Bug 16053: Add tests
[koha.git] / C4 / Search.pm
blobfcbb660f428cb4373146c91ce0ebb768a2c56418
1 package C4::Search;
3 # This file is part of Koha.
5 # Koha is free software; you can redistribute it and/or modify it
6 # under the terms of the GNU General Public License as published by
7 # the Free Software Foundation; either version 3 of the License, or
8 # (at your option) any later version.
10 # Koha is distributed in the hope that it will be useful, but
11 # WITHOUT ANY WARRANTY; without even the implied warranty of
12 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 # GNU General Public License for more details.
15 # You should have received a copy of the GNU General Public License
16 # along with Koha; if not, see <http://www.gnu.org/licenses>.
18 use strict;
19 #use warnings; FIXME - Bug 2505
20 require Exporter;
21 use C4::Context;
22 use C4::Biblio; # GetMarcFromKohaField, GetBiblioData
23 use C4::Koha; # getFacets
24 use Koha::DateUtils;
25 use Lingua::Stem;
26 use C4::Search::PazPar2;
27 use XML::Simple;
28 use C4::Members qw(GetHideLostItemsPreference);
29 use C4::XSLT;
30 use C4::Branch;
31 use C4::Reserves; # GetReserveStatus
32 use C4::Debug;
33 use C4::Charset;
34 use Koha::Libraries;
35 use YAML;
36 use URI::Escape;
37 use Business::ISBN;
38 use MARC::Record;
39 use MARC::Field;
40 use vars qw($VERSION @ISA @EXPORT @EXPORT_OK %EXPORT_TAGS $DEBUG);
42 # set the version for version checking
43 BEGIN {
44 $VERSION = 3.07.00.049;
45 $DEBUG = ($ENV{DEBUG}) ? 1 : 0;
48 =head1 NAME
50 C4::Search - Functions for searching the Koha catalog.
52 =head1 SYNOPSIS
54 See opac/opac-search.pl or catalogue/search.pl for example of usage
56 =head1 DESCRIPTION
58 This module provides searching functions for Koha's bibliographic databases
60 =head1 FUNCTIONS
62 =cut
64 @ISA = qw(Exporter);
65 @EXPORT = qw(
66 &FindDuplicate
67 &SimpleSearch
68 &searchResults
69 &getRecords
70 &buildQuery
71 &GetDistinctValues
72 &enabled_staff_search_views
75 # make all your functions, whether exported or not;
77 =head2 FindDuplicate
79 ($biblionumber,$biblionumber,$title) = FindDuplicate($record);
81 This function attempts to find duplicate records using a hard-coded, fairly simplistic algorithm
83 =cut
85 sub FindDuplicate {
86 my ($record) = @_;
87 my $dbh = C4::Context->dbh;
88 my $result = TransformMarcToKoha( $dbh, $record, '' );
89 my $sth;
90 my $query;
91 my $search;
92 my $type;
93 my ( $biblionumber, $title );
95 # search duplicate on ISBN, easy and fast..
96 # ... normalize first
97 if ( $result->{isbn} ) {
98 $result->{isbn} =~ s/\(.*$//;
99 $result->{isbn} =~ s/\s+$//;
100 $query = "isbn:$result->{isbn}";
102 else {
103 my $QParser;
104 $QParser = C4::Context->queryparser if (C4::Context->preference('UseQueryParser'));
105 my $titleindex;
106 my $authorindex;
107 my $op;
109 if ($QParser) {
110 $titleindex = 'title|exact';
111 $authorindex = 'author|exact';
112 $op = '&&';
113 $QParser->custom_data->{'QueryAutoTruncate'} = C4::Context->preference('QueryAutoTruncate');
114 } else {
115 $titleindex = 'ti,ext';
116 $authorindex = 'au,ext';
117 $op = 'and';
120 $result->{title} =~ s /\\//g;
121 $result->{title} =~ s /\"//g;
122 $result->{title} =~ s /\(//g;
123 $result->{title} =~ s /\)//g;
125 # FIXME: instead of removing operators, could just do
126 # quotes around the value
127 $result->{title} =~ s/(and|or|not)//g;
128 $query = "$titleindex:\"$result->{title}\"";
129 if ( $result->{author} ) {
130 $result->{author} =~ s /\\//g;
131 $result->{author} =~ s /\"//g;
132 $result->{author} =~ s /\(//g;
133 $result->{author} =~ s /\)//g;
135 # remove valid operators
136 $result->{author} =~ s/(and|or|not)//g;
137 $query .= " $op $authorindex:\"$result->{author}\"";
141 my ( $error, $searchresults, undef ) = SimpleSearch($query); # FIXME :: hardcoded !
142 my @results;
143 if (!defined $error) {
144 foreach my $possible_duplicate_record (@{$searchresults}) {
145 my $marcrecord = new_record_from_zebra(
146 'biblioserver',
147 $possible_duplicate_record
150 my $result = TransformMarcToKoha( $dbh, $marcrecord, '' );
152 # FIXME :: why 2 $biblionumber ?
153 if ($result) {
154 push @results, $result->{'biblionumber'};
155 push @results, $result->{'title'};
159 return @results;
162 =head2 SimpleSearch
164 ( $error, $results, $total_hits ) = SimpleSearch( $query, $offset, $max_results, [@servers] );
166 This function provides a simple search API on the bibliographic catalog
168 =over 2
170 =item C<input arg:>
172 * $query can be a simple keyword or a complete CCL query
173 * @servers is optional. Defaults to biblioserver as found in koha-conf.xml
174 * $offset - If present, represents the number of records at the beginning to omit. Defaults to 0
175 * $max_results - if present, determines the maximum number of records to fetch. undef is All. defaults to undef.
178 =item C<Return:>
180 Returns an array consisting of three elements
181 * $error is undefined unless an error is detected
182 * $results is a reference to an array of records.
183 * $total_hits is the number of hits that would have been returned with no limit
185 If an error is returned the two other return elements are undefined. If error itself is undefined
186 the other two elements are always defined
188 =item C<usage in the script:>
190 =back
192 my ( $error, $marcresults, $total_hits ) = SimpleSearch($query);
194 if (defined $error) {
195 $template->param(query_error => $error);
196 warn "error: ".$error;
197 output_html_with_http_headers $input, $cookie, $template->output;
198 exit;
201 my $hits = @{$marcresults};
202 my @results;
204 for my $r ( @{$marcresults} ) {
205 my $marcrecord = MARC::File::USMARC::decode($r);
206 my $biblio = TransformMarcToKoha(C4::Context->dbh,$marcrecord,q{});
208 #build the iarray of hashs for the template.
209 push @results, {
210 title => $biblio->{'title'},
211 subtitle => $biblio->{'subtitle'},
212 biblionumber => $biblio->{'biblionumber'},
213 author => $biblio->{'author'},
214 publishercode => $biblio->{'publishercode'},
215 publicationyear => $biblio->{'publicationyear'},
220 $template->param(result=>\@results);
222 =cut
224 sub SimpleSearch {
225 my ( $query, $offset, $max_results, $servers ) = @_;
227 return ( 'No query entered', undef, undef ) unless $query;
228 # FIXME hardcoded value. See catalog/search.pl & opac-search.pl too.
229 my @servers = defined ( $servers ) ? @$servers : ( 'biblioserver' );
230 my @zoom_queries;
231 my @tmpresults;
232 my @zconns;
233 my $results = [];
234 my $total_hits = 0;
236 my $QParser;
237 $QParser = C4::Context->queryparser if (C4::Context->preference('UseQueryParser') && ! ($query =~ m/\w,\w|\w=\w/));
238 if ($QParser) {
239 $QParser->custom_data->{'QueryAutoTruncate'} = C4::Context->preference('QueryAutoTruncate');
242 # Initialize & Search Zebra
243 for ( my $i = 0 ; $i < @servers ; $i++ ) {
244 eval {
245 $zconns[$i] = C4::Context->Zconn( $servers[$i], 1 );
246 if ($QParser) {
247 $query =~ s/=/:/g;
248 $QParser->parse( $query );
249 $query = $QParser->target_syntax($servers[$i]);
250 $zoom_queries[$i] = new ZOOM::Query::PQF( $query, $zconns[$i]);
251 } else {
252 $query =~ s/:/=/g;
253 $zoom_queries[$i] = new ZOOM::Query::CCL2RPN( $query, $zconns[$i]);
255 $tmpresults[$i] = $zconns[$i]->search( $zoom_queries[$i] );
257 # error handling
258 my $error =
259 $zconns[$i]->errmsg() . " ("
260 . $zconns[$i]->errcode() . ") "
261 . $zconns[$i]->addinfo() . " "
262 . $zconns[$i]->diagset();
264 return ( $error, undef, undef ) if $zconns[$i]->errcode();
266 if ($@) {
268 # caught a ZOOM::Exception
269 my $error =
270 $@->message() . " ("
271 . $@->code() . ") "
272 . $@->addinfo() . " "
273 . $@->diagset();
274 warn $error." for query: $query";
275 return ( $error, undef, undef );
279 _ZOOM_event_loop(
280 \@zconns,
281 \@tmpresults,
282 sub {
283 my ($i, $size) = @_;
284 my $first_record = defined($offset) ? $offset + 1 : 1;
285 my $hits = $tmpresults[ $i - 1 ]->size();
286 $total_hits += $hits;
287 my $last_record = $hits;
288 if ( defined $max_results && $offset + $max_results < $hits ) {
289 $last_record = $offset + $max_results;
292 for my $j ( $first_record .. $last_record ) {
293 my $record = eval {
294 $tmpresults[ $i - 1 ]->record( $j - 1 )->raw()
295 ; # 0 indexed
297 push @{$results}, $record if defined $record;
302 foreach my $zoom_query (@zoom_queries) {
303 $zoom_query->destroy();
306 return ( undef, $results, $total_hits );
309 =head2 getRecords
311 ( undef, $results_hashref, \@facets_loop ) = getRecords (
313 $koha_query, $simple_query, $sort_by_ref, $servers_ref,
314 $results_per_page, $offset, $expanded_facet, $branches,$itemtypes,
315 $query_type, $scan
318 The all singing, all dancing, multi-server, asynchronous, scanning,
319 searching, record nabbing, facet-building
321 See verbse embedded documentation.
323 =cut
325 sub getRecords {
326 my (
327 $koha_query, $simple_query, $sort_by_ref, $servers_ref,
328 $results_per_page, $offset, $expanded_facet, $branches,
329 $itemtypes, $query_type, $scan, $opac
330 ) = @_;
332 my @servers = @$servers_ref;
333 my @sort_by = @$sort_by_ref;
335 # Initialize variables for the ZOOM connection and results object
336 my $zconn;
337 my @zconns;
338 my @results;
339 my $results_hashref = ();
341 # Initialize variables for the faceted results objects
342 my $facets_counter = {};
343 my $facets_info = {};
344 my $facets = getFacets();
346 my @facets_loop; # stores the ref to array of hashes for template facets loop
348 ### LOOP THROUGH THE SERVERS
349 for ( my $i = 0 ; $i < @servers ; $i++ ) {
350 $zconns[$i] = C4::Context->Zconn( $servers[$i], 1 );
352 # perform the search, create the results objects
353 # if this is a local search, use the $koha-query, if it's a federated one, use the federated-query
354 my $query_to_use = ($servers[$i] =~ /biblioserver/) ? $koha_query : $simple_query;
356 #$query_to_use = $simple_query if $scan;
357 warn $simple_query if ( $scan and $DEBUG );
359 # Check if we've got a query_type defined, if so, use it
360 eval {
361 if ($query_type) {
362 if ($query_type =~ /^ccl/) {
363 $query_to_use =~ s/\:/\=/g; # change : to = last minute (FIXME)
364 $results[$i] = $zconns[$i]->search(new ZOOM::Query::CCL2RPN($query_to_use, $zconns[$i]));
365 } elsif ($query_type =~ /^cql/) {
366 $results[$i] = $zconns[$i]->search(new ZOOM::Query::CQL($query_to_use, $zconns[$i]));
367 } elsif ($query_type =~ /^pqf/) {
368 $results[$i] = $zconns[$i]->search(new ZOOM::Query::PQF($query_to_use, $zconns[$i]));
369 } else {
370 warn "Unknown query_type '$query_type'. Results undetermined.";
372 } elsif ($scan) {
373 $results[$i] = $zconns[$i]->scan( new ZOOM::Query::CCL2RPN($query_to_use, $zconns[$i]));
374 } else {
375 $results[$i] = $zconns[$i]->search(new ZOOM::Query::CCL2RPN($query_to_use, $zconns[$i]));
378 if ($@) {
379 warn "WARNING: query problem with $query_to_use " . $@;
382 # Concatenate the sort_by limits and pass them to the results object
383 # Note: sort will override rank
384 my $sort_by;
385 foreach my $sort (@sort_by) {
386 if ( $sort eq "author_az" || $sort eq "author_asc" ) {
387 $sort_by .= "1=1003 <i ";
389 elsif ( $sort eq "author_za" || $sort eq "author_dsc" ) {
390 $sort_by .= "1=1003 >i ";
392 elsif ( $sort eq "popularity_asc" ) {
393 $sort_by .= "1=9003 <i ";
395 elsif ( $sort eq "popularity_dsc" ) {
396 $sort_by .= "1=9003 >i ";
398 elsif ( $sort eq "call_number_asc" ) {
399 $sort_by .= "1=8007 <i ";
401 elsif ( $sort eq "call_number_dsc" ) {
402 $sort_by .= "1=8007 >i ";
404 elsif ( $sort eq "pubdate_asc" ) {
405 $sort_by .= "1=31 <i ";
407 elsif ( $sort eq "pubdate_dsc" ) {
408 $sort_by .= "1=31 >i ";
410 elsif ( $sort eq "acqdate_asc" ) {
411 $sort_by .= "1=32 <i ";
413 elsif ( $sort eq "acqdate_dsc" ) {
414 $sort_by .= "1=32 >i ";
416 elsif ( $sort eq "title_az" || $sort eq "title_asc" ) {
417 $sort_by .= "1=4 <i ";
419 elsif ( $sort eq "title_za" || $sort eq "title_dsc" ) {
420 $sort_by .= "1=4 >i ";
422 else {
423 warn "Ignoring unrecognized sort '$sort' requested" if $sort_by;
426 if ( $sort_by && !$scan && $results[$i] ) {
427 if ( $results[$i]->sort( "yaz", $sort_by ) < 0 ) {
428 warn "WARNING sort $sort_by failed";
431 } # finished looping through servers
433 # The big moment: asynchronously retrieve results from all servers
434 _ZOOM_event_loop(
435 \@zconns,
436 \@results,
437 sub {
438 my ( $i, $size ) = @_;
439 my $results_hash;
441 # loop through the results
442 $results_hash->{'hits'} = $size;
443 my $times;
444 if ( $offset + $results_per_page <= $size ) {
445 $times = $offset + $results_per_page;
447 else {
448 $times = $size;
451 for ( my $j = $offset ; $j < $times ; $j++ ) {
452 my $records_hash;
453 my $record;
455 ## Check if it's an index scan
456 if ($scan) {
457 my ( $term, $occ ) = $results[ $i - 1 ]->display_term($j);
459 # here we create a minimal MARC record and hand it off to the
460 # template just like a normal result ... perhaps not ideal, but
461 # it works for now
462 my $tmprecord = MARC::Record->new();
463 $tmprecord->encoding('UTF-8');
464 my $tmptitle;
465 my $tmpauthor;
467 # the minimal record in author/title (depending on MARC flavour)
468 if ( C4::Context->preference("marcflavour") eq
469 "UNIMARC" )
471 $tmptitle = MARC::Field->new(
472 '200', ' ', ' ',
473 a => $term,
474 f => $occ
476 $tmprecord->append_fields($tmptitle);
478 else {
479 $tmptitle =
480 MARC::Field->new( '245', ' ', ' ', a => $term, );
481 $tmpauthor =
482 MARC::Field->new( '100', ' ', ' ', a => $occ, );
483 $tmprecord->append_fields($tmptitle);
484 $tmprecord->append_fields($tmpauthor);
486 $results_hash->{'RECORDS'}[$j] =
487 $tmprecord->as_usmarc();
490 # not an index scan
491 else {
492 $record = $results[ $i - 1 ]->record($j)->raw();
493 # warn "RECORD $j:".$record;
494 $results_hash->{'RECORDS'}[$j] = $record;
498 $results_hashref->{ $servers[ $i - 1 ] } = $results_hash;
500 # Fill the facets while we're looping, but only for the
501 # biblioserver and not for a scan
502 if ( !$scan && $servers[ $i - 1 ] =~ /biblioserver/ ) {
503 $facets_counter = GetFacets( $results[ $i - 1 ] );
504 $facets_info = _get_facets_info( $facets );
507 # BUILD FACETS
508 if ( $servers[ $i - 1 ] =~ /biblioserver/ ) {
509 for my $link_value (
510 sort { $facets_counter->{$b} <=> $facets_counter->{$a} }
511 keys %$facets_counter
514 my $expandable;
515 my $number_of_facets;
516 my @this_facets_array;
517 for my $one_facet (
518 sort {
519 $facets_counter->{$link_value}
520 ->{$b} <=> $facets_counter->{$link_value}
521 ->{$a}
522 } keys %{ $facets_counter->{$link_value} }
525 $number_of_facets++;
526 if ( ( $number_of_facets <= 5 )
527 || ( $expanded_facet eq $link_value )
528 || ( $facets_info->{$link_value}->{'expanded'} )
532 # Sanitize the link value : parenthesis, question and exclamation mark will cause errors with CCL
533 my $facet_link_value = $one_facet;
534 $facet_link_value =~ s/[()!?¡¿؟]/ /g;
536 # fix the length that will display in the label,
537 my $facet_label_value = $one_facet;
538 my $facet_max_length = C4::Context->preference(
539 'FacetLabelTruncationLength')
540 || 20;
541 $facet_label_value =
542 substr( $one_facet, 0, $facet_max_length )
543 . "..."
544 if length($facet_label_value) >
545 $facet_max_length;
547 # if it's a branch, label by the name, not the code,
548 if ( $link_value =~ /branch/ ) {
549 if ( defined $branches
550 && ref($branches) eq "HASH"
551 && defined $branches->{$one_facet}
552 && ref( $branches->{$one_facet} ) eq
553 "HASH" )
555 $facet_label_value =
556 $branches->{$one_facet}
557 ->{'branchname'};
559 else {
560 $facet_label_value = "*";
564 # if it's a itemtype, label by the name, not the code,
565 if ( $link_value =~ /itype/ ) {
566 if ( defined $itemtypes
567 && ref($itemtypes) eq "HASH"
568 && defined $itemtypes->{$one_facet}
569 && ref( $itemtypes->{$one_facet} ) eq
570 "HASH" )
572 $facet_label_value =
573 $itemtypes->{$one_facet}
574 ->{translated_description};
578 # also, if it's a location code, use the name instead of the code
579 if ( $link_value =~ /location/ ) {
580 $facet_label_value =
581 GetKohaAuthorisedValueLib( 'LOC',
582 $one_facet, $opac );
585 # but we're down with the whole label being in the link's title.
586 push @this_facets_array,
588 facet_count =>
589 $facets_counter->{$link_value}
590 ->{$one_facet},
591 facet_label_value => $facet_label_value,
592 facet_title_value => $one_facet,
593 facet_link_value => $facet_link_value,
594 type_link_value => $link_value,
596 if ($facet_label_value);
600 # handle expanded option
601 unless ( $facets_info->{$link_value}->{'expanded'} ) {
602 $expandable = 1
603 if ( ( $number_of_facets > 5 )
604 && ( $expanded_facet ne $link_value ) );
606 push @facets_loop,
608 type_link_value => $link_value,
609 type_id => $link_value . "_id",
610 "type_label_"
611 . $facets_info->{$link_value}->{'label_value'} =>
613 facets => \@this_facets_array,
614 expandable => $expandable,
615 expand => $link_value,
617 unless (
619 $facets_info->{$link_value}->{'label_value'} =~
620 /Libraries/
622 and ( Koha::Libraries->search->count == 1 )
628 return ( undef, $results_hashref, \@facets_loop );
631 sub GetFacets {
633 my $rs = shift;
634 my $facets;
636 my $indexing_mode = C4::Context->config('zebra_bib_index_mode') // 'dom';
637 my $use_zebra_facets = C4::Context->config('use_zebra_facets') // 0;
639 if ( $indexing_mode eq 'dom' &&
640 $use_zebra_facets ) {
641 $facets = _get_facets_from_zebra( $rs );
642 } else {
643 $facets = _get_facets_from_records( $rs );
646 return $facets;
649 sub _get_facets_from_records {
651 my $rs = shift;
653 my $facets_maxrecs = C4::Context->preference('maxRecordsForFacets') // 20;
654 my $facets_config = getFacets();
655 my $facets = {};
656 my $size = $rs->size();
657 my $jmax = $size > $facets_maxrecs
658 ? $facets_maxrecs
659 : $size;
661 for ( my $j = 0 ; $j < $jmax ; $j++ ) {
663 my $marc_record = new_record_from_zebra (
664 'biblioserver',
665 $rs->record( $j )->raw()
668 if ( ! defined $marc_record ) {
669 warn "ERROR DECODING RECORD - $@: " .
670 $rs->record( $j )->raw();
671 next;
674 _get_facets_data_from_record( $marc_record, $facets_config, $facets );
677 return $facets;
680 =head2 _get_facets_data_from_record
682 C4::Search::_get_facets_data_from_record( $marc_record, $facets, $facets_counter );
684 Internal function that extracts facets information from a MARC::Record object
685 and populates $facets_counter for using in getRecords.
687 $facets is expected to be filled with C4::Koha::getFacets output (i.e. the configured
688 facets for Zebra).
690 =cut
692 sub _get_facets_data_from_record {
694 my ( $marc_record, $facets, $facets_counter ) = @_;
696 for my $facet (@$facets) {
698 my @used_datas = ();
700 foreach my $tag ( @{ $facet->{ tags } } ) {
702 # tag number is the first three digits
703 my $tag_num = substr( $tag, 0, 3 );
704 # subfields are the remainder
705 my $subfield_letters = substr( $tag, 3 );
707 my @fields = $marc_record->field( $tag_num );
708 foreach my $field (@fields) {
709 # If $field->indicator(1) eq 'z', it means it is a 'see from'
710 # field introduced because of IncludeSeeFromInSearches, so skip it
711 next if $field->indicator(1) eq 'z';
713 my $data = $field->as_string( $subfield_letters, $facet->{ sep } );
715 unless ( grep { /^\Q$data\E$/ } @used_datas ) {
716 push @used_datas, $data;
717 $facets_counter->{ $facet->{ idx } }->{ $data }++;
724 =head2 _get_facets_from_zebra
726 my $facets = _get_facets_from_zebra( $result_set )
728 Retrieves facets for a specified result set. It loops through the facets defined
729 in C4::Koha::getFacets and returns a hash with the following structure:
731 { facet_idx => {
732 facet_value => count
737 =cut
739 sub _get_facets_from_zebra {
741 my $rs = shift;
743 # save current elementSetName
744 my $elementSetName = $rs->option( 'elementSetName' );
746 my $facets_loop = getFacets();
747 my $facets_data = {};
748 # loop through defined facets and fill the facets hashref
749 foreach my $facet ( @$facets_loop ) {
751 my $idx = $facet->{ idx };
752 my $sep = $facet->{ sep };
753 my $facet_values = _get_facet_from_result_set( $idx, $rs, $sep );
754 if ( $facet_values ) {
755 # we've actually got a result
756 $facets_data->{ $idx } = $facet_values;
759 # set elementSetName to its previous value to avoid side effects
760 $rs->option( elementSetName => $elementSetName );
762 return $facets_data;
765 =head2 _get_facet_from_result_set
767 my $facet_values =
768 C4::Search::_get_facet_from_result_set( $facet_idx, $result_set, $sep )
770 Internal function that extracts facet information for a specific index ($facet_idx) and
771 returns a hash containing facet values and count:
774 $facet_value => $count ,
778 Warning: this function has the side effect of changing the elementSetName for the result
779 set. It is a helper function for the main loop, which takes care of backing it up for
780 restoring.
782 =cut
784 sub _get_facet_from_result_set {
786 my $facet_idx = shift;
787 my $rs = shift;
788 my $sep = shift;
790 my $internal_sep = '<*>';
791 my $facetMaxCount = C4::Context->preference('FacetMaxCount') // 20;
793 return if ( ! defined $facet_idx || ! defined $rs );
794 # zebra's facet element, untokenized index
795 my $facet_element = 'zebra::facet::' . $facet_idx . ':0:' . $facetMaxCount;
796 # configure zebra results for retrieving the desired facet
797 $rs->option( elementSetName => $facet_element );
798 # get the facet record from result set
799 my $facet = $rs->record( 0 )->raw;
800 # if the facet has no restuls...
801 return if !defined $facet;
802 # TODO: benchmark DOM vs. SAX performance
803 my $facet_dom = XML::LibXML->load_xml(
804 string => ($facet)
806 my @terms = $facet_dom->getElementsByTagName('term');
807 return if ! @terms;
809 my $facets = {};
810 foreach my $term ( @terms ) {
811 my $facet_value = $term->textContent;
812 $facet_value =~ s/\Q$internal_sep\E/$sep/ if defined $sep;
813 $facets->{ $facet_value } = $term->getAttribute( 'occur' );
816 return $facets;
819 =head2 _get_facets_info
821 my $facets_info = C4::Search::_get_facets_info( $facets )
823 Internal function that extracts facets information and properly builds
824 the data structure needed to render facet labels.
826 =cut
828 sub _get_facets_info {
830 my $facets = shift;
832 my $facets_info = {};
834 for my $facet ( @$facets ) {
835 $facets_info->{ $facet->{ idx } }->{ label_value } = $facet->{ label };
836 $facets_info->{ $facet->{ idx } }->{ expanded } = $facet->{ expanded };
839 return $facets_info;
842 sub pazGetRecords {
843 my (
844 $koha_query, $simple_query, $sort_by_ref, $servers_ref,
845 $results_per_page, $offset, $expanded_facet, $branches,
846 $query_type, $scan
847 ) = @_;
849 my $paz = C4::Search::PazPar2->new(C4::Context->config('pazpar2url'));
850 $paz->init();
851 $paz->search($simple_query);
852 sleep 1; # FIXME: WHY?
854 # do results
855 my $results_hashref = {};
856 my $stats = XMLin($paz->stat);
857 my $results = XMLin($paz->show($offset, $results_per_page, 'work-title:1'), forcearray => 1);
859 # for a grouped search result, the number of hits
860 # is the number of groups returned; 'bib_hits' will have
861 # the total number of bibs.
862 $results_hashref->{'biblioserver'}->{'hits'} = $results->{'merged'}->[0];
863 $results_hashref->{'biblioserver'}->{'bib_hits'} = $stats->{'hits'};
865 HIT: foreach my $hit (@{ $results->{'hit'} }) {
866 my $recid = $hit->{recid}->[0];
868 my $work_title = $hit->{'md-work-title'}->[0];
869 my $work_author;
870 if (exists $hit->{'md-work-author'}) {
871 $work_author = $hit->{'md-work-author'}->[0];
873 my $group_label = (defined $work_author) ? "$work_title / $work_author" : $work_title;
875 my $result_group = {};
876 $result_group->{'group_label'} = $group_label;
877 $result_group->{'group_merge_key'} = $recid;
879 my $count = 1;
880 if (exists $hit->{count}) {
881 $count = $hit->{count}->[0];
883 $result_group->{'group_count'} = $count;
885 for (my $i = 0; $i < $count; $i++) {
886 # FIXME -- may need to worry about diacritics here
887 my $rec = $paz->record($recid, $i);
888 push @{ $result_group->{'RECORDS'} }, $rec;
891 push @{ $results_hashref->{'biblioserver'}->{'GROUPS'} }, $result_group;
894 # pass through facets
895 my $termlist_xml = $paz->termlist('author,subject');
896 my $terms = XMLin($termlist_xml, forcearray => 1);
897 my @facets_loop = ();
898 #die Dumper($results);
899 # foreach my $list (sort keys %{ $terms->{'list'} }) {
900 # my @facets = ();
901 # foreach my $facet (sort @{ $terms->{'list'}->{$list}->{'term'} } ) {
902 # push @facets, {
903 # facet_label_value => $facet->{'name'}->[0],
904 # };
906 # push @facets_loop, ( {
907 # type_label => $list,
908 # facets => \@facets,
909 # } );
912 return ( undef, $results_hashref, \@facets_loop );
915 # TRUNCATION
916 sub _detect_truncation {
917 my ( $operand, $index ) = @_;
918 my ( @nontruncated, @righttruncated, @lefttruncated, @rightlefttruncated,
919 @regexpr );
920 $operand =~ s/^ //g;
921 my @wordlist = split( /\s/, $operand );
922 foreach my $word (@wordlist) {
923 if ( $word =~ s/^\*([^\*]+)\*$/$1/ ) {
924 push @rightlefttruncated, $word;
926 elsif ( $word =~ s/^\*([^\*]+)$/$1/ ) {
927 push @lefttruncated, $word;
929 elsif ( $word =~ s/^([^\*]+)\*$/$1/ ) {
930 push @righttruncated, $word;
932 elsif ( index( $word, "*" ) < 0 ) {
933 push @nontruncated, $word;
935 else {
936 push @regexpr, $word;
939 return (
940 \@nontruncated, \@righttruncated, \@lefttruncated,
941 \@rightlefttruncated, \@regexpr
945 # STEMMING
946 sub _build_stemmed_operand {
947 my ($operand,$lang) = @_;
948 require Lingua::Stem::Snowball ;
949 my $stemmed_operand=q{};
951 # If operand contains a digit, it is almost certainly an identifier, and should
952 # not be stemmed. This is particularly relevant for ISBNs and ISSNs, which
953 # can contain the letter "X" - for example, _build_stemmend_operand would reduce
954 # "014100018X" to "x ", which for a MARC21 database would bring up irrelevant
955 # results (e.g., "23 x 29 cm." from the 300$c). Bug 2098.
956 return $operand if $operand =~ /\d/;
958 # FIXME: the locale should be set based on the user's language and/or search choice
959 #warn "$lang";
960 # Make sure we only use the first two letters from the language code
961 $lang = lc(substr($lang, 0, 2));
962 # The language codes for the two variants of Norwegian will now be "nb" and "nn",
963 # none of which Lingua::Stem::Snowball can use, so we need to "translate" them
964 if ($lang eq 'nb' || $lang eq 'nn') {
965 $lang = 'no';
967 my $stemmer = Lingua::Stem::Snowball->new( lang => $lang,
968 encoding => "UTF-8" );
970 my @words = split( / /, $operand );
971 my @stems = $stemmer->stem(\@words);
972 for my $stem (@stems) {
973 $stemmed_operand .= "$stem";
974 $stemmed_operand .= "?"
975 unless ( $stem =~ /(and$|or$|not$)/ ) || ( length($stem) < 3 );
976 $stemmed_operand .= " ";
978 warn "STEMMED OPERAND: $stemmed_operand" if $DEBUG;
979 return $stemmed_operand;
982 # FIELD WEIGHTING
983 sub _build_weighted_query {
985 # FIELD WEIGHTING - This is largely experimental stuff. What I'm committing works
986 # pretty well but could work much better if we had a smarter query parser
987 my ( $operand, $stemmed_operand, $index ) = @_;
988 my $stemming = C4::Context->preference("QueryStemming") || 0;
989 my $weight_fields = C4::Context->preference("QueryWeightFields") || 0;
990 my $fuzzy_enabled = C4::Context->preference("QueryFuzzy") || 0;
991 $operand =~ s/"/ /g; # Bug 7518: searches with quotation marks don't work
993 my $weighted_query .= "(rk=("; # Specifies that we're applying rank
995 # Keyword, or, no index specified
996 if ( ( $index eq 'kw' ) || ( !$index ) ) {
997 $weighted_query .=
998 "Title-cover,ext,r1=\"$operand\""; # exact title-cover
999 $weighted_query .= " or ti,ext,r2=\"$operand\""; # exact title
1000 $weighted_query .= " or Title-cover,phr,r3=\"$operand\""; # phrase title
1001 $weighted_query .= " or ti,wrdl,r4=\"$operand\""; # words in title
1002 #$weighted_query .= " or any,ext,r4=$operand"; # exact any
1003 #$weighted_query .=" or kw,wrdl,r5=\"$operand\""; # word list any
1004 $weighted_query .= " or wrdl,fuzzy,r8=\"$operand\""
1005 if $fuzzy_enabled; # add fuzzy, word list
1006 $weighted_query .= " or wrdl,right-Truncation,r9=\"$stemmed_operand\""
1007 if ( $stemming and $stemmed_operand )
1008 ; # add stemming, right truncation
1009 $weighted_query .= " or wrdl,r9=\"$operand\"";
1011 # embedded sorting: 0 a-z; 1 z-a
1012 # $weighted_query .= ") or (sort1,aut=1";
1015 # Barcode searches should skip this process
1016 elsif ( $index eq 'bc' ) {
1017 $weighted_query .= "bc=\"$operand\"";
1020 # Authority-number searches should skip this process
1021 elsif ( $index eq 'an' ) {
1022 $weighted_query .= "an=\"$operand\"";
1025 # If the index already has more than one qualifier, wrap the operand
1026 # in quotes and pass it back (assumption is that the user knows what they
1027 # are doing and won't appreciate us mucking up their query
1028 elsif ( $index =~ ',' ) {
1029 $weighted_query .= " $index=\"$operand\"";
1032 #TODO: build better cases based on specific search indexes
1033 else {
1034 $weighted_query .= " $index,ext,r1=\"$operand\""; # exact index
1035 #$weighted_query .= " or (title-sort-az=0 or $index,startswithnt,st-word,r3=$operand #)";
1036 $weighted_query .= " or $index,phr,r3=\"$operand\""; # phrase index
1037 $weighted_query .= " or $index,wrdl,r6=\"$operand\""; # word list index
1038 $weighted_query .= " or $index,wrdl,fuzzy,r8=\"$operand\""
1039 if $fuzzy_enabled; # add fuzzy, word list
1040 $weighted_query .= " or $index,wrdl,rt,r9=\"$stemmed_operand\""
1041 if ( $stemming and $stemmed_operand ); # add stemming, right truncation
1044 $weighted_query .= "))"; # close rank specification
1045 return $weighted_query;
1048 =head2 getIndexes
1050 Return an array with available indexes.
1052 =cut
1054 sub getIndexes{
1055 my @indexes = (
1056 # biblio indexes
1057 'ab',
1058 'Abstract',
1059 'acqdate',
1060 'allrecords',
1061 'an',
1062 'Any',
1063 'at',
1064 'au',
1065 'aub',
1066 'aud',
1067 'audience',
1068 'auo',
1069 'aut',
1070 'Author',
1071 'Author-in-order ',
1072 'Author-personal-bibliography',
1073 'Authority-Number',
1074 'authtype',
1075 'bc',
1076 'Bib-level',
1077 'biblionumber',
1078 'bio',
1079 'biography',
1080 'callnum',
1081 'cfn',
1082 'Chronological-subdivision',
1083 'cn-bib-source',
1084 'cn-bib-sort',
1085 'cn-class',
1086 'cn-item',
1087 'cn-prefix',
1088 'cn-suffix',
1089 'cpn',
1090 'Code-institution',
1091 'Conference-name',
1092 'Conference-name-heading',
1093 'Conference-name-see',
1094 'Conference-name-seealso',
1095 'Content-type',
1096 'Control-number',
1097 'copydate',
1098 'Corporate-name',
1099 'Corporate-name-heading',
1100 'Corporate-name-see',
1101 'Corporate-name-seealso',
1102 'Country-publication',
1103 'ctype',
1104 'curriculum',
1105 'date-entered-on-file',
1106 'Date-of-acquisition',
1107 'Date-of-publication',
1108 'Date-time-last-modified',
1109 'Dewey-classification',
1110 'Dissertation-information',
1111 'diss',
1112 'dtlm',
1113 'EAN',
1114 'extent',
1115 'fic',
1116 'fiction',
1117 'Form-subdivision',
1118 'format',
1119 'Geographic-subdivision',
1120 'he',
1121 'Heading',
1122 'Heading-use-main-or-added-entry',
1123 'Heading-use-series-added-entry ',
1124 'Heading-use-subject-added-entry',
1125 'Host-item',
1126 'id-other',
1127 'Illustration-code',
1128 'Index-term-genre',
1129 'Index-term-uncontrolled',
1130 'ISBN',
1131 'isbn',
1132 'ISSN',
1133 'issn',
1134 'itemtype',
1135 'kw',
1136 'Koha-Auth-Number',
1137 'l-format',
1138 'language',
1139 'language-original',
1140 'lc-card',
1141 'LC-card-number',
1142 'lcn',
1143 'lex',
1144 'llength',
1145 'ln',
1146 'ln-audio',
1147 'ln-subtitle',
1148 'Local-classification',
1149 'Local-number',
1150 'Match-heading',
1151 'Match-heading-see-from',
1152 'Material-type',
1153 'mc-itemtype',
1154 'mc-rtype',
1155 'mus',
1156 'name',
1157 'Music-number',
1158 'Name-geographic',
1159 'Name-geographic-heading',
1160 'Name-geographic-see',
1161 'Name-geographic-seealso',
1162 'nb',
1163 'Note',
1164 'notes',
1165 'ns',
1166 'nt',
1167 'pb',
1168 'Personal-name',
1169 'Personal-name-heading',
1170 'Personal-name-see',
1171 'Personal-name-seealso',
1172 'pl',
1173 'Place-publication',
1174 'pn',
1175 'popularity',
1176 'pubdate',
1177 'Publisher',
1178 'Provider',
1179 'pv',
1180 'Record-control-number',
1181 'rcn',
1182 'Record-type',
1183 'rtype',
1184 'se',
1185 'See',
1186 'See-also',
1187 'sn',
1188 'Stock-number',
1189 'su',
1190 'Subject',
1191 'Subject-heading-thesaurus',
1192 'Subject-name-personal',
1193 'Subject-subdivision',
1194 'Summary',
1195 'Suppress',
1196 'su-geo',
1197 'su-na',
1198 'su-to',
1199 'su-ut',
1200 'ut',
1201 'Term-genre-form',
1202 'Term-genre-form-heading',
1203 'Term-genre-form-see',
1204 'Term-genre-form-seealso',
1205 'ti',
1206 'Title',
1207 'Title-cover',
1208 'Title-series',
1209 'Title-uniform',
1210 'Title-uniform-heading',
1211 'Title-uniform-see',
1212 'Title-uniform-seealso',
1213 'totalissues',
1214 'yr',
1216 # items indexes
1217 'acqsource',
1218 'barcode',
1219 'bc',
1220 'branch',
1221 'ccode',
1222 'classification-source',
1223 'cn-sort',
1224 'coded-location-qualifier',
1225 'copynumber',
1226 'damaged',
1227 'datelastborrowed',
1228 'datelastseen',
1229 'holdingbranch',
1230 'homebranch',
1231 'issues',
1232 'item',
1233 'itemnumber',
1234 'itype',
1235 'Local-classification',
1236 'location',
1237 'lost',
1238 'materials-specified',
1239 'mc-ccode',
1240 'mc-itype',
1241 'mc-loc',
1242 'notforloan',
1243 'Number-local-acquisition',
1244 'onloan',
1245 'price',
1246 'renewals',
1247 'replacementprice',
1248 'replacementpricedate',
1249 'reserves',
1250 'restricted',
1251 'stack',
1252 'stocknumber',
1253 'inv',
1254 'uri',
1255 'withdrawn',
1257 # subject related
1260 return \@indexes;
1263 =head2 _handle_exploding_index
1265 my $query = _handle_exploding_index($index, $term)
1267 Callback routine to generate the search for "exploding" indexes (i.e.
1268 those indexes which are turned into multiple or-connected searches based
1269 on authority data).
1271 =cut
1273 sub _handle_exploding_index {
1274 my ($QParser, $filter, $params, $negate, $server) = @_;
1275 my $index = $filter;
1276 my $term = join(' ', @$params);
1278 return unless ($index =~ m/(su-br|su-na|su-rl)/ && $term);
1280 my $marcflavour = C4::Context->preference('marcflavour');
1282 my $codesubfield = $marcflavour eq 'UNIMARC' ? '5' : 'w';
1283 my $wantedcodes = '';
1284 my @subqueries = ( "\@attr 1=Subject \@attr 4=1 \"$term\"");
1285 my ($error, $results, $total_hits) = SimpleSearch( "he:$term", undef, undef, [ "authorityserver" ] );
1286 foreach my $auth (@$results) {
1287 my $record = MARC::Record->new_from_usmarc($auth);
1288 my @references = $record->field('5..');
1289 if (@references) {
1290 if ($index eq 'su-br') {
1291 $wantedcodes = 'g';
1292 } elsif ($index eq 'su-na') {
1293 $wantedcodes = 'h';
1294 } elsif ($index eq 'su-rl') {
1295 $wantedcodes = '';
1297 foreach my $reference (@references) {
1298 my $codes = $reference->subfield($codesubfield);
1299 push @subqueries, '@attr 1=Subject @attr 4=1 "' . $reference->as_string('abcdefghijlmnopqrstuvxyz') . '"' if (($codes && $codes eq $wantedcodes) || !$wantedcodes);
1303 my $query = ' @or ' x (scalar(@subqueries) - 1) . join(' ', @subqueries);
1304 return $query;
1307 =head2 parseQuery
1309 ( $operators, $operands, $indexes, $limits,
1310 $sort_by, $scan, $lang ) =
1311 buildQuery ( $operators, $operands, $indexes, $limits, $sort_by, $scan, $lang);
1313 Shim function to ease the transition from buildQuery to a new QueryParser.
1314 This function is called at the beginning of buildQuery, and modifies
1315 buildQuery's input. If it can handle the input, it returns a query that
1316 buildQuery will not try to parse.
1317 =cut
1319 sub parseQuery {
1320 my ( $operators, $operands, $indexes, $limits, $sort_by, $scan, $lang) = @_;
1322 my @operators = $operators ? @$operators : ();
1323 my @indexes = $indexes ? @$indexes : ();
1324 my @operands = $operands ? @$operands : ();
1325 my @limits = $limits ? @$limits : ();
1326 my @sort_by = $sort_by ? @$sort_by : ();
1328 my $query = $operands[0];
1329 my $index;
1330 my $term;
1331 my $query_desc;
1333 my $QParser;
1334 $QParser = C4::Context->queryparser if (C4::Context->preference('UseQueryParser') || $query =~ s/^qp=//);
1335 undef $QParser if ($query =~ m/^(ccl=|pqf=|cql=)/ || grep (/\w,\w|\w=\w/, @operands, @indexes) );
1336 undef $QParser if (scalar @limits > 0);
1338 if ($QParser)
1340 $QParser->custom_data->{'QueryAutoTruncate'} = C4::Context->preference('QueryAutoTruncate');
1341 $query = '';
1342 for ( my $ii = 0 ; $ii <= @operands ; $ii++ ) {
1343 next unless $operands[$ii];
1344 $query .= $operators[ $ii - 1 ] eq 'or' ? ' || ' : ' && '
1345 if ($query);
1346 if ( $operands[$ii] =~ /^[^"]\W*[-|_\w]*:\w.*[^"]$/ ) {
1347 $query .= $operands[$ii];
1349 elsif ( $indexes[$ii] =~ m/su-/ ) {
1350 $query .= $indexes[$ii] . '(' . $operands[$ii] . ')';
1352 else {
1353 $query .=
1354 ( $indexes[$ii] ? "$indexes[$ii]:" : '' ) . $operands[$ii];
1357 foreach my $limit (@limits) {
1359 if ( scalar(@sort_by) > 0 ) {
1360 my $modifier_re =
1361 '#(' . join( '|', @{ $QParser->modifiers } ) . ')';
1362 $query =~ s/$modifier_re//g;
1363 foreach my $modifier (@sort_by) {
1364 $query .= " #$modifier";
1368 $query_desc = $query;
1369 $query_desc =~ s/\s+/ /g;
1370 if ( C4::Context->preference("QueryWeightFields") ) {
1372 $QParser->add_bib1_filter_map( 'su-br' => 'biblioserver' =>
1373 { 'target_syntax_callback' => \&_handle_exploding_index } );
1374 $QParser->add_bib1_filter_map( 'su-na' => 'biblioserver' =>
1375 { 'target_syntax_callback' => \&_handle_exploding_index } );
1376 $QParser->add_bib1_filter_map( 'su-rl' => 'biblioserver' =>
1377 { 'target_syntax_callback' => \&_handle_exploding_index } );
1378 $QParser->parse($query);
1379 $operands[0] = "pqf=" . $QParser->target_syntax('biblioserver');
1381 else {
1382 require Koha::QueryParser::Driver::PQF;
1383 my $modifier_re = '#(' . join( '|', @{Koha::QueryParser::Driver::PQF->modifiers}) . ')';
1384 s/$modifier_re//g for @operands;
1387 return ( $operators, \@operands, $indexes, $limits, $sort_by, $scan, $lang, $query_desc);
1390 =head2 buildQuery
1392 ( $error, $query,
1393 $simple_query, $query_cgi,
1394 $query_desc, $limit,
1395 $limit_cgi, $limit_desc,
1396 $query_type ) = buildQuery ( $operators, $operands, $indexes, $limits, $sort_by, $scan, $lang);
1398 Build queries and limits in CCL, CGI, Human,
1399 handle truncation, stemming, field weighting, fuzziness, etc.
1401 See verbose embedded documentation.
1404 =cut
1406 sub buildQuery {
1407 my ( $operators, $operands, $indexes, $limits, $sort_by, $scan, $lang) = @_;
1409 warn "---------\nEnter buildQuery\n---------" if $DEBUG;
1411 my $query_desc;
1412 ( $operators, $operands, $indexes, $limits, $sort_by, $scan, $lang, $query_desc) = parseQuery($operators, $operands, $indexes, $limits, $sort_by, $scan, $lang);
1414 # dereference
1415 my @operators = $operators ? @$operators : ();
1416 my @indexes = $indexes ? @$indexes : ();
1417 my @operands = $operands ? @$operands : ();
1418 my @limits = $limits ? @$limits : ();
1419 my @sort_by = $sort_by ? @$sort_by : ();
1421 my $stemming = C4::Context->preference("QueryStemming") || 0;
1422 my $auto_truncation = C4::Context->preference("QueryAutoTruncate") || 0;
1423 my $weight_fields = C4::Context->preference("QueryWeightFields") || 0;
1424 my $fuzzy_enabled = C4::Context->preference("QueryFuzzy") || 0;
1426 my $query = $operands[0];
1427 my $simple_query = $operands[0];
1429 # initialize the variables we're passing back
1430 my $query_cgi;
1431 my $query_type;
1433 my $limit;
1434 my $limit_cgi;
1435 my $limit_desc;
1437 my $cclq = 0;
1438 my $cclindexes = getIndexes();
1439 if ( $query !~ /\s*(ccl=|pqf=|cql=)/ ) {
1440 while ( !$cclq && $query =~ /(?:^|\W)([\w-]+)(,[\w-]+)*[:=]/g ) {
1441 my $dx = lc($1);
1442 $cclq = grep { lc($_) eq $dx } @$cclindexes;
1444 $query = "ccl=$query" if $cclq;
1447 # for handling ccl, cql, pqf queries in diagnostic mode, skip the rest of the steps
1448 # DIAGNOSTIC ONLY!!
1449 if ( $query =~ /^ccl=/ ) {
1450 my $q=$';
1451 # This is needed otherwise ccl= and &limit won't work together, and
1452 # this happens when selecting a subject on the opac-detail page
1453 @limits = grep {!/^$/} @limits;
1454 if ( @limits ) {
1455 $q .= ' and '.join(' and ', @limits);
1457 return ( undef, $q, $q, "q=ccl=".uri_escape_utf8($q), $q, '', '', '', 'ccl' );
1459 if ( $query =~ /^cql=/ ) {
1460 return ( undef, $', $', "q=cql=".uri_escape_utf8($'), $', '', '', '', 'cql' );
1462 if ( $query =~ /^pqf=/ ) {
1463 if ($query_desc) {
1464 $query_cgi = "q=".uri_escape_utf8($query_desc);
1465 } else {
1466 $query_desc = $';
1467 $query_cgi = "q=pqf=".uri_escape_utf8($');
1469 return ( undef, $', $', $query_cgi, $query_desc, '', '', '', 'pqf' );
1472 # pass nested queries directly
1473 # FIXME: need better handling of some of these variables in this case
1474 # Nested queries aren't handled well and this implementation is flawed and causes users to be
1475 # unable to search for anything containing () commenting out, will be rewritten for 3.4.0
1476 # if ( $query =~ /(\(|\))/ ) {
1477 # return (
1478 # undef, $query, $simple_query, $query_cgi,
1479 # $query, $limit, $limit_cgi, $limit_desc,
1480 # 'ccl'
1481 # );
1484 # Form-based queries are non-nested and fixed depth, so we can easily modify the incoming
1485 # query operands and indexes and add stemming, truncation, field weighting, etc.
1486 # Once we do so, we'll end up with a value in $query, just like if we had an
1487 # incoming $query from the user
1488 else {
1489 $query = ""
1490 ; # clear it out so we can populate properly with field-weighted, stemmed, etc. query
1491 my $previous_operand
1492 ; # a flag used to keep track if there was a previous query
1493 # if there was, we can apply the current operator
1494 # for every operand
1495 for ( my $i = 0 ; $i <= @operands ; $i++ ) {
1497 # COMBINE OPERANDS, INDEXES AND OPERATORS
1498 if ( $operands[$i] ) {
1499 $operands[$i]=~s/^\s+//;
1501 # A flag to determine whether or not to add the index to the query
1502 my $indexes_set;
1504 # If the user is sophisticated enough to specify an index, turn off field weighting, and stemming handling
1505 if ( $operands[$i] =~ /\w(:|=)/ || $scan ) {
1506 $weight_fields = 0;
1507 $stemming = 0;
1508 } else {
1509 $operands[$i] =~ s/\?/{?}/g; # need to escape question marks
1511 my $operand = $operands[$i];
1512 my $index = $indexes[$i];
1514 # Add index-specific attributes
1516 #Afaik, this 'yr' condition will only ever be met in the staff client advanced search
1517 #for "Publication date", since typing 'yr:YYYY' into the search box produces a CCL query,
1518 #which is processed higher up in this sub. Other than that, year searches are typically
1519 #handled as limits which are not processed her either.
1521 # Date of Publication
1522 if ( $index =~ /yr/ ) {
1523 #weight_fields/relevance search causes errors with date ranges
1524 #In the case of YYYY-, it will only return records with a 'yr' of YYYY (not the range)
1525 #In the case of YYYY-YYYY, it will return no results
1526 $stemming = $auto_truncation = $weight_fields = $fuzzy_enabled = 0;
1529 # Date of Acquisition
1530 elsif ( $index =~ /acqdate/ ) {
1531 #stemming and auto_truncation would have zero impact since it already is YYYY-MM-DD format
1532 #Weight_fields probably SHOULD be turned OFF, otherwise you'll get records floating to the
1533 #top of the results just because they have lots of item records matching that date.
1534 #Fuzzy actually only applies during _build_weighted_query, and is reset there anyway, so
1535 #irrelevant here
1536 $stemming = $auto_truncation = $weight_fields = $fuzzy_enabled = 0;
1538 # ISBN,ISSN,Standard Number, don't need special treatment
1539 elsif ( $index eq 'nb' || $index eq 'ns' ) {
1541 $stemming, $auto_truncation,
1542 $weight_fields, $fuzzy_enabled
1543 ) = ( 0, 0, 0, 0 );
1545 if ( $index eq 'nb' ) {
1546 if ( C4::Context->preference("SearchWithISBNVariations") ) {
1547 my @isbns = C4::Koha::GetVariationsOfISBN( $operand );
1548 $operands[$i] = $operand = '(nb=' . join(' OR nb=', @isbns) . ')';
1549 $indexes[$i] = $index = '';
1554 if(not $index){
1555 $index = 'kw';
1558 # Set default structure attribute (word list)
1559 my $struct_attr = q{};
1560 unless ( $indexes_set || !$index || $index =~ /,(st-|phr|ext|wrdl)/ || $index =~ /^(nb|ns)$/ ) {
1561 $struct_attr = ",wrdl";
1564 # Some helpful index variants
1565 my $index_plus = $index . $struct_attr . ':';
1566 my $index_plus_comma = $index . $struct_attr . ',';
1568 if ($auto_truncation){
1569 unless ( $index =~ /,(st-|phr|ext)/ ) {
1570 #FIXME only valid with LTR scripts
1571 $operand=join(" ",map{
1572 (index($_,"*")>0?"$_":"$_*")
1573 }split (/\s+/,$operand));
1574 warn $operand if $DEBUG;
1578 # Detect Truncation
1579 my $truncated_operand;
1580 my( $nontruncated, $righttruncated, $lefttruncated,
1581 $rightlefttruncated, $regexpr
1582 ) = _detect_truncation( $operand, $index );
1583 warn
1584 "TRUNCATION: NON:>@$nontruncated< RIGHT:>@$righttruncated< LEFT:>@$lefttruncated< RIGHTLEFT:>@$rightlefttruncated< REGEX:>@$regexpr<"
1585 if $DEBUG;
1587 # Apply Truncation
1588 if (
1589 scalar(@$righttruncated) + scalar(@$lefttruncated) +
1590 scalar(@$rightlefttruncated) > 0 )
1593 # Don't field weight or add the index to the query, we do it here
1594 $indexes_set = 1;
1595 undef $weight_fields;
1596 my $previous_truncation_operand;
1597 if (scalar @$nontruncated) {
1598 $truncated_operand .= "$index_plus @$nontruncated ";
1599 $previous_truncation_operand = 1;
1601 if (scalar @$righttruncated) {
1602 $truncated_operand .= "and " if $previous_truncation_operand;
1603 $truncated_operand .= $index_plus_comma . "rtrn:@$righttruncated ";
1604 $previous_truncation_operand = 1;
1606 if (scalar @$lefttruncated) {
1607 $truncated_operand .= "and " if $previous_truncation_operand;
1608 $truncated_operand .= $index_plus_comma . "ltrn:@$lefttruncated ";
1609 $previous_truncation_operand = 1;
1611 if (scalar @$rightlefttruncated) {
1612 $truncated_operand .= "and " if $previous_truncation_operand;
1613 $truncated_operand .= $index_plus_comma . "rltrn:@$rightlefttruncated ";
1614 $previous_truncation_operand = 1;
1617 $operand = $truncated_operand if $truncated_operand;
1618 warn "TRUNCATED OPERAND: >$truncated_operand<" if $DEBUG;
1620 # Handle Stemming
1621 my $stemmed_operand;
1622 $stemmed_operand = _build_stemmed_operand($operand, $lang)
1623 if $stemming;
1625 warn "STEMMED OPERAND: >$stemmed_operand<" if $DEBUG;
1627 # Handle Field Weighting
1628 my $weighted_operand;
1629 if ($weight_fields) {
1630 $weighted_operand = _build_weighted_query( $operand, $stemmed_operand, $index );
1631 $operand = $weighted_operand;
1632 $indexes_set = 1;
1635 warn "FIELD WEIGHTED OPERAND: >$weighted_operand<" if $DEBUG;
1637 ($query,$query_cgi,$query_desc,$previous_operand) = _build_initial_query({
1638 query => $query,
1639 query_cgi => $query_cgi,
1640 query_desc => $query_desc,
1641 operator => ($operators[ $i - 1 ]) ? $operators[ $i - 1 ] : '',
1642 parsed_operand => $operand,
1643 original_operand => ($operands[$i]) ? $operands[$i] : '',
1644 index => $index,
1645 index_plus => $index_plus,
1646 indexes_set => $indexes_set,
1647 previous_operand => $previous_operand,
1650 } #/if $operands
1651 } # /for
1653 warn "QUERY BEFORE LIMITS: >$query<" if $DEBUG;
1655 # add limits
1656 my %group_OR_limits;
1657 my $availability_limit;
1658 foreach my $this_limit (@limits) {
1659 next unless $this_limit;
1660 if ( $this_limit =~ /available/ ) {
1662 ## 'available' is defined as (items.onloan is NULL) and (items.itemlost = 0)
1663 ## In English:
1664 ## all records not indexed in the onloan register (zebra) and all records with a value of lost equal to 0
1665 $availability_limit .=
1666 "( ( allrecords,AlwaysMatches='' not onloan,AlwaysMatches='') and (lost,st-numeric=0) )"; #or ( allrecords,AlwaysMatches='' not lost,AlwaysMatches='')) )";
1667 $limit_cgi .= "&limit=available";
1668 $limit_desc .= "";
1671 # group_OR_limits, prefixed by mc-
1672 # OR every member of the group
1673 elsif ( $this_limit =~ /mc/ ) {
1674 my ($k,$v) = split(/:/, $this_limit,2);
1675 if ( $k !~ /mc-i(tem)?type/ ) {
1676 # in case the mc-ccode value has complicating chars like ()'s inside it we wrap in quotes
1677 $this_limit =~ tr/"//d;
1678 $this_limit = $k.":'".$v."'";
1681 $group_OR_limits{$k} .= " or " if $group_OR_limits{$k};
1682 $limit_desc .= " or " if $group_OR_limits{$k};
1683 $group_OR_limits{$k} .= "$this_limit";
1684 $limit_cgi .= "&limit=" . uri_escape_utf8($this_limit);
1685 $limit_desc .= " $this_limit";
1688 # Regular old limits
1689 else {
1690 $limit .= " and " if $limit || $query;
1691 $limit .= "$this_limit";
1692 $limit_cgi .= "&limit=" . uri_escape_utf8($this_limit);
1693 if ($this_limit =~ /^branch:(.+)/) {
1694 my $branchcode = $1;
1695 my $branchname = GetBranchName($branchcode);
1696 if (defined $branchname) {
1697 $limit_desc .= " branch:$branchname";
1698 } else {
1699 $limit_desc .= " $this_limit";
1701 } else {
1702 $limit_desc .= " $this_limit";
1706 foreach my $k (keys (%group_OR_limits)) {
1707 $limit .= " and " if ( $query || $limit );
1708 $limit .= "($group_OR_limits{$k})";
1710 if ($availability_limit) {
1711 $limit .= " and " if ( $query || $limit );
1712 $limit .= "($availability_limit)";
1715 # Normalize the query and limit strings
1716 # This is flawed , means we can't search anything with : in it
1717 # if user wants to do ccl or cql, start the query with that
1718 # $query =~ s/:/=/g;
1719 #NOTE: We use several several different regexps here as you can't have variable length lookback assertions
1720 $query =~ s/(?<=(ti|au|pb|su|an|kw|mc|nb|ns)):/=/g;
1721 $query =~ s/(?<=(wrdl)):/=/g;
1722 $query =~ s/(?<=(trn|phr)):/=/g;
1723 $query =~ s/(?<=(st-numeric)):/=/g;
1724 $query =~ s/(?<=(st-year)):/=/g;
1725 $query =~ s/(?<=(st-date-normalized)):/=/g;
1726 $limit =~ s/:/=/g;
1727 for ( $query, $query_desc, $limit, $limit_desc ) {
1728 s/ +/ /g; # remove extra spaces
1729 s/^ //g; # remove any beginning spaces
1730 s/ $//g; # remove any ending spaces
1731 s/==/=/g; # remove double == from query
1733 $query_cgi =~ s/^&//; # remove unnecessary & from beginning of the query cgi
1735 for ($query_cgi,$simple_query) {
1736 s/"//g;
1738 # append the limit to the query
1739 $query .= " " . $limit;
1741 # Warnings if DEBUG
1742 if ($DEBUG) {
1743 warn "QUERY:" . $query;
1744 warn "QUERY CGI:" . $query_cgi;
1745 warn "QUERY DESC:" . $query_desc;
1746 warn "LIMIT:" . $limit;
1747 warn "LIMIT CGI:" . $limit_cgi;
1748 warn "LIMIT DESC:" . $limit_desc;
1749 warn "---------\nLeave buildQuery\n---------";
1751 return (
1752 undef, $query, $simple_query, $query_cgi,
1753 $query_desc, $limit, $limit_cgi, $limit_desc,
1754 $query_type
1758 =head2 _build_initial_query
1760 ($query, $query_cgi, $query_desc, $previous_operand) = _build_initial_query($initial_query_params);
1762 Build a section of the initial query containing indexes, operators, and operands.
1764 =cut
1766 sub _build_initial_query {
1767 my ($params) = @_;
1769 my $operator = "";
1770 if ($params->{previous_operand}){
1771 #If there is a previous operand, add a supplied operator or the default 'and'
1772 $operator = ($params->{operator}) ? " ".($params->{operator})." " : ' and ';
1775 #NOTE: indexes_set is typically set when doing truncation or field weighting
1776 my $operand = ($params->{indexes_set}) ? $params->{parsed_operand} : $params->{index_plus}.$params->{parsed_operand};
1778 #e.g. "kw,wrdl:test"
1779 #e.g. " and kw,wrdl:test"
1780 $params->{query} .= $operator . $operand;
1782 $params->{query_cgi} .= "&op=".uri_escape_utf8($operator) if $operator;
1783 $params->{query_cgi} .= "&idx=".uri_escape_utf8($params->{index}) if $params->{index};
1784 $params->{query_cgi} .= "&q=".uri_escape_utf8($params->{original_operand}) if $params->{original_operand};
1786 #e.g. " and kw,wrdl: test"
1787 $params->{query_desc} .= $operator . $params->{index_plus} . " " . $params->{original_operand};
1789 $params->{previous_operand} = 1 unless $params->{previous_operand}; #If there is no previous operand, mark this as one
1791 return ($params->{query}, $params->{query_cgi}, $params->{query_desc}, $params->{previous_operand});
1794 =head2 searchResults
1796 my @search_results = searchResults($search_context, $searchdesc, $hits,
1797 $results_per_page, $offset, $scan,
1798 @marcresults);
1800 Format results in a form suitable for passing to the template
1802 =cut
1804 # IMO this subroutine is pretty messy still -- it's responsible for
1805 # building the HTML output for the template
1806 sub searchResults {
1807 my ( $search_context, $searchdesc, $hits, $results_per_page, $offset, $scan, $marcresults ) = @_;
1808 my $dbh = C4::Context->dbh;
1809 my @newresults;
1811 require C4::Items;
1813 $search_context = 'opac' if !$search_context || $search_context ne 'intranet';
1814 my ($is_opac, $hidelostitems);
1815 if ($search_context eq 'opac') {
1816 $hidelostitems = C4::Context->preference('hidelostitems');
1817 $is_opac = 1;
1820 #Build branchnames hash
1821 #find branchname
1822 #get branch information.....
1823 my %branches;
1824 my $bsth =$dbh->prepare("SELECT branchcode,branchname FROM branches"); # FIXME : use C4::Branch::GetBranches
1825 $bsth->execute();
1826 while ( my $bdata = $bsth->fetchrow_hashref ) {
1827 $branches{ $bdata->{'branchcode'} } = $bdata->{'branchname'};
1829 # FIXME - We build an authorised values hash here, using the default framework
1830 # though it is possible to have different authvals for different fws.
1832 my $shelflocations =GetKohaAuthorisedValues('items.location','');
1834 # get notforloan authorised value list (see $shelflocations FIXME)
1835 my $notforloan_authorised_value = GetAuthValCode('items.notforloan','');
1837 #Get itemtype hash
1838 my %itemtypes = %{ GetItemTypes() };
1840 #search item field code
1841 my ($itemtag, undef) = &GetMarcFromKohaField( "items.itemnumber", "" );
1843 ## find column names of items related to MARC
1844 my %subfieldstosearch;
1845 my @columns = Koha::Database->new()->schema()->resultset('Item')->result_source->columns;
1846 for my $column ( @columns ) {
1847 my ( $tagfield, $tagsubfield ) =
1848 &GetMarcFromKohaField( "items." . $column, "" );
1849 if ( defined $tagsubfield ) {
1850 $subfieldstosearch{$column} = $tagsubfield;
1854 # handle which records to actually retrieve
1855 my $times;
1856 if ( $hits && $offset + $results_per_page <= $hits ) {
1857 $times = $offset + $results_per_page;
1859 else {
1860 $times = $hits; # FIXME: if $hits is undefined, why do we want to equal it?
1863 my $marcflavour = C4::Context->preference("marcflavour");
1864 # We get the biblionumber position in MARC
1865 my ($bibliotag,$bibliosubf)=GetMarcFromKohaField('biblio.biblionumber','');
1867 # loop through all of the records we've retrieved
1868 for ( my $i = $offset ; $i <= $times - 1 ; $i++ ) {
1870 my $marcrecord;
1871 if ($scan) {
1872 # For Scan searches we built USMARC data
1873 $marcrecord = MARC::Record->new_from_usmarc( $marcresults->[$i]);
1874 } else {
1875 # Normal search, render from Zebra's output
1876 $marcrecord = new_record_from_zebra(
1877 'biblioserver',
1878 $marcresults->[$i]
1881 if ( ! defined $marcrecord ) {
1882 warn "ERROR DECODING RECORD - $@: " . $marcresults->[$i];
1883 next;
1887 my $fw = $scan
1888 ? undef
1889 : $bibliotag < 10
1890 ? GetFrameworkCode($marcrecord->field($bibliotag)->data)
1891 : GetFrameworkCode($marcrecord->subfield($bibliotag,$bibliosubf));
1893 SetUTF8Flag($marcrecord);
1894 my $oldbiblio = TransformMarcToKoha( $dbh, $marcrecord, $fw );
1895 $oldbiblio->{subtitle} = GetRecordValue('subtitle', $marcrecord, $fw);
1896 $oldbiblio->{result_number} = $i + 1;
1898 # add imageurl to itemtype if there is one
1899 $oldbiblio->{imageurl} = getitemtypeimagelocation( $search_context, $itemtypes{ $oldbiblio->{itemtype} }->{imageurl} );
1901 $oldbiblio->{'authorised_value_images'} = ($search_context eq 'opac' && C4::Context->preference('AuthorisedValueImages')) || ($search_context eq 'intranet' && C4::Context->preference('StaffAuthorisedValueImages')) ? C4::Items::get_authorised_value_images( C4::Biblio::get_biblio_authorised_values( $oldbiblio->{'biblionumber'}, $marcrecord ) ) : [];
1902 $oldbiblio->{normalized_upc} = GetNormalizedUPC( $marcrecord,$marcflavour);
1903 $oldbiblio->{normalized_ean} = GetNormalizedEAN( $marcrecord,$marcflavour);
1904 $oldbiblio->{normalized_oclc} = GetNormalizedOCLCNumber($marcrecord,$marcflavour);
1905 $oldbiblio->{normalized_isbn} = GetNormalizedISBN(undef,$marcrecord,$marcflavour);
1906 $oldbiblio->{content_identifier_exists} = 1 if ($oldbiblio->{normalized_isbn} or $oldbiblio->{normalized_oclc} or $oldbiblio->{normalized_ean} or $oldbiblio->{normalized_upc});
1908 # edition information, if any
1909 $oldbiblio->{edition} = $oldbiblio->{editionstatement};
1910 $oldbiblio->{description} = $itemtypes{ $oldbiblio->{itemtype} }->{translated_description};
1911 # Build summary if there is one (the summary is defined in the itemtypes table)
1912 # FIXME: is this used anywhere, I think it can be commented out? -- JF
1913 if ( $itemtypes{ $oldbiblio->{itemtype} }->{summary} ) {
1914 my $summary = $itemtypes{ $oldbiblio->{itemtype} }->{summary};
1915 my @fields = $marcrecord->fields();
1917 my $newsummary;
1918 foreach my $line ( "$summary\n" =~ /(.*)\n/g ){
1919 my $tags = {};
1920 foreach my $tag ( $line =~ /\[(\d{3}[\w|\d])\]/ ) {
1921 $tag =~ /(.{3})(.)/;
1922 if($marcrecord->field($1)){
1923 my @abc = $marcrecord->field($1)->subfield($2);
1924 $tags->{$tag} = $#abc + 1 ;
1928 # We catch how many times to repeat this line
1929 my $max = 0;
1930 foreach my $tag (keys(%$tags)){
1931 $max = $tags->{$tag} if($tags->{$tag} > $max);
1934 # we replace, and repeat each line
1935 for (my $i = 0 ; $i < $max ; $i++){
1936 my $newline = $line;
1938 foreach my $tag ( $newline =~ /\[(\d{3}[\w|\d])\]/g ) {
1939 $tag =~ /(.{3})(.)/;
1941 if($marcrecord->field($1)){
1942 my @repl = $marcrecord->field($1)->subfield($2);
1943 my $subfieldvalue = $repl[$i];
1944 $newline =~ s/\[$tag\]/$subfieldvalue/g;
1947 $newsummary .= "$newline\n";
1951 $newsummary =~ s/\[(.*?)]//g;
1952 $newsummary =~ s/\n/<br\/>/g;
1953 $oldbiblio->{summary} = $newsummary;
1956 # Pull out the items fields
1957 my @fields = $marcrecord->field($itemtag);
1958 my $marcflavor = C4::Context->preference("marcflavour");
1959 # adding linked items that belong to host records
1960 my $analyticsfield = '773';
1961 if ($marcflavor eq 'MARC21' || $marcflavor eq 'NORMARC') {
1962 $analyticsfield = '773';
1963 } elsif ($marcflavor eq 'UNIMARC') {
1964 $analyticsfield = '461';
1966 foreach my $hostfield ( $marcrecord->field($analyticsfield)) {
1967 my $hostbiblionumber = $hostfield->subfield("0");
1968 my $linkeditemnumber = $hostfield->subfield("9");
1969 if(!$hostbiblionumber eq undef){
1970 my $hostbiblio = GetMarcBiblio($hostbiblionumber, 1);
1971 my ($itemfield, undef) = GetMarcFromKohaField( 'items.itemnumber', GetFrameworkCode($hostbiblionumber) );
1972 if(!$hostbiblio eq undef){
1973 my @hostitems = $hostbiblio->field($itemfield);
1974 foreach my $hostitem (@hostitems){
1975 if ($hostitem->subfield("9") eq $linkeditemnumber){
1976 my $linkeditem =$hostitem;
1977 # append linked items if they exist
1978 if (!$linkeditem eq undef){
1979 push (@fields, $linkeditem);}
1986 # Setting item statuses for display
1987 my @available_items_loop;
1988 my @onloan_items_loop;
1989 my @other_items_loop;
1991 my $available_items;
1992 my $onloan_items;
1993 my $other_items;
1995 my $ordered_count = 0;
1996 my $available_count = 0;
1997 my $onloan_count = 0;
1998 my $longoverdue_count = 0;
1999 my $other_count = 0;
2000 my $withdrawn_count = 0;
2001 my $itemlost_count = 0;
2002 my $hideatopac_count = 0;
2003 my $itembinding_count = 0;
2004 my $itemdamaged_count = 0;
2005 my $item_in_transit_count = 0;
2006 my $can_place_holds = 0;
2007 my $item_onhold_count = 0;
2008 my $notforloan_count = 0;
2009 my $items_count = scalar(@fields);
2010 my $maxitems_pref = C4::Context->preference('maxItemsinSearchResults');
2011 my $maxitems = $maxitems_pref ? $maxitems_pref - 1 : 1;
2012 my @hiddenitems; # hidden itemnumbers based on OpacHiddenItems syspref
2014 # loop through every item
2015 foreach my $field (@fields) {
2016 my $item;
2018 # populate the items hash
2019 foreach my $code ( keys %subfieldstosearch ) {
2020 $item->{$code} = $field->subfield( $subfieldstosearch{$code} );
2022 $item->{description} = $itemtypes{ $item->{itype} }{translated_description};
2024 # OPAC hidden items
2025 if ($is_opac) {
2026 # hidden because lost
2027 if ($hidelostitems && $item->{itemlost}) {
2028 $hideatopac_count++;
2029 next;
2031 # hidden based on OpacHiddenItems syspref
2032 my @hi = C4::Items::GetHiddenItemnumbers($item);
2033 if (scalar @hi) {
2034 push @hiddenitems, @hi;
2035 $hideatopac_count++;
2036 next;
2040 my $hbranch = C4::Context->preference('StaffSearchResultsDisplayBranch');
2041 my $otherbranch = $hbranch eq 'homebranch' ? 'holdingbranch' : 'homebranch';
2043 # set item's branch name, use HomeOrHoldingBranch syspref first, fall back to the other one
2044 if ($item->{$hbranch}) {
2045 $item->{'branchname'} = $branches{$item->{$hbranch}};
2047 elsif ($item->{$otherbranch}) { # Last resort
2048 $item->{'branchname'} = $branches{$item->{$otherbranch}};
2051 my $prefix = $item->{$hbranch} . '--' . $item->{location} . $item->{itype} . $item->{itemcallnumber};
2052 # For each grouping of items (onloan, available, unavailable), we build a key to store relevant info about that item
2053 my $userenv = C4::Context->userenv;
2054 if ( $item->{onloan}
2055 && !( C4::Members::GetHideLostItemsPreference( $userenv->{'number'} ) && $item->{itemlost} ) )
2057 $onloan_count++;
2058 my $key = $prefix . $item->{onloan} . $item->{barcode};
2059 $onloan_items->{$key}->{due_date} = output_pref( { dt => dt_from_string( $item->{onloan} ), dateonly => 1 } );
2060 $onloan_items->{$key}->{count}++ if $item->{$hbranch};
2061 $onloan_items->{$key}->{branchname} = $item->{branchname};
2062 $onloan_items->{$key}->{location} = $shelflocations->{ $item->{location} };
2063 $onloan_items->{$key}->{itemcallnumber} = $item->{itemcallnumber};
2064 $onloan_items->{$key}->{description} = $item->{description};
2065 $onloan_items->{$key}->{imageurl} =
2066 getitemtypeimagelocation( $search_context, $itemtypes{ $item->{itype} }->{imageurl} );
2068 # if something's checked out and lost, mark it as 'long overdue'
2069 if ( $item->{itemlost} ) {
2070 $onloan_items->{$key}->{longoverdue}++;
2071 $longoverdue_count++;
2073 else { # can place holds as long as item isn't lost
2074 $can_place_holds = 1;
2078 # items not on loan, but still unavailable ( lost, withdrawn, damaged )
2079 else {
2081 $item->{notforloan}=1 if !$item->{notforloan} && $itemtypes{ C4::Context->preference("item-level_itypes")? $item->{itype}: $oldbiblio->{itemtype} }->{notforloan};
2083 # item is on order
2084 if ( $item->{notforloan} < 0 ) {
2085 $ordered_count++;
2086 } elsif ( $item->{notforloan} > 0 ) {
2087 $notforloan_count++;
2090 # is item in transit?
2091 my $transfertwhen = '';
2092 my ($transfertfrom, $transfertto);
2094 # is item on the reserve shelf?
2095 my $reservestatus = '';
2097 unless ($item->{withdrawn}
2098 || $item->{itemlost}
2099 || $item->{damaged}
2100 || $item->{notforloan}
2101 || ( C4::Context->preference('MaxSearchResultsItemsPerRecordStatusCheck')
2102 && $items_count > C4::Context->preference('MaxSearchResultsItemsPerRecordStatusCheck') ) ) {
2104 # A couple heuristics to limit how many times
2105 # we query the database for item transfer information, sacrificing
2106 # accuracy in some cases for speed;
2108 # 1. don't query if item has one of the other statuses
2109 # 2. don't check transit status if the bib has
2110 # more than 20 items
2112 # FIXME: to avoid having the query the database like this, and to make
2113 # the in transit status count as unavailable for search limiting,
2114 # should map transit status to record indexed in Zebra.
2116 ($transfertwhen, $transfertfrom, $transfertto) = C4::Circulation::GetTransfers($item->{itemnumber});
2117 $reservestatus = C4::Reserves::GetReserveStatus( $item->{itemnumber} );
2120 # item is withdrawn, lost, damaged, not for loan, reserved or in transit
2121 if ( $item->{withdrawn}
2122 || $item->{itemlost}
2123 || $item->{damaged}
2124 || $item->{notforloan}
2125 || $reservestatus eq 'Waiting'
2126 || ($transfertwhen ne ''))
2128 $withdrawn_count++ if $item->{withdrawn};
2129 $itemlost_count++ if $item->{itemlost};
2130 $itemdamaged_count++ if $item->{damaged};
2131 $item_in_transit_count++ if $transfertwhen ne '';
2132 $item_onhold_count++ if $reservestatus eq 'Waiting';
2133 $item->{status} = $item->{withdrawn} . "-" . $item->{itemlost} . "-" . $item->{damaged} . "-" . $item->{notforloan};
2135 # can place a hold on a item if
2136 # not lost nor withdrawn
2137 # not damaged unless AllowHoldsOnDamagedItems is true
2138 # item is either for loan or on order (notforloan < 0)
2139 $can_place_holds = 1
2140 if (
2141 !$item->{itemlost}
2142 && !$item->{withdrawn}
2143 && ( !$item->{damaged} || C4::Context->preference('AllowHoldsOnDamagedItems') )
2144 && ( !$item->{notforloan} || $item->{notforloan} < 0 )
2147 $other_count++;
2149 my $key = $prefix . $item->{status};
2150 foreach (qw(withdrawn itemlost damaged branchname itemcallnumber)) {
2151 $other_items->{$key}->{$_} = $item->{$_};
2153 $other_items->{$key}->{intransit} = ( $transfertwhen ne '' ) ? 1 : 0;
2154 $other_items->{$key}->{onhold} = ($reservestatus) ? 1 : 0;
2155 $other_items->{$key}->{notforloan} = GetAuthorisedValueDesc('','',$item->{notforloan},'','',$notforloan_authorised_value) if $notforloan_authorised_value and $item->{notforloan};
2156 $other_items->{$key}->{count}++ if $item->{$hbranch};
2157 $other_items->{$key}->{location} = $shelflocations->{ $item->{location} };
2158 $other_items->{$key}->{description} = $item->{description};
2159 $other_items->{$key}->{imageurl} = getitemtypeimagelocation( $search_context, $itemtypes{ $item->{itype} }->{imageurl} );
2161 # item is available
2162 else {
2163 $can_place_holds = 1;
2164 $available_count++;
2165 $available_items->{$prefix}->{count}++ if $item->{$hbranch};
2166 foreach (qw(branchname itemcallnumber description)) {
2167 $available_items->{$prefix}->{$_} = $item->{$_};
2169 $available_items->{$prefix}->{location} = $shelflocations->{ $item->{location} };
2170 $available_items->{$prefix}->{imageurl} = getitemtypeimagelocation( $search_context, $itemtypes{ $item->{itype} }->{imageurl} );
2173 } # notforloan, item level and biblioitem level
2175 # if all items are hidden, do not show the record
2176 if ($items_count > 0 && $hideatopac_count == $items_count) {
2177 next;
2180 my ( $availableitemscount, $onloanitemscount, $otheritemscount );
2181 for my $key ( sort keys %$onloan_items ) {
2182 (++$onloanitemscount > $maxitems) and last;
2183 push @onloan_items_loop, $onloan_items->{$key};
2185 for my $key ( sort keys %$other_items ) {
2186 (++$otheritemscount > $maxitems) and last;
2187 push @other_items_loop, $other_items->{$key};
2189 for my $key ( sort keys %$available_items ) {
2190 (++$availableitemscount > $maxitems) and last;
2191 push @available_items_loop, $available_items->{$key}
2194 # XSLT processing of some stuff
2195 my $interface = $search_context eq 'opac' ? 'OPAC' : '';
2196 if (!$scan && C4::Context->preference($interface . "XSLTResultsDisplay")) {
2197 $oldbiblio->{XSLTResultsRecord} = XSLTParse4Display($oldbiblio->{biblionumber}, $marcrecord, $interface."XSLTResultsDisplay", 1, \@hiddenitems);
2198 # the last parameter tells Koha to clean up the problematic ampersand entities that Zebra outputs
2201 # if biblio level itypes are used and itemtype is notforloan, it can't be reserved either
2202 if (!C4::Context->preference("item-level_itypes")) {
2203 if ($itemtypes{ $oldbiblio->{itemtype} }->{notforloan}) {
2204 $can_place_holds = 0;
2207 $oldbiblio->{norequests} = 1 unless $can_place_holds;
2208 $oldbiblio->{itemsplural} = 1 if $items_count > 1;
2209 $oldbiblio->{items_count} = $items_count;
2210 $oldbiblio->{available_items_loop} = \@available_items_loop;
2211 $oldbiblio->{onloan_items_loop} = \@onloan_items_loop;
2212 $oldbiblio->{other_items_loop} = \@other_items_loop;
2213 $oldbiblio->{availablecount} = $available_count;
2214 $oldbiblio->{availableplural} = 1 if $available_count > 1;
2215 $oldbiblio->{onloancount} = $onloan_count;
2216 $oldbiblio->{onloanplural} = 1 if $onloan_count > 1;
2217 $oldbiblio->{othercount} = $other_count;
2218 $oldbiblio->{otherplural} = 1 if $other_count > 1;
2219 $oldbiblio->{withdrawncount} = $withdrawn_count;
2220 $oldbiblio->{itemlostcount} = $itemlost_count;
2221 $oldbiblio->{damagedcount} = $itemdamaged_count;
2222 $oldbiblio->{intransitcount} = $item_in_transit_count;
2223 $oldbiblio->{onholdcount} = $item_onhold_count;
2224 $oldbiblio->{orderedcount} = $ordered_count;
2225 $oldbiblio->{notforloancount} = $notforloan_count;
2227 if (C4::Context->preference("AlternateHoldingsField") && $items_count == 0) {
2228 my $fieldspec = C4::Context->preference("AlternateHoldingsField");
2229 my $subfields = substr $fieldspec, 3;
2230 my $holdingsep = C4::Context->preference("AlternateHoldingsSeparator") || ' ';
2231 my @alternateholdingsinfo = ();
2232 my @holdingsfields = $marcrecord->field(substr $fieldspec, 0, 3);
2233 my $alternateholdingscount = 0;
2235 for my $field (@holdingsfields) {
2236 my %holding = ( holding => '' );
2237 my $havesubfield = 0;
2238 for my $subfield ($field->subfields()) {
2239 if ((index $subfields, $$subfield[0]) >= 0) {
2240 $holding{'holding'} .= $holdingsep if (length $holding{'holding'} > 0);
2241 $holding{'holding'} .= $$subfield[1];
2242 $havesubfield++;
2245 if ($havesubfield) {
2246 push(@alternateholdingsinfo, \%holding);
2247 $alternateholdingscount++;
2251 $oldbiblio->{'ALTERNATEHOLDINGS'} = \@alternateholdingsinfo;
2252 $oldbiblio->{'alternateholdings_count'} = $alternateholdingscount;
2255 push( @newresults, $oldbiblio );
2258 return @newresults;
2261 =head2 SearchAcquisitions
2262 Search for acquisitions
2263 =cut
2265 sub SearchAcquisitions{
2266 my ($datebegin, $dateend, $itemtypes,$criteria, $orderby) = @_;
2268 my $dbh=C4::Context->dbh;
2269 # Variable initialization
2270 my $str=qq|
2271 SELECT marcxml
2272 FROM biblio
2273 LEFT JOIN biblioitems ON biblioitems.biblionumber=biblio.biblionumber
2274 LEFT JOIN items ON items.biblionumber=biblio.biblionumber
2275 WHERE dateaccessioned BETWEEN ? AND ?
2278 my (@params,@loopcriteria);
2280 push @params, $datebegin->output("iso");
2281 push @params, $dateend->output("iso");
2283 if (scalar(@$itemtypes)>0 and $criteria ne "itemtype" ){
2284 if(C4::Context->preference("item-level_itypes")){
2285 $str .= "AND items.itype IN (?".( ',?' x scalar @$itemtypes - 1 ).") ";
2286 }else{
2287 $str .= "AND biblioitems.itemtype IN (?".( ',?' x scalar @$itemtypes - 1 ).") ";
2289 push @params, @$itemtypes;
2292 if ($criteria =~/itemtype/){
2293 if(C4::Context->preference("item-level_itypes")){
2294 $str .= "AND items.itype=? ";
2295 }else{
2296 $str .= "AND biblioitems.itemtype=? ";
2299 if(scalar(@$itemtypes) == 0){
2300 my $itypes = GetItemTypes();
2301 for my $key (keys %$itypes){
2302 push @$itemtypes, $key;
2306 @loopcriteria= @$itemtypes;
2307 }elsif ($criteria=~/itemcallnumber/){
2308 $str .= "AND (items.itemcallnumber LIKE CONCAT(?,'%')
2309 OR items.itemcallnumber is NULL
2310 OR items.itemcallnumber = '')";
2312 @loopcriteria = ("AA".."ZZ", "") unless (scalar(@loopcriteria)>0);
2313 }else {
2314 $str .= "AND biblio.title LIKE CONCAT(?,'%') ";
2315 @loopcriteria = ("A".."z") unless (scalar(@loopcriteria)>0);
2318 if ($orderby =~ /date_desc/){
2319 $str.=" ORDER BY dateaccessioned DESC";
2320 } else {
2321 $str.=" ORDER BY title";
2324 my $qdataacquisitions=$dbh->prepare($str);
2326 my @loopacquisitions;
2327 foreach my $value(@loopcriteria){
2328 push @params,$value;
2329 my %cell;
2330 $cell{"title"}=$value;
2331 $cell{"titlecode"}=$value;
2333 eval{$qdataacquisitions->execute(@params);};
2335 if ($@){ warn "recentacquisitions Error :$@";}
2336 else {
2337 my @loopdata;
2338 while (my $data=$qdataacquisitions->fetchrow_hashref){
2339 push @loopdata, {"summary"=>GetBiblioSummary( $data->{'marcxml'} ) };
2341 $cell{"loopdata"}=\@loopdata;
2343 push @loopacquisitions,\%cell if (scalar(@{$cell{loopdata}})>0);
2344 pop @params;
2346 $qdataacquisitions->finish;
2347 return \@loopacquisitions;
2350 =head2 enabled_staff_search_views
2352 %hash = enabled_staff_search_views()
2354 This function returns a hash that contains three flags obtained from the system
2355 preferences, used to determine whether a particular staff search results view
2356 is enabled.
2358 =over 2
2360 =item C<Output arg:>
2362 * $hash{can_view_MARC} is true only if the MARC view is enabled
2363 * $hash{can_view_ISBD} is true only if the ISBD view is enabled
2364 * $hash{can_view_labeledMARC} is true only if the Labeled MARC view is enabled
2366 =item C<usage in the script:>
2368 =back
2370 $template->param ( C4::Search::enabled_staff_search_views );
2372 =cut
2374 sub enabled_staff_search_views
2376 return (
2377 can_view_MARC => C4::Context->preference('viewMARC'), # 1 if the staff search allows the MARC view
2378 can_view_ISBD => C4::Context->preference('viewISBD'), # 1 if the staff search allows the ISBD view
2379 can_view_labeledMARC => C4::Context->preference('viewLabeledMARC'), # 1 if the staff search allows the Labeled MARC view
2383 =head2 z3950_search_args
2385 $arrayref = z3950_search_args($matchpoints)
2387 This function returns an array reference that contains the search parameters to be
2388 passed to the Z39.50 search script (z3950_search.pl). The array elements
2389 are hash refs whose keys are name and value, and whose values are the
2390 name of a search parameter, the value of that search parameter and the URL encoded
2391 value of that parameter.
2393 The search parameter names are lccn, isbn, issn, title, author, dewey and subject.
2395 The search parameter values are obtained from the bibliographic record whose
2396 data is in a hash reference in $matchpoints, as returned by Biblio::GetBiblioData().
2398 If $matchpoints is a scalar, it is assumed to be an unnamed query descriptor, e.g.
2399 a general purpose search argument. In this case, the returned array contains only
2400 entry: the key is 'title' and the value is derived from $matchpoints.
2402 If a search parameter value is undefined or empty, it is not included in the returned
2403 array.
2405 The returned array reference may be passed directly to the template parameters.
2407 =over 2
2409 =item C<Output arg:>
2411 * $array containing hash refs as described above
2413 =item C<usage in the script:>
2415 =back
2417 $data = Biblio::GetBiblioData($bibno);
2418 $template->param ( MYLOOP => C4::Search::z3950_search_args($data) )
2420 *OR*
2422 $template->param ( MYLOOP => C4::Search::z3950_search_args($searchscalar) )
2424 =cut
2426 sub z3950_search_args {
2427 my $bibrec = shift;
2429 my $isbn_string = ref( $bibrec ) ? $bibrec->{title} : $bibrec;
2430 my $isbn = Business::ISBN->new( $isbn_string );
2432 if (defined $isbn && $isbn->is_valid)
2434 if ( ref($bibrec) ) {
2435 $bibrec->{isbn} = $isbn_string;
2436 $bibrec->{title} = undef;
2437 } else {
2438 $bibrec = { isbn => $isbn_string };
2441 else {
2442 $bibrec = { title => $bibrec } if !ref $bibrec;
2444 my $array = [];
2445 for my $field (qw/ lccn isbn issn title author dewey subject /)
2447 push @$array, { name => $field, value => $bibrec->{$field} }
2448 if defined $bibrec->{$field};
2450 return $array;
2453 =head2 GetDistinctValues($field);
2455 C<$field> is a reference to the fields array
2457 =cut
2459 sub GetDistinctValues {
2460 my ($fieldname,$string)=@_;
2461 # returns a reference to a hash of references to branches...
2462 if ($fieldname=~/\./){
2463 my ($table,$column)=split /\./, $fieldname;
2464 my $dbh = C4::Context->dbh;
2465 warn "select DISTINCT($column) as value, count(*) as cnt from $table group by lib order by $column " if $DEBUG;
2466 my $sth = $dbh->prepare("select DISTINCT($column) as value, count(*) as cnt from $table ".($string?" where $column like \"$string%\"":"")."group by value order by $column ");
2467 $sth->execute;
2468 my $elements=$sth->fetchall_arrayref({});
2469 return $elements;
2471 else {
2472 $string||= qq("");
2473 my @servers=qw<biblioserver authorityserver>;
2474 my (@zconns,@results);
2475 for ( my $i = 0 ; $i < @servers ; $i++ ) {
2476 $zconns[$i] = C4::Context->Zconn( $servers[$i], 1 );
2477 $results[$i] =
2478 $zconns[$i]->scan(
2479 ZOOM::Query::CCL2RPN->new( qq"$fieldname $string", $zconns[$i])
2482 # The big moment: asynchronously retrieve results from all servers
2483 my @elements;
2484 _ZOOM_event_loop(
2485 \@zconns,
2486 \@results,
2487 sub {
2488 my ( $i, $size ) = @_;
2489 for ( my $j = 0 ; $j < $size ; $j++ ) {
2490 my %hashscan;
2491 @hashscan{qw(value cnt)} =
2492 $results[ $i - 1 ]->display_term($j);
2493 push @elements, \%hashscan;
2497 return \@elements;
2501 =head2 _ZOOM_event_loop
2503 _ZOOM_event_loop(\@zconns, \@results, sub {
2504 my ( $i, $size ) = @_;
2505 ....
2506 } );
2508 Processes a ZOOM event loop and passes control to a closure for
2509 processing the results, and destroying the resultsets.
2511 =cut
2513 sub _ZOOM_event_loop {
2514 my ($zconns, $results, $callback) = @_;
2515 while ( ( my $i = ZOOM::event( $zconns ) ) != 0 ) {
2516 my $ev = $zconns->[ $i - 1 ]->last_event();
2517 if ( $ev == ZOOM::Event::ZEND ) {
2518 next unless $results->[ $i - 1 ];
2519 my $size = $results->[ $i - 1 ]->size();
2520 if ( $size > 0 ) {
2521 $callback->($i, $size);
2526 foreach my $result (@$results) {
2527 $result->destroy();
2531 =head2 new_record_from_zebra
2533 Given raw data from a Zebra result set, return a MARC::Record object
2535 This helper function is needed to take into account all the involved
2536 system preferences and configuration variables to properly create the
2537 MARC::Record object.
2539 If we are using GRS-1, then the raw data we get from Zebra should be USMARC
2540 data. If we are using DOM, then it has to be MARCXML.
2542 =cut
2544 sub new_record_from_zebra {
2546 my $server = shift;
2547 my $raw_data = shift;
2548 # Set the default indexing modes
2549 my $index_mode = ( $server eq 'biblioserver' )
2550 ? C4::Context->config('zebra_bib_index_mode') // 'dom'
2551 : C4::Context->config('zebra_auth_index_mode') // 'dom';
2553 my $marc_record = eval {
2554 if ( $index_mode eq 'dom' ) {
2555 MARC::Record->new_from_xml( $raw_data, 'UTF-8' );
2556 } else {
2557 MARC::Record->new_from_usmarc( $raw_data );
2561 if ($@) {
2562 return;
2563 } else {
2564 return $marc_record;
2569 END { } # module clean-up code here (global destructor)
2572 __END__
2574 =head1 AUTHOR
2576 Koha Development Team <http://koha-community.org/>
2578 =cut