3 # This file is part of Koha.
5 # Koha is free software; you can redistribute it and/or modify it under the
6 # terms of the GNU General Public License as published by the Free Software
7 # Foundation; either version 2 of the License, or (at your option) any later
10 # Koha is distributed in the hope that it will be useful, but WITHOUT ANY
11 # WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
12 # A PARTICULAR PURPOSE. See the GNU General Public License for more details.
14 # You should have received a copy of the GNU General Public License along with
15 # Koha; if not, write to the Free Software Foundation, Inc., 59 Temple Place,
16 # Suite 330, Boston, MA 02111-1307 USA
19 #use warnings; FIXME - Bug 2505
22 use C4::Biblio; # GetMarcFromKohaField, GetBiblioData
23 use C4::Koha; # getFacets
25 use C4::Search::PazPar2;
27 use C4::Dates qw(format_date);
28 use C4::Members qw(GetHideLostItemsPreference);
31 use C4::Reserves; # GetReserveStatus
40 use vars qw($VERSION @ISA @EXPORT @EXPORT_OK %EXPORT_TAGS $DEBUG);
42 # set the version for version checking
44 $VERSION = 3.07.00.049;
45 $DEBUG = ($ENV{DEBUG}) ? 1 : 0;
50 C4::Search - Functions for searching the Koha catalog.
54 See opac/opac-search.pl or catalogue/search.pl for example of usage
58 This module provides searching functions for Koha's bibliographic databases
73 &enabled_staff_search_views
77 # make all your functions, whether exported or not;
81 ($biblionumber,$biblionumber,$title) = FindDuplicate($record);
83 This function attempts to find duplicate records using a hard-coded, fairly simplistic algorithm
89 my $dbh = C4::Context->dbh;
90 my $result = TransformMarcToKoha( $dbh, $record, '' );
95 my ( $biblionumber, $title );
97 # search duplicate on ISBN, easy and fast..
99 if ( $result->{isbn} ) {
100 $result->{isbn} =~ s/\(.*$//;
101 $result->{isbn} =~ s/\s+$//;
102 $query = "isbn:$result->{isbn}";
106 $QParser = C4::Context->queryparser if (C4::Context->preference('UseQueryParser'));
112 $titleindex = 'title|exact';
113 $authorindex = 'author|exact';
116 $titleindex = 'ti,ext';
117 $authorindex = 'au,ext';
121 $result->{title} =~ s /\\//g;
122 $result->{title} =~ s /\"//g;
123 $result->{title} =~ s /\(//g;
124 $result->{title} =~ s /\)//g;
126 # FIXME: instead of removing operators, could just do
127 # quotes around the value
128 $result->{title} =~ s/(and|or|not)//g;
129 $query = "$titleindex:\"$result->{title}\"";
130 if ( $result->{author} ) {
131 $result->{author} =~ s /\\//g;
132 $result->{author} =~ s /\"//g;
133 $result->{author} =~ s /\(//g;
134 $result->{author} =~ s /\)//g;
136 # remove valid operators
137 $result->{author} =~ s/(and|or|not)//g;
138 $query .= " $op $authorindex:\"$result->{author}\"";
142 my ( $error, $searchresults, undef ) = SimpleSearch($query); # FIXME :: hardcoded !
144 if (!defined $error) {
145 foreach my $possible_duplicate_record (@{$searchresults}) {
147 MARC::Record->new_from_usmarc($possible_duplicate_record);
148 my $result = TransformMarcToKoha( $dbh, $marcrecord, '' );
150 # FIXME :: why 2 $biblionumber ?
152 push @results, $result->{'biblionumber'};
153 push @results, $result->{'title'};
162 ( $error, $results, $total_hits ) = SimpleSearch( $query, $offset, $max_results, [@servers] );
164 This function provides a simple search API on the bibliographic catalog
170 * $query can be a simple keyword or a complete CCL query
171 * @servers is optional. Defaults to biblioserver as found in koha-conf.xml
172 * $offset - If present, represents the number of records at the beggining to omit. Defaults to 0
173 * $max_results - if present, determines the maximum number of records to fetch. undef is All. defaults to undef.
178 Returns an array consisting of three elements
179 * $error is undefined unless an error is detected
180 * $results is a reference to an array of records.
181 * $total_hits is the number of hits that would have been returned with no limit
183 If an error is returned the two other return elements are undefined. If error itself is undefined
184 the other two elements are always defined
186 =item C<usage in the script:>
190 my ( $error, $marcresults, $total_hits ) = SimpleSearch($query);
192 if (defined $error) {
193 $template->param(query_error => $error);
194 warn "error: ".$error;
195 output_html_with_http_headers $input, $cookie, $template->output;
199 my $hits = @{$marcresults};
202 for my $r ( @{$marcresults} ) {
203 my $marcrecord = MARC::File::USMARC::decode($r);
204 my $biblio = TransformMarcToKoha(C4::Context->dbh,$marcrecord,q{});
206 #build the iarray of hashs for the template.
208 title => $biblio->{'title'},
209 subtitle => $biblio->{'subtitle'},
210 biblionumber => $biblio->{'biblionumber'},
211 author => $biblio->{'author'},
212 publishercode => $biblio->{'publishercode'},
213 publicationyear => $biblio->{'publicationyear'},
218 $template->param(result=>\@results);
223 my ( $query, $offset, $max_results, $servers ) = @_;
225 return ( 'No query entered', undef, undef ) unless $query;
226 # FIXME hardcoded value. See catalog/search.pl & opac-search.pl too.
227 my @servers = defined ( $servers ) ? @$servers : ( 'biblioserver' );
235 $QParser = C4::Context->queryparser if (C4::Context->preference('UseQueryParser') && ! ($query =~ m/\w,\w|\w=\w/));
237 # Initialize & Search Zebra
238 for ( my $i = 0 ; $i < @servers ; $i++ ) {
240 $zconns[$i] = C4::Context->Zconn( $servers[$i], 1 );
243 $QParser->parse( $query );
244 $query = $QParser->target_syntax($servers[$i]);
245 $zoom_queries[$i] = new ZOOM::Query::PQF( $query, $zconns[$i]);
248 $zoom_queries[$i] = new ZOOM::Query::CCL2RPN( $query, $zconns[$i]);
250 $tmpresults[$i] = $zconns[$i]->search( $zoom_queries[$i] );
254 $zconns[$i]->errmsg() . " ("
255 . $zconns[$i]->errcode() . ") "
256 . $zconns[$i]->addinfo() . " "
257 . $zconns[$i]->diagset();
259 return ( $error, undef, undef ) if $zconns[$i]->errcode();
263 # caught a ZOOM::Exception
267 . $@->addinfo() . " "
269 warn $error." for query: $query";
270 return ( $error, undef, undef );
279 my $first_record = defined($offset) ? $offset + 1 : 1;
280 my $hits = $tmpresults[ $i - 1 ]->size();
281 $total_hits += $hits;
282 my $last_record = $hits;
283 if ( defined $max_results && $offset + $max_results < $hits ) {
284 $last_record = $offset + $max_results;
287 for my $j ( $first_record .. $last_record ) {
289 $tmpresults[ $i - 1 ]->record( $j - 1 )->raw()
291 push @{$results}, $record;
296 foreach my $zoom_query (@zoom_queries) {
297 $zoom_query->destroy();
300 return ( undef, $results, $total_hits );
305 ( undef, $results_hashref, \@facets_loop ) = getRecords (
307 $koha_query, $simple_query, $sort_by_ref, $servers_ref,
308 $results_per_page, $offset, $expanded_facet, $branches,$itemtypes,
312 The all singing, all dancing, multi-server, asynchronous, scanning,
313 searching, record nabbing, facet-building
315 See verbse embedded documentation.
321 $koha_query, $simple_query, $sort_by_ref, $servers_ref,
322 $results_per_page, $offset, $expanded_facet, $branches,
323 $itemtypes, $query_type, $scan, $opac
326 my @servers = @$servers_ref;
327 my @sort_by = @$sort_by_ref;
329 # Initialize variables for the ZOOM connection and results object
333 my $results_hashref = ();
335 # Initialize variables for the faceted results objects
336 my $facets_counter = ();
337 my $facets_info = ();
338 my $facets = getFacets();
339 my $facets_maxrecs = C4::Context->preference('maxRecordsForFacets')||20;
341 my @facets_loop; # stores the ref to array of hashes for template facets loop
343 ### LOOP THROUGH THE SERVERS
344 for ( my $i = 0 ; $i < @servers ; $i++ ) {
345 $zconns[$i] = C4::Context->Zconn( $servers[$i], 1 );
347 # perform the search, create the results objects
348 # if this is a local search, use the $koha-query, if it's a federated one, use the federated-query
349 my $query_to_use = ($servers[$i] =~ /biblioserver/) ? $koha_query : $simple_query;
351 #$query_to_use = $simple_query if $scan;
352 warn $simple_query if ( $scan and $DEBUG );
354 # Check if we've got a query_type defined, if so, use it
357 if ($query_type =~ /^ccl/) {
358 $query_to_use =~ s/\:/\=/g; # change : to = last minute (FIXME)
359 $results[$i] = $zconns[$i]->search(new ZOOM::Query::CCL2RPN($query_to_use, $zconns[$i]));
360 } elsif ($query_type =~ /^cql/) {
361 $results[$i] = $zconns[$i]->search(new ZOOM::Query::CQL($query_to_use, $zconns[$i]));
362 } elsif ($query_type =~ /^pqf/) {
363 $results[$i] = $zconns[$i]->search(new ZOOM::Query::PQF($query_to_use, $zconns[$i]));
365 warn "Unknown query_type '$query_type'. Results undetermined.";
368 $results[$i] = $zconns[$i]->scan( new ZOOM::Query::CCL2RPN($query_to_use, $zconns[$i]));
370 $results[$i] = $zconns[$i]->search(new ZOOM::Query::CCL2RPN($query_to_use, $zconns[$i]));
374 warn "WARNING: query problem with $query_to_use " . $@;
377 # Concatenate the sort_by limits and pass them to the results object
378 # Note: sort will override rank
380 foreach my $sort (@sort_by) {
381 if ( $sort eq "author_az" || $sort eq "author_asc" ) {
382 $sort_by .= "1=1003 <i ";
384 elsif ( $sort eq "author_za" || $sort eq "author_dsc" ) {
385 $sort_by .= "1=1003 >i ";
387 elsif ( $sort eq "popularity_asc" ) {
388 $sort_by .= "1=9003 <i ";
390 elsif ( $sort eq "popularity_dsc" ) {
391 $sort_by .= "1=9003 >i ";
393 elsif ( $sort eq "call_number_asc" ) {
394 $sort_by .= "1=8007 <i ";
396 elsif ( $sort eq "call_number_dsc" ) {
397 $sort_by .= "1=8007 >i ";
399 elsif ( $sort eq "pubdate_asc" ) {
400 $sort_by .= "1=31 <i ";
402 elsif ( $sort eq "pubdate_dsc" ) {
403 $sort_by .= "1=31 >i ";
405 elsif ( $sort eq "acqdate_asc" ) {
406 $sort_by .= "1=32 <i ";
408 elsif ( $sort eq "acqdate_dsc" ) {
409 $sort_by .= "1=32 >i ";
411 elsif ( $sort eq "title_az" || $sort eq "title_asc" ) {
412 $sort_by .= "1=4 <i ";
414 elsif ( $sort eq "title_za" || $sort eq "title_dsc" ) {
415 $sort_by .= "1=4 >i ";
418 warn "Ignoring unrecognized sort '$sort' requested" if $sort_by;
421 if ($sort_by && !$scan) {
422 if ( $results[$i]->sort( "yaz", $sort_by ) < 0 ) {
423 warn "WARNING sort $sort_by failed";
426 } # finished looping through servers
428 # The big moment: asynchronously retrieve results from all servers
433 my ( $i, $size ) = @_;
436 # loop through the results
437 $results_hash->{'hits'} = $size;
439 if ( $offset + $results_per_page <= $size ) {
440 $times = $offset + $results_per_page;
445 for ( my $j = $offset ; $j < $times ; $j++ ) {
449 ## Check if it's an index scan
451 my ( $term, $occ ) = $results[ $i - 1 ]->term($j);
453 # here we create a minimal MARC record and hand it off to the
454 # template just like a normal result ... perhaps not ideal, but
456 my $tmprecord = MARC::Record->new();
457 $tmprecord->encoding('UTF-8');
461 # the minimal record in author/title (depending on MARC flavour)
462 if ( C4::Context->preference("marcflavour") eq
465 $tmptitle = MARC::Field->new(
470 $tmprecord->append_fields($tmptitle);
474 MARC::Field->new( '245', ' ', ' ', a => $term, );
476 MARC::Field->new( '100', ' ', ' ', a => $occ, );
477 $tmprecord->append_fields($tmptitle);
478 $tmprecord->append_fields($tmpauthor);
480 $results_hash->{'RECORDS'}[$j] =
481 $tmprecord->as_usmarc();
486 $record = $results[ $i - 1 ]->record($j)->raw();
488 # warn "RECORD $j:".$record;
489 $results_hash->{'RECORDS'}[$j] = $record;
493 $results_hashref->{ $servers[ $i - 1 ] } = $results_hash;
495 # Fill the facets while we're looping, but only for the biblioserver and not for a scan
496 if ( !$scan && $servers[ $i - 1 ] =~ /biblioserver/ ) {
499 $size > $facets_maxrecs ? $facets_maxrecs : $size;
500 for my $facet (@$facets) {
501 for ( my $j = 0 ; $j < $jmax ; $j++ ) {
503 $results[ $i - 1 ]->record($j)->render();
505 foreach my $tag ( @{ $facet->{tags} } ) {
508 my $tag_num = substr( $tag, 0, 3 );
509 my $letters = substr( $tag, 3 );
511 '\n' . $tag_num . ' ([^z][^\n]+)';
512 $field_pattern = '\n' . $tag_num . ' ([^\n]+)'
513 if ( int($tag_num) < 10 );
515 ( $render_record =~ /$field_pattern/g );
516 foreach my $field_token (@field_tokens) {
517 my @subf = ( $field_token =~
518 /\$([a-zA-Z0-9]) ([^\$]+)/g );
520 for ( my $i = 0 ; $i < @subf ; $i += 2 ) {
521 if ( $letters =~ $subf[$i] ) {
522 my $value = $subf[ $i + 1 ];
525 push @values, $value;
528 my $data = join( $facet->{sep}, @values );
529 unless ( $data ~~ @used_datas ) {
530 $facets_counter->{ $facet->{idx} }
532 push @used_datas, $data;
537 $facets_info->{ $facet->{idx} }->{label_value} =
539 $facets_info->{ $facet->{idx} }->{expanded} =
544 # warn "connection ", $i-1, ": $size hits";
545 # warn $results[$i-1]->record(0)->render() if $size > 0;
548 if ( $servers[ $i - 1 ] =~ /biblioserver/ ) {
550 sort { $facets_counter->{$b} <=> $facets_counter->{$a} }
551 keys %$facets_counter
555 my $number_of_facets;
556 my @this_facets_array;
559 $facets_counter->{$link_value}
560 ->{$b} <=> $facets_counter->{$link_value}
562 } keys %{ $facets_counter->{$link_value} }
566 if ( ( $number_of_facets < 6 )
567 || ( $expanded_facet eq $link_value )
568 || ( $facets_info->{$link_value}->{'expanded'} )
572 # Sanitize the link value : parenthesis, question and exclamation mark will cause errors with CCL
573 my $facet_link_value = $one_facet;
574 $facet_link_value =~ s/[()!?¡¿؟]/ /g;
576 # fix the length that will display in the label,
577 my $facet_label_value = $one_facet;
578 my $facet_max_length = C4::Context->preference(
579 'FacetLabelTruncationLength')
582 substr( $one_facet, 0, $facet_max_length )
584 if length($facet_label_value) >
587 # if it's a branch, label by the name, not the code,
588 if ( $link_value =~ /branch/ ) {
589 if ( defined $branches
590 && ref($branches) eq "HASH"
591 && defined $branches->{$one_facet}
592 && ref( $branches->{$one_facet} ) eq
596 $branches->{$one_facet}
600 $facet_label_value = "*";
604 # if it's a itemtype, label by the name, not the code,
605 if ( $link_value =~ /itype/ ) {
606 if ( defined $itemtypes
607 && ref($itemtypes) eq "HASH"
608 && defined $itemtypes->{$one_facet}
609 && ref( $itemtypes->{$one_facet} ) eq
613 $itemtypes->{$one_facet}
618 # also, if it's a location code, use the name instead of the code
619 if ( $link_value =~ /location/ ) {
621 GetKohaAuthorisedValueLib( 'LOC',
625 # but we're down with the whole label being in the link's title.
626 push @this_facets_array,
629 $facets_counter->{$link_value}
631 facet_label_value => $facet_label_value,
632 facet_title_value => $one_facet,
633 facet_link_value => $facet_link_value,
634 type_link_value => $link_value,
636 if ($facet_label_value);
640 # handle expanded option
641 unless ( $facets_info->{$link_value}->{'expanded'} ) {
643 if ( ( $number_of_facets > 6 )
644 && ( $expanded_facet ne $link_value ) );
648 type_link_value => $link_value,
649 type_id => $link_value . "_id",
651 . $facets_info->{$link_value}->{'label_value'} =>
653 facets => \@this_facets_array,
654 expandable => $expandable,
655 expand => $link_value,
659 $facets_info->{$link_value}->{'label_value'} =~
662 and ( C4::Context->preference('singleBranchMode') )
668 return ( undef, $results_hashref, \@facets_loop );
673 $koha_query, $simple_query, $sort_by_ref, $servers_ref,
674 $results_per_page, $offset, $expanded_facet, $branches,
678 my $paz = C4::Search::PazPar2->new(C4::Context->config('pazpar2url'));
680 $paz->search($simple_query);
681 sleep 1; # FIXME: WHY?
684 my $results_hashref = {};
685 my $stats = XMLin($paz->stat);
686 my $results = XMLin($paz->show($offset, $results_per_page, 'work-title:1'), forcearray => 1);
688 # for a grouped search result, the number of hits
689 # is the number of groups returned; 'bib_hits' will have
690 # the total number of bibs.
691 $results_hashref->{'biblioserver'}->{'hits'} = $results->{'merged'}->[0];
692 $results_hashref->{'biblioserver'}->{'bib_hits'} = $stats->{'hits'};
694 HIT: foreach my $hit (@{ $results->{'hit'} }) {
695 my $recid = $hit->{recid}->[0];
697 my $work_title = $hit->{'md-work-title'}->[0];
699 if (exists $hit->{'md-work-author'}) {
700 $work_author = $hit->{'md-work-author'}->[0];
702 my $group_label = (defined $work_author) ? "$work_title / $work_author" : $work_title;
704 my $result_group = {};
705 $result_group->{'group_label'} = $group_label;
706 $result_group->{'group_merge_key'} = $recid;
709 if (exists $hit->{count}) {
710 $count = $hit->{count}->[0];
712 $result_group->{'group_count'} = $count;
714 for (my $i = 0; $i < $count; $i++) {
715 # FIXME -- may need to worry about diacritics here
716 my $rec = $paz->record($recid, $i);
717 push @{ $result_group->{'RECORDS'} }, $rec;
720 push @{ $results_hashref->{'biblioserver'}->{'GROUPS'} }, $result_group;
723 # pass through facets
724 my $termlist_xml = $paz->termlist('author,subject');
725 my $terms = XMLin($termlist_xml, forcearray => 1);
726 my @facets_loop = ();
727 #die Dumper($results);
728 # foreach my $list (sort keys %{ $terms->{'list'} }) {
730 # foreach my $facet (sort @{ $terms->{'list'}->{$list}->{'term'} } ) {
732 # facet_label_value => $facet->{'name'}->[0],
735 # push @facets_loop, ( {
736 # type_label => $list,
737 # facets => \@facets,
741 return ( undef, $results_hashref, \@facets_loop );
745 sub _remove_stopwords {
746 my ( $operand, $index ) = @_;
747 my @stopwords_removed;
749 # phrase and exact-qualified indexes shouldn't have stopwords removed
750 if ( $index !~ m/phr|ext/ ) {
752 # remove stopwords from operand : parse all stopwords & remove them (case insensitive)
753 # we use IsAlpha unicode definition, to deal correctly with diacritics.
754 # otherwise, a French word like "leçon" woudl be split into "le" "çon", "le"
755 # is a stopword, we'd get "çon" and wouldn't find anything...
757 foreach ( keys %{ C4::Context->stopwords } ) {
758 next if ( $_ =~ /(and|or|not)/ ); # don't remove operators
759 if ( my ($matched) = ($operand =~
760 /([^\X\p{isAlnum}]\Q$_\E[^\X\p{isAlnum}]|[^\X\p{isAlnum}]\Q$_\E$|^\Q$_\E[^\X\p{isAlnum}])/gi))
762 $operand =~ s/\Q$matched\E/ /gi;
763 push @stopwords_removed, $_;
767 return ( $operand, \@stopwords_removed );
771 sub _detect_truncation {
772 my ( $operand, $index ) = @_;
773 my ( @nontruncated, @righttruncated, @lefttruncated, @rightlefttruncated,
776 my @wordlist = split( /\s/, $operand );
777 foreach my $word (@wordlist) {
778 if ( $word =~ s/^\*([^\*]+)\*$/$1/ ) {
779 push @rightlefttruncated, $word;
781 elsif ( $word =~ s/^\*([^\*]+)$/$1/ ) {
782 push @lefttruncated, $word;
784 elsif ( $word =~ s/^([^\*]+)\*$/$1/ ) {
785 push @righttruncated, $word;
787 elsif ( index( $word, "*" ) < 0 ) {
788 push @nontruncated, $word;
791 push @regexpr, $word;
795 \@nontruncated, \@righttruncated, \@lefttruncated,
796 \@rightlefttruncated, \@regexpr
801 sub _build_stemmed_operand {
802 my ($operand,$lang) = @_;
803 require Lingua::Stem::Snowball ;
804 my $stemmed_operand=q{};
806 # If operand contains a digit, it is almost certainly an identifier, and should
807 # not be stemmed. This is particularly relevant for ISBNs and ISSNs, which
808 # can contain the letter "X" - for example, _build_stemmend_operand would reduce
809 # "014100018X" to "x ", which for a MARC21 database would bring up irrelevant
810 # results (e.g., "23 x 29 cm." from the 300$c). Bug 2098.
811 return $operand if $operand =~ /\d/;
813 # FIXME: the locale should be set based on the user's language and/or search choice
815 # Make sure we only use the first two letters from the language code
816 $lang = lc(substr($lang, 0, 2));
817 # The language codes for the two variants of Norwegian will now be "nb" and "nn",
818 # none of which Lingua::Stem::Snowball can use, so we need to "translate" them
819 if ($lang eq 'nb' || $lang eq 'nn') {
822 my $stemmer = Lingua::Stem::Snowball->new( lang => $lang,
823 encoding => "UTF-8" );
825 my @words = split( / /, $operand );
826 my @stems = $stemmer->stem(\@words);
827 for my $stem (@stems) {
828 $stemmed_operand .= "$stem";
829 $stemmed_operand .= "?"
830 unless ( $stem =~ /(and$|or$|not$)/ ) || ( length($stem) < 3 );
831 $stemmed_operand .= " ";
833 warn "STEMMED OPERAND: $stemmed_operand" if $DEBUG;
834 return $stemmed_operand;
838 sub _build_weighted_query {
840 # FIELD WEIGHTING - This is largely experimental stuff. What I'm committing works
841 # pretty well but could work much better if we had a smarter query parser
842 my ( $operand, $stemmed_operand, $index ) = @_;
843 my $stemming = C4::Context->preference("QueryStemming") || 0;
844 my $weight_fields = C4::Context->preference("QueryWeightFields") || 0;
845 my $fuzzy_enabled = C4::Context->preference("QueryFuzzy") || 0;
847 my $weighted_query .= "(rk=("; # Specifies that we're applying rank
849 # Keyword, or, no index specified
850 if ( ( $index eq 'kw' ) || ( !$index ) ) {
852 "Title-cover,ext,r1=\"$operand\""; # exact title-cover
853 $weighted_query .= " or ti,ext,r2=\"$operand\""; # exact title
854 $weighted_query .= " or Title-cover,phr,r3=\"$operand\""; # phrase title
855 #$weighted_query .= " or any,ext,r4=$operand"; # exact any
856 #$weighted_query .=" or kw,wrdl,r5=\"$operand\""; # word list any
857 $weighted_query .= " or wrdl,fuzzy,r8=\"$operand\""
858 if $fuzzy_enabled; # add fuzzy, word list
859 $weighted_query .= " or wrdl,right-Truncation,r9=\"$stemmed_operand\""
860 if ( $stemming and $stemmed_operand )
861 ; # add stemming, right truncation
862 $weighted_query .= " or wrdl,r9=\"$operand\"";
864 # embedded sorting: 0 a-z; 1 z-a
865 # $weighted_query .= ") or (sort1,aut=1";
868 # Barcode searches should skip this process
869 elsif ( $index eq 'bc' ) {
870 $weighted_query .= "bc=\"$operand\"";
873 # Authority-number searches should skip this process
874 elsif ( $index eq 'an' ) {
875 $weighted_query .= "an=\"$operand\"";
878 # If the index already has more than one qualifier, wrap the operand
879 # in quotes and pass it back (assumption is that the user knows what they
880 # are doing and won't appreciate us mucking up their query
881 elsif ( $index =~ ',' ) {
882 $weighted_query .= " $index=\"$operand\"";
885 #TODO: build better cases based on specific search indexes
887 $weighted_query .= " $index,ext,r1=\"$operand\""; # exact index
888 #$weighted_query .= " or (title-sort-az=0 or $index,startswithnt,st-word,r3=$operand #)";
889 $weighted_query .= " or $index,phr,r3=\"$operand\""; # phrase index
891 " or $index,rt,wrdl,r3=\"$operand\""; # word list index
894 $weighted_query .= "))"; # close rank specification
895 return $weighted_query;
900 Return an array with available indexes.
922 'Author-personal-bibliography',
932 'Chronological-subdivision',
942 'Conference-name-heading',
943 'Conference-name-see',
944 'Conference-name-seealso',
949 'Corporate-name-heading',
950 'Corporate-name-see',
951 'Corporate-name-seealso',
953 'date-entered-on-file',
954 'Date-of-acquisition',
955 'Date-of-publication',
956 'Dewey-classification',
963 'Geographic-subdivision',
966 'Heading-use-main-or-added-entry',
967 'Heading-use-series-added-entry ',
968 'Heading-use-subject-added-entry',
986 'Local-classification',
989 'Match-heading-see-from',
997 'Name-geographic-heading',
998 'Name-geographic-see',
999 'Name-geographic-seealso',
1007 'Personal-name-heading',
1008 'Personal-name-see',
1009 'Personal-name-seealso',
1011 'Place-publication',
1016 'Record-control-number',
1027 'Subject-heading-thesaurus',
1028 'Subject-name-personal',
1029 'Subject-subdivision',
1039 'Term-genre-form-heading',
1040 'Term-genre-form-see',
1041 'Term-genre-form-seealso',
1048 'Title-uniform-heading',
1049 'Title-uniform-see',
1050 'Title-uniform-seealso',
1060 'classification-source',
1062 'coded-location-qualifier',
1073 'Local-classification',
1076 'materials-specified',
1085 'replacementpricedate',
1100 =head2 _handle_exploding_index
1102 my $query = _handle_exploding_index($index, $term)
1104 Callback routine to generate the search for "exploding" indexes (i.e.
1105 those indexes which are turned into multiple or-connected searches based
1110 sub _handle_exploding_index {
1111 my ($QParser, $filter, $params, $negate, $server) = @_;
1112 my $index = $filter;
1113 my $term = join(' ', @$params);
1115 return unless ($index =~ m/(su-br|su-na|su-rl)/ && $term);
1117 my $marcflavour = C4::Context->preference('marcflavour');
1119 my $codesubfield = $marcflavour eq 'UNIMARC' ? '5' : 'w';
1120 my $wantedcodes = '';
1121 my @subqueries = ( "\@attr 1=Subject \@attr 4=1 \"$term\"");
1122 my ($error, $results, $total_hits) = SimpleSearch( "he:$term", undef, undef, [ "authorityserver" ] );
1123 foreach my $auth (@$results) {
1124 my $record = MARC::Record->new_from_usmarc($auth);
1125 my @references = $record->field('5..');
1127 if ($index eq 'su-br') {
1129 } elsif ($index eq 'su-na') {
1131 } elsif ($index eq 'su-rl') {
1134 foreach my $reference (@references) {
1135 my $codes = $reference->subfield($codesubfield);
1136 push @subqueries, '@attr 1=Subject @attr 4=1 "' . $reference->as_string('abcdefghijlmnopqrstuvxyz') . '"' if (($codes && $codes eq $wantedcodes) || !$wantedcodes);
1140 my $query = ' @or ' x (scalar(@subqueries) - 1) . join(' ', @subqueries);
1146 ( $operators, $operands, $indexes, $limits,
1147 $sort_by, $scan, $lang ) =
1148 buildQuery ( $operators, $operands, $indexes, $limits, $sort_by, $scan, $lang);
1150 Shim function to ease the transition from buildQuery to a new QueryParser.
1151 This function is called at the beginning of buildQuery, and modifies
1152 buildQuery's input. If it can handle the input, it returns a query that
1153 buildQuery will not try to parse.
1157 my ( $operators, $operands, $indexes, $limits, $sort_by, $scan, $lang) = @_;
1159 my @operators = $operators ? @$operators : ();
1160 my @indexes = $indexes ? @$indexes : ();
1161 my @operands = $operands ? @$operands : ();
1162 my @limits = $limits ? @$limits : ();
1163 my @sort_by = $sort_by ? @$sort_by : ();
1165 my $query = $operands[0];
1171 $QParser = C4::Context->queryparser if (C4::Context->preference('UseQueryParser') || $query =~ s/^qp=//);
1172 undef $QParser if ($query =~ m/^(ccl=|pqf=|cql=)/ || grep (/\w,\w|\w=\w/, @operands, @indexes) );
1173 undef $QParser if (scalar @limits > 0);
1178 for ( my $ii = 0 ; $ii <= @operands ; $ii++ ) {
1179 next unless $operands[$ii];
1180 $query .= $operators[ $ii - 1 ] eq 'or' ? ' || ' : ' && '
1182 if ( $indexes[$ii] =~ m/su-/ ) {
1183 $query .= $indexes[$ii] . '(' . $operands[$ii] . ')';
1187 ( $indexes[$ii] ? "$indexes[$ii]:" : '' ) . $operands[$ii];
1190 foreach my $limit (@limits) {
1192 if ( scalar(@sort_by) > 0 ) {
1194 '#(' . join( '|', @{ $QParser->modifiers } ) . ')';
1195 $query =~ s/$modifier_re//g;
1196 foreach my $modifier (@sort_by) {
1197 $query .= " #$modifier";
1201 $query_desc = $query;
1202 $query_desc =~ s/\s+/ /g;
1203 if ( C4::Context->preference("QueryWeightFields") ) {
1205 $QParser->add_bib1_filter_map( 'su-br' => 'biblioserver' =>
1206 { 'target_syntax_callback' => \&_handle_exploding_index } );
1207 $QParser->add_bib1_filter_map( 'su-na' => 'biblioserver' =>
1208 { 'target_syntax_callback' => \&_handle_exploding_index } );
1209 $QParser->add_bib1_filter_map( 'su-rl' => 'biblioserver' =>
1210 { 'target_syntax_callback' => \&_handle_exploding_index } );
1211 $QParser->parse($query);
1212 $operands[0] = "pqf=" . $QParser->target_syntax('biblioserver');
1215 require Koha::QueryParser::Driver::PQF;
1216 my $modifier_re = '#(' . join( '|', @{Koha::QueryParser::Driver::PQF->modifiers}) . ')';
1217 s/$modifier_re//g for @operands;
1220 return ( $operators, \@operands, $indexes, $limits, $sort_by, $scan, $lang, $query_desc);
1226 $simple_query, $query_cgi,
1227 $query_desc, $limit,
1228 $limit_cgi, $limit_desc,
1229 $stopwords_removed, $query_type ) = buildQuery ( $operators, $operands, $indexes, $limits, $sort_by, $scan, $lang);
1231 Build queries and limits in CCL, CGI, Human,
1232 handle truncation, stemming, field weighting, stopwords, fuzziness, etc.
1234 See verbose embedded documentation.
1240 my ( $operators, $operands, $indexes, $limits, $sort_by, $scan, $lang) = @_;
1242 warn "---------\nEnter buildQuery\n---------" if $DEBUG;
1245 ( $operators, $operands, $indexes, $limits, $sort_by, $scan, $lang, $query_desc) = parseQuery($operators, $operands, $indexes, $limits, $sort_by, $scan, $lang);
1248 my @operators = $operators ? @$operators : ();
1249 my @indexes = $indexes ? @$indexes : ();
1250 my @operands = $operands ? @$operands : ();
1251 my @limits = $limits ? @$limits : ();
1252 my @sort_by = $sort_by ? @$sort_by : ();
1254 my $stemming = C4::Context->preference("QueryStemming") || 0;
1255 my $auto_truncation = C4::Context->preference("QueryAutoTruncate") || 0;
1256 my $weight_fields = C4::Context->preference("QueryWeightFields") || 0;
1257 my $fuzzy_enabled = C4::Context->preference("QueryFuzzy") || 0;
1258 my $remove_stopwords = C4::Context->preference("QueryRemoveStopwords") || 0;
1260 # no stemming/weight/fuzzy in NoZebra
1261 if ( C4::Context->preference("NoZebra") ) {
1265 $auto_truncation = 0;
1268 my $query = $operands[0];
1269 my $simple_query = $operands[0];
1271 # initialize the variables we're passing back
1279 my $stopwords_removed; # flag to determine if stopwords have been removed
1282 my $cclindexes = getIndexes();
1283 if ( $query !~ /\s*ccl=/ ) {
1284 while ( !$cclq && $query =~ /(?:^|\W)([\w-]+)(,[\w-]+)*[:=]/g ) {
1286 $cclq = grep { lc($_) eq $dx } @$cclindexes;
1288 $query = "ccl=$query" if $cclq;
1291 # for handling ccl, cql, pqf queries in diagnostic mode, skip the rest of the steps
1293 if ( $query =~ /^ccl=/ ) {
1295 # This is needed otherwise ccl= and &limit won't work together, and
1296 # this happens when selecting a subject on the opac-detail page
1297 @limits = grep {!/^$/} @limits;
1299 $q .= ' and '.join(' and ', @limits);
1301 return ( undef, $q, $q, "q=ccl=".uri_escape($q), $q, '', '', '', '', 'ccl' );
1303 if ( $query =~ /^cql=/ ) {
1304 return ( undef, $', $', "q=cql=".uri_escape($'), $', '', '', '', '', 'cql' );
1306 if ( $query =~ /^pqf=/ ) {
1308 $query_cgi = "q=".uri_escape($query_desc);
1311 $query_cgi = "q=pqf=".uri_escape($');
1313 return ( undef, $', $', $query_cgi, $query_desc, '', '', '', '', 'pqf' );
1316 # pass nested queries directly
1317 # FIXME: need better handling of some of these variables in this case
1318 # Nested queries aren't handled well and this implementation is flawed and causes users to be
1319 # unable to search for anything containing () commenting out, will be rewritten for 3.4.0
1320 # if ( $query =~ /(\(|\))/ ) {
1322 # undef, $query, $simple_query, $query_cgi,
1323 # $query, $limit, $limit_cgi, $limit_desc,
1324 # $stopwords_removed, 'ccl'
1328 # Form-based queries are non-nested and fixed depth, so we can easily modify the incoming
1329 # query operands and indexes and add stemming, truncation, field weighting, etc.
1330 # Once we do so, we'll end up with a value in $query, just like if we had an
1331 # incoming $query from the user
1334 ; # clear it out so we can populate properly with field-weighted, stemmed, etc. query
1335 my $previous_operand
1336 ; # a flag used to keep track if there was a previous query
1337 # if there was, we can apply the current operator
1339 for ( my $i = 0 ; $i <= @operands ; $i++ ) {
1341 # COMBINE OPERANDS, INDEXES AND OPERATORS
1342 if ( $operands[$i] ) {
1343 $operands[$i]=~s/^\s+//;
1345 # A flag to determine whether or not to add the index to the query
1348 # If the user is sophisticated enough to specify an index, turn off field weighting, stemming, and stopword handling
1349 if ( $operands[$i] =~ /\w(:|=)/ || $scan ) {
1352 $remove_stopwords = 0;
1354 $operands[$i] =~ s/\?/{?}/g; # need to escape question marks
1356 my $operand = $operands[$i];
1357 my $index = $indexes[$i];
1359 # Add index-specific attributes
1360 # Date of Publication
1361 if ( $index eq 'yr' ) {
1362 $index .= ",st-numeric";
1364 $stemming = $auto_truncation = $weight_fields = $fuzzy_enabled = $remove_stopwords = 0;
1367 # Date of Acquisition
1368 elsif ( $index eq 'acqdate' ) {
1369 $index .= ",st-date-normalized";
1371 $stemming = $auto_truncation = $weight_fields = $fuzzy_enabled = $remove_stopwords = 0;
1373 # ISBN,ISSN,Standard Number, don't need special treatment
1374 elsif ( $index eq 'nb' || $index eq 'ns' ) {
1376 $stemming, $auto_truncation,
1377 $weight_fields, $fuzzy_enabled,
1379 ) = ( 0, 0, 0, 0, 0 );
1387 # Set default structure attribute (word list)
1388 my $struct_attr = q{};
1389 unless ( $indexes_set || !$index || $index =~ /(st-|phr|ext|wrdl|nb|ns)/ ) {
1390 $struct_attr = ",wrdl";
1393 # Some helpful index variants
1394 my $index_plus = $index . $struct_attr . ':';
1395 my $index_plus_comma = $index . $struct_attr . ',';
1398 if ($remove_stopwords) {
1399 ( $operand, $stopwords_removed ) =
1400 _remove_stopwords( $operand, $index );
1401 warn "OPERAND w/out STOPWORDS: >$operand<" if $DEBUG;
1402 warn "REMOVED STOPWORDS: @$stopwords_removed"
1403 if ( $stopwords_removed && $DEBUG );
1406 if ($auto_truncation){
1407 unless ( $index =~ /(st-|phr|ext)/ ) {
1408 #FIXME only valid with LTR scripts
1409 $operand=join(" ",map{
1410 (index($_,"*")>0?"$_":"$_*")
1411 }split (/\s+/,$operand));
1412 warn $operand if $DEBUG;
1417 my $truncated_operand;
1418 my( $nontruncated, $righttruncated, $lefttruncated,
1419 $rightlefttruncated, $regexpr
1420 ) = _detect_truncation( $operand, $index );
1422 "TRUNCATION: NON:>@$nontruncated< RIGHT:>@$righttruncated< LEFT:>@$lefttruncated< RIGHTLEFT:>@$rightlefttruncated< REGEX:>@$regexpr<"
1427 scalar(@$righttruncated) + scalar(@$lefttruncated) +
1428 scalar(@$rightlefttruncated) > 0 )
1431 # Don't field weight or add the index to the query, we do it here
1433 undef $weight_fields;
1434 my $previous_truncation_operand;
1435 if (scalar @$nontruncated) {
1436 $truncated_operand .= "$index_plus @$nontruncated ";
1437 $previous_truncation_operand = 1;
1439 if (scalar @$righttruncated) {
1440 $truncated_operand .= "and " if $previous_truncation_operand;
1441 $truncated_operand .= $index_plus_comma . "rtrn:@$righttruncated ";
1442 $previous_truncation_operand = 1;
1444 if (scalar @$lefttruncated) {
1445 $truncated_operand .= "and " if $previous_truncation_operand;
1446 $truncated_operand .= $index_plus_comma . "ltrn:@$lefttruncated ";
1447 $previous_truncation_operand = 1;
1449 if (scalar @$rightlefttruncated) {
1450 $truncated_operand .= "and " if $previous_truncation_operand;
1451 $truncated_operand .= $index_plus_comma . "rltrn:@$rightlefttruncated ";
1452 $previous_truncation_operand = 1;
1455 $operand = $truncated_operand if $truncated_operand;
1456 warn "TRUNCATED OPERAND: >$truncated_operand<" if $DEBUG;
1459 my $stemmed_operand;
1460 $stemmed_operand = _build_stemmed_operand($operand, $lang)
1463 warn "STEMMED OPERAND: >$stemmed_operand<" if $DEBUG;
1465 # Handle Field Weighting
1466 my $weighted_operand;
1467 if ($weight_fields) {
1468 $weighted_operand = _build_weighted_query( $operand, $stemmed_operand, $index );
1469 $operand = $weighted_operand;
1473 warn "FIELD WEIGHTED OPERAND: >$weighted_operand<" if $DEBUG;
1475 # If there's a previous operand, we need to add an operator
1476 if ($previous_operand) {
1478 # User-specified operator
1479 if ( $operators[ $i - 1 ] ) {
1480 $query .= " $operators[$i-1] ";
1481 $query .= " $index_plus " unless $indexes_set;
1482 $query .= " $operand";
1483 $query_cgi .= "&op=".uri_escape($operators[$i-1]);
1484 $query_cgi .= "&idx=".uri_escape($index) if $index;
1485 $query_cgi .= "&q=".uri_escape($operands[$i]) if $operands[$i];
1487 " $operators[$i-1] $index_plus $operands[$i]";
1490 # Default operator is and
1493 $query .= "$index_plus " unless $indexes_set;
1494 $query .= "$operand";
1495 $query_cgi .= "&op=and&idx=".uri_escape($index) if $index;
1496 $query_cgi .= "&q=".uri_escape($operands[$i]) if $operands[$i];
1497 $query_desc .= " and $index_plus $operands[$i]";
1501 # There isn't a pervious operand, don't need an operator
1504 # Field-weighted queries already have indexes set
1505 $query .= " $index_plus " unless $indexes_set;
1507 $query_desc .= " $index_plus $operands[$i]";
1508 $query_cgi .= "&idx=".uri_escape($index) if $index;
1509 $query_cgi .= "&q=".uri_escape($operands[$i]) if $operands[$i];
1510 $previous_operand = 1;
1515 warn "QUERY BEFORE LIMITS: >$query<" if $DEBUG;
1518 my %group_OR_limits;
1519 my $availability_limit;
1520 foreach my $this_limit (@limits) {
1521 next unless $this_limit;
1522 if ( $this_limit =~ /available/ ) {
1524 ## 'available' is defined as (items.onloan is NULL) and (items.itemlost = 0)
1526 ## all records not indexed in the onloan register (zebra) and all records with a value of lost equal to 0
1527 $availability_limit .=
1528 "( ( allrecords,AlwaysMatches='' not onloan,AlwaysMatches='') and (lost,st-numeric=0) )"; #or ( allrecords,AlwaysMatches='' not lost,AlwaysMatches='')) )";
1529 $limit_cgi .= "&limit=available";
1533 # group_OR_limits, prefixed by mc-
1534 # OR every member of the group
1535 elsif ( $this_limit =~ /mc/ ) {
1536 my ($k,$v) = split(/:/, $this_limit,2);
1537 if ( $k !~ /mc-i(tem)?type/ ) {
1538 # in case the mc-ccode value has complicating chars like ()'s inside it we wrap in quotes
1539 $this_limit =~ tr/"//d;
1540 $this_limit = $k.":\"".$v."\"";
1543 $group_OR_limits{$k} .= " or " if $group_OR_limits{$k};
1544 $limit_desc .= " or " if $group_OR_limits{$k};
1545 $group_OR_limits{$k} .= "$this_limit";
1546 $limit_cgi .= "&limit=$this_limit";
1547 $limit_desc .= " $this_limit";
1550 # Regular old limits
1552 $limit .= " and " if $limit || $query;
1553 $limit .= "$this_limit";
1554 $limit_cgi .= "&limit=$this_limit";
1555 if ($this_limit =~ /^branch:(.+)/) {
1556 my $branchcode = $1;
1557 my $branchname = GetBranchName($branchcode);
1558 if (defined $branchname) {
1559 $limit_desc .= " branch:$branchname";
1561 $limit_desc .= " $this_limit";
1564 $limit_desc .= " $this_limit";
1568 foreach my $k (keys (%group_OR_limits)) {
1569 $limit .= " and " if ( $query || $limit );
1570 $limit .= "($group_OR_limits{$k})";
1572 if ($availability_limit) {
1573 $limit .= " and " if ( $query || $limit );
1574 $limit .= "($availability_limit)";
1577 # Normalize the query and limit strings
1578 # This is flawed , means we can't search anything with : in it
1579 # if user wants to do ccl or cql, start the query with that
1580 # $query =~ s/:/=/g;
1581 $query =~ s/(?<=(ti|au|pb|su|an|kw|mc|nb|ns)):/=/g;
1582 $query =~ s/(?<=(wrdl)):/=/g;
1583 $query =~ s/(?<=(trn|phr)):/=/g;
1585 for ( $query, $query_desc, $limit, $limit_desc ) {
1586 s/ +/ /g; # remove extra spaces
1587 s/^ //g; # remove any beginning spaces
1588 s/ $//g; # remove any ending spaces
1589 s/==/=/g; # remove double == from query
1591 $query_cgi =~ s/^&//; # remove unnecessary & from beginning of the query cgi
1593 for ($query_cgi,$simple_query) {
1596 # append the limit to the query
1597 $query .= " " . $limit;
1601 warn "QUERY:" . $query;
1602 warn "QUERY CGI:" . $query_cgi;
1603 warn "QUERY DESC:" . $query_desc;
1604 warn "LIMIT:" . $limit;
1605 warn "LIMIT CGI:" . $limit_cgi;
1606 warn "LIMIT DESC:" . $limit_desc;
1607 warn "---------\nLeave buildQuery\n---------";
1610 undef, $query, $simple_query, $query_cgi,
1611 $query_desc, $limit, $limit_cgi, $limit_desc,
1612 $stopwords_removed, $query_type
1616 =head2 searchResults
1618 my @search_results = searchResults($search_context, $searchdesc, $hits,
1619 $results_per_page, $offset, $scan,
1622 Format results in a form suitable for passing to the template
1626 # IMO this subroutine is pretty messy still -- it's responsible for
1627 # building the HTML output for the template
1629 my ( $search_context, $searchdesc, $hits, $results_per_page, $offset, $scan, $marcresults ) = @_;
1630 my $dbh = C4::Context->dbh;
1635 $search_context = 'opac' if !$search_context || $search_context ne 'intranet';
1636 my ($is_opac, $hidelostitems);
1637 if ($search_context eq 'opac') {
1638 $hidelostitems = C4::Context->preference('hidelostitems');
1642 #Build branchnames hash
1644 #get branch information.....
1646 my $bsth =$dbh->prepare("SELECT branchcode,branchname FROM branches"); # FIXME : use C4::Branch::GetBranches
1648 while ( my $bdata = $bsth->fetchrow_hashref ) {
1649 $branches{ $bdata->{'branchcode'} } = $bdata->{'branchname'};
1651 # FIXME - We build an authorised values hash here, using the default framework
1652 # though it is possible to have different authvals for different fws.
1654 my $shelflocations =GetKohaAuthorisedValues('items.location','');
1656 # get notforloan authorised value list (see $shelflocations FIXME)
1657 my $notforloan_authorised_value = GetAuthValCode('items.notforloan','');
1659 #Build itemtype hash
1660 #find itemtype & itemtype image
1664 "SELECT itemtype,description,imageurl,summary,notforloan FROM itemtypes"
1667 while ( my $bdata = $bsth->fetchrow_hashref ) {
1668 foreach (qw(description imageurl summary notforloan)) {
1669 $itemtypes{ $bdata->{'itemtype'} }->{$_} = $bdata->{$_};
1673 #search item field code
1674 my ($itemtag, undef) = &GetMarcFromKohaField( "items.itemnumber", "" );
1676 ## find column names of items related to MARC
1677 my $sth2 = $dbh->prepare("SHOW COLUMNS FROM items");
1679 my %subfieldstosearch;
1680 while ( ( my $column ) = $sth2->fetchrow ) {
1681 my ( $tagfield, $tagsubfield ) =
1682 &GetMarcFromKohaField( "items." . $column, "" );
1683 $subfieldstosearch{$column} = $tagsubfield;
1686 # handle which records to actually retrieve
1688 if ( $hits && $offset + $results_per_page <= $hits ) {
1689 $times = $offset + $results_per_page;
1692 $times = $hits; # FIXME: if $hits is undefined, why do we want to equal it?
1695 my $marcflavour = C4::Context->preference("marcflavour");
1696 # We get the biblionumber position in MARC
1697 my ($bibliotag,$bibliosubf)=GetMarcFromKohaField('biblio.biblionumber','');
1699 # loop through all of the records we've retrieved
1700 for ( my $i = $offset ; $i <= $times - 1 ; $i++ ) {
1701 my $marcrecord = MARC::File::USMARC::decode( $marcresults->[$i] );
1705 ? GetFrameworkCode($marcrecord->field($bibliotag)->data)
1706 : GetFrameworkCode($marcrecord->subfield($bibliotag,$bibliosubf));
1707 my $oldbiblio = TransformMarcToKoha( $dbh, $marcrecord, $fw );
1708 $oldbiblio->{subtitle} = GetRecordValue('subtitle', $marcrecord, $fw);
1709 $oldbiblio->{result_number} = $i + 1;
1711 # add imageurl to itemtype if there is one
1712 $oldbiblio->{imageurl} = getitemtypeimagelocation( $search_context, $itemtypes{ $oldbiblio->{itemtype} }->{imageurl} );
1714 $oldbiblio->{'authorised_value_images'} = ($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 ) ) : [];
1715 $oldbiblio->{normalized_upc} = GetNormalizedUPC( $marcrecord,$marcflavour);
1716 $oldbiblio->{normalized_ean} = GetNormalizedEAN( $marcrecord,$marcflavour);
1717 $oldbiblio->{normalized_oclc} = GetNormalizedOCLCNumber($marcrecord,$marcflavour);
1718 $oldbiblio->{normalized_isbn} = GetNormalizedISBN(undef,$marcrecord,$marcflavour);
1719 $oldbiblio->{content_identifier_exists} = 1 if ($oldbiblio->{normalized_isbn} or $oldbiblio->{normalized_oclc} or $oldbiblio->{normalized_ean} or $oldbiblio->{normalized_upc});
1721 # edition information, if any
1722 $oldbiblio->{edition} = $oldbiblio->{editionstatement};
1723 $oldbiblio->{description} = $itemtypes{ $oldbiblio->{itemtype} }->{description};
1724 # Build summary if there is one (the summary is defined in the itemtypes table)
1725 # FIXME: is this used anywhere, I think it can be commented out? -- JF
1726 if ( $itemtypes{ $oldbiblio->{itemtype} }->{summary} ) {
1727 my $summary = $itemtypes{ $oldbiblio->{itemtype} }->{summary};
1728 my @fields = $marcrecord->fields();
1731 foreach my $line ( "$summary\n" =~ /(.*)\n/g ){
1733 foreach my $tag ( $line =~ /\[(\d{3}[\w|\d])\]/ ) {
1734 $tag =~ /(.{3})(.)/;
1735 if($marcrecord->field($1)){
1736 my @abc = $marcrecord->field($1)->subfield($2);
1737 $tags->{$tag} = $#abc + 1 ;
1741 # We catch how many times to repeat this line
1743 foreach my $tag (keys(%$tags)){
1744 $max = $tags->{$tag} if($tags->{$tag} > $max);
1747 # we replace, and repeat each line
1748 for (my $i = 0 ; $i < $max ; $i++){
1749 my $newline = $line;
1751 foreach my $tag ( $newline =~ /\[(\d{3}[\w|\d])\]/g ) {
1752 $tag =~ /(.{3})(.)/;
1754 if($marcrecord->field($1)){
1755 my @repl = $marcrecord->field($1)->subfield($2);
1756 my $subfieldvalue = $repl[$i];
1758 if (! utf8::is_utf8($subfieldvalue)) {
1759 utf8::decode($subfieldvalue);
1762 $newline =~ s/\[$tag\]/$subfieldvalue/g;
1765 $newsummary .= "$newline\n";
1769 $newsummary =~ s/\[(.*?)]//g;
1770 $newsummary =~ s/\n/<br\/>/g;
1771 $oldbiblio->{summary} = $newsummary;
1774 # Pull out the items fields
1775 my @fields = $marcrecord->field($itemtag);
1776 my $marcflavor = C4::Context->preference("marcflavour");
1777 # adding linked items that belong to host records
1778 my $analyticsfield = '773';
1779 if ($marcflavor eq 'MARC21' || $marcflavor eq 'NORMARC') {
1780 $analyticsfield = '773';
1781 } elsif ($marcflavor eq 'UNIMARC') {
1782 $analyticsfield = '461';
1784 foreach my $hostfield ( $marcrecord->field($analyticsfield)) {
1785 my $hostbiblionumber = $hostfield->subfield("0");
1786 my $linkeditemnumber = $hostfield->subfield("9");
1787 if(!$hostbiblionumber eq undef){
1788 my $hostbiblio = GetMarcBiblio($hostbiblionumber, 1);
1789 my ($itemfield, undef) = GetMarcFromKohaField( 'items.itemnumber', GetFrameworkCode($hostbiblionumber) );
1790 if(!$hostbiblio eq undef){
1791 my @hostitems = $hostbiblio->field($itemfield);
1792 foreach my $hostitem (@hostitems){
1793 if ($hostitem->subfield("9") eq $linkeditemnumber){
1794 my $linkeditem =$hostitem;
1795 # append linked items if they exist
1796 if (!$linkeditem eq undef){
1797 push (@fields, $linkeditem);}
1804 # Setting item statuses for display
1805 my @available_items_loop;
1806 my @onloan_items_loop;
1807 my @other_items_loop;
1809 my $available_items;
1813 my $ordered_count = 0;
1814 my $available_count = 0;
1815 my $onloan_count = 0;
1816 my $longoverdue_count = 0;
1817 my $other_count = 0;
1818 my $wthdrawn_count = 0;
1819 my $itemlost_count = 0;
1820 my $hideatopac_count = 0;
1821 my $itembinding_count = 0;
1822 my $itemdamaged_count = 0;
1823 my $item_in_transit_count = 0;
1824 my $can_place_holds = 0;
1825 my $item_onhold_count = 0;
1826 my $items_count = scalar(@fields);
1827 my $maxitems_pref = C4::Context->preference('maxItemsinSearchResults');
1828 my $maxitems = $maxitems_pref ? $maxitems_pref - 1 : 1;
1829 my @hiddenitems; # hidden itemnumbers based on OpacHiddenItems syspref
1831 # loop through every item
1832 foreach my $field (@fields) {
1835 # populate the items hash
1836 foreach my $code ( keys %subfieldstosearch ) {
1837 $item->{$code} = $field->subfield( $subfieldstosearch{$code} );
1839 $item->{description} = $itemtypes{ $item->{itype} }{description};
1843 # hidden because lost
1844 if ($hidelostitems && $item->{itemlost}) {
1845 $hideatopac_count++;
1848 # hidden based on OpacHiddenItems syspref
1849 my @hi = C4::Items::GetHiddenItemnumbers($item);
1851 push @hiddenitems, @hi;
1852 $hideatopac_count++;
1857 my $hbranch = C4::Context->preference('HomeOrHoldingBranch') eq 'homebranch' ? 'homebranch' : 'holdingbranch';
1858 my $otherbranch = C4::Context->preference('HomeOrHoldingBranch') eq 'homebranch' ? 'holdingbranch' : 'homebranch';
1860 # set item's branch name, use HomeOrHoldingBranch syspref first, fall back to the other one
1861 if ($item->{$hbranch}) {
1862 $item->{'branchname'} = $branches{$item->{$hbranch}};
1864 elsif ($item->{$otherbranch}) { # Last resort
1865 $item->{'branchname'} = $branches{$item->{$otherbranch}};
1868 my $prefix = $item->{$hbranch} . '--' . $item->{location} . $item->{itype} . $item->{itemcallnumber};
1869 # For each grouping of items (onloan, available, unavailable), we build a key to store relevant info about that item
1870 my $userenv = C4::Context->userenv;
1871 if ( $item->{onloan} && !(C4::Members::GetHideLostItemsPreference($userenv->{'number'}) && $item->{itemlost}) ) {
1873 my $key = $prefix . $item->{onloan} . $item->{barcode};
1874 $onloan_items->{$key}->{due_date} = format_date($item->{onloan});
1875 $onloan_items->{$key}->{count}++ if $item->{$hbranch};
1876 $onloan_items->{$key}->{branchname} = $item->{branchname};
1877 $onloan_items->{$key}->{location} = $shelflocations->{ $item->{location} };
1878 $onloan_items->{$key}->{itemcallnumber} = $item->{itemcallnumber};
1879 $onloan_items->{$key}->{description} = $item->{description};
1880 $onloan_items->{$key}->{imageurl} = getitemtypeimagelocation( $search_context, $itemtypes{ $item->{itype} }->{imageurl} );
1881 # if something's checked out and lost, mark it as 'long overdue'
1882 if ( $item->{itemlost} ) {
1883 $onloan_items->{$prefix}->{longoverdue}++;
1884 $longoverdue_count++;
1885 } else { # can place holds as long as item isn't lost
1886 $can_place_holds = 1;
1890 # items not on loan, but still unavailable ( lost, withdrawn, damaged )
1894 if ( $item->{notforloan} < 0 ) {
1898 # is item in transit?
1899 my $transfertwhen = '';
1900 my ($transfertfrom, $transfertto);
1902 # is item on the reserve shelf?
1903 my $reservestatus = '';
1905 unless ($item->{wthdrawn}
1906 || $item->{itemlost}
1908 || $item->{notforloan}
1909 || $items_count > 20) {
1911 # A couple heuristics to limit how many times
1912 # we query the database for item transfer information, sacrificing
1913 # accuracy in some cases for speed;
1915 # 1. don't query if item has one of the other statuses
1916 # 2. don't check transit status if the bib has
1917 # more than 20 items
1919 # FIXME: to avoid having the query the database like this, and to make
1920 # the in transit status count as unavailable for search limiting,
1921 # should map transit status to record indexed in Zebra.
1923 ($transfertwhen, $transfertfrom, $transfertto) = C4::Circulation::GetTransfers($item->{itemnumber});
1924 $reservestatus = C4::Reserves::GetReserveStatus( $item->{itemnumber}, $oldbiblio->{biblionumber} );
1927 # item is withdrawn, lost, damaged, not for loan, reserved or in transit
1928 if ( $item->{wthdrawn}
1929 || $item->{itemlost}
1931 || $item->{notforloan}
1932 || $reservestatus eq 'Waiting'
1933 || ($transfertwhen ne ''))
1935 $wthdrawn_count++ if $item->{wthdrawn};
1936 $itemlost_count++ if $item->{itemlost};
1937 $itemdamaged_count++ if $item->{damaged};
1938 $item_in_transit_count++ if $transfertwhen ne '';
1939 $item_onhold_count++ if $reservestatus eq 'Waiting';
1940 $item->{status} = $item->{wthdrawn} . "-" . $item->{itemlost} . "-" . $item->{damaged} . "-" . $item->{notforloan};
1942 # can place hold on item ?
1943 if ( !$item->{itemlost} ) {
1944 if ( !$item->{wthdrawn} ){
1945 if ( $item->{damaged} ){
1946 if ( C4::Context->preference('AllowHoldsOnDamagedItems') ){
1947 # can place a hold on a damaged item if AllowHoldsOnDamagedItems is true
1948 if ( ( !$item->{notforloan} || $item->{notforloan} < 0 ) ){
1949 # item is either for loan or has notforloan < 0
1950 $can_place_holds = 1;
1953 } elsif ( $item->{notforloan} < 0 ) {
1954 # item is not damaged and notforloan is < 0
1955 $can_place_holds = 1;
1962 my $key = $prefix . $item->{status};
1963 foreach (qw(wthdrawn itemlost damaged branchname itemcallnumber)) {
1964 $other_items->{$key}->{$_} = $item->{$_};
1966 $other_items->{$key}->{intransit} = ( $transfertwhen ne '' ) ? 1 : 0;
1967 $other_items->{$key}->{onhold} = ($reservestatus) ? 1 : 0;
1968 $other_items->{$key}->{notforloan} = GetAuthorisedValueDesc('','',$item->{notforloan},'','',$notforloan_authorised_value) if $notforloan_authorised_value and $item->{notforloan};
1969 $other_items->{$key}->{count}++ if $item->{$hbranch};
1970 $other_items->{$key}->{location} = $shelflocations->{ $item->{location} };
1971 $other_items->{$key}->{description} = $item->{description};
1972 $other_items->{$key}->{imageurl} = getitemtypeimagelocation( $search_context, $itemtypes{ $item->{itype} }->{imageurl} );
1976 $can_place_holds = 1;
1978 $available_items->{$prefix}->{count}++ if $item->{$hbranch};
1979 foreach (qw(branchname itemcallnumber description)) {
1980 $available_items->{$prefix}->{$_} = $item->{$_};
1982 $available_items->{$prefix}->{location} = $shelflocations->{ $item->{location} };
1983 $available_items->{$prefix}->{imageurl} = getitemtypeimagelocation( $search_context, $itemtypes{ $item->{itype} }->{imageurl} );
1986 } # notforloan, item level and biblioitem level
1988 # if all items are hidden, do not show the record
1989 if ($items_count > 0 && $hideatopac_count == $items_count) {
1993 my ( $availableitemscount, $onloanitemscount, $otheritemscount );
1994 for my $key ( sort keys %$onloan_items ) {
1995 (++$onloanitemscount > $maxitems) and last;
1996 push @onloan_items_loop, $onloan_items->{$key};
1998 for my $key ( sort keys %$other_items ) {
1999 (++$otheritemscount > $maxitems) and last;
2000 push @other_items_loop, $other_items->{$key};
2002 for my $key ( sort keys %$available_items ) {
2003 (++$availableitemscount > $maxitems) and last;
2004 push @available_items_loop, $available_items->{$key}
2007 # XSLT processing of some stuff
2009 SetUTF8Flag($marcrecord);
2010 warn $marcrecord->as_formatted if $DEBUG;
2011 my $interface = $search_context eq 'opac' ? 'OPAC' : '';
2012 if (!$scan && C4::Context->preference($interface . "XSLTResultsDisplay")) {
2013 $oldbiblio->{XSLTResultsRecord} = XSLTParse4Display($oldbiblio->{biblionumber}, $marcrecord, $interface."XSLTResultsDisplay", 1, \@hiddenitems);
2014 # the last parameter tells Koha to clean up the problematic ampersand entities that Zebra outputs
2017 # if biblio level itypes are used and itemtype is notforloan, it can't be reserved either
2018 if (!C4::Context->preference("item-level_itypes")) {
2019 if ($itemtypes{ $oldbiblio->{itemtype} }->{notforloan}) {
2020 $can_place_holds = 0;
2023 $oldbiblio->{norequests} = 1 unless $can_place_holds;
2024 $oldbiblio->{itemsplural} = 1 if $items_count > 1;
2025 $oldbiblio->{items_count} = $items_count;
2026 $oldbiblio->{available_items_loop} = \@available_items_loop;
2027 $oldbiblio->{onloan_items_loop} = \@onloan_items_loop;
2028 $oldbiblio->{other_items_loop} = \@other_items_loop;
2029 $oldbiblio->{availablecount} = $available_count;
2030 $oldbiblio->{availableplural} = 1 if $available_count > 1;
2031 $oldbiblio->{onloancount} = $onloan_count;
2032 $oldbiblio->{onloanplural} = 1 if $onloan_count > 1;
2033 $oldbiblio->{othercount} = $other_count;
2034 $oldbiblio->{otherplural} = 1 if $other_count > 1;
2035 $oldbiblio->{wthdrawncount} = $wthdrawn_count;
2036 $oldbiblio->{itemlostcount} = $itemlost_count;
2037 $oldbiblio->{damagedcount} = $itemdamaged_count;
2038 $oldbiblio->{intransitcount} = $item_in_transit_count;
2039 $oldbiblio->{onholdcount} = $item_onhold_count;
2040 $oldbiblio->{orderedcount} = $ordered_count;
2042 if (C4::Context->preference("AlternateHoldingsField") && $items_count == 0) {
2043 my $fieldspec = C4::Context->preference("AlternateHoldingsField");
2044 my $subfields = substr $fieldspec, 3;
2045 my $holdingsep = C4::Context->preference("AlternateHoldingsSeparator") || ' ';
2046 my @alternateholdingsinfo = ();
2047 my @holdingsfields = $marcrecord->field(substr $fieldspec, 0, 3);
2048 my $alternateholdingscount = 0;
2050 for my $field (@holdingsfields) {
2051 my %holding = ( holding => '' );
2052 my $havesubfield = 0;
2053 for my $subfield ($field->subfields()) {
2054 if ((index $subfields, $$subfield[0]) >= 0) {
2055 $holding{'holding'} .= $holdingsep if (length $holding{'holding'} > 0);
2056 $holding{'holding'} .= $$subfield[1];
2060 if ($havesubfield) {
2061 push(@alternateholdingsinfo, \%holding);
2062 $alternateholdingscount++;
2066 $oldbiblio->{'ALTERNATEHOLDINGS'} = \@alternateholdingsinfo;
2067 $oldbiblio->{'alternateholdings_count'} = $alternateholdingscount;
2070 push( @newresults, $oldbiblio );
2076 =head2 SearchAcquisitions
2077 Search for acquisitions
2080 sub SearchAcquisitions{
2081 my ($datebegin, $dateend, $itemtypes,$criteria, $orderby) = @_;
2083 my $dbh=C4::Context->dbh;
2084 # Variable initialization
2088 LEFT JOIN biblioitems ON biblioitems.biblionumber=biblio.biblionumber
2089 LEFT JOIN items ON items.biblionumber=biblio.biblionumber
2090 WHERE dateaccessioned BETWEEN ? AND ?
2093 my (@params,@loopcriteria);
2095 push @params, $datebegin->output("iso");
2096 push @params, $dateend->output("iso");
2098 if (scalar(@$itemtypes)>0 and $criteria ne "itemtype" ){
2099 if(C4::Context->preference("item-level_itypes")){
2100 $str .= "AND items.itype IN (?".( ',?' x scalar @$itemtypes - 1 ).") ";
2102 $str .= "AND biblioitems.itemtype IN (?".( ',?' x scalar @$itemtypes - 1 ).") ";
2104 push @params, @$itemtypes;
2107 if ($criteria =~/itemtype/){
2108 if(C4::Context->preference("item-level_itypes")){
2109 $str .= "AND items.itype=? ";
2111 $str .= "AND biblioitems.itemtype=? ";
2114 if(scalar(@$itemtypes) == 0){
2115 my $itypes = GetItemTypes();
2116 for my $key (keys %$itypes){
2117 push @$itemtypes, $key;
2121 @loopcriteria= @$itemtypes;
2122 }elsif ($criteria=~/itemcallnumber/){
2123 $str .= "AND (items.itemcallnumber LIKE CONCAT(?,'%')
2124 OR items.itemcallnumber is NULL
2125 OR items.itemcallnumber = '')";
2127 @loopcriteria = ("AA".."ZZ", "") unless (scalar(@loopcriteria)>0);
2129 $str .= "AND biblio.title LIKE CONCAT(?,'%') ";
2130 @loopcriteria = ("A".."z") unless (scalar(@loopcriteria)>0);
2133 if ($orderby =~ /date_desc/){
2134 $str.=" ORDER BY dateaccessioned DESC";
2136 $str.=" ORDER BY title";
2139 my $qdataacquisitions=$dbh->prepare($str);
2141 my @loopacquisitions;
2142 foreach my $value(@loopcriteria){
2143 push @params,$value;
2145 $cell{"title"}=$value;
2146 $cell{"titlecode"}=$value;
2148 eval{$qdataacquisitions->execute(@params);};
2150 if ($@){ warn "recentacquisitions Error :$@";}
2153 while (my $data=$qdataacquisitions->fetchrow_hashref){
2154 push @loopdata, {"summary"=>GetBiblioSummary( $data->{'marcxml'} ) };
2156 $cell{"loopdata"}=\@loopdata;
2158 push @loopacquisitions,\%cell if (scalar(@{$cell{loopdata}})>0);
2161 $qdataacquisitions->finish;
2162 return \@loopacquisitions;
2165 =head2 enabled_staff_search_views
2167 %hash = enabled_staff_search_views()
2169 This function returns a hash that contains three flags obtained from the system
2170 preferences, used to determine whether a particular staff search results view
2175 =item C<Output arg:>
2177 * $hash{can_view_MARC} is true only if the MARC view is enabled
2178 * $hash{can_view_ISBD} is true only if the ISBD view is enabled
2179 * $hash{can_view_labeledMARC} is true only if the Labeled MARC view is enabled
2181 =item C<usage in the script:>
2185 $template->param ( C4::Search::enabled_staff_search_views );
2189 sub enabled_staff_search_views
2192 can_view_MARC => C4::Context->preference('viewMARC'), # 1 if the staff search allows the MARC view
2193 can_view_ISBD => C4::Context->preference('viewISBD'), # 1 if the staff search allows the ISBD view
2194 can_view_labeledMARC => C4::Context->preference('viewLabeledMARC'), # 1 if the staff search allows the Labeled MARC view
2198 sub AddSearchHistory{
2199 my ($borrowernumber,$session,$query_desc,$query_cgi, $total)=@_;
2200 my $dbh = C4::Context->dbh;
2202 # Add the request the user just made
2203 my $sql = "INSERT INTO search_history(userid, sessionid, query_desc, query_cgi, total, time) VALUES(?, ?, ?, ?, ?, NOW())";
2204 my $sth = $dbh->prepare($sql);
2205 $sth->execute($borrowernumber, $session, $query_desc, $query_cgi, $total);
2206 return $dbh->last_insert_id(undef, 'search_history', undef,undef,undef);
2209 sub GetSearchHistory{
2210 my ($borrowernumber,$session)=@_;
2211 my $dbh = C4::Context->dbh;
2213 # Add the request the user just made
2214 my $query = "SELECT FROM search_history WHERE (userid=? OR sessionid=?)";
2215 my $sth = $dbh->prepare($query);
2216 $sth->execute($borrowernumber, $session);
2217 return $sth->fetchall_hashref({});
2220 =head2 z3950_search_args
2222 $arrayref = z3950_search_args($matchpoints)
2224 This function returns an array reference that contains the search parameters to be
2225 passed to the Z39.50 search script (z3950_search.pl). The array elements
2226 are hash refs whose keys are name, value and encvalue, and whose values are the
2227 name of a search parameter, the value of that search parameter and the URL encoded
2228 value of that parameter.
2230 The search parameter names are lccn, isbn, issn, title, author, dewey and subject.
2232 The search parameter values are obtained from the bibliographic record whose
2233 data is in a hash reference in $matchpoints, as returned by Biblio::GetBiblioData().
2235 If $matchpoints is a scalar, it is assumed to be an unnamed query descriptor, e.g.
2236 a general purpose search argument. In this case, the returned array contains only
2237 entry: the key is 'title' and the value and encvalue are derived from $matchpoints.
2239 If a search parameter value is undefined or empty, it is not included in the returned
2242 The returned array reference may be passed directly to the template parameters.
2246 =item C<Output arg:>
2248 * $array containing hash refs as described above
2250 =item C<usage in the script:>
2254 $data = Biblio::GetBiblioData($bibno);
2255 $template->param ( MYLOOP => C4::Search::z3950_search_args($data) )
2259 $template->param ( MYLOOP => C4::Search::z3950_search_args($searchscalar) )
2263 sub z3950_search_args {
2265 my $isbn = Business::ISBN->new($bibrec);
2267 if (defined $isbn && $isbn->is_valid)
2269 $bibrec = { isbn => $bibrec } if !ref $bibrec;
2272 $bibrec = { title => $bibrec } if !ref $bibrec;
2275 for my $field (qw/ lccn isbn issn title author dewey subject /)
2277 my $encvalue = URI::Escape::uri_escape_utf8($bibrec->{$field});
2278 push @$array, { name=>$field, value=>$bibrec->{$field}, encvalue=>$encvalue } if defined $bibrec->{$field};
2283 =head2 GetDistinctValues($field);
2285 C<$field> is a reference to the fields array
2289 sub GetDistinctValues {
2290 my ($fieldname,$string)=@_;
2291 # returns a reference to a hash of references to branches...
2292 if ($fieldname=~/\./){
2293 my ($table,$column)=split /\./, $fieldname;
2294 my $dbh = C4::Context->dbh;
2295 warn "select DISTINCT($column) as value, count(*) as cnt from $table group by lib order by $column " if $DEBUG;
2296 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 ");
2298 my $elements=$sth->fetchall_arrayref({});
2303 my @servers=qw<biblioserver authorityserver>;
2304 my (@zconns,@results);
2305 for ( my $i = 0 ; $i < @servers ; $i++ ) {
2306 $zconns[$i] = C4::Context->Zconn( $servers[$i], 1 );
2309 ZOOM::Query::CCL2RPN->new( qq"$fieldname $string", $zconns[$i])
2312 # The big moment: asynchronously retrieve results from all servers
2318 my ( $i, $size ) = @_;
2319 for ( my $j = 0 ; $j < $size ; $j++ ) {
2321 @hashscan{qw(value cnt)} =
2322 $results[ $i - 1 ]->display_term($j);
2323 push @elements, \%hashscan;
2331 =head2 _ZOOM_event_loop
2333 _ZOOM_event_loop(\@zconns, \@results, sub {
2334 my ( $i, $size ) = @_;
2338 Processes a ZOOM event loop and passes control to a closure for
2339 processing the results, and destroying the resultsets.
2343 sub _ZOOM_event_loop {
2344 my ($zconns, $results, $callback) = @_;
2345 while ( ( my $i = ZOOM::event( $zconns ) ) != 0 ) {
2346 my $ev = $zconns->[ $i - 1 ]->last_event();
2347 if ( $ev == ZOOM::Event::ZEND ) {
2348 next unless $results->[ $i - 1 ];
2349 my $size = $results->[ $i - 1 ]->size();
2351 $callback->($i, $size);
2356 foreach my $result (@$results) {
2362 END { } # module clean-up code here (global destructor)
2369 Koha Development Team <http://koha-community.org/>