing>
Weighting method used to distribute terms in the cloud. We could have two
values: Logarithmic and Linear. Now it's Logarithmic by default.
=item B<Order>
Now terms are outputted in the lexical order. They could be sorted
by their weight.
=back
=cut