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;
33 $builder = Koha::SearchEngine::Elasticsearch->new();
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(Class::Accessor);
44 use List::MoreUtils qw/ each_array /;
48 use Data::Dumper; # TODO remove
52 my $simple_query = $builder->build_query("hello", %options)
54 This will build a query that can be issued to elasticsearch from the provided
55 string input. This expects a lucene style search form (see
56 L<http://www.elasticsearch.org/guide/en/elasticsearch/reference/current/query-dsl-query-string-query.html#query-string-syntax>
59 It'll make an attempt to respect the various query options.
61 Additional options can be provided with the C<%options> hash.
67 This should be an arrayref of hashrefs, each containing a C<field> and an
68 C<direction> (optional, defaults to C<asc>.) The results will be sorted
69 according to these values. Valid values for C<direction> are 'asc' and 'desc'.
76 my ( $self, $query, %options ) = @_;
78 my $stemming = C4::Context->preference("QueryStemming") || 0;
79 my $auto_truncation = C4::Context->preference("QueryAutoTruncate") || 0;
80 my $weight_fields = C4::Context->preference("QueryWeightFields") || 0;
81 my $fuzzy_enabled = C4::Context->preference("QueryFuzzy") || 0;
83 $query = '*' unless defined $query;
89 fuzziness => $fuzzy_enabled ? 'auto' : '0',
90 default_operator => "AND",
91 default_field => "_all",
95 if ( $options{sort} ) {
96 foreach my $sort ( @{ $options{sort} } ) {
97 my ( $f, $d ) = @$sort{qw/ field direction /};
98 die "Invalid sort direction, $d"
99 if $d && ( $d ne 'asc' && $d ne 'desc' );
100 $d = 'asc' unless $d;
102 # TODO account for fields that don't have a 'phrase' type
103 push @{ $res->{sort} }, { "$f.phrase" => { order => $d } };
107 # See _convert_facets in Search.pm for how these get turned into
108 # things that Koha can use.
110 author => { terms => { field => "author__facet" } },
111 subject => { terms => { field => "subject__facet" } },
112 itype => { terms => { field => "itype__facet" } },
117 =head2 build_browse_query
119 my $browse_query = $builder->build_browse_query($field, $query);
121 This performs a "starts with" style query on a particular field. The field
122 to be searched must have been indexed with an appropriate mapping as a
126 # XXX this isn't really a browse query like we want in the end
127 sub build_browse_query {
128 my ( $self, $field, $query ) = @_;
130 my $fuzzy_enabled = C4::Context->preference("QueryFuzzy") || 0;
132 return { query => '*' } if !defined $query;
134 # TODO this should come from Koha::Elasticsearch
135 my %field_whitelist = (
139 $field = 'title' if !exists $field_whitelist{$field};
143 match_phrase_prefix => {
147 fuzziness => $fuzzy_enabled ? 'auto' : '0',
151 sort => [ { "$field.phrase" => { order => "asc" } } ],
155 =head2 build_query_compat
158 $error, $query, $simple_query, $query_cgi,
159 $query_desc, $limit, $limit_cgi, $limit_desc,
160 $stopwords_removed, $query_type
162 = $builder->build_query_compat( \@operators, \@operands, \@indexes,
163 \@limits, \@sort_by, $scan, $lang );
165 This handles a search using the same api as L<C4::Search::buildQuery> does.
167 A very simple query will go in with C<$operands> set to ['query'], and
168 C<$sort_by> set to ['pubdate_dsc']. This simple case will return with
169 C<$query> set to something that can perform the search, C<$simple_query>
170 set to just the search term, C<$query_cgi> set to something that can
171 reproduce this search, and C<$query_desc> set to something else.
175 sub build_query_compat {
176 my ( $self, $operators, $operands, $indexes, $orig_limits, $sort_by, $scan,
180 #die Dumper ( $self, $operators, $operands, $indexes, $limits, $sort_by, $scan, $lang );
181 my @sort_params = $self->_convert_sort_fields(@$sort_by);
182 my @index_params = $self->_convert_index_fields(@$indexes);
183 my $limits = $self->_fix_limit_special_cases($orig_limits);
185 # Merge the indexes in with the search terms and the operands so that
186 # each search thing is a handy unit.
187 unshift @$operators, undef; # The first one can't have an op
189 my $ea = each_array( @$operands, @$operators, @index_params );
190 while ( my ( $oand, $otor, $index ) = $ea->() ) {
191 next if ( !defined($oand) || $oand eq '' );
192 push @search_params, {
193 operand => $self->_clean_search_term($oand), # the search terms
194 operator => defined($otor) ? uc $otor : undef, # AND and so on
195 $index ? %$index : (),
199 # We build a string query from limits and the queries. An alternative
200 # would be to pass them separately into build_query and let it build
201 # them into a structured ES query itself. Maybe later, though that'd be
203 my $query_str = join( ' AND ',
204 join( ' ', $self->_create_query_string(@search_params) ),
205 $self->_join_queries( $self->_convert_index_strings(@$limits) ) );
207 # If there's no query on the left, let's remove the junk left behind
208 $query_str =~ s/^ AND //;
210 $options{sort} = \@sort_params;
211 my $query = $self->build_query( $query_str, %options );
214 # We roughly emulate the CGI parameters of the zebra query builder
215 my $query_cgi = 'idx=kw&q=' . uri_escape( $operands->[0] ) if @$operands;
216 my $simple_query = $operands->[0] if @$operands == 1;
217 my $query_desc = $simple_query;
218 my $limit = 'and ' . join( ' and ', @$limits );
220 '&limit=' . join( '&limit=', map { uri_escape($_) } @$orig_limits );
221 my $limit_desc = "@$limits";
224 undef, $query, $simple_query, $query_cgi, $query_desc,
225 $limit, $limit_cgi, $limit_desc, undef, undef
229 =head2 build_authorities_query
231 my $query = $builder->build_authorities_query(\%search);
233 This takes a nice description of an authority search and turns it into a black-box
234 query that can then be passed to the appropriate searcher.
236 The search description is a hashref that looks something like:
241 where => 'Heading', # search the main entry
242 operator => 'exact', # require an exact match
243 value => 'frogs', # the search string
246 where => '', # search all entries
247 operator => '', # default keyword, right truncation
255 authtypecode => 'TOPIC_TERM',
260 sub build_authorities_query {
261 my ($self, $search) = @_;
263 warn "Search is " . Dumper $search;
264 # Start by making the query parts
267 foreach my $s ( @{ $search->{searches} } ) {
268 my ($wh, $op, $val) = @{ $s }{qw(where operator value)};
269 $wh = '_all' if $wh eq '';
270 if ($op eq 'is' || $op eq '=') {
271 # look for something that matches completely
272 # note, '=' is about numerical vals. May need special handling.
273 # _allphrase is a special field that only groups the exact
274 # matches. Also, we lowercase our search because the ES
275 # index lowercases its values, and term searches don't get the
276 # search analyzer applied to them.
277 push @filter_parts, { term => { "$wh.phrase" => lc $val }};
278 } elsif ($op eq 'exact') {
279 # left and right truncation, otherwise an exact phrase
280 push @query_parts, { match_phrase => { $wh => $val }};
281 } elsif ($op eq 'start') {
283 push @query_parts, { wildcard => { "$wh.phrase" => lc "$val*" }};
285 # regular wordlist stuff
286 push @query_parts, { match => { $wh => $val }};
289 # Merge the query and filter parts appropriately
290 # 'should' behaves like 'or', if we want 'and', use 'must'
291 my $query_part = { bool => { should => \@query_parts } };
292 my $filter_part = { bool => { should => \@filter_parts }};
293 # extract the sort stuff
294 my %sort = ( sort => [ $search->{sort} ] ) if exists $search->{sort};
297 $query = { query => { filtered => { filter => $filter_part, query => $query_part }}};
299 $query = { query => $query_part };
301 $query = { %$query, %sort };
306 =head2 build_authorities_query_compat
309 $builder->build_authorities_query_compat( \@marclist, \@and_or,
310 \@excluding, \@operator, \@value, $authtypecode, $orderby );
312 This builds a query for searching for authorities, in the style of
313 L<C4::AuthoritiesMarc::SearchAuthorities>.
321 An arrayref containing where the particular term should be searched for.
322 Options are: mainmainentry, mainentry, match, match-heading, see-from, and
323 thesaurus. If left blank, any field is used.
327 Totally ignored. It is never used in L<C4::AuthoritiesMarc::SearchAuthorities>.
335 What form of search to do. Options are: is (phrase, no trunction, whole field
336 must match), = (number exact match), exact (phrase, but with left and right
337 truncation). If left blank, then word list, right truncted, anywhere is used.
341 The actual user-provided string value to search for.
345 The authority type code to search within. If blank, then all will be searched.
349 The order to sort the results by. Options are Relevance, HeadingAsc,
350 HeadingDsc, AuthidAsc, AuthidDsc.
354 marclist, operator, and value must be the same length, and the values at
355 index /i/ all relate to each other.
357 This returns a query, which is a black box object that can be passed to the
358 appropriate search object.
362 sub build_authorities_query_compat {
363 my ( $self, $marclist, $and_or, $excluding, $operator, $value,
364 $authtypecode, $orderby )
367 # This turns the old-style many-options argument form into a more
368 # extensible hash form that is understood by L<build_authorities_query>.
371 my %koha_to_index_name = (
372 mainmainentry => 'Heading-Main',
373 mainentry => 'Heading',
375 'match-heading' => 'Match-heading',
376 'see-from' => 'Match-heading-see-from',
377 thesaurus => 'Subject-heading-thesaurus',
381 # Make sure everything exists
382 foreach my $m (@$marclist) {
383 confess "Invalid marclist field provided: $m" unless exists $koha_to_index_name{$m};
385 for ( my $i = 0 ; $i < @$value ; $i++ ) {
388 where => $koha_to_index_name{$marclist->[$i]},
389 operator => $operator->[$i],
390 value => $value->[$i],
396 ( $orderby =~ /^Heading/ ) ? 'Heading'
397 : ( $orderby =~ /^Auth/ ) ? 'Local-Number'
400 my $sort_order = ( $orderby =~ /Asc$/ ) ? 'asc' : 'desc';
401 %sort = ( $sort_field => $sort_order, );
404 searches => \@searches,
405 authtypecode => $authtypecode,
407 $search{sort} = \%sort if %sort;
408 my $query = $self->build_authorities_query( \%search );
412 =head2 _convert_sort_fields
414 my @sort_params = _convert_sort_fields(@sort_by)
416 Converts the zebra-style sort index information into elasticsearch-style.
418 C<@sort_by> is the same as presented to L<build_query_compat>, and it returns
419 something that can be sent to L<build_query>.
423 sub _convert_sort_fields {
424 my ( $self, @sort_by ) = @_;
426 # Turn the sorting into something we care about.
427 my %sort_field_convert = (
428 acqdate => 'acqdate',
430 call_number => 'callnum',
431 popularity => 'issues',
432 relevance => undef, # default
434 pubdate => 'pubdate',
436 my %sort_order_convert =
437 ( qw( dsc desc ), qw( asc asc ), qw( az asc ), qw( za desc ) );
439 # Convert the fields and orders, drop anything we don't know about.
440 grep { $_->{field} } map {
441 my ( $f, $d ) = split /_/;
443 field => $sort_field_convert{$f},
444 direction => $sort_order_convert{$d}
449 =head2 _convert_index_fields
451 my @index_params = $self->_convert_index_fields(@indexes);
453 Converts zebra-style search index notation into elasticsearch-style.
455 C<@indexes> is an array of index names, as presented to L<build_query_compat>,
456 and it returns something that can be sent to L<build_query>.
458 B<TODO>: this will pull from the elasticsearch mappings table to figure out
463 our %index_field_convert = (
469 'se' => 'title-series',
470 'callnum' => 'callnum',
471 'mc-itype' => 'itype',
473 'branch' => 'homebranch',
479 sub _convert_index_fields {
480 my ( $self, @indexes ) = @_;
482 my %index_type_convert =
483 ( __default => undef, phr => 'phrase', rtrn => 'right-truncate' );
485 # Convert according to our table, drop anything that doesn't convert
486 grep { $_->{field} } map {
487 my ( $f, $t ) = split /,/;
489 field => $index_field_convert{$f},
490 type => $index_type_convert{ $t // '__default' }
495 =head2 _convert_index_strings
497 my @searches = $self->_convert_index_strings(@searches);
499 Similar to L<_convert_index_fields>, this takes strings of the form
500 B<field:search term> and rewrites the field from zebra-style to
501 elasticsearch-style. Anything it doesn't understand is returned verbatim.
505 sub _convert_index_strings {
506 my ( $self, @searches ) = @_;
509 foreach my $s (@searches) {
511 my ( $field, $term ) = $s =~ /^\s*([\w,-]*?):(.*)/;
512 unless ( defined($field) && defined($term) ) {
516 my ($conv) = $self->_convert_index_fields($field);
517 unless ( defined($conv) ) {
521 push @res, $conv->{field} . ":"
522 . $self->_modify_string_by_type( %$conv, operand => $term );
527 =head2 _modify_string_by_type
529 my $str = $self->_modify_string_by_type(%index_field);
531 If you have a search term (operand) and a type (phrase, right-truncated), this
532 will convert the string to have the function in lucene search terms, e.g.
533 wrapping quotes around it.
537 sub _modify_string_by_type {
538 my ( $self, %idx ) = @_;
540 my $type = $idx{type} || '';
541 my $str = $idx{operand};
542 return $str unless $str; # Empty or undef, we can't use it.
544 $str .= '*' if $type eq 'right-truncate';
545 $str = '"' . $str . '"' if $type eq 'phrase';
549 =head2 _convert_index_strings_freeform
551 my $search = $self->_convert_index_strings_freeform($search);
553 This is similar to L<_convert_index_strings>, however it'll search out the
554 things to change within the string. So it can handle strings such as
555 C<(su:foo) AND (su:bar)>, converting the C<su> appropriately.
559 sub _convert_index_strings_freeform {
560 my ( $self, $search ) = @_;
562 while ( my ( $zeb, $es ) = each %index_field_convert ) {
563 $search =~ s/\b$zeb:/$es:/g;
570 my $query_str = $self->_join_queries(@query_parts);
572 This takes a list of query parts, that might be search terms on their own, or
573 booleaned together, or specifying fields, or whatever, wraps them in
574 parentheses, and ANDs them all together. Suitable for feeding to the ES
580 my ( $self, @parts ) = @_;
582 @parts = grep { defined($_) && $_ ne '' } @parts;
583 return () unless @parts;
584 return $parts[0] if @parts < 2;
585 join ' AND ', map { "($_)" } @parts;
590 my @phrased_queries = $self->_make_phrases(@query_parts);
592 This takes the supplied queries and forces them to be phrases by wrapping
593 quotes around them. It understands field prefixes, e.g. 'subject:' and puts
594 the quotes outside of them if they're there.
599 my ( $self, @parts ) = @_;
600 map { s/^\s*(\w*?:)(.*)$/$1"$2"/r } @parts;
603 =head2 _create_query_string
605 my @query_strings = $self->_create_query_string(@queries);
607 Given a list of hashrefs, it will turn them into a lucene-style query string.
608 The hash should contain field, type (both for the indexes), operator, and
613 sub _create_query_string {
614 my ( $self, @queries ) = @_;
617 my $otor = $_->{operator} ? $_->{operator} . ' ' : '';
618 my $field = $_->{field} ? $_->{field} . ':' : '';
620 my $oand = $self->_modify_string_by_type(%$_);
621 "$otor($field$oand)";
625 =head2 _clean_search_term
627 my $term = $self->_clean_search_term($term);
629 This cleans a search term by removing any funny characters that may upset
630 ES and give us an error. It also calls L<_convert_index_strings_freeform>
631 to ensure those parts are correct.
635 sub _clean_search_term {
636 my ( $self, $term ) = @_;
638 $term = $self->_convert_index_strings_freeform($term);
643 =head2 _fix_limit_special_cases
645 my $limits = $self->_fix_limit_special_cases($limits);
647 This converts any special cases that the limit specifications have into things
648 that are more readily processable by the rest of the code.
650 The argument should be an arrayref, and it'll return an arrayref.
654 sub _fix_limit_special_cases {
655 my ( $self, $limits ) = @_;
658 foreach my $l (@$limits) {
660 # This is set up by opac-search.pl
661 if ( $l =~ /^yr,st-numeric,ge=/ ) {
662 my ( $start, $end ) =
663 ( $l =~ /^yr,st-numeric,ge=(.*) and yr,st-numeric,le=(.*)$/ );
664 next unless defined($start) && defined($end);
665 push @new_lim, "copydate:[$start TO $end]";
667 elsif ( $l =~ /^yr,st-numeric=/ ) {
668 my ($date) = ( $l =~ /^yr,st-numeric=(.*)$/ );
669 next unless defined($date);
670 push @new_lim, "copydate:$date";
672 elsif ( $l =~ /^available$/ ) {
673 push @new_lim, 'onloan:false';