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);
207 if ( $params->{suppress} ) { push @$limits, "suppress:0"; }
209 # Merge the indexes in with the search terms and the operands so that
210 # each search thing is a handy unit.
211 unshift @$operators, undef; # The first one can't have an op
213 my $ea = each_array( @$operands, @$operators, @index_params );
214 while ( my ( $oand, $otor, $index ) = $ea->() ) {
215 next if ( !defined($oand) || $oand eq '' );
216 push @search_params, {
217 operand => $self->_clean_search_term($oand), # the search terms
218 operator => defined($otor) ? uc $otor : undef, # AND and so on
219 $index ? %$index : (),
223 # We build a string query from limits and the queries. An alternative
224 # would be to pass them separately into build_query and let it build
225 # them into a structured ES query itself. Maybe later, though that'd be
227 my $query_str = join( ' AND ',
228 join( ' ', $self->_create_query_string(@search_params) ) || (),
229 $self->_join_queries( $self->_convert_index_strings(@$limits) ) || () );
231 # If there's no query on the left, let's remove the junk left behind
232 $query_str =~ s/^ AND //;
234 $options{sort} = \@sort_params;
235 $options{expanded_facet} = $params->{expanded_facet};
236 my $query = $self->build_query( $query_str, %options );
239 # We roughly emulate the CGI parameters of the zebra query builder
241 $query_cgi = 'idx=kw&q=' . uri_escape_utf8( $operands->[0] ) if @$operands;
243 $simple_query = $operands->[0] if @$operands == 1;
244 my $query_desc = $simple_query;
245 my $limit = $self->_join_queries( $self->_convert_index_strings(@$limits));
246 my $limit_cgi = ( $orig_limits and @$orig_limits )
247 ? '&limit=' . join( '&limit=', map { uri_escape_utf8($_) } @$orig_limits )
250 $limit_desc = "$limit" if $limit;
252 undef, $query, $simple_query, $query_cgi, $query_desc,
253 $limit, $limit_cgi, $limit_desc, undef, undef
257 =head2 build_authorities_query
259 my $query = $builder->build_authorities_query(\%search);
261 This takes a nice description of an authority search and turns it into a black-box
262 query that can then be passed to the appropriate searcher.
264 The search description is a hashref that looks something like:
269 where => 'Heading', # search the main entry
270 operator => 'exact', # require an exact match
271 value => 'frogs', # the search string
274 where => '', # search all entries
275 operator => '', # default keyword, right truncation
283 authtypecode => 'TOPIC_TERM',
288 sub build_authorities_query {
289 my ( $self, $search ) = @_;
291 # Start by making the query parts
294 foreach my $s ( @{ $search->{searches} } ) {
295 my ( $wh, $op, $val ) = @{$s}{qw(where operator value)};
296 $wh = '_all' if $wh eq '';
297 if ( $op eq 'is' || $op eq '=' ) {
299 # look for something that matches completely
300 # note, '=' is about numerical vals. May need special handling.
301 # _allphrase is a special field that only groups the exact
302 # matches. Also, we lowercase our search because the ES
303 # index lowercases its values, and term searches don't get the
304 # search analyzer applied to them.
305 push @filter_parts, { term => { "$wh.phrase" => lc $val } };
307 elsif ( $op eq 'exact' ) {
309 # left and right truncation, otherwise an exact phrase
310 push @query_parts, { match_phrase => { $wh => $val } };
312 elsif ( $op eq 'start' ) {
315 push @query_parts, { wildcard => { "$wh.phrase" => lc "$val*" } };
318 # regular wordlist stuff
319 push @query_parts, { match => { $wh => $val } };
323 # Merge the query and filter parts appropriately
324 # 'should' behaves like 'or', if we want 'and', use 'must'
325 my $query_part = { bool => { should => \@query_parts } };
326 my $filter_part = { bool => { should => \@filter_parts } };
328 # We need to add '.phrase' to all the sort headings otherwise it'll sort
329 # based on the tokenised form.
331 if ( exists $search->{sort} ) {
332 foreach my $k ( keys %{ $search->{sort} } ) {
333 my $f = $self->_sort_field($k);
334 $s{"$f.phrase"} = $search->{sort}{$k};
336 $search->{sort} = \%s;
339 # extract the sort stuff
341 %sort = ( sort => [ $search->{sort} ] ) if exists $search->{sort};
346 { filtered => { filter => $filter_part, query => $query_part } }
350 $query = { query => $query_part };
352 $query = { %$query, %sort };
357 =head2 build_authorities_query_compat
360 $builder->build_authorities_query_compat( \@marclist, \@and_or,
361 \@excluding, \@operator, \@value, $authtypecode, $orderby );
363 This builds a query for searching for authorities, in the style of
364 L<C4::AuthoritiesMarc::SearchAuthorities>.
372 An arrayref containing where the particular term should be searched for.
373 Options are: mainmainentry, mainentry, match, match-heading, see-from, and
374 thesaurus. If left blank, any field is used.
378 Totally ignored. It is never used in L<C4::AuthoritiesMarc::SearchAuthorities>.
386 What form of search to do. Options are: is (phrase, no trunction, whole field
387 must match), = (number exact match), exact (phrase, but with left and right
388 truncation). If left blank, then word list, right truncted, anywhere is used.
392 The actual user-provided string value to search for.
396 The authority type code to search within. If blank, then all will be searched.
400 The order to sort the results by. Options are Relevance, HeadingAsc,
401 HeadingDsc, AuthidAsc, AuthidDsc.
405 marclist, operator, and value must be the same length, and the values at
406 index /i/ all relate to each other.
408 This returns a query, which is a black box object that can be passed to the
409 appropriate search object.
413 our $koha_to_index_name = {
414 mainmainentry => 'Heading-Main',
415 mainentry => 'Heading',
417 'match-heading' => 'Match-heading',
418 'see-from' => 'Match-heading-see-from',
419 thesaurus => 'Subject-heading-thesaurus',
423 sub build_authorities_query_compat {
424 my ( $self, $marclist, $and_or, $excluding, $operator, $value,
425 $authtypecode, $orderby )
428 # This turns the old-style many-options argument form into a more
429 # extensible hash form that is understood by L<build_authorities_query>.
432 # Make sure everything exists
433 foreach my $m (@$marclist) {
434 Koha::Exceptions::WrongParameter->throw("Invalid marclist field provided: $m")
435 unless exists $koha_to_index_name->{$m};
437 for ( my $i = 0 ; $i < @$value ; $i++ ) {
438 next unless $value->[$i]; #clean empty form values, ES doesn't like undefined searches
441 where => $koha_to_index_name->{$marclist->[$i]},
442 operator => $operator->[$i],
443 value => $value->[$i],
449 ( $orderby =~ /^Heading/ ) ? 'Heading'
450 : ( $orderby =~ /^Auth/ ) ? 'Local-Number'
453 my $sort_order = ( $orderby =~ /Asc$/ ) ? 'asc' : 'desc';
454 %sort = ( $sort_field => $sort_order, );
457 searches => \@searches,
458 authtypecode => $authtypecode,
460 $search{sort} = \%sort if %sort;
461 my $query = $self->build_authorities_query( \%search );
465 =head2 _convert_sort_fields
467 my @sort_params = _convert_sort_fields(@sort_by)
469 Converts the zebra-style sort index information into elasticsearch-style.
471 C<@sort_by> is the same as presented to L<build_query_compat>, and it returns
472 something that can be sent to L<build_query>.
476 sub _convert_sort_fields {
477 my ( $self, @sort_by ) = @_;
479 # Turn the sorting into something we care about.
480 my %sort_field_convert = (
481 acqdate => 'acqdate',
483 call_number => 'callnum',
484 popularity => 'issues',
485 relevance => undef, # default
487 pubdate => 'pubdate',
489 my %sort_order_convert =
490 ( qw( dsc desc ), qw( asc asc ), qw( az asc ), qw( za desc ) );
492 # Convert the fields and orders, drop anything we don't know about.
493 grep { $_->{field} } map {
494 my ( $f, $d ) = split /_/;
496 field => $sort_field_convert{$f},
497 direction => $sort_order_convert{$d}
502 =head2 _convert_index_fields
504 my @index_params = $self->_convert_index_fields(@indexes);
506 Converts zebra-style search index notation into elasticsearch-style.
508 C<@indexes> is an array of index names, as presented to L<build_query_compat>,
509 and it returns something that can be sent to L<build_query>.
511 B<TODO>: this will pull from the elasticsearch mappings table to figure out
516 our %index_field_convert = (
522 'se' => 'title-series',
523 'callnum' => 'callnum',
526 'branch' => 'homebranch',
530 'hi' => 'Host-Item-Number',
533 sub _convert_index_fields {
534 my ( $self, @indexes ) = @_;
536 my %index_type_convert =
537 ( __default => undef, phr => 'phrase', rtrn => 'right-truncate' );
539 # Convert according to our table, drop anything that doesn't convert.
540 # If a field starts with mc- we save it as it's used (and removed) later
541 # when joining things, to indicate we make it an 'OR' join.
542 # (Sorry, this got a bit ugly after special cases were found.)
543 grep { $_->{field} } map {
544 my ( $f, $t ) = split /,/;
551 field => $index_field_convert{$f},
552 type => $index_type_convert{ $t // '__default' }
554 $r->{field} = ($mc . $r->{field}) if $mc && $r->{field};
559 =head2 _convert_index_strings
561 my @searches = $self->_convert_index_strings(@searches);
563 Similar to L<_convert_index_fields>, this takes strings of the form
564 B<field:search term> and rewrites the field from zebra-style to
565 elasticsearch-style. Anything it doesn't understand is returned verbatim.
569 sub _convert_index_strings {
570 my ( $self, @searches ) = @_;
572 foreach my $s (@searches) {
574 my ( $field, $term ) = $s =~ /^\s*([\w,-]*?):(.*)/;
575 unless ( defined($field) && defined($term) ) {
579 my ($conv) = $self->_convert_index_fields($field);
580 unless ( defined($conv) ) {
584 push @res, $conv->{field} . ":"
585 . $self->_modify_string_by_type( %$conv, operand => $term );
590 =head2 _convert_index_strings_freeform
592 my $search = $self->_convert_index_strings_freeform($search);
594 This is similar to L<_convert_index_strings>, however it'll search out the
595 things to change within the string. So it can handle strings such as
596 C<(su:foo) AND (su:bar)>, converting the C<su> appropriately.
598 If there is something of the form "su,complete-subfield" or something, the
599 second part is stripped off as we can't yet handle that. Making it work
600 will have to wait for a real query parser.
604 sub _convert_index_strings_freeform {
605 my ( $self, $search ) = @_;
606 while ( my ( $zeb, $es ) = each %index_field_convert ) {
607 $search =~ s/\b$zeb(?:,[\w\-]*)?:/$es:/g;
612 =head2 _modify_string_by_type
614 my $str = $self->_modify_string_by_type(%index_field);
616 If you have a search term (operand) and a type (phrase, right-truncated), this
617 will convert the string to have the function in lucene search terms, e.g.
618 wrapping quotes around it.
622 sub _modify_string_by_type {
623 my ( $self, %idx ) = @_;
625 my $type = $idx{type} || '';
626 my $str = $idx{operand};
627 return $str unless $str; # Empty or undef, we can't use it.
629 $str .= '*' if $type eq 'right-truncate';
630 $str = '"' . $str . '"' if $type eq 'phrase';
636 my $query_str = $self->_join_queries(@query_parts);
638 This takes a list of query parts, that might be search terms on their own, or
639 booleaned together, or specifying fields, or whatever, wraps them in
640 parentheses, and ANDs them all together. Suitable for feeding to the ES
643 Note: doesn't AND them together if they specify an index that starts with "mc"
644 as that was a special case in the original code for dealing with multiple
645 choice options (you can't search for something that has an itype of A and
646 and itype of B otherwise.)
651 my ( $self, @parts ) = @_;
653 my @norm_parts = grep { defined($_) && $_ ne '' && $_ !~ /^mc-/ } @parts;
655 map { s/^mc-//r } grep { defined($_) && $_ ne '' && $_ =~ /^mc-/ } @parts;
656 return () unless @norm_parts + @mc_parts;
657 return ( @norm_parts, @mc_parts )[0] if @norm_parts + @mc_parts == 1;
659 @mc_parts ? '(' . ( join ' OR ', map { "($_)" } @mc_parts ) . ')' : ();
661 # Handy trick: $x || () inside a join means that if $x ends up as an
662 # empty string, it gets replaced with (), which makes join ignore it.
663 # (bad effect: this'll also happen to '0', this hopefully doesn't matter
666 join( ' AND ', map { "($_)" } @norm_parts ) || (),
672 my @phrased_queries = $self->_make_phrases(@query_parts);
674 This takes the supplied queries and forces them to be phrases by wrapping
675 quotes around them. It understands field prefixes, e.g. 'subject:' and puts
676 the quotes outside of them if they're there.
681 my ( $self, @parts ) = @_;
682 map { s/^\s*(\w*?:)(.*)$/$1"$2"/r } @parts;
685 =head2 _create_query_string
687 my @query_strings = $self->_create_query_string(@queries);
689 Given a list of hashrefs, it will turn them into a lucene-style query string.
690 The hash should contain field, type (both for the indexes), operator, and
695 sub _create_query_string {
696 my ( $self, @queries ) = @_;
699 my $otor = $_->{operator} ? $_->{operator} . ' ' : '';
700 my $field = $_->{field} ? $_->{field} . ':' : '';
702 my $oand = $self->_modify_string_by_type(%$_);
703 "$otor($field$oand)";
707 =head2 _clean_search_term
709 my $term = $self->_clean_search_term($term);
711 This cleans a search term by removing any funny characters that may upset
712 ES and give us an error. It also calls L<_convert_index_strings_freeform>
713 to ensure those parts are correct.
717 sub _clean_search_term {
718 my ( $self, $term ) = @_;
720 my $auto_truncation = C4::Context->preference("QueryAutoTruncate") || 0;
722 # Some hardcoded searches (like with authorities) produce things like
723 # 'an=123', when it ought to be 'an:123' for our purposes.
725 $term = $self->_convert_index_strings_freeform($term);
727 $term = $self->_truncate_terms($term) if ($auto_truncation);
731 =head2 _fix_limit_special_cases
733 my $limits = $self->_fix_limit_special_cases($limits);
735 This converts any special cases that the limit specifications have into things
736 that are more readily processable by the rest of the code.
738 The argument should be an arrayref, and it'll return an arrayref.
742 sub _fix_limit_special_cases {
743 my ( $self, $limits ) = @_;
746 foreach my $l (@$limits) {
748 # This is set up by opac-search.pl
749 if ( $l =~ /^yr,st-numeric,ge=/ ) {
750 my ( $start, $end ) =
751 ( $l =~ /^yr,st-numeric,ge=(.*) and yr,st-numeric,le=(.*)$/ );
752 next unless defined($start) && defined($end);
753 push @new_lim, "copydate:[$start TO $end]";
755 elsif ( $l =~ /^yr,st-numeric=/ ) {
756 my ($date) = ( $l =~ /^yr,st-numeric=(.*)$/ );
757 next unless defined($date);
758 push @new_lim, "copydate:$date";
760 elsif ( $l =~ /^available$/ ) {
761 push @new_lim, 'onloan:0';
772 my $field = $self->_sort_field($field);
774 Given a field name, this works out what the actual name of the version to sort
775 on should be. Often it's the same, sometimes it involves sticking "__sort" on
776 the end. Maybe it'll be something else in the future, who knows?
782 if ($self->sort_fields()->{$f}) {
788 =head2 _truncate_terms
790 my $query = $self->_truncate_terms($query);
792 Given a string query this function appends '*' wildcard to all terms except
793 operands and double quoted strings.
797 sub _truncate_terms {
798 my ( $self, $query ) = @_;
800 # '"donald duck" title:"the mouse" and peter" get split into
801 # ['', '"donald duck"', '', ' ', '', 'title:"the mouse"', '', ' ', 'and', ' ', 'pete']
802 my @tokens = split /((?:[\w\-.]+:)?"[^"]+"|\s+)/, $query;
804 # Filter out empty tokens
805 my @words = grep { $_ !~ /^\s*$/ } @tokens;
807 # Append '*' to words if needed, ie. if it's not surrounded by quotes, not
808 # terminated by '*' and not a keyword
811 (/"$/ or /\*$/ or grep {lc($w) eq $_} qw/and or not/) ? $_ : "$_*";
814 return join ' ', @terms;