Koha/C4/Search.pm
Daniel Sweeney afde76f1c1 Added upc to content_identifier_exists criteria.
Signed-off-by: Galen Charlton <galen.charlton@liblime.com>
2009-03-27 17:23:51 -05:00

2274 lines
86 KiB
Perl

package C4::Search;
# 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 2 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, write to the Free Software Foundation, Inc., 59 Temple Place,
# Suite 330, Boston, MA 02111-1307 USA
use strict;
# use warnings; # FIXME
require Exporter;
use C4::Context;
use C4::Biblio; # GetMarcFromKohaField
use C4::Koha; # getFacets
use Lingua::Stem;
use C4::Search::PazPar2;
use XML::Simple;
use C4::Dates qw(format_date);
use C4::XSLT;
use C4::Branch;
use vars qw($VERSION @ISA @EXPORT @EXPORT_OK %EXPORT_TAGS $DEBUG);
# set the version for version checking
BEGIN {
$VERSION = 3.01;
$DEBUG = ($ENV{DEBUG}) ? 1 : 0;
}
=head1 NAME
C4::Search - Functions for searching the Koha catalog.
=head1 SYNOPSIS
See opac/opac-search.pl or catalogue/search.pl for example of usage
=head1 DESCRIPTION
This module provides searching functions for Koha's bibliographic databases
=head1 FUNCTIONS
=cut
@ISA = qw(Exporter);
@EXPORT = qw(
&findseealso
&FindDuplicate
&SimpleSearch
&searchResults
&getRecords
&buildQuery
&NZgetRecords
&ModBiblios
);
# make all your functions, whether exported or not;
=head2 findseealso($dbh,$fields);
C<$dbh> is a link to the DB handler.
use C4::Context;
my $dbh =C4::Context->dbh;
C<$fields> is a reference to the fields array
This function modifies the @$fields array and adds related fields to search on.
FIXME: this function is probably deprecated in Koha 3
=cut
sub findseealso {
my ( $dbh, $fields ) = @_;
my $tagslib = GetMarcStructure(1);
for ( my $i = 0 ; $i <= $#{$fields} ; $i++ ) {
my ($tag) = substr( @$fields[$i], 1, 3 );
my ($subfield) = substr( @$fields[$i], 4, 1 );
@$fields[$i] .= ',' . $tagslib->{$tag}->{$subfield}->{seealso}
if ( $tagslib->{$tag}->{$subfield}->{seealso} );
}
}
=head2 FindDuplicate
($biblionumber,$biblionumber,$title) = FindDuplicate($record);
This function attempts to find duplicate records using a hard-coded, fairly simplistic algorithm
=cut
sub FindDuplicate {
my ($record) = @_;
my $dbh = C4::Context->dbh;
my $result = TransformMarcToKoha( $dbh, $record, '' );
my $sth;
my $query;
my $search;
my $type;
my ( $biblionumber, $title );
# search duplicate on ISBN, easy and fast..
# ... normalize first
if ( $result->{isbn} ) {
$result->{isbn} =~ s/\(.*$//;
$result->{isbn} =~ s/\s+$//;
$query = "isbn=$result->{isbn}";
}
else {
$result->{title} =~ s /\\//g;
$result->{title} =~ s /\"//g;
$result->{title} =~ s /\(//g;
$result->{title} =~ s /\)//g;
# FIXME: instead of removing operators, could just do
# quotes around the value
$result->{title} =~ s/(and|or|not)//g;
$query = "ti,ext=$result->{title}";
$query .= " and itemtype=$result->{itemtype}"
if ( $result->{itemtype} );
if ( $result->{author} ) {
$result->{author} =~ s /\\//g;
$result->{author} =~ s /\"//g;
$result->{author} =~ s /\(//g;
$result->{author} =~ s /\)//g;
# remove valid operators
$result->{author} =~ s/(and|or|not)//g;
$query .= " and au,ext=$result->{author}";
}
}
# FIXME: add error handling
my ( $error, $searchresults ) = SimpleSearch($query); # FIXME :: hardcoded !
my @results;
foreach my $possible_duplicate_record (@$searchresults) {
my $marcrecord =
MARC::Record->new_from_usmarc($possible_duplicate_record);
my $result = TransformMarcToKoha( $dbh, $marcrecord, '' );
# FIXME :: why 2 $biblionumber ?
if ($result) {
push @results, $result->{'biblionumber'};
push @results, $result->{'title'};
}
}
return @results;
}
=head2 SimpleSearch
( $error, $results, $total_hits ) = SimpleSearch( $query, $offset, $max_results, [@servers] );
This function provides a simple search API on the bibliographic catalog
=over 2
=item C<input arg:>
* $query can be a simple keyword or a complete CCL query
* @servers is optional. Defaults to biblioserver as found in koha-conf.xml
* $offset - If present, represents the number of records at the beggining to omit. Defaults to 0
* $max_results - if present, determines the maximum number of records to fetch. undef is All. defaults to undef.
=item C<Output:>
* $error is a empty unless an error is detected
* \@results is an array of records.
* $total_hits is the number of hits that would have been returned with no limit
=item C<usage in the script:>
=back
my ( $error, $marcresults, $total_hits ) = SimpleSearch($query);
if (defined $error) {
$template->param(query_error => $error);
warn "error: ".$error;
output_html_with_http_headers $input, $cookie, $template->output;
exit;
}
my $hits = scalar @$marcresults;
my @results;
for my $i (0..$hits) {
my %resultsloop;
my $marcrecord = MARC::File::USMARC::decode($marcresults->[$i]);
my $biblio = TransformMarcToKoha(C4::Context->dbh,$marcrecord,'');
#build the hash for the template.
$resultsloop{highlight} = ($i % 2)?(1):(0);
$resultsloop{title} = $biblio->{'title'};
$resultsloop{subtitle} = $biblio->{'subtitle'};
$resultsloop{biblionumber} = $biblio->{'biblionumber'};
$resultsloop{author} = $biblio->{'author'};
$resultsloop{publishercode} = $biblio->{'publishercode'};
$resultsloop{publicationyear} = $biblio->{'publicationyear'};
push @results, \%resultsloop;
}
$template->param(result=>\@results);
=cut
sub SimpleSearch {
my ( $query, $offset, $max_results, $servers ) = @_;
if ( C4::Context->preference('NoZebra') ) {
my $result = NZorder( NZanalyse($query) )->{'biblioserver'};
my $search_result =
( $result->{hits}
&& $result->{hits} > 0 ? $result->{'RECORDS'} : [] );
return ( undef, $search_result, scalar($search_result) );
}
else {
# FIXME hardcoded value. See catalog/search.pl & opac-search.pl too.
my @servers = defined ( $servers ) ? @$servers : ( "biblioserver" );
my @results;
my @zoom_queries;
my @tmpresults;
my @zconns;
my $total_hits;
return ( "No query entered", undef, undef ) unless $query;
# Initialize & Search Zebra
for ( my $i = 0 ; $i < @servers ; $i++ ) {
eval {
$zconns[$i] = C4::Context->Zconn( $servers[$i], 1 );
$zoom_queries[$i] = new ZOOM::Query::CCL2RPN( $query, $zconns[$i]);
$tmpresults[$i] = $zconns[$i]->search( $zoom_queries[$i] );
# error handling
my $error =
$zconns[$i]->errmsg() . " ("
. $zconns[$i]->errcode() . ") "
. $zconns[$i]->addinfo() . " "
. $zconns[$i]->diagset();
return ( $error, undef, undef ) if $zconns[$i]->errcode();
};
if ($@) {
# caught a ZOOM::Exception
my $error =
$@->message() . " ("
. $@->code() . ") "
. $@->addinfo() . " "
. $@->diagset();
warn $error;
return ( $error, undef, undef );
}
}
while ( ( my $i = ZOOM::event( \@zconns ) ) != 0 ) {
my $event = $zconns[ $i - 1 ]->last_event();
if ( $event == ZOOM::Event::ZEND ) {
my $first_record = defined( $offset ) ? $offset+1 : 1;
my $hits = $tmpresults[ $i - 1 ]->size();
$total_hits += $hits;
my $last_record = $hits;
if ( defined $max_results && $offset + $max_results < $hits ) {
$last_record = $offset + $max_results;
}
for my $j ( $first_record..$last_record ) {
my $record = $tmpresults[ $i - 1 ]->record( $j-1 )->raw(); # 0 indexed
push @results, $record;
}
}
}
foreach my $result (@tmpresults) {
$result->destroy();
}
foreach my $zoom_query (@zoom_queries) {
$zoom_query->destroy();
}
return ( undef, \@results, $total_hits );
}
}
=head2 getRecords
( undef, $results_hashref, \@facets_loop ) = getRecords (
$koha_query, $simple_query, $sort_by_ref, $servers_ref,
$results_per_page, $offset, $expanded_facet, $branches,
$query_type, $scan
);
The all singing, all dancing, multi-server, asynchronous, scanning,
searching, record nabbing, facet-building
See verbse embedded documentation.
=cut
sub getRecords {
my (
$koha_query, $simple_query, $sort_by_ref, $servers_ref,
$results_per_page, $offset, $expanded_facet, $branches,
$query_type, $scan
) = @_;
my @servers = @$servers_ref;
my @sort_by = @$sort_by_ref;
# Initialize variables for the ZOOM connection and results object
my $zconn;
my @zconns;
my @results;
my $results_hashref = ();
# Initialize variables for the faceted results objects
my $facets_counter = ();
my $facets_info = ();
my $facets = getFacets();
my @facets_loop
; # stores the ref to array of hashes for template facets loop
### LOOP THROUGH THE SERVERS
for ( my $i = 0 ; $i < @servers ; $i++ ) {
$zconns[$i] = C4::Context->Zconn( $servers[$i], 1 );
# perform the search, create the results objects
# if this is a local search, use the $koha-query, if it's a federated one, use the federated-query
my $query_to_use = ($servers[$i] =~ /biblioserver/) ? $koha_query : $simple_query;
#$query_to_use = $simple_query if $scan;
warn $simple_query if ( $scan and $DEBUG );
# Check if we've got a query_type defined, if so, use it
eval {
if ($query_type)
{
if ( $query_type =~ /^ccl/ ) {
$query_to_use =~
s/\:/\=/g; # change : to = last minute (FIXME)
$results[$i] =
$zconns[$i]->search(
new ZOOM::Query::CCL2RPN( $query_to_use, $zconns[$i] )
);
}
elsif ( $query_type =~ /^cql/ ) {
$results[$i] =
$zconns[$i]->search(
new ZOOM::Query::CQL( $query_to_use, $zconns[$i] ) );
}
elsif ( $query_type =~ /^pqf/ ) {
$results[$i] =
$zconns[$i]->search(
new ZOOM::Query::PQF( $query_to_use, $zconns[$i] ) );
}
}
else {
if ($scan) {
$results[$i] =
$zconns[$i]->scan(
new ZOOM::Query::CCL2RPN( $query_to_use, $zconns[$i] )
);
}
else {
$results[$i] =
$zconns[$i]->search(
new ZOOM::Query::CCL2RPN( $query_to_use, $zconns[$i] )
);
}
}
};
if ($@) {
warn "WARNING: query problem with $query_to_use " . $@;
}
# Concatenate the sort_by limits and pass them to the results object
# Note: sort will override rank
my $sort_by;
foreach my $sort (@sort_by) {
if ( $sort eq "author_az" ) {
$sort_by .= "1=1003 <i ";
}
elsif ( $sort eq "author_za" ) {
$sort_by .= "1=1003 >i ";
}
elsif ( $sort eq "popularity_asc" ) {
$sort_by .= "1=9003 <i ";
}
elsif ( $sort eq "popularity_dsc" ) {
$sort_by .= "1=9003 >i ";
}
elsif ( $sort eq "call_number_asc" ) {
$sort_by .= "1=20 <i ";
}
elsif ( $sort eq "call_number_dsc" ) {
$sort_by .= "1=20 >i ";
}
elsif ( $sort eq "pubdate_asc" ) {
$sort_by .= "1=31 <i ";
}
elsif ( $sort eq "pubdate_dsc" ) {
$sort_by .= "1=31 >i ";
}
elsif ( $sort eq "acqdate_asc" ) {
$sort_by .= "1=32 <i ";
}
elsif ( $sort eq "acqdate_dsc" ) {
$sort_by .= "1=32 >i ";
}
elsif ( $sort eq "title_az" ) {
$sort_by .= "1=4 <i ";
}
elsif ( $sort eq "title_za" ) {
$sort_by .= "1=4 >i ";
}
}
if ($sort_by) {
if ( $results[$i]->sort( "yaz", $sort_by ) < 0 ) {
warn "WARNING sort $sort_by failed";
}
}
} # finished looping through servers
# The big moment: asynchronously retrieve results from all servers
while ( ( my $i = ZOOM::event( \@zconns ) ) != 0 ) {
my $ev = $zconns[ $i - 1 ]->last_event();
if ( $ev == ZOOM::Event::ZEND ) {
next unless $results[ $i - 1 ];
my $size = $results[ $i - 1 ]->size();
if ( $size > 0 ) {
my $results_hash;
# loop through the results
$results_hash->{'hits'} = $size;
my $times;
if ( $offset + $results_per_page <= $size ) {
$times = $offset + $results_per_page;
}
else {
$times = $size;
}
for ( my $j = $offset ; $j < $times ; $j++ ) {
my $records_hash;
my $record;
my $facet_record;
## Check if it's an index scan
if ($scan) {
my ( $term, $occ ) = $results[ $i - 1 ]->term($j);
# here we create a minimal MARC record and hand it off to the
# template just like a normal result ... perhaps not ideal, but
# it works for now
my $tmprecord = MARC::Record->new();
$tmprecord->encoding('UTF-8');
my $tmptitle;
my $tmpauthor;
# the minimal record in author/title (depending on MARC flavour)
if (C4::Context->preference("marcflavour") eq "UNIMARC") {
$tmptitle = MARC::Field->new('200',' ',' ', a => $term, f => $occ);
$tmprecord->append_fields($tmptitle);
} else {
$tmptitle = MARC::Field->new('245',' ',' ', a => $term,);
$tmpauthor = MARC::Field->new('100',' ',' ', a => $occ,);
$tmprecord->append_fields($tmptitle);
$tmprecord->append_fields($tmpauthor);
}
$results_hash->{'RECORDS'}[$j] = $tmprecord->as_usmarc();
}
# not an index scan
else {
$record = $results[ $i - 1 ]->record($j)->raw();
# warn "RECORD $j:".$record;
$results_hash->{'RECORDS'}[$j] = $record;
# Fill the facets while we're looping, but only for the biblioserver
$facet_record = MARC::Record->new_from_usmarc($record)
if $servers[ $i - 1 ] =~ /biblioserver/;
#warn $servers[$i-1]."\n".$record; #.$facet_record->title();
if ($facet_record) {
for ( my $k = 0 ; $k <= @$facets ; $k++ ) {
if ( $facets->[$k] ) {
my @fields;
for my $tag ( @{ $facets->[$k]->{'tags'} } )
{
push @fields,
$facet_record->field($tag);
}
for my $field (@fields) {
my @subfields = $field->subfields();
for my $subfield (@subfields) {
my ( $code, $data ) = @$subfield;
if ( $code eq
$facets->[$k]->{'subfield'} )
{
$facets_counter->{ $facets->[$k]
->{'link_value'} }
->{$data}++;
}
}
}
$facets_info->{ $facets->[$k]
->{'link_value'} }->{'label_value'} =
$facets->[$k]->{'label_value'};
$facets_info->{ $facets->[$k]
->{'link_value'} }->{'expanded'} =
$facets->[$k]->{'expanded'};
}
}
}
}
}
$results_hashref->{ $servers[ $i - 1 ] } = $results_hash;
}
# warn "connection ", $i-1, ": $size hits";
# warn $results[$i-1]->record(0)->render() if $size > 0;
# BUILD FACETS
if ( $servers[ $i - 1 ] =~ /biblioserver/ ) {
for my $link_value (
sort { $facets_counter->{$b} <=> $facets_counter->{$a} }
keys %$facets_counter )
{
my $expandable;
my $number_of_facets;
my @this_facets_array;
for my $one_facet (
sort {
$facets_counter->{$link_value}
->{$b} <=> $facets_counter->{$link_value}->{$a}
} keys %{ $facets_counter->{$link_value} }
)
{
$number_of_facets++;
if ( ( $number_of_facets < 6 )
|| ( $expanded_facet eq $link_value )
|| ( $facets_info->{$link_value}->{'expanded'} ) )
{
# Sanitize the link value ), ( will cause errors with CCL,
my $facet_link_value = $one_facet;
$facet_link_value =~ s/(\(|\))/ /g;
# fix the length that will display in the label,
my $facet_label_value = $one_facet;
$facet_label_value =
substr( $one_facet, 0, 20 ) . "..."
unless length($facet_label_value) <= 20;
# if it's a branch, label by the name, not the code,
if ( $link_value =~ /branch/ ) {
$facet_label_value =
$branches->{$one_facet}->{'branchname'};
}
# but we're down with the whole label being in the link's title.
my $facet_title_value = $one_facet;
push @this_facets_array,
(
{
facet_count =>
$facets_counter->{$link_value}
->{$one_facet},
facet_label_value => $facet_label_value,
facet_title_value => $facet_title_value,
facet_link_value => $facet_link_value,
type_link_value => $link_value,
},
);
}
}
# handle expanded option
unless ( $facets_info->{$link_value}->{'expanded'} ) {
$expandable = 1
if ( ( $number_of_facets > 6 )
&& ( $expanded_facet ne $link_value ) );
}
push @facets_loop,
(
{
type_link_value => $link_value,
type_id => $link_value . "_id",
"type_label_" . $facets_info->{$link_value}->{'label_value'} => 1,
facets => \@this_facets_array,
expandable => $expandable,
expand => $link_value,
}
) unless ( ($facets_info->{$link_value}->{'label_value'} =~ /Libraries/) and (C4::Context->preference('singleBranchMode')) );
}
}
}
}
return ( undef, $results_hashref, \@facets_loop );
}
sub pazGetRecords {
my (
$koha_query, $simple_query, $sort_by_ref, $servers_ref,
$results_per_page, $offset, $expanded_facet, $branches,
$query_type, $scan
) = @_;
my $paz = C4::Search::PazPar2->new(C4::Context->config('pazpar2url'));
$paz->init();
$paz->search($simple_query);
sleep 1;
# do results
my $results_hashref = {};
my $stats = XMLin($paz->stat);
my $results = XMLin($paz->show($offset, $results_per_page, 'work-title:1'), forcearray => 1);
# for a grouped search result, the number of hits
# is the number of groups returned; 'bib_hits' will have
# the total number of bibs.
$results_hashref->{'biblioserver'}->{'hits'} = $results->{'merged'}->[0];
$results_hashref->{'biblioserver'}->{'bib_hits'} = $stats->{'hits'};
HIT: foreach my $hit (@{ $results->{'hit'} }) {
my $recid = $hit->{recid}->[0];
my $work_title = $hit->{'md-work-title'}->[0];
my $work_author;
if (exists $hit->{'md-work-author'}) {
$work_author = $hit->{'md-work-author'}->[0];
}
my $group_label = (defined $work_author) ? "$work_title / $work_author" : $work_title;
my $result_group = {};
$result_group->{'group_label'} = $group_label;
$result_group->{'group_merge_key'} = $recid;
my $count = 1;
if (exists $hit->{count}) {
$count = $hit->{count}->[0];
}
$result_group->{'group_count'} = $count;
for (my $i = 0; $i < $count; $i++) {
# FIXME -- may need to worry about diacritics here
my $rec = $paz->record($recid, $i);
push @{ $result_group->{'RECORDS'} }, $rec;
}
push @{ $results_hashref->{'biblioserver'}->{'GROUPS'} }, $result_group;
}
# pass through facets
my $termlist_xml = $paz->termlist('author,subject');
my $terms = XMLin($termlist_xml, forcearray => 1);
my @facets_loop = ();
#die Dumper($results);
# foreach my $list (sort keys %{ $terms->{'list'} }) {
# my @facets = ();
# foreach my $facet (sort @{ $terms->{'list'}->{$list}->{'term'} } ) {
# push @facets, {
# facet_label_value => $facet->{'name'}->[0],
# };
# }
# push @facets_loop, ( {
# type_label => $list,
# facets => \@facets,
# } );
# }
return ( undef, $results_hashref, \@facets_loop );
}
# STOPWORDS
sub _remove_stopwords {
my ( $operand, $index ) = @_;
my @stopwords_removed;
# phrase and exact-qualified indexes shouldn't have stopwords removed
if ( $index !~ m/phr|ext/ ) {
# remove stopwords from operand : parse all stopwords & remove them (case insensitive)
# we use IsAlpha unicode definition, to deal correctly with diacritics.
# otherwise, a French word like "leçon" woudl be split into "le" "çon", "le"
# is a stopword, we'd get "çon" and wouldn't find anything...
foreach ( keys %{ C4::Context->stopwords } ) {
next if ( $_ =~ /(and|or|not)/ ); # don't remove operators
if ( $operand =~
/(\P{IsAlpha}$_\P{IsAlpha}|^$_\P{IsAlpha}|\P{IsAlpha}$_$|^$_$)/ )
{
$operand =~ s/\P{IsAlpha}$_\P{IsAlpha}/ /gi;
$operand =~ s/^$_\P{IsAlpha}/ /gi;
$operand =~ s/\P{IsAlpha}$_$/ /gi;
$operand =~ s/$1//gi;
push @stopwords_removed, $_;
}
}
}
return ( $operand, \@stopwords_removed );
}
# TRUNCATION
sub _detect_truncation {
my ( $operand, $index ) = @_;
my ( @nontruncated, @righttruncated, @lefttruncated, @rightlefttruncated,
@regexpr );
$operand =~ s/^ //g;
my @wordlist = split( /\s/, $operand );
foreach my $word (@wordlist) {
if ( $word =~ s/^\*([^\*]+)\*$/$1/ ) {
push @rightlefttruncated, $word;
}
elsif ( $word =~ s/^\*([^\*]+)$/$1/ ) {
push @lefttruncated, $word;
}
elsif ( $word =~ s/^([^\*]+)\*$/$1/ ) {
push @righttruncated, $word;
}
elsif ( index( $word, "*" ) < 0 ) {
push @nontruncated, $word;
}
else {
push @regexpr, $word;
}
}
return (
\@nontruncated, \@righttruncated, \@lefttruncated,
\@rightlefttruncated, \@regexpr
);
}
# STEMMING
sub _build_stemmed_operand {
my ($operand) = @_;
my $stemmed_operand;
# If operand contains a digit, it is almost certainly an identifier, and should
# not be stemmed. This is particularly relevant for ISBNs and ISSNs, which
# can contain the letter "X" - for example, _build_stemmend_operand would reduce
# "014100018X" to "x ", which for a MARC21 database would bring up irrelevant
# results (e.g., "23 x 29 cm." from the 300$c). Bug 2098.
return $operand if $operand =~ /\d/;
# FIXME: the locale should be set based on the user's language and/or search choice
my $stemmer = Lingua::Stem->new( -locale => 'EN-US' );
# FIXME: these should be stored in the db so the librarian can modify the behavior
$stemmer->add_exceptions(
{
'and' => 'and',
'or' => 'or',
'not' => 'not',
}
);
my @words = split( / /, $operand );
my $stems = $stemmer->stem(@words);
for my $stem (@$stems) {
$stemmed_operand .= "$stem";
$stemmed_operand .= "?"
unless ( $stem =~ /(and$|or$|not$)/ ) || ( length($stem) < 3 );
$stemmed_operand .= " ";
}
warn "STEMMED OPERAND: $stemmed_operand" if $DEBUG;
return $stemmed_operand;
}
# FIELD WEIGHTING
sub _build_weighted_query {
# FIELD WEIGHTING - This is largely experimental stuff. What I'm committing works
# pretty well but could work much better if we had a smarter query parser
my ( $operand, $stemmed_operand, $index ) = @_;
my $stemming = C4::Context->preference("QueryStemming") || 0;
my $weight_fields = C4::Context->preference("QueryWeightFields") || 0;
my $fuzzy_enabled = C4::Context->preference("QueryFuzzy") || 0;
my $weighted_query .= "(rk=("; # Specifies that we're applying rank
# Keyword, or, no index specified
if ( ( $index eq 'kw' ) || ( !$index ) ) {
$weighted_query .=
"Title-cover,ext,r1=\"$operand\""; # exact title-cover
$weighted_query .= " or ti,ext,r2=\"$operand\""; # exact title
$weighted_query .= " or ti,phr,r3=\"$operand\""; # phrase title
#$weighted_query .= " or any,ext,r4=$operand"; # exact any
#$weighted_query .=" or kw,wrdl,r5=\"$operand\""; # word list any
$weighted_query .= " or wrdl,fuzzy,r8=\"$operand\""
if $fuzzy_enabled; # add fuzzy, word list
$weighted_query .= " or wrdl,right-Truncation,r9=\"$stemmed_operand\""
if ( $stemming and $stemmed_operand )
; # add stemming, right truncation
$weighted_query .= " or wrdl,r9=\"$operand\"";
# embedded sorting: 0 a-z; 1 z-a
# $weighted_query .= ") or (sort1,aut=1";
}
# Barcode searches should skip this process
elsif ( $index eq 'bc' ) {
$weighted_query .= "bc=\"$operand\"";
}
# Authority-number searches should skip this process
elsif ( $index eq 'an' ) {
$weighted_query .= "an=\"$operand\"";
}
# If the index already has more than one qualifier, wrap the operand
# in quotes and pass it back (assumption is that the user knows what they
# are doing and won't appreciate us mucking up their query
elsif ( $index =~ ',' ) {
$weighted_query .= " $index=\"$operand\"";
}
#TODO: build better cases based on specific search indexes
else {
$weighted_query .= " $index,ext,r1=\"$operand\""; # exact index
#$weighted_query .= " or (title-sort-az=0 or $index,startswithnt,st-word,r3=$operand #)";
$weighted_query .= " or $index,phr,r3=\"$operand\""; # phrase index
$weighted_query .=
" or $index,rt,wrdl,r3=\"$operand\""; # word list index
}
$weighted_query .= "))"; # close rank specification
return $weighted_query;
}
=head2 buildQuery
( $error, $query,
$simple_query, $query_cgi,
$query_desc, $limit,
$limit_cgi, $limit_desc,
$stopwords_removed, $query_type ) = getRecords ( $operators, $operands, $indexes, $limits, $sort_by, $scan);
Build queries and limits in CCL, CGI, Human,
handle truncation, stemming, field weighting, stopwords, fuzziness, etc.
See verbose embedded documentation.
=cut
sub buildQuery {
my ( $operators, $operands, $indexes, $limits, $sort_by, $scan ) = @_;
warn "---------\nEnter buildQuery\n---------" if $DEBUG;
# dereference
my @operators = @$operators if $operators;
my @indexes = @$indexes if $indexes;
my @operands = @$operands if $operands;
my @limits = @$limits if $limits;
my @sort_by = @$sort_by if $sort_by;
my $stemming = C4::Context->preference("QueryStemming") || 0;
my $auto_truncation = C4::Context->preference("QueryAutoTruncate") || 0;
my $weight_fields = C4::Context->preference("QueryWeightFields") || 0;
my $fuzzy_enabled = C4::Context->preference("QueryFuzzy") || 0;
my $remove_stopwords = C4::Context->preference("QueryRemoveStopwords") || 0;
# no stemming/weight/fuzzy in NoZebra
if ( C4::Context->preference("NoZebra") ) {
$stemming = 0;
$weight_fields = 0;
$fuzzy_enabled = 0;
}
my $query = $operands[0];
my $simple_query = $operands[0];
# initialize the variables we're passing back
my $query_cgi;
my $query_desc;
my $query_type;
my $limit;
my $limit_cgi;
my $limit_desc;
my $stopwords_removed; # flag to determine if stopwords have been removed
# for handling ccl, cql, pqf queries in diagnostic mode, skip the rest of the steps
# DIAGNOSTIC ONLY!!
if ( $query =~ /^ccl=/ ) {
return ( undef, $', $', $', $', '', '', '', '', 'ccl' );
}
if ( $query =~ /^cql=/ ) {
return ( undef, $', $', $', $', '', '', '', '', 'cql' );
}
if ( $query =~ /^pqf=/ ) {
return ( undef, $', $', $', $', '', '', '', '', 'pqf' );
}
# pass nested queries directly
# FIXME: need better handling of some of these variables in this case
if ( $query =~ /(\(|\))/ ) {
return (
undef, $query, $simple_query, $query_cgi,
$query, $limit, $limit_cgi, $limit_desc,
$stopwords_removed, 'ccl'
);
}
# Form-based queries are non-nested and fixed depth, so we can easily modify the incoming
# query operands and indexes and add stemming, truncation, field weighting, etc.
# Once we do so, we'll end up with a value in $query, just like if we had an
# incoming $query from the user
else {
$query = ""
; # clear it out so we can populate properly with field-weighted, stemmed, etc. query
my $previous_operand
; # a flag used to keep track if there was a previous query
# if there was, we can apply the current operator
# for every operand
for ( my $i = 0 ; $i <= @operands ; $i++ ) {
# COMBINE OPERANDS, INDEXES AND OPERATORS
if ( $operands[$i] ) {
# A flag to determine whether or not to add the index to the query
my $indexes_set;
# If the user is sophisticated enough to specify an index, turn off field weighting, stemming, and stopword handling
if ( $operands[$i] =~ /(:|=)/ || $scan ) {
$weight_fields = 0;
$stemming = 0;
$remove_stopwords = 0;
}
my $operand = $operands[$i];
my $index = $indexes[$i];
# Add index-specific attributes
# Date of Publication
if ( $index eq 'yr' ) {
$index .= ",st-numeric";
$indexes_set++;
$stemming = $auto_truncation = $weight_fields = $fuzzy_enabled = $remove_stopwords = 0;
}
# Date of Acquisition
elsif ( $index eq 'acqdate' ) {
$index .= ",st-date-normalized";
$indexes_set++;
$stemming = $auto_truncation = $weight_fields = $fuzzy_enabled = $remove_stopwords = 0;
}
# ISBN,ISSN,Standard Number, don't need special treatment
elsif ( $index eq 'nb' || $index eq 'ns' ) {
$indexes_set++;
(
$stemming, $auto_truncation,
$weight_fields, $fuzzy_enabled,
$remove_stopwords
) = ( 0, 0, 0, 0, 0 );
}
# Set default structure attribute (word list)
my $struct_attr;
unless ( $indexes_set || !$index || $index =~ /(st-|phr|ext|wrdl)/ ) {
$struct_attr = ",wrdl";
}
# Some helpful index variants
my $index_plus = $index . $struct_attr . ":" if $index;
my $index_plus_comma = $index . $struct_attr . "," if $index;
# Remove Stopwords
if ($remove_stopwords) {
( $operand, $stopwords_removed ) =
_remove_stopwords( $operand, $index );
warn "OPERAND w/out STOPWORDS: >$operand<" if $DEBUG;
warn "REMOVED STOPWORDS: @$stopwords_removed"
if ( $stopwords_removed && $DEBUG );
}
# Detect Truncation
my $truncated_operand;
my( $nontruncated, $righttruncated, $lefttruncated,
$rightlefttruncated, $regexpr
) = _detect_truncation( $operand, $index );
warn
"TRUNCATION: NON:>@$nontruncated< RIGHT:>@$righttruncated< LEFT:>@$lefttruncated< RIGHTLEFT:>@$rightlefttruncated< REGEX:>@$regexpr<"
if $DEBUG;
# Apply Truncation
if (
scalar(@$righttruncated) + scalar(@$lefttruncated) +
scalar(@$rightlefttruncated) > 0 )
{
# Don't field weight or add the index to the query, we do it here
$indexes_set = 1;
undef $weight_fields;
my $previous_truncation_operand;
if ( scalar(@$nontruncated) > 0 ) {
$truncated_operand .= "$index_plus @$nontruncated ";
$previous_truncation_operand = 1;
}
if ( scalar(@$righttruncated) > 0 ) {
$truncated_operand .= "and "
if $previous_truncation_operand;
$truncated_operand .=
"$index_plus_comma" . "rtrn:@$righttruncated ";
$previous_truncation_operand = 1;
}
if ( scalar(@$lefttruncated) > 0 ) {
$truncated_operand .= "and "
if $previous_truncation_operand;
$truncated_operand .=
"$index_plus_comma" . "ltrn:@$lefttruncated ";
$previous_truncation_operand = 1;
}
if ( scalar(@$rightlefttruncated) > 0 ) {
$truncated_operand .= "and "
if $previous_truncation_operand;
$truncated_operand .=
"$index_plus_comma" . "rltrn:@$rightlefttruncated ";
$previous_truncation_operand = 1;
}
}
$operand = $truncated_operand if $truncated_operand;
warn "TRUNCATED OPERAND: >$truncated_operand<" if $DEBUG;
# Handle Stemming
my $stemmed_operand;
$stemmed_operand = _build_stemmed_operand($operand)
if $stemming;
warn "STEMMED OPERAND: >$stemmed_operand<" if $DEBUG;
# Handle Field Weighting
my $weighted_operand;
$weighted_operand =
_build_weighted_query( $operand, $stemmed_operand, $index )
if $weight_fields;
warn "FIELD WEIGHTED OPERAND: >$weighted_operand<" if $DEBUG;
$operand = $weighted_operand if $weight_fields;
$indexes_set = 1 if $weight_fields;
# If there's a previous operand, we need to add an operator
if ($previous_operand) {
# User-specified operator
if ( $operators[ $i - 1 ] ) {
$query .= " $operators[$i-1] ";
$query .= " $index_plus " unless $indexes_set;
$query .= " $operand";
$query_cgi .= "&op=$operators[$i-1]";
$query_cgi .= "&idx=$index" if $index;
$query_cgi .= "&q=$operands[$i]" if $operands[$i];
$query_desc .=
" $operators[$i-1] $index_plus $operands[$i]";
}
# Default operator is and
else {
$query .= " and ";
$query .= "$index_plus " unless $indexes_set;
$query .= "$operand";
$query_cgi .= "&op=and&idx=$index" if $index;
$query_cgi .= "&q=$operands[$i]" if $operands[$i];
$query_desc .= " and $index_plus $operands[$i]";
}
}
# There isn't a pervious operand, don't need an operator
else {
# Field-weighted queries already have indexes set
$query .= " $index_plus " unless $indexes_set;
$query .= $operand;
$query_desc .= " $index_plus $operands[$i]";
$query_cgi .= "&idx=$index" if $index;
$query_cgi .= "&q=$operands[$i]" if $operands[$i];
$previous_operand = 1;
}
} #/if $operands
} # /for
}
warn "QUERY BEFORE LIMITS: >$query<" if $DEBUG;
# add limits
my $group_OR_limits;
my $availability_limit;
foreach my $this_limit (@limits) {
if ( $this_limit =~ /available/ ) {
# 'available' is defined as (items.onloan is NULL) and (items.itemlost = 0)
# In English:
# all records not indexed in the onloan register (zebra) and all records with a value of lost equal to 0
$availability_limit .=
"( ( allrecords,AlwaysMatches='' not onloan,AlwaysMatches='') and (lost,st-numeric=0) )"; #or ( allrecords,AlwaysMatches='' not lost,AlwaysMatches='')) )";
$limit_cgi .= "&limit=available";
$limit_desc .= "";
}
# group_OR_limits, prefixed by mc-
# OR every member of the group
elsif ( $this_limit =~ /mc/ ) {
$group_OR_limits .= " or " if $group_OR_limits;
$limit_desc .= " or " if $group_OR_limits;
$group_OR_limits .= "$this_limit";
$limit_cgi .= "&limit=$this_limit";
$limit_desc .= " $this_limit";
}
# Regular old limits
else {
$limit .= " and " if $limit || $query;
$limit .= "$this_limit";
$limit_cgi .= "&limit=$this_limit";
if ($this_limit =~ /^branch:(.+)/) {
my $branchcode = $1;
my $branchname = GetBranchName($branchcode);
if (defined $branchname) {
$limit_desc .= " branch:$branchname";
} else {
$limit_desc .= " $this_limit";
}
} else {
$limit_desc .= " $this_limit";
}
}
}
if ($group_OR_limits) {
$limit .= " and " if ( $query || $limit );
$limit .= "($group_OR_limits)";
}
if ($availability_limit) {
$limit .= " and " if ( $query || $limit );
$limit .= "($availability_limit)";
}
# Normalize the query and limit strings
$query =~ s/:/=/g;
$limit =~ s/:/=/g;
for ( $query, $query_desc, $limit, $limit_desc ) {
$_ =~ s/ / /g; # remove extra spaces
$_ =~ s/^ //g; # remove any beginning spaces
$_ =~ s/ $//g; # remove any ending spaces
$_ =~ s/==/=/g; # remove double == from query
}
$query_cgi =~ s/^&//; # remove unnecessary & from beginning of the query cgi
for ($query_cgi,$simple_query) {
$_ =~ s/"//g;
}
# append the limit to the query
$query .= " " . $limit;
# Warnings if DEBUG
if ($DEBUG) {
warn "QUERY:" . $query;
warn "QUERY CGI:" . $query_cgi;
warn "QUERY DESC:" . $query_desc;
warn "LIMIT:" . $limit;
warn "LIMIT CGI:" . $limit_cgi;
warn "LIMIT DESC:" . $limit_desc;
warn "---------\nLeave buildQuery\n---------";
}
return (
undef, $query, $simple_query, $query_cgi,
$query_desc, $limit, $limit_cgi, $limit_desc,
$stopwords_removed, $query_type
);
}
=head2 searchResults
Format results in a form suitable for passing to the template
=cut
# IMO this subroutine is pretty messy still -- it's responsible for
# building the HTML output for the template
sub searchResults {
my ( $searchdesc, $hits, $results_per_page, $offset, $scan, @marcresults ) = @_;
my $dbh = C4::Context->dbh;
my $even = 1;
my @newresults;
# add search-term highlighting via <span>s on the search terms
my $span_terms_hashref;
for my $span_term ( split( / /, $searchdesc ) ) {
$span_term =~ s/(.*=|\)|\(|\+|\.|\*)//g;
$span_terms_hashref->{$span_term}++;
}
#Build branchnames hash
#find branchname
#get branch information.....
my %branches;
my $bsth =
$dbh->prepare("SELECT branchcode,branchname FROM branches")
; # FIXME : use C4::Koha::GetBranches
$bsth->execute();
while ( my $bdata = $bsth->fetchrow_hashref ) {
$branches{ $bdata->{'branchcode'} } = $bdata->{'branchname'};
}
# FIXME - We build an authorised values hash here, using the default framework
# though it is possible to have different authvals for different fws.
my $shelflocations =GetKohaAuthorisedValues('items.location','');
# get notforloan authorised value list (see $shelflocations FIXME)
my $notforloan_authorised_value = GetAuthValCode('items.notforloan','');
#Build itemtype hash
#find itemtype & itemtype image
my %itemtypes;
$bsth =
$dbh->prepare(
"SELECT itemtype,description,imageurl,summary,notforloan FROM itemtypes"
);
$bsth->execute();
while ( my $bdata = $bsth->fetchrow_hashref ) {
foreach (qw(description imageurl summary notforloan)) {
$itemtypes{ $bdata->{'itemtype'} }->{$_} = $bdata->{$_};
}
}
#search item field code
my $sth =
$dbh->prepare(
"SELECT tagfield FROM marc_subfield_structure WHERE kohafield LIKE 'items.itemnumber'"
);
$sth->execute;
my ($itemtag) = $sth->fetchrow;
## find column names of items related to MARC
my $sth2 = $dbh->prepare("SHOW COLUMNS FROM items");
$sth2->execute;
my %subfieldstosearch;
while ( ( my $column ) = $sth2->fetchrow ) {
my ( $tagfield, $tagsubfield ) =
&GetMarcFromKohaField( "items." . $column, "" );
$subfieldstosearch{$column} = $tagsubfield;
}
# handle which records to actually retrieve
my $times;
if ( $hits && $offset + $results_per_page <= $hits ) {
$times = $offset + $results_per_page;
}
else {
$times = $hits; # FIXME: if $hits is undefined, why do we want to equal it?
}
my $marcflavour = C4::Context->preference("marcflavour");
# loop through all of the records we've retrieved
for ( my $i = $offset ; $i <= $times - 1 ; $i++ ) {
my $marcrecord = MARC::File::USMARC::decode( $marcresults[$i] );
my $oldbiblio = TransformMarcToKoha( $dbh, $marcrecord, '' );
$oldbiblio->{subtitle} = C4::Biblio::get_koha_field_from_marc('bibliosubtitle', 'subtitle', $marcrecord, '');
$oldbiblio->{result_number} = $i + 1;
# add imageurl to itemtype if there is one
$oldbiblio->{imageurl} = getitemtypeimagelocation( 'opac', $itemtypes{ $oldbiblio->{itemtype} }->{imageurl} );
$oldbiblio->{'authorised_value_images'} = C4::Items::get_authorised_value_images( C4::Biblio::get_biblio_authorised_values( $oldbiblio->{'biblionumber'}, $marcrecord ) );
$oldbiblio->{normalized_upc} = GetNormalizedUPC($marcrecord,$marcflavour);
$oldbiblio->{normalized_ean} = GetNormalizedEAN($marcrecord,$marcflavour);
$oldbiblio->{normalized_oclc} = GetNormalizedOCLCNumber($marcrecord,$marcflavour);
$oldbiblio->{normalized_isbn} = GetNormalizedISBN(undef,$marcrecord,$marcflavour);
$oldbiblio->{content_identifier_exists} = 1 if ($oldbiblio->{normalized_isbn} or $oldbiblio->{normalized_oclc} or $oldbiblio->{normalized_ean} or $oldbiblio->{normalized_upc});
$oldbiblio->{description} = $itemtypes{ $oldbiblio->{itemtype} }->{description};
# Build summary if there is one (the summary is defined in the itemtypes table)
# FIXME: is this used anywhere, I think it can be commented out? -- JF
if ( $itemtypes{ $oldbiblio->{itemtype} }->{summary} ) {
my $summary = $itemtypes{ $oldbiblio->{itemtype} }->{summary};
my @fields = $marcrecord->fields();
foreach my $field (@fields) {
my $tag = $field->tag();
my $tagvalue = $field->as_string();
$summary =~
s/\[(.?.?.?.?)$tag\*(.*?)]/$1$tagvalue$2\[$1$tag$2]/g;
unless ( $tag < 10 ) {
my @subf = $field->subfields;
for my $i ( 0 .. $#subf ) {
my $subfieldcode = $subf[$i][0];
my $subfieldvalue = $subf[$i][1];
my $tagsubf = $tag . $subfieldcode;
$summary =~
s/\[(.?.?.?.?)$tagsubf(.*?)]/$1$subfieldvalue$2\[$1$tagsubf$2]/g;
}
}
}
# FIXME: yuk
$summary =~ s/\[(.*?)]//g;
$summary =~ s/\n/<br\/>/g;
$oldbiblio->{summary} = $summary;
}
# save an author with no <span> tag, for the <a href=search.pl?q=<!--tmpl_var name="author"-->> link
$oldbiblio->{'author_nospan'} = $oldbiblio->{'author'};
$oldbiblio->{'title_nospan'} = $oldbiblio->{'title'};
# Add search-term highlighting to the whole record where they match using <span>s
if (C4::Context->preference("OpacHighlightedWords")){
my $searchhighlightblob;
for my $highlight_field ( $marcrecord->fields ) {
# FIXME: need to skip title, subtitle, author, etc., as they are handled below
next if $highlight_field->tag() =~ /(^00)/; # skip fixed fields
for my $subfield ($highlight_field->subfields()) {
my $match;
next if $subfield->[0] eq '9';
my $field = $subfield->[1];
for my $term ( keys %$span_terms_hashref ) {
if ( ( $field =~ /$term/i ) && (( length($term) > 3 ) || ($field =~ / $term /i)) ) {
$field =~ s/$term/<span class=\"term\">$&<\/span>/gi;
$match++;
}
}
$searchhighlightblob .= $field . " ... " if $match;
}
}
$searchhighlightblob = ' ... '.$searchhighlightblob if $searchhighlightblob;
$oldbiblio->{'searchhighlightblob'} = $searchhighlightblob;
}
# Add search-term highlighting to the title, subtitle, etc. fields
for my $term ( keys %$span_terms_hashref ) {
my $old_term = $term;
if ( length($term) > 3 ) {
$term =~ s/(.*=|\)|\(|\+|\.|\?|\[|\]|\\|\*)//g;
foreach(qw(title subtitle author publishercode place pages notes size)) {
$oldbiblio->{$_} =~ s/$term/<span class=\"term\">$&<\/span>/gi;
}
}
}
($i % 2) and $oldbiblio->{'toggle'} = 1;
# Pull out the items fields
my @fields = $marcrecord->field($itemtag);
# Setting item statuses for display
my @available_items_loop;
my @onloan_items_loop;
my @other_items_loop;
my $available_items;
my $onloan_items;
my $other_items;
my $ordered_count = 0;
my $available_count = 0;
my $onloan_count = 0;
my $longoverdue_count = 0;
my $other_count = 0;
my $wthdrawn_count = 0;
my $itemlost_count = 0;
my $itembinding_count = 0;
my $itemdamaged_count = 0;
my $item_in_transit_count = 0;
my $can_place_holds = 0;
my $items_count = scalar(@fields);
my $maxitems =
( C4::Context->preference('maxItemsinSearchResults') )
? C4::Context->preference('maxItemsinSearchResults') - 1
: 1;
# loop through every item
foreach my $field (@fields) {
my $item;
# populate the items hash
foreach my $code ( keys %subfieldstosearch ) {
$item->{$code} = $field->subfield( $subfieldstosearch{$code} );
}
my $hbranch = C4::Context->preference('HomeOrHoldingBranch') eq 'homebranch' ? 'homebranch' : 'holdingbranch';
my $otherbranch = C4::Context->preference('HomeOrHoldingBranch') eq 'homebranch' ? 'holdingbranch' : 'homebranch';
# set item's branch name, use HomeOrHoldingBranch syspref first, fall back to the other one
if ($item->{$hbranch}) {
$item->{'branchname'} = $branches{$item->{$hbranch}};
}
elsif ($item->{$otherbranch}) { # Last resort
$item->{'branchname'} = $branches{$item->{$otherbranch}};
}
my $prefix = $item->{$hbranch} . '--' . $item->{location} . $item->{itype} . $item->{itemcallnumber};
# For each grouping of items (onloan, available, unavailable), we build a key to store relevant info about that item
if ( $item->{onloan} ) {
$onloan_count++;
my $key = $prefix . $item->{due_date};
$onloan_items->{$key}->{due_date} = format_date($item->{onloan});
$onloan_items->{$key}->{count}++ if $item->{$hbranch};
$onloan_items->{$key}->{branchname} = $item->{branchname};
$onloan_items->{$key}->{location} = $shelflocations->{ $item->{location} };
$onloan_items->{$key}->{itemcallnumber} = $item->{itemcallnumber};
$onloan_items->{$key}->{imageurl} = getitemtypeimagelocation( 'opac', $itemtypes{ $item->{itype} }->{imageurl} );
# if something's checked out and lost, mark it as 'long overdue'
if ( $item->{itemlost} ) {
$onloan_items->{$prefix}->{longoverdue}++;
$longoverdue_count++;
} else { # can place holds as long as item isn't lost
$can_place_holds = 1;
}
}
# items not on loan, but still unavailable ( lost, withdrawn, damaged )
else {
# item is on order
if ( $item->{notforloan} == -1 ) {
$ordered_count++;
}
# is item in transit?
my $transfertwhen = '';
my ($transfertfrom, $transfertto);
unless ($item->{wthdrawn}
|| $item->{itemlost}
|| $item->{damaged}
|| $item->{notforloan}
|| $items_count > 20) {
# A couple heuristics to limit how many times
# we query the database for item transfer information, sacrificing
# accuracy in some cases for speed;
#
# 1. don't query if item has one of the other statuses
# 2. don't check transit status if the bib has
# more than 20 items
#
# FIXME: to avoid having the query the database like this, and to make
# the in transit status count as unavailable for search limiting,
# should map transit status to record indexed in Zebra.
#
($transfertwhen, $transfertfrom, $transfertto) = C4::Circulation::GetTransfers($item->{itemnumber});
}
# item is withdrawn, lost or damaged
if ( $item->{wthdrawn}
|| $item->{itemlost}
|| $item->{damaged}
|| $item->{notforloan}
|| ($transfertwhen ne ''))
{
$wthdrawn_count++ if $item->{wthdrawn};
$itemlost_count++ if $item->{itemlost};
$itemdamaged_count++ if $item->{damaged};
$item_in_transit_count++ if $transfertwhen ne '';
$item->{status} = $item->{wthdrawn} . "-" . $item->{itemlost} . "-" . $item->{damaged} . "-" . $item->{notforloan};
$other_count++;
my $key = $prefix . $item->{status};
foreach (qw(wthdrawn itemlost damaged branchname itemcallnumber)) {
$other_items->{$key}->{$_} = $item->{$_};
}
$other_items->{$key}->{intransit} = ($transfertwhen ne '') ? 1 : 0;
$other_items->{$key}->{notforloan} = GetAuthorisedValueDesc('','',$item->{notforloan},'','',$notforloan_authorised_value) if $notforloan_authorised_value;
$other_items->{$key}->{count}++ if $item->{$hbranch};
$other_items->{$key}->{location} = $shelflocations->{ $item->{location} };
$other_items->{$key}->{imageurl} = getitemtypeimagelocation( 'opac', $itemtypes{ $item->{itype} }->{imageurl} );
}
# item is available
else {
$can_place_holds = 1;
$available_count++;
$available_items->{$prefix}->{count}++ if $item->{$hbranch};
foreach (qw(branchname itemcallnumber)) {
$available_items->{$prefix}->{$_} = $item->{$_};
}
$available_items->{$prefix}->{location} = $shelflocations->{ $item->{location} };
$available_items->{$prefix}->{imageurl} = getitemtypeimagelocation( 'opac', $itemtypes{ $item->{itype} }->{imageurl} );
}
}
} # notforloan, item level and biblioitem level
my ( $availableitemscount, $onloanitemscount, $otheritemscount );
$maxitems =
( C4::Context->preference('maxItemsinSearchResults') )
? C4::Context->preference('maxItemsinSearchResults') - 1
: 1;
for my $key ( sort keys %$onloan_items ) {
(++$onloanitemscount > $maxitems) and last;
push @onloan_items_loop, $onloan_items->{$key};
}
for my $key ( sort keys %$other_items ) {
(++$otheritemscount > $maxitems) and last;
push @other_items_loop, $other_items->{$key};
}
for my $key ( sort keys %$available_items ) {
(++$availableitemscount > $maxitems) and last;
push @available_items_loop, $available_items->{$key}
}
# XSLT processing of some stuff
if (C4::Context->preference("XSLTResultsDisplay") && !$scan) {
my $newxmlrecord = XSLTParse4Display($oldbiblio->{biblionumber}, $marcrecord, C4::Context->config('opachtdocs')."/prog/en/xslt/MARC21slim2OPACResults.xsl");
$oldbiblio->{XSLTResultsRecord} = $newxmlrecord;
}
# last check for norequest : if itemtype is notforloan, it can't be reserved either, whatever the items
$can_place_holds = 0
if $itemtypes{ $oldbiblio->{itemtype} }->{notforloan};
$oldbiblio->{norequests} = 1 unless $can_place_holds;
$oldbiblio->{itemsplural} = 1 if $items_count > 1;
$oldbiblio->{items_count} = $items_count;
$oldbiblio->{available_items_loop} = \@available_items_loop;
$oldbiblio->{onloan_items_loop} = \@onloan_items_loop;
$oldbiblio->{other_items_loop} = \@other_items_loop;
$oldbiblio->{availablecount} = $available_count;
$oldbiblio->{availableplural} = 1 if $available_count > 1;
$oldbiblio->{onloancount} = $onloan_count;
$oldbiblio->{onloanplural} = 1 if $onloan_count > 1;
$oldbiblio->{othercount} = $other_count;
$oldbiblio->{otherplural} = 1 if $other_count > 1;
$oldbiblio->{wthdrawncount} = $wthdrawn_count;
$oldbiblio->{itemlostcount} = $itemlost_count;
$oldbiblio->{damagedcount} = $itemdamaged_count;
$oldbiblio->{intransitcount} = $item_in_transit_count;
$oldbiblio->{orderedcount} = $ordered_count;
push( @newresults, $oldbiblio );
}
return @newresults;
}
#----------------------------------------------------------------------
#
# Non-Zebra GetRecords#
#----------------------------------------------------------------------
=head2 NZgetRecords
NZgetRecords has the same API as zera getRecords, even if some parameters are not managed
=cut
sub NZgetRecords {
my (
$query, $simple_query, $sort_by_ref, $servers_ref,
$results_per_page, $offset, $expanded_facet, $branches,
$query_type, $scan
) = @_;
warn "query =$query" if $DEBUG;
my $result = NZanalyse($query);
warn "results =$result" if $DEBUG;
return ( undef,
NZorder( $result, @$sort_by_ref[0], $results_per_page, $offset ),
undef );
}
=head2 NZanalyse
NZanalyse : get a CQL string as parameter, and returns a list of biblionumber;title,biblionumber;title,...
the list is built from an inverted index in the nozebra SQL table
note that title is here only for convenience : the sorting will be very fast when requested on title
if the sorting is requested on something else, we will have to reread all results, and that may be longer.
=cut
sub NZanalyse {
my ( $string, $server ) = @_;
# warn "---------" if $DEBUG;
warn " NZanalyse" if $DEBUG;
# warn "---------" if $DEBUG;
# $server contains biblioserver or authorities, depending on what we search on.
#warn "querying : $string on $server";
$server = 'biblioserver' unless $server;
# if we have a ", replace the content to discard temporarily any and/or/not inside
my $commacontent;
if ( $string =~ /"/ ) {
$string =~ s/"(.*?)"/__X__/;
$commacontent = $1;
warn "commacontent : $commacontent" if $DEBUG;
}
# split the query string in 3 parts : X AND Y means : $left="X", $operand="AND" and $right="Y"
# then, call again NZanalyse with $left and $right
# (recursive until we find a leaf (=> something without and/or/not)
# delete repeated operator... Would then go in infinite loop
while ( $string =~ s/( and| or| not| AND| OR| NOT)\1/$1/g ) {
}
#process parenthesis before.
if ( $string =~ /^\s*\((.*)\)(( and | or | not | AND | OR | NOT )(.*))?/ ) {
my $left = $1;
my $right = $4;
my $operator = lc($3); # FIXME: and/or/not are operators, not operands
warn
"dealing w/parenthesis before recursive sub call. left :$left operator:$operator right:$right"
if $DEBUG;
my $leftresult = NZanalyse( $left, $server );
if ($operator) {
my $rightresult = NZanalyse( $right, $server );
# OK, we have the results for right and left part of the query
# depending of operand, intersect, union or exclude both lists
# to get a result list
if ( $operator eq ' and ' ) {
return NZoperatorAND($leftresult,$rightresult);
}
elsif ( $operator eq ' or ' ) {
# just merge the 2 strings
return $leftresult . $rightresult;
}
elsif ( $operator eq ' not ' ) {
return NZoperatorNOT($leftresult,$rightresult);
}
}
else {
# this error is impossible, because of the regexp that isolate the operand, but just in case...
return $leftresult;
}
}
warn "string :" . $string if $DEBUG;
my $left = "";
my $right = "";
my $operator = "";
if ($string =~ /(.*?)( and | or | not | AND | OR | NOT )(.*)/) {
$left = $1;
$right = $3;
$operator = lc($2); # FIXME: and/or/not are operators, not operands
}
warn "no parenthesis. left : $left operator: $operator right: $right"
if $DEBUG;
# it's not a leaf, we have a and/or/not
if ($operator) {
# reintroduce comma content if needed
$right =~ s/__X__/"$commacontent"/ if $commacontent;
$left =~ s/__X__/"$commacontent"/ if $commacontent;
warn "node : $left / $operator / $right\n" if $DEBUG;
my $leftresult = NZanalyse( $left, $server );
my $rightresult = NZanalyse( $right, $server );
warn " leftresult : $leftresult" if $DEBUG;
warn " rightresult : $rightresult" if $DEBUG;
# OK, we have the results for right and left part of the query
# depending of operand, intersect, union or exclude both lists
# to get a result list
if ( $operator eq ' and ' ) {
warn "NZAND";
return NZoperatorAND($leftresult,$rightresult);
}
elsif ( $operator eq ' or ' ) {
# just merge the 2 strings
return $leftresult . $rightresult;
}
elsif ( $operator eq ' not ' ) {
return NZoperatorNOT($leftresult,$rightresult);
}
else {
# this error is impossible, because of the regexp that isolate the operand, but just in case...
die "error : operand unknown : $operator for $string";
}
# it's a leaf, do the real SQL query and return the result
}
else {
$string =~ s/__X__/"$commacontent"/ if $commacontent;
$string =~ s/-|\.|\?|,|;|!|'|\(|\)|\[|\]|{|}|"|&|\+|\*|\// /g;
#remove trailing blank at the beginning
$string =~ s/^ //g;
warn "leaf:$string" if $DEBUG;
# parse the string in in operator/operand/value again
my $left = "";
my $operator = "";
my $right = "";
if ($string =~ /(.*)(>=|<=)(.*)/) {
$left = $1;
$operator = $2;
$right = $3;
} else {
$left = $string;
}
# warn "handling leaf... left:$left operator:$operator right:$right"
# if $DEBUG;
unless ($operator) {
if ($string =~ /(.*)(>|<|=)(.*)/) {
$left = $1;
$operator = $2;
$right = $3;
warn
"handling unless (operator)... left:$left operator:$operator right:$right"
if $DEBUG;
} else {
$left = $string;
}
}
my $results;
# strip adv, zebra keywords, currently not handled in nozebra: wrdl, ext, phr...
$left =~ s/ .*$//;
# automatic replace for short operators
$left = 'title' if $left =~ '^ti$';
$left = 'author' if $left =~ '^au$';
$left = 'publisher' if $left =~ '^pb$';
$left = 'subject' if $left =~ '^su$';
$left = 'koha-Auth-Number' if $left =~ '^an$';
$left = 'keyword' if $left =~ '^kw$';
warn "handling leaf... left:$left operator:$operator right:$right" if $DEBUG;
if ( $operator && $left ne 'keyword' ) {
#do a specific search
my $dbh = C4::Context->dbh;
$operator = 'LIKE' if $operator eq '=' and $right =~ /%/;
my $sth =
$dbh->prepare(
"SELECT biblionumbers,value FROM nozebra WHERE server=? AND indexname=? AND value $operator ?"
);
warn "$left / $operator / $right\n" if $DEBUG;
# split each word, query the DB and build the biblionumbers result
#sanitizing leftpart
$left =~ s/^\s+|\s+$//;
foreach ( split / /, $right ) {
my $biblionumbers;
$_ =~ s/^\s+|\s+$//;
next unless $_;
warn "EXECUTE : $server, $left, $_" if $DEBUG;
$sth->execute( $server, $left, $_ )
or warn "execute failed: $!";
while ( my ( $line, $value ) = $sth->fetchrow ) {
# if we are dealing with a numeric value, use only numeric results (in case of >=, <=, > or <)
# otherwise, fill the result
$biblionumbers .= $line
unless ( $right =~ /^\d+$/ && $value =~ /\D/ );
warn "result : $value "
. ( $right =~ /\d/ ) . "=="
. ( $value =~ /\D/?$line:"" ) if $DEBUG; #= $line";
}
# do a AND with existing list if there is one, otherwise, use the biblionumbers list as 1st result list
if ($results) {
warn "NZAND" if $DEBUG;
$results = NZoperatorAND($biblionumbers,$results);
}
else {
$results = $biblionumbers;
}
}
}
else {
#do a complete search (all indexes), if index='kw' do complete search too.
my $dbh = C4::Context->dbh;
my $sth =
$dbh->prepare(
"SELECT biblionumbers FROM nozebra WHERE server=? AND value LIKE ?"
);
# split each word, query the DB and build the biblionumbers result
foreach ( split / /, $string ) {
next if C4::Context->stopwords->{ uc($_) }; # skip if stopword
warn "search on all indexes on $_" if $DEBUG;
my $biblionumbers;
next unless $_;
$sth->execute( $server, $_ );
while ( my $line = $sth->fetchrow ) {
$biblionumbers .= $line;
}
# do a AND with existing list if there is one, otherwise, use the biblionumbers list as 1st result list
if ($results) {
$results = NZoperatorAND($biblionumbers,$results);
}
else {
warn "NEW RES for $_ = $biblionumbers" if $DEBUG;
$results = $biblionumbers;
}
}
}
warn "return : $results for LEAF : $string" if $DEBUG;
return $results;
}
warn "---------\nLeave NZanalyse\n---------" if $DEBUG;
}
sub NZoperatorAND{
my ($rightresult, $leftresult)=@_;
my @leftresult = split /;/, $leftresult;
warn " @leftresult / $rightresult \n" if $DEBUG;
# my @rightresult = split /;/,$leftresult;
my $finalresult;
# parse the left results, and if the biblionumber exist in the right result, save it in finalresult
# the result is stored twice, to have the same weight for AND than OR.
# example : TWO : 61,61,64,121 (two is twice in the biblio #61) / TOWER : 61,64,130
# result : 61,61,61,61,64,64 for two AND tower : 61 has more weight than 64
foreach (@leftresult) {
my $value = $_;
my $countvalue;
( $value, $countvalue ) = ( $1, $2 ) if ($value=~/(.*)-(\d+)$/);
if ( $rightresult =~ /\Q$value\E-(\d+);/ ) {
$countvalue = ( $1 > $countvalue ? $countvalue : $1 );
$finalresult .=
"$value-$countvalue;$value-$countvalue;";
}
}
warn "NZAND DONE : $finalresult \n" if $DEBUG;
return $finalresult;
}
sub NZoperatorOR{
my ($rightresult, $leftresult)=@_;
return $rightresult.$leftresult;
}
sub NZoperatorNOT{
my ($leftresult, $rightresult)=@_;
my @leftresult = split /;/, $leftresult;
# my @rightresult = split /;/,$leftresult;
my $finalresult;
foreach (@leftresult) {
my $value=$_;
$value=$1 if $value=~m/(.*)-\d+$/;
unless ($rightresult =~ "$value-") {
$finalresult .= "$_;";
}
}
return $finalresult;
}
=head2 NZorder
$finalresult = NZorder($biblionumbers, $ordering,$results_per_page,$offset);
TODO :: Description
=cut
sub NZorder {
my ( $biblionumbers, $ordering, $results_per_page, $offset ) = @_;
warn "biblionumbers = $biblionumbers and ordering = $ordering\n" if $DEBUG;
# order title asc by default
# $ordering = '1=36 <i' unless $ordering;
$results_per_page = 20 unless $results_per_page;
$offset = 0 unless $offset;
my $dbh = C4::Context->dbh;
#
# order by POPULARITY
#
if ( $ordering =~ /popularity/ ) {
my %result;
my %popularity;
# popularity is not in MARC record, it's builded from a specific query
my $sth =
$dbh->prepare("select sum(issues) from items where biblionumber=?");
foreach ( split /;/, $biblionumbers ) {
my ( $biblionumber, $title ) = split /,/, $_;
$result{$biblionumber} = GetMarcBiblio($biblionumber);
$sth->execute($biblionumber);
my $popularity = $sth->fetchrow || 0;
# hint : the key is popularity.title because we can have
# many results with the same popularity. In this cas, sub-ordering is done by title
# we also have biblionumber to avoid bug for 2 biblios with the same title & popularity
# (un-frequent, I agree, but we won't forget anything that way ;-)
$popularity{ sprintf( "%10d", $popularity ) . $title
. $biblionumber } = $biblionumber;
}
# sort the hash and return the same structure as GetRecords (Zebra querying)
my $result_hash;
my $numbers = 0;
if ( $ordering eq 'popularity_dsc' ) { # sort popularity DESC
foreach my $key ( sort { $b cmp $a } ( keys %popularity ) ) {
$result_hash->{'RECORDS'}[ $numbers++ ] =
$result{ $popularity{$key} }->as_usmarc();
}
}
else { # sort popularity ASC
foreach my $key ( sort ( keys %popularity ) ) {
$result_hash->{'RECORDS'}[ $numbers++ ] =
$result{ $popularity{$key} }->as_usmarc();
}
}
my $finalresult = ();
$result_hash->{'hits'} = $numbers;
$finalresult->{'biblioserver'} = $result_hash;
return $finalresult;
#
# ORDER BY author
#
}
elsif ( $ordering =~ /author/ ) {
my %result;
foreach ( split /;/, $biblionumbers ) {
my ( $biblionumber, $title ) = split /,/, $_;
my $record = GetMarcBiblio($biblionumber);
my $author;
if ( C4::Context->preference('marcflavour') eq 'UNIMARC' ) {
$author = $record->subfield( '200', 'f' );
$author = $record->subfield( '700', 'a' ) unless $author;
}
else {
$author = $record->subfield( '100', 'a' );
}
# hint : the result is sorted by title.biblionumber because we can have X biblios with the same title
# and we don't want to get only 1 result for each of them !!!
$result{ $author . $biblionumber } = $record;
}
# sort the hash and return the same structure as GetRecords (Zebra querying)
my $result_hash;
my $numbers = 0;
if ( $ordering eq 'author_za' ) { # sort by author desc
foreach my $key ( sort { $b cmp $a } ( keys %result ) ) {
$result_hash->{'RECORDS'}[ $numbers++ ] =
$result{$key}->as_usmarc();
}
}
else { # sort by author ASC
foreach my $key ( sort ( keys %result ) ) {
$result_hash->{'RECORDS'}[ $numbers++ ] =
$result{$key}->as_usmarc();
}
}
my $finalresult = ();
$result_hash->{'hits'} = $numbers;
$finalresult->{'biblioserver'} = $result_hash;
return $finalresult;
#
# ORDER BY callnumber
#
}
elsif ( $ordering =~ /callnumber/ ) {
my %result;
foreach ( split /;/, $biblionumbers ) {
my ( $biblionumber, $title ) = split /,/, $_;
my $record = GetMarcBiblio($biblionumber);
my $callnumber;
my $frameworkcode = GetFrameworkCode($biblionumber);
my ( $callnumber_tag, $callnumber_subfield ) = GetMarcFromKohaField( 'items.itemcallnumber', $frameworkcode);
( $callnumber_tag, $callnumber_subfield ) = GetMarcFromKohaField('biblioitems.callnumber', $frameworkcode)
unless $callnumber_tag;
if ( C4::Context->preference('marcflavour') eq 'UNIMARC' ) {
$callnumber = $record->subfield( '200', 'f' );
} else {
$callnumber = $record->subfield( '100', 'a' );
}
# hint : the result is sorted by title.biblionumber because we can have X biblios with the same title
# and we don't want to get only 1 result for each of them !!!
$result{ $callnumber . $biblionumber } = $record;
}
# sort the hash and return the same structure as GetRecords (Zebra querying)
my $result_hash;
my $numbers = 0;
if ( $ordering eq 'call_number_dsc' ) { # sort by title desc
foreach my $key ( sort { $b cmp $a } ( keys %result ) ) {
$result_hash->{'RECORDS'}[ $numbers++ ] =
$result{$key}->as_usmarc();
}
}
else { # sort by title ASC
foreach my $key ( sort { $a cmp $b } ( keys %result ) ) {
$result_hash->{'RECORDS'}[ $numbers++ ] =
$result{$key}->as_usmarc();
}
}
my $finalresult = ();
$result_hash->{'hits'} = $numbers;
$finalresult->{'biblioserver'} = $result_hash;
return $finalresult;
}
elsif ( $ordering =~ /pubdate/ ) { #pub year
my %result;
foreach ( split /;/, $biblionumbers ) {
my ( $biblionumber, $title ) = split /,/, $_;
my $record = GetMarcBiblio($biblionumber);
my ( $publicationyear_tag, $publicationyear_subfield ) =
GetMarcFromKohaField( 'biblioitems.publicationyear', '' );
my $publicationyear =
$record->subfield( $publicationyear_tag,
$publicationyear_subfield );
# hint : the result is sorted by title.biblionumber because we can have X biblios with the same title
# and we don't want to get only 1 result for each of them !!!
$result{ $publicationyear . $biblionumber } = $record;
}
# sort the hash and return the same structure as GetRecords (Zebra querying)
my $result_hash;
my $numbers = 0;
if ( $ordering eq 'pubdate_dsc' ) { # sort by pubyear desc
foreach my $key ( sort { $b cmp $a } ( keys %result ) ) {
$result_hash->{'RECORDS'}[ $numbers++ ] =
$result{$key}->as_usmarc();
}
}
else { # sort by pub year ASC
foreach my $key ( sort ( keys %result ) ) {
$result_hash->{'RECORDS'}[ $numbers++ ] =
$result{$key}->as_usmarc();
}
}
my $finalresult = ();
$result_hash->{'hits'} = $numbers;
$finalresult->{'biblioserver'} = $result_hash;
return $finalresult;
#
# ORDER BY title
#
}
elsif ( $ordering =~ /title/ ) {
# the title is in the biblionumbers string, so we just need to build a hash, sort it and return
my %result;
foreach ( split /;/, $biblionumbers ) {
my ( $biblionumber, $title ) = split /,/, $_;
# hint : the result is sorted by title.biblionumber because we can have X biblios with the same title
# and we don't want to get only 1 result for each of them !!!
# hint & speed improvement : we can order without reading the record
# so order, and read records only for the requested page !
$result{ $title . $biblionumber } = $biblionumber;
}
# sort the hash and return the same structure as GetRecords (Zebra querying)
my $result_hash;
my $numbers = 0;
if ( $ordering eq 'title_az' ) { # sort by title desc
foreach my $key ( sort ( keys %result ) ) {
$result_hash->{'RECORDS'}[ $numbers++ ] = $result{$key};
}
}
else { # sort by title ASC
foreach my $key ( sort { $b cmp $a } ( keys %result ) ) {
$result_hash->{'RECORDS'}[ $numbers++ ] = $result{$key};
}
}
# limit the $results_per_page to result size if it's more
$results_per_page = $numbers - 1 if $numbers < $results_per_page;
# for the requested page, replace biblionumber by the complete record
# speed improvement : avoid reading too much things
for (
my $counter = $offset ;
$counter <= $offset + $results_per_page ;
$counter++
)
{
$result_hash->{'RECORDS'}[$counter] =
GetMarcBiblio( $result_hash->{'RECORDS'}[$counter] )->as_usmarc;
}
my $finalresult = ();
$result_hash->{'hits'} = $numbers;
$finalresult->{'biblioserver'} = $result_hash;
return $finalresult;
}
else {
#
# order by ranking
#
# we need 2 hashes to order by ranking : the 1st one to count the ranking, the 2nd to order by ranking
my %result;
my %count_ranking;
foreach ( split /;/, $biblionumbers ) {
my ( $biblionumber, $title ) = split /,/, $_;
$title =~ /(.*)-(\d)/;
# get weight
my $ranking = $2;
# note that we + the ranking because ranking is calculated on weight of EACH term requested.
# if we ask for "two towers", and "two" has weight 2 in biblio N, and "towers" has weight 4 in biblio N
# biblio N has ranking = 6
$count_ranking{$biblionumber} += $ranking;
}
# build the result by "inverting" the count_ranking hash
# hing : as usual, we don't order by ranking only, to avoid having only 1 result for each rank. We build an hash on concat(ranking,biblionumber) instead
# warn "counting";
foreach ( keys %count_ranking ) {
$result{ sprintf( "%10d", $count_ranking{$_} ) . '-' . $_ } = $_;
}
# sort the hash and return the same structure as GetRecords (Zebra querying)
my $result_hash;
my $numbers = 0;
foreach my $key ( sort { $b cmp $a } ( keys %result ) ) {
$result_hash->{'RECORDS'}[ $numbers++ ] = $result{$key};
}
# limit the $results_per_page to result size if it's more
$results_per_page = $numbers - 1 if $numbers < $results_per_page;
# for the requested page, replace biblionumber by the complete record
# speed improvement : avoid reading too much things
for (
my $counter = $offset ;
$counter <= $offset + $results_per_page ;
$counter++
)
{
$result_hash->{'RECORDS'}[$counter] =
GetMarcBiblio( $result_hash->{'RECORDS'}[$counter] )->as_usmarc
if $result_hash->{'RECORDS'}[$counter];
}
my $finalresult = ();
$result_hash->{'hits'} = $numbers;
$finalresult->{'biblioserver'} = $result_hash;
return $finalresult;
}
}
=head2 ModBiblios
($countchanged,$listunchanged) = ModBiblios($listbiblios, $tagsubfield,$initvalue,$targetvalue,$test);
this function changes all the values $initvalue in subfield $tag$subfield in any record in $listbiblios
test parameter if set donot perform change to records in database.
=over 2
=item C<input arg:>
* $listbiblios is an array ref to marcrecords to be changed
* $tagsubfield is the reference of the subfield to change.
* $initvalue is the value to search the record for
* $targetvalue is the value to set the subfield to
* $test is to be set only not to perform changes in database.
=item C<Output arg:>
* $countchanged counts all the changes performed.
* $listunchanged contains the list of all the biblionumbers of records unchanged.
=item C<usage in the script:>
=back
my ($countchanged, $listunchanged) = EditBiblios($results->{RECORD}, $tagsubfield,$initvalue,$targetvalue);;
#If one wants to display unchanged records, you should get biblios foreach @$listunchanged
$template->param(countchanged => $countchanged, loopunchanged=>$listunchanged);
=cut
sub ModBiblios {
my ( $listbiblios, $tagsubfield, $initvalue, $targetvalue, $test ) = @_;
my $countmatched;
my @unmatched;
my ( $tag, $subfield ) = ( $1, $2 )
if ( $tagsubfield =~ /^(\d{1,3})([a-z0-9A-Z@])?$/ );
if ( ( length($tag) < 3 ) && $subfield =~ /0-9/ ) {
$tag = $tag . $subfield;
undef $subfield;
}
my ( $bntag, $bnsubf ) = GetMarcFromKohaField('biblio.biblionumber', '');
my ( $itemtag, $itemsubf ) = GetMarcFromKohaField('items.itemnumber', '');
if ($tag eq $itemtag) {
# do not allow the embedded item tag to be
# edited from here
warn "Attempting to edit item tag via C4::Search::ModBiblios -- not allowed";
return (0, []);
}
foreach my $usmarc (@$listbiblios) {
my $record;
$record = eval { MARC::Record->new_from_usmarc($usmarc) };
my $biblionumber;
if ($@) {
# usmarc is not a valid usmarc May be a biblionumber
# FIXME - sorry, please let's figure out whether
# this function is to be passed a list of
# record numbers or a list of MARC::Record
# objects. The former is probably better
# because the MARC records supplied by Zebra
# may be not current.
$record = GetMarcBiblio($usmarc);
$biblionumber = $usmarc;
}
else {
if ( $bntag >= 010 ) {
$biblionumber = $record->subfield( $bntag, $bnsubf );
}
else {
$biblionumber = $record->field($bntag)->data;
}
}
#GetBiblionumber is to be written.
#Could be replaced by TransformMarcToKoha (But Would be longer)
if ( $record->field($tag) ) {
my $modify = 0;
foreach my $field ( $record->field($tag) ) {
if ($subfield) {
if (
$field->delete_subfield(
'code' => $subfield,
'match' => qr($initvalue)
)
)
{
$countmatched++;
$modify = 1;
$field->update( $subfield, $targetvalue )
if ($targetvalue);
}
}
else {
if ( $tag >= 010 ) {
if ( $field->delete_field($field) ) {
$countmatched++;
$modify = 1;
}
}
else {
$field->data = $targetvalue
if ( $field->data =~ qr($initvalue) );
}
}
}
# warn $record->as_formatted;
if ($modify) {
ModBiblio( $record, $biblionumber,
GetFrameworkCode($biblionumber) )
unless ($test);
}
else {
push @unmatched, $biblionumber;
}
}
else {
push @unmatched, $biblionumber;
}
}
return ( $countmatched, \@unmatched );
}
END { } # module clean-up code here (global destructor)
1;
__END__
=head1 AUTHOR
Koha Developement team <info@koha.org>
=cut