1 package Koha::SearchEngine::Elasticsearch::QueryBuilder;
3 # This file is part of Koha.
5 # Copyright 2014 Catalyst IT Ltd.
7 # Koha is free software; you can redistribute it and/or modify it
8 # under the terms of the GNU General Public License as published by
9 # the Free Software Foundation; either version 3 of the License, or
10 # (at your option) any later version.
12 # Koha is distributed in the hope that it will be useful, but
13 # WITHOUT ANY WARRANTY; without even the implied warranty of
14 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 # GNU General Public License for more details.
17 # You should have received a copy of the GNU General Public License
18 # along with Koha; if not, see <http://www.gnu.org/licenses>.
22 Koha::SearchEngine::Elasticsearch::QueryBuilder - constructs elasticsearch
23 query objects from user-supplied queries
27 This provides the functions that take a user-supplied search query, and
28 provides something that can be given to elasticsearch to get answers.
32 use Koha::SearchEngine::Elasticsearch::QueryBuilder;
33 $builder = Koha::SearchEngine::Elasticsearch->new({ index => $index });
34 my $simple_query = $builder->build_query("hello");
35 # This is currently undocumented because the original code is undocumented
36 my $adv_query = $builder->build_advanced_query($indexes, $operands, $operators);
42 use base qw(Koha::SearchEngine::Elasticsearch);
45 use List::MoreUtils qw/ each_array /;
54 my $simple_query = $builder->build_query("hello", %options)
56 This will build a query that can be issued to elasticsearch from the provided
57 string input. This expects a lucene style search form (see
58 L<http://www.elasticsearch.org/guide/en/elasticsearch/reference/current/query-dsl-query-string-query.html#query-string-syntax>
61 It'll make an attempt to respect the various query options.
63 Additional options can be provided with the C<%options> hash.
69 This should be an arrayref of hashrefs, each containing a C<field> and an
70 C<direction> (optional, defaults to C<asc>.) The results will be sorted
71 according to these values. Valid values for C<direction> are 'asc' and 'desc'.
78 my ( $self, $query, %options ) = @_;
80 my $stemming = C4::Context->preference("QueryStemming") || 0;
81 my $auto_truncation = C4::Context->preference("QueryAutoTruncate") || 0;
82 my $weight_fields = C4::Context->preference("QueryWeightFields") || 0;
83 my $fuzzy_enabled = C4::Context->preference("QueryFuzzy") || 0;
85 $query = '*' unless defined $query;
91 fuzziness => $fuzzy_enabled ? 'auto' : '0',
92 default_operator => 'AND',
93 default_field => '_all',
94 lenient => JSON::true,
98 if ( $options{sort} ) {
99 foreach my $sort ( @{ $options{sort} } ) {
100 my ( $f, $d ) = @$sort{qw/ field direction /};
101 die "Invalid sort direction, $d"
102 if $d && ( $d ne 'asc' && $d ne 'desc' );
103 $d = 'asc' unless $d;
105 # TODO account for fields that don't have a 'phrase' type
107 $f = $self->_sort_field($f);
108 push @{ $res->{sort} }, { "$f.phrase" => { order => $d } };
112 # See _convert_facets in Search.pm for how these get turned into
113 # things that Koha can use.
114 $res->{aggregations} = {
115 author => { terms => { field => "author__facet" } },
116 subject => { terms => { field => "subject__facet" } },
117 itype => { terms => { field => "itype__facet" } },
118 location => { terms => { field => "location__facet" } },
119 'su-geo' => { terms => { field => "su-geo__facet" } },
120 se => { terms => { field => "se__facet" } },
121 ccode => { terms => { field => "ccode__facet" } },
124 my $display_library_facets = C4::Context->preference('DisplayLibraryFacets');
125 if ( $display_library_facets eq 'both'
126 or $display_library_facets eq 'home' ) {
127 $res->{aggregations}{homebranch} = { terms => { field => "homebranch__facet" } };
129 if ( $display_library_facets eq 'both'
130 or $display_library_facets eq 'holding' ) {
131 $res->{aggregations}{holdingbranch} = { terms => { field => "holdingbranch__facet" } };
133 if ( my $ef = $options{expanded_facet} ) {
134 $res->{aggregations}{$ef}{terms}{size} = C4::Context->preference('FacetMaxCount');
139 =head2 build_browse_query
141 my $browse_query = $builder->build_browse_query($field, $query);
143 This performs a "starts with" style query on a particular field. The field
144 to be searched must have been indexed with an appropriate mapping as a
145 "phrase" subfield, which pretty much everything has.
149 # XXX this isn't really a browse query like we want in the end
150 sub build_browse_query {
151 my ( $self, $field, $query ) = @_;
153 my $fuzzy_enabled = C4::Context->preference("QueryFuzzy") || 0;
155 return { query => '*' } if !defined $query;
157 # TODO this should come from Koha::SearchEngine::Elasticsearch
158 my %field_whitelist = (
162 $field = 'title' if !exists $field_whitelist{$field};
163 my $sort = $self->_sort_field($field);
166 match_phrase_prefix => {
170 fuzziness => $fuzzy_enabled ? 'auto' : '0',
174 sort => [ { "$sort.phrase" => { order => "asc" } } ],
178 =head2 build_query_compat
181 $error, $query, $simple_query, $query_cgi,
182 $query_desc, $limit, $limit_cgi, $limit_desc,
183 $stopwords_removed, $query_type
185 = $builder->build_query_compat( \@operators, \@operands, \@indexes,
186 \@limits, \@sort_by, $scan, $lang );
188 This handles a search using the same api as L<C4::Search::buildQuery> does.
190 A very simple query will go in with C<$operands> set to ['query'], and
191 C<$sort_by> set to ['pubdate_dsc']. This simple case will return with
192 C<$query> set to something that can perform the search, C<$simple_query>
193 set to just the search term, C<$query_cgi> set to something that can
194 reproduce this search, and C<$query_desc> set to something else.
198 sub build_query_compat {
199 my ( $self, $operators, $operands, $indexes, $orig_limits, $sort_by, $scan,
203 #die Dumper ( $self, $operators, $operands, $indexes, $orig_limits, $sort_by, $scan, $lang );
204 my @sort_params = $self->_convert_sort_fields(@$sort_by);
205 my @index_params = $self->_convert_index_fields(@$indexes);
206 my $limits = $self->_fix_limit_special_cases($orig_limits);
208 # Merge the indexes in with the search terms and the operands so that
209 # each search thing is a handy unit.
210 unshift @$operators, undef; # The first one can't have an op
212 my $ea = each_array( @$operands, @$operators, @index_params );
213 while ( my ( $oand, $otor, $index ) = $ea->() ) {
214 next if ( !defined($oand) || $oand eq '' );
215 push @search_params, {
216 operand => $self->_clean_search_term($oand), # the search terms
217 operator => defined($otor) ? uc $otor : undef, # AND and so on
218 $index ? %$index : (),
222 # We build a string query from limits and the queries. An alternative
223 # would be to pass them separately into build_query and let it build
224 # them into a structured ES query itself. Maybe later, though that'd be
226 my $query_str = join( ' AND ',
227 join( ' ', $self->_create_query_string(@search_params) ) || (),
228 $self->_join_queries( $self->_convert_index_strings(@$limits) ) || () );
230 # If there's no query on the left, let's remove the junk left behind
231 $query_str =~ s/^ AND //;
233 $options{sort} = \@sort_params;
234 $options{expanded_facet} = $params->{expanded_facet};
235 my $query = $self->build_query( $query_str, %options );
238 # We roughly emulate the CGI parameters of the zebra query builder
240 $query_cgi = 'idx=kw&q=' . uri_escape_utf8( $operands->[0] ) if @$operands;
242 $simple_query = $operands->[0] if @$operands == 1;
243 my $query_desc = $simple_query;
244 my $limit = $self->_join_queries( $self->_convert_index_strings(@$limits));
245 my $limit_cgi = ( $orig_limits and @$orig_limits )
246 ? '&limit=' . join( '&limit=', map { uri_escape_utf8($_) } @$orig_limits )
249 $limit_desc = "$limit" if $limit;
251 undef, $query, $simple_query, $query_cgi, $query_desc,
252 $limit, $limit_cgi, $limit_desc, undef, undef
256 =head2 build_authorities_query
258 my $query = $builder->build_authorities_query(\%search);
260 This takes a nice description of an authority search and turns it into a black-box
261 query that can then be passed to the appropriate searcher.
263 The search description is a hashref that looks something like:
268 where => 'Heading', # search the main entry
269 operator => 'exact', # require an exact match
270 value => 'frogs', # the search string
273 where => '', # search all entries
274 operator => '', # default keyword, right truncation
282 authtypecode => 'TOPIC_TERM',
287 sub build_authorities_query {
288 my ( $self, $search ) = @_;
290 # Start by making the query parts
293 foreach my $s ( @{ $search->{searches} } ) {
294 my ( $wh, $op, $val ) = @{$s}{qw(where operator value)};
295 $wh = '_all' if $wh eq '';
296 if ( $op eq 'is' || $op eq '=' ) {
298 # look for something that matches completely
299 # note, '=' is about numerical vals. May need special handling.
300 # _allphrase is a special field that only groups the exact
301 # matches. Also, we lowercase our search because the ES
302 # index lowercases its values, and term searches don't get the
303 # search analyzer applied to them.
304 push @filter_parts, { term => { "$wh.phrase" => lc $val } };
306 elsif ( $op eq 'exact' ) {
308 # left and right truncation, otherwise an exact phrase
309 push @query_parts, { match_phrase => { $wh => $val } };
311 elsif ( $op eq 'start' ) {
314 push @query_parts, { wildcard => { "$wh.phrase" => lc "$val*" } };
317 # regular wordlist stuff
318 push @query_parts, { match => { $wh => $val } };
322 # Merge the query and filter parts appropriately
323 # 'should' behaves like 'or', if we want 'and', use 'must'
324 my $query_part = { bool => { should => \@query_parts } };
325 my $filter_part = { bool => { should => \@filter_parts } };
327 # We need to add '.phrase' to all the sort headings otherwise it'll sort
328 # based on the tokenised form.
330 if ( exists $search->{sort} ) {
331 foreach my $k ( keys %{ $search->{sort} } ) {
332 my $f = $self->_sort_field($k);
333 $s{"$f.phrase"} = $search->{sort}{$k};
335 $search->{sort} = \%s;
338 # extract the sort stuff
340 %sort = ( sort => [ $search->{sort} ] ) if exists $search->{sort};
345 { filtered => { filter => $filter_part, query => $query_part } }
349 $query = { query => $query_part };
351 $query = { %$query, %sort };
356 =head2 build_authorities_query_compat
359 $builder->build_authorities_query_compat( \@marclist, \@and_or,
360 \@excluding, \@operator, \@value, $authtypecode, $orderby );
362 This builds a query for searching for authorities, in the style of
363 L<C4::AuthoritiesMarc::SearchAuthorities>.
371 An arrayref containing where the particular term should be searched for.
372 Options are: mainmainentry, mainentry, match, match-heading, see-from, and
373 thesaurus. If left blank, any field is used.
377 Totally ignored. It is never used in L<C4::AuthoritiesMarc::SearchAuthorities>.
385 What form of search to do. Options are: is (phrase, no trunction, whole field
386 must match), = (number exact match), exact (phrase, but with left and right
387 truncation). If left blank, then word list, right truncted, anywhere is used.
391 The actual user-provided string value to search for.
395 The authority type code to search within. If blank, then all will be searched.
399 The order to sort the results by. Options are Relevance, HeadingAsc,
400 HeadingDsc, AuthidAsc, AuthidDsc.
404 marclist, operator, and value must be the same length, and the values at
405 index /i/ all relate to each other.
407 This returns a query, which is a black box object that can be passed to the
408 appropriate search object.
412 our $koha_to_index_name = {
413 mainmainentry => 'Heading-Main',
414 mainentry => 'Heading',
416 'match-heading' => 'Match-heading',
417 'see-from' => 'Match-heading-see-from',
418 thesaurus => 'Subject-heading-thesaurus',
422 sub build_authorities_query_compat {
423 my ( $self, $marclist, $and_or, $excluding, $operator, $value,
424 $authtypecode, $orderby )
427 # This turns the old-style many-options argument form into a more
428 # extensible hash form that is understood by L<build_authorities_query>.
431 # Make sure everything exists
432 foreach my $m (@$marclist) {
433 Koha::Exceptions::WrongParameter->throw("Invalid marclist field provided: $m")
434 unless exists $koha_to_index_name->{$m};
436 for ( my $i = 0 ; $i < @$value ; $i++ ) {
439 where => $koha_to_index_name->{$marclist->[$i]},
440 operator => $operator->[$i],
441 value => $value->[$i],
447 ( $orderby =~ /^Heading/ ) ? 'Heading'
448 : ( $orderby =~ /^Auth/ ) ? 'Local-Number'
451 my $sort_order = ( $orderby =~ /Asc$/ ) ? 'asc' : 'desc';
452 %sort = ( $sort_field => $sort_order, );
455 searches => \@searches,
456 authtypecode => $authtypecode,
458 $search{sort} = \%sort if %sort;
459 my $query = $self->build_authorities_query( \%search );
463 =head2 _convert_sort_fields
465 my @sort_params = _convert_sort_fields(@sort_by)
467 Converts the zebra-style sort index information into elasticsearch-style.
469 C<@sort_by> is the same as presented to L<build_query_compat>, and it returns
470 something that can be sent to L<build_query>.
474 sub _convert_sort_fields {
475 my ( $self, @sort_by ) = @_;
477 # Turn the sorting into something we care about.
478 my %sort_field_convert = (
479 acqdate => 'acqdate',
481 call_number => 'callnum',
482 popularity => 'issues',
483 relevance => undef, # default
485 pubdate => 'pubdate',
487 my %sort_order_convert =
488 ( qw( dsc desc ), qw( asc asc ), qw( az asc ), qw( za desc ) );
490 # Convert the fields and orders, drop anything we don't know about.
491 grep { $_->{field} } map {
492 my ( $f, $d ) = split /_/;
494 field => $sort_field_convert{$f},
495 direction => $sort_order_convert{$d}
500 =head2 _convert_index_fields
502 my @index_params = $self->_convert_index_fields(@indexes);
504 Converts zebra-style search index notation into elasticsearch-style.
506 C<@indexes> is an array of index names, as presented to L<build_query_compat>,
507 and it returns something that can be sent to L<build_query>.
509 B<TODO>: this will pull from the elasticsearch mappings table to figure out
514 our %index_field_convert = (
520 'se' => 'title-series',
521 'callnum' => 'callnum',
524 'branch' => 'homebranch',
528 'hi' => 'Host-Item-Number',
531 sub _convert_index_fields {
532 my ( $self, @indexes ) = @_;
534 my %index_type_convert =
535 ( __default => undef, phr => 'phrase', rtrn => 'right-truncate' );
537 # Convert according to our table, drop anything that doesn't convert.
538 # If a field starts with mc- we save it as it's used (and removed) later
539 # when joining things, to indicate we make it an 'OR' join.
540 # (Sorry, this got a bit ugly after special cases were found.)
541 grep { $_->{field} } map {
542 my ( $f, $t ) = split /,/;
549 field => $index_field_convert{$f},
550 type => $index_type_convert{ $t // '__default' }
552 $r->{field} = ($mc . $r->{field}) if $mc && $r->{field};
557 =head2 _convert_index_strings
559 my @searches = $self->_convert_index_strings(@searches);
561 Similar to L<_convert_index_fields>, this takes strings of the form
562 B<field:search term> and rewrites the field from zebra-style to
563 elasticsearch-style. Anything it doesn't understand is returned verbatim.
567 sub _convert_index_strings {
568 my ( $self, @searches ) = @_;
570 foreach my $s (@searches) {
572 my ( $field, $term ) = $s =~ /^\s*([\w,-]*?):(.*)/;
573 unless ( defined($field) && defined($term) ) {
577 my ($conv) = $self->_convert_index_fields($field);
578 unless ( defined($conv) ) {
582 push @res, $conv->{field} . ":"
583 . $self->_modify_string_by_type( %$conv, operand => $term );
588 =head2 _convert_index_strings_freeform
590 my $search = $self->_convert_index_strings_freeform($search);
592 This is similar to L<_convert_index_strings>, however it'll search out the
593 things to change within the string. So it can handle strings such as
594 C<(su:foo) AND (su:bar)>, converting the C<su> appropriately.
596 If there is something of the form "su,complete-subfield" or something, the
597 second part is stripped off as we can't yet handle that. Making it work
598 will have to wait for a real query parser.
602 sub _convert_index_strings_freeform {
603 my ( $self, $search ) = @_;
604 while ( my ( $zeb, $es ) = each %index_field_convert ) {
605 $search =~ s/\b$zeb(?:,[\w-]*)?:/$es:/g;
610 =head2 _modify_string_by_type
612 my $str = $self->_modify_string_by_type(%index_field);
614 If you have a search term (operand) and a type (phrase, right-truncated), this
615 will convert the string to have the function in lucene search terms, e.g.
616 wrapping quotes around it.
620 sub _modify_string_by_type {
621 my ( $self, %idx ) = @_;
623 my $type = $idx{type} || '';
624 my $str = $idx{operand};
625 return $str unless $str; # Empty or undef, we can't use it.
627 $str .= '*' if $type eq 'right-truncate';
628 $str = '"' . $str . '"' if $type eq 'phrase';
634 my $query_str = $self->_join_queries(@query_parts);
636 This takes a list of query parts, that might be search terms on their own, or
637 booleaned together, or specifying fields, or whatever, wraps them in
638 parentheses, and ANDs them all together. Suitable for feeding to the ES
641 Note: doesn't AND them together if they specify an index that starts with "mc"
642 as that was a special case in the original code for dealing with multiple
643 choice options (you can't search for something that has an itype of A and
644 and itype of B otherwise.)
649 my ( $self, @parts ) = @_;
651 my @norm_parts = grep { defined($_) && $_ ne '' && $_ !~ /^mc-/ } @parts;
653 map { s/^mc-//r } grep { defined($_) && $_ ne '' && $_ =~ /^mc-/ } @parts;
654 return () unless @norm_parts + @mc_parts;
655 return ( @norm_parts, @mc_parts )[0] if @norm_parts + @mc_parts == 1;
657 @mc_parts ? '(' . ( join ' OR ', map { "($_)" } @mc_parts ) . ')' : ();
659 # Handy trick: $x || () inside a join means that if $x ends up as an
660 # empty string, it gets replaced with (), which makes join ignore it.
661 # (bad effect: this'll also happen to '0', this hopefully doesn't matter
664 join( ' AND ', map { "($_)" } @norm_parts ) || (),
670 my @phrased_queries = $self->_make_phrases(@query_parts);
672 This takes the supplied queries and forces them to be phrases by wrapping
673 quotes around them. It understands field prefixes, e.g. 'subject:' and puts
674 the quotes outside of them if they're there.
679 my ( $self, @parts ) = @_;
680 map { s/^\s*(\w*?:)(.*)$/$1"$2"/r } @parts;
683 =head2 _create_query_string
685 my @query_strings = $self->_create_query_string(@queries);
687 Given a list of hashrefs, it will turn them into a lucene-style query string.
688 The hash should contain field, type (both for the indexes), operator, and
693 sub _create_query_string {
694 my ( $self, @queries ) = @_;
697 my $otor = $_->{operator} ? $_->{operator} . ' ' : '';
698 my $field = $_->{field} ? $_->{field} . ':' : '';
700 my $oand = $self->_modify_string_by_type(%$_);
701 "$otor($field$oand)";
705 =head2 _clean_search_term
707 my $term = $self->_clean_search_term($term);
709 This cleans a search term by removing any funny characters that may upset
710 ES and give us an error. It also calls L<_convert_index_strings_freeform>
711 to ensure those parts are correct.
715 sub _clean_search_term {
716 my ( $self, $term ) = @_;
718 my $auto_truncation = C4::Context->preference("QueryAutoTruncate") || 0;
720 # Some hardcoded searches (like with authorities) produce things like
721 # 'an=123', when it ought to be 'an:123' for our purposes.
723 $term = $self->_convert_index_strings_freeform($term);
725 $term = $self->_truncate_terms($term) if ($auto_truncation);
729 =head2 _fix_limit_special_cases
731 my $limits = $self->_fix_limit_special_cases($limits);
733 This converts any special cases that the limit specifications have into things
734 that are more readily processable by the rest of the code.
736 The argument should be an arrayref, and it'll return an arrayref.
740 sub _fix_limit_special_cases {
741 my ( $self, $limits ) = @_;
744 foreach my $l (@$limits) {
746 # This is set up by opac-search.pl
747 if ( $l =~ /^yr,st-numeric,ge=/ ) {
748 my ( $start, $end ) =
749 ( $l =~ /^yr,st-numeric,ge=(.*) and yr,st-numeric,le=(.*)$/ );
750 next unless defined($start) && defined($end);
751 push @new_lim, "copydate:[$start TO $end]";
753 elsif ( $l =~ /^yr,st-numeric=/ ) {
754 my ($date) = ( $l =~ /^yr,st-numeric=(.*)$/ );
755 next unless defined($date);
756 push @new_lim, "copydate:$date";
758 elsif ( $l =~ /^available$/ ) {
759 push @new_lim, 'onloan:0';
770 my $field = $self->_sort_field($field);
772 Given a field name, this works out what the actual name of the version to sort
773 on should be. Often it's the same, sometimes it involves sticking "__sort" on
774 the end. Maybe it'll be something else in the future, who knows?
780 if ($self->sort_fields()->{$f}) {
786 =head2 _truncate_terms
788 my $query = $self->_truncate_terms($query);
790 Given a string query this function appends '*' wildcard to all terms except
795 sub _truncate_terms {
796 my ( $self, $query ) = @_;
797 my @stops = qw/and or not/;
799 my @split_query = split /[\(\s\)]/, $query;
800 foreach my $term (@split_query) {
801 next if ( $term eq '' || $term eq ' ' );
802 $term .= "*" unless ( ( grep { lc($term) =~ /^$_$/ } @stops ) || ( $term =~ /\*$/ ) );
803 push @new_terms, $term;
805 $query = join ' ', @new_terms;