Add colon after prefix when term starts with a colon

[xapian.git] / xapian-core / queryparser / queryparser.lemony

%include {
/* queryparser.lemony: build a Xapian::Query object from a user query string.
 *
 * Copyright (C) 2004,2005,2006,2007,2008,2009,2010,2011,2012,2013,2015,2016 Olly Betts
 * Copyright (C) 2007,2008,2009 Lemur Consulting Ltd
 * Copyright (C) 2010 Adam Sjøgren
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
 * published by the Free Software Foundation; either version 2 of the
 * License, or (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301
 * USA
 */

#include <config.h>

#include "queryparser_internal.h"

#include "api/queryinternal.h"
#include "omassert.h"
#include "str.h"
#include "stringutils.h"
#include "xapian/error.h"
#include "xapian/unicode.h"

// Include the list of token values lemon generates.
#include "queryparser_token.h"

#include "cjk-tokenizer.h"

#include <algorithm>
#include <cstring>
#include <limits>
#include <list>
#include <string>
#include <vector>

using namespace std;

using namespace Xapian;

inline bool
U_isupper(unsigned ch) {
    return (ch < 128 && C_isupper(static_cast<unsigned char>(ch)));
}

inline bool
U_isdigit(unsigned ch) {
    return (ch < 128 && C_isdigit(static_cast<unsigned char>(ch)));
}

inline bool
U_isalpha(unsigned ch) {
    return (ch < 128 && C_isalpha(static_cast<unsigned char>(ch)));
}

using Xapian::Unicode::is_whitespace;

inline bool
is_not_whitespace(unsigned ch) {
    return !is_whitespace(ch);
}

using Xapian::Unicode::is_wordchar;

inline bool
is_not_wordchar(unsigned ch) {
    return !is_wordchar(ch);
}

inline bool
is_digit(unsigned ch) {
    return (Unicode::get_category(ch) == Unicode::DECIMAL_DIGIT_NUMBER);
}

// FIXME: we used to keep trailing "-" (e.g. Cl-) but it's of dubious utility
// and there's the risk of hyphens getting stuck onto the end of terms...
inline bool
is_suffix(unsigned ch) {
    return ch == '+' || ch == '#';
}

inline bool
is_double_quote(unsigned ch) {
    // We simply treat all double quotes as equivalent, which is a bit crude,
    // but it isn't clear that it would actually better to require them to
    // match up exactly.
    //
    // 0x201c is Unicode opening double quote.
    // 0x201d is Unicode closing double quote.
    return ch == '"' || ch == 0x201c || ch == 0x201d;
}

inline bool
prefix_needs_colon(const string & prefix, unsigned ch)
{
    if (!U_isupper(ch) && ch != ':') return false;
    string::size_type len = prefix.length();
    return (len > 1 && prefix[len - 1] != ':');
}

using Unicode::is_currency;

inline bool
is_positional(Xapian::Query::op op)
{
    return (op == Xapian::Query::OP_PHRASE || op == Xapian::Query::OP_NEAR);
}

class Terms;

/** Class used to pass information about a token from lexer to parser.
 *
 *  Generally an instance of this class carries term information, but it can be
 *  used for a range query, and with some operators (e.g. the distance in
 *  NEAR/3 or ADJ/3, etc).
 */
class Term {
    State * state;

  public:
    string name;
    const FieldInfo * field_info;
    string unstemmed;
    QueryParser::stem_strategy stem;
    termpos pos;
    Query query;

    Term(const string &name_, termpos pos_)
        : name(name_), stem(QueryParser::STEM_NONE), pos(pos_) { }
    explicit Term(const string &name_)
        : name(name_), stem(QueryParser::STEM_NONE), pos(0) { }
    Term(const string &name_, const FieldInfo * field_info_)
        : name(name_), field_info(field_info_),
          stem(QueryParser::STEM_NONE), pos(0) { }
    explicit Term(termpos pos_) : stem(QueryParser::STEM_NONE), pos(pos_) { }
    Term(State * state_, const string &name_, const FieldInfo * field_info_,
         const string &unstemmed_,
         QueryParser::stem_strategy stem_ = QueryParser::STEM_NONE,
         termpos pos_ = 0)
        : state(state_), name(name_), field_info(field_info_),
          unstemmed(unstemmed_), stem(stem_), pos(pos_) { }
    // For RANGE tokens.
    Term(const Xapian::Query & q, const string & grouping)
        : name(grouping), query(q) { }

    string make_term(const string & prefix) const;

    void need_positions() {
        if (stem == QueryParser::STEM_SOME) stem = QueryParser::STEM_NONE;
    }

    termpos get_termpos() const { return pos; }

    string get_grouping() const {
        return field_info->grouping;
    }

    Query * as_wildcarded_query(State * state) const;

    /** Build a query for a term at the very end of the query string when
     *  FLAG_PARTIAL is in use.
     *
     *  This query should match documents containing any terms which start with
     *  the characters specified, but should give a higher score to exact
     *  matches (since the user might have finished typing - we simply don't
     *  know).
     */
    Query * as_partial_query(State * state_) const;

    /** Build a query for a string of CJK characters. */
    Query * as_cjk_query() const;

    /** Handle a CJK character string in a positional context. */
    void as_positional_cjk_term(Terms * terms) const;

    /// Range query.
    Query as_range_query() const;

    Query get_query() const;

    Query get_query_with_synonyms() const;

    Query get_query_with_auto_synonyms() const;
};

/// Parser State shared between the lexer and the parser.
class State {
    QueryParser::Internal * qpi;

  public:
    Query query;
    const char * error;
    unsigned flags;

    State(QueryParser::Internal * qpi_, unsigned flags_)
        : qpi(qpi_), error(NULL), flags(flags_) { }

    string stem_term(const string &term) {
        return qpi->stemmer(term);
    }

    void add_to_stoplist(const Term * term) {
        qpi->stoplist.push_back(term->name);
    }

    void add_to_unstem(const string & term, const string & unstemmed) {
        qpi->unstem.insert(make_pair(term, unstemmed));
    }

    Term * range(const string &a, const string &b) {
        for (auto i : qpi->rangeprocs) {
            Xapian::Query range_query = (i.proc)->check_range(a, b);
            Xapian::Query::op op = range_query.get_type();
            switch (op) {
                case Xapian::Query::OP_INVALID:
                    break;
                case Xapian::Query::OP_VALUE_RANGE:
                case Xapian::Query::OP_VALUE_GE:
                case Xapian::Query::OP_VALUE_LE:
                    if (i.default_grouping) {
                        Xapian::Internal::QueryValueBase * base =
                            static_cast<Xapian::Internal::QueryValueBase*>(
                                range_query.internal.get());
                        Xapian::valueno slot = base->get_slot();
                        return new Term(range_query, str(slot));
                    }
                    // FALLTHRU
                case Xapian::Query::LEAF_TERM:
                    return new Term(range_query, i.grouping);
                default:
                    return new Term(range_query, string());
            }
        }
        return NULL;
    }

    Query::op default_op() const { return qpi->default_op; }

    bool is_stopword(const Term *term) const {
        return qpi->stopper.get() && (*qpi->stopper)(term->name);
    }

    Database get_database() const {
        return qpi->db;
    }

    const Stopper * get_stopper() const {
        return qpi->stopper.get();
    }

    size_t stoplist_size() const {
        return qpi->stoplist.size();
    }

    void stoplist_resize(size_t s) {
        qpi->stoplist.resize(s);
    }

    Xapian::termcount get_max_wildcard_expansion() const {
        return qpi->max_wildcard_expansion;
    }

    int get_max_wildcard_type() const {
        return qpi->max_wildcard_type;
    }

    Xapian::termcount get_max_partial_expansion() const {
        return qpi->max_partial_expansion;
    }

    int get_max_partial_type() const {
        return qpi->max_partial_type;
    }
};

string
Term::make_term(const string & prefix) const
{
    string term;
    if (stem == QueryParser::STEM_SOME || stem == QueryParser::STEM_ALL_Z)
        term += 'Z';
    if (!prefix.empty()) {
        term += prefix;
        if (prefix_needs_colon(prefix, name[0])) term += ':';
    }
    if (stem != QueryParser::STEM_NONE) {
        term += state->stem_term(name);
    } else {
        term += name;
    }

    if (!unstemmed.empty())
        state->add_to_unstem(term, unstemmed);
    return term;
}

// Iterator shim to allow building a synonym query from a TermIterator pair.
class SynonymIterator {
    Xapian::TermIterator i;

    Xapian::termpos pos;

    const Xapian::Query * first;

  public:
    SynonymIterator(const Xapian::TermIterator & i_,
                    Xapian::termpos pos_ = 0,
                    const Xapian::Query * first_ = NULL)
        : i(i_), pos(pos_), first(first_) { }

    SynonymIterator & operator++() {
        if (first)
            first = NULL;
        else
            ++i;
        return *this;
    }

    const Xapian::Query operator*() const {
        if (first) return *first;
        return Xapian::Query(*i, 1, pos);
    }

    bool operator==(const SynonymIterator & o) const {
        return i == o.i && first == o.first;
    }

    bool operator!=(const SynonymIterator & o) const {
        return !(*this == o);
    }

    typedef std::input_iterator_tag iterator_category;
    typedef Xapian::Query value_type;
    typedef Xapian::termcount_diff difference_type;
    typedef Xapian::Query * pointer;
    typedef Xapian::Query & reference;
};

Query
Term::get_query_with_synonyms() const
{
    // Handle single-word synonyms with each prefix.
    const list<string> & prefixes = field_info->prefixes;
    if (prefixes.empty()) {
        // FIXME: handle multiple here
        Assert(!field_info->procs.empty());
        return (**field_info->procs.begin())(name);
    }

    Query q = get_query();

    list<string>::const_iterator piter;
    for (piter = prefixes.begin(); piter != prefixes.end(); ++piter) {
        // First try the unstemmed term:
        string term;
        if (!piter->empty()) {
            term += *piter;
            if (prefix_needs_colon(*piter, name[0])) term += ':';
        }
        term += name;

        Xapian::Database db = state->get_database();
        Xapian::TermIterator syn = db.synonyms_begin(term);
        Xapian::TermIterator end = db.synonyms_end(term);
        if (syn == end && stem != QueryParser::STEM_NONE) {
            // If that has no synonyms, try the stemmed form:
            term = 'Z';
            if (!piter->empty()) {
                term += *piter;
                if (prefix_needs_colon(*piter, name[0])) term += ':';
            }
            term += state->stem_term(name);
            syn = db.synonyms_begin(term);
            end = db.synonyms_end(term);
        }
        q = Query(q.OP_SYNONYM,
                  SynonymIterator(syn, pos, &q),
                  SynonymIterator(end));
    }
    return q;
}

Query
Term::get_query_with_auto_synonyms() const
{
    const unsigned MASK_ENABLE_AUTO_SYNONYMS =
        QueryParser::FLAG_AUTO_SYNONYMS |
        QueryParser::FLAG_AUTO_MULTIWORD_SYNONYMS;
    if (state->flags & MASK_ENABLE_AUTO_SYNONYMS)
        return get_query_with_synonyms();

    return get_query();
}

static void
add_to_query(Query *& q, Query::op op, Query * term)
{
    Assert(term);
    if (q) {
        *q = Query(op, *q, *term);
        delete term;
    } else {
        q = term;
    }
}

static void
add_to_query(Query *& q, Query::op op, const Query & term)
{
    if (q) {
        *q = Query(op, *q, term);
    } else {
        q = new Query(term);
    }
}

Query
Term::get_query() const
{
    const list<string> & prefixes = field_info->prefixes;
    if (prefixes.empty()) {
        // FIXME: handle multiple here
        Assert(!field_info->procs.empty());
        return (**field_info->procs.begin())(name);
    }
    list<string>::const_iterator piter = prefixes.begin();
    Query q(make_term(*piter), 1, pos);
    while (++piter != prefixes.end()) {
        q = Query(Query::OP_OR, q, Query(make_term(*piter), 1, pos));
    }
    return q;
}

Query *
Term::as_wildcarded_query(State * state_) const
{
    const list<string> & prefixes = field_info->prefixes;
    list<string>::const_iterator piter;
    Xapian::termcount max = state_->get_max_wildcard_expansion();
    int max_type = state_->get_max_wildcard_type();
    vector<Query> subqs;
    subqs.reserve(prefixes.size());
    for (piter = prefixes.begin(); piter != prefixes.end(); ++piter) {
        string root = *piter;
        root += name;
        // Combine with OP_OR, and apply OP_SYNONYM afterwards.
        subqs.push_back(Query(Query::OP_WILDCARD, root, max, max_type,
                              Query::OP_OR));
    }
    Query * q = new Query(Query::OP_SYNONYM, subqs.begin(), subqs.end());
    delete this;
    return q;
}

Query *
Term::as_partial_query(State * state_) const
{
    Xapian::termcount max = state_->get_max_partial_expansion();
    int max_type = state_->get_max_partial_type();
    vector<Query> subqs_partial; // A synonym of all the partial terms.
    vector<Query> subqs_full; // A synonym of all the full terms.

    const list<string> & prefixes = field_info->prefixes;
    list<string>::const_iterator piter;
    for (piter = prefixes.begin(); piter != prefixes.end(); ++piter) {
        string root = *piter;
        root += name;
        // Combine with OP_OR, and apply OP_SYNONYM afterwards.
        subqs_partial.push_back(Query(Query::OP_WILDCARD, root, max, max_type,
                                      Query::OP_OR));
        // Add the term, as it would normally be handled, as an alternative.
        subqs_full.push_back(Query(make_term(*piter), 1, pos));
    }
    Query * q = new Query(Query::OP_OR,
                          Query(Query::OP_SYNONYM,
                                subqs_partial.begin(), subqs_partial.end()),
                          Query(Query::OP_SYNONYM,
                                subqs_full.begin(), subqs_full.end()));
    delete this;
    return q;
}

Query *
Term::as_cjk_query() const
{
    vector<Query> prefix_subqs;
    vector<Query> cjk_subqs;
    const list<string> & prefixes = field_info->prefixes;
    list<string>::const_iterator piter;
    for (piter = prefixes.begin(); piter != prefixes.end(); ++piter) {
        const string& prefix = *piter;
        for (CJKTokenIterator tk(name); tk != CJKTokenIterator(); ++tk) {
            cjk_subqs.push_back(Query(prefix + *tk, 1, pos));
        }
        prefix_subqs.push_back(Query(Query::OP_AND,
                                     cjk_subqs.begin(), cjk_subqs.end()));
        cjk_subqs.clear();
    }
    Query * q = new Query(Query::OP_OR,
                          prefix_subqs.begin(), prefix_subqs.end());
    delete this;
    return q;
}

Query
Term::as_range_query() const
{
    Query q = query;
    delete this;
    return q;
}

inline bool
is_phrase_generator(unsigned ch)
{
    // These characters generate a phrase search.
    // Ordered mostly by frequency of calls to this function done when
    // running the testcases in api_queryparser.cc.
    return (ch && ch < 128 && strchr(".-/:\\@", ch) != NULL);
}

inline bool
is_stem_preventer(unsigned ch)
{
    return (ch && ch < 128 && strchr("(/\\@<>=*[{\"", ch) != NULL);
}

inline bool
should_stem(const string & term)
{
    const unsigned int SHOULD_STEM_MASK =
        (1 << Unicode::LOWERCASE_LETTER) |
        (1 << Unicode::TITLECASE_LETTER) |
        (1 << Unicode::MODIFIER_LETTER) |
        (1 << Unicode::OTHER_LETTER);
    Utf8Iterator u(term);
    return ((SHOULD_STEM_MASK >> Unicode::get_category(*u)) & 1);
}

/** Value representing "ignore this" when returned by check_infix() or
 *  check_infix_digit().
 */
const unsigned UNICODE_IGNORE = numeric_limits<unsigned>::max();

inline unsigned check_infix(unsigned ch) {
    if (ch == '\'' || ch == '&' || ch == 0xb7 || ch == 0x5f4 || ch == 0x2027) {
        // Unicode includes all these except '&' in its word boundary rules,
        // as well as 0x2019 (which we handle below) and ':' (for Swedish
        // apparently, but we ignore this for now as it's problematic in
        // real world cases).
        return ch;
    }
    if (ch >= 0x200b) {
        // 0x2019 is Unicode apostrophe and single closing quote.
        // 0x201b is Unicode single opening quote with the tail rising.
        if (ch == 0x2019 || ch == 0x201b)
            return '\'';
        if (ch <= 0x200d || ch == 0x2060 || ch == 0xfeff)
            return UNICODE_IGNORE;
    }
    return 0;
}

inline unsigned check_infix_digit(unsigned ch) {
    // This list of characters comes from Unicode's word identifying algorithm.
    switch (ch) {
        case ',':
        case '.':
        case ';':
        case 0x037e: // GREEK QUESTION MARK
        case 0x0589: // ARMENIAN FULL STOP
        case 0x060D: // ARABIC DATE SEPARATOR
        case 0x07F8: // NKO COMMA
        case 0x2044: // FRACTION SLASH
        case 0xFE10: // PRESENTATION FORM FOR VERTICAL COMMA
        case 0xFE13: // PRESENTATION FORM FOR VERTICAL COLON
        case 0xFE14: // PRESENTATION FORM FOR VERTICAL SEMICOLON
            return ch;
    }
    if (ch >= 0x200b && (ch <= 0x200d || ch == 0x2060 || ch == 0xfeff))
        return UNICODE_IGNORE;
    return 0;
}

struct yyParser;

// Prototype the functions lemon generates.
static yyParser *ParseAlloc();
static void ParseFree(yyParser *);
static void Parse(yyParser *, int, Term *, State *);
static void yy_parse_failed(yyParser *);

void
QueryParser::Internal::add_prefix(const string &field, const string &prefix)
{
    map<string, FieldInfo>::iterator p = field_map.find(field);
    if (p == field_map.end()) {
        field_map.insert(make_pair(field, FieldInfo(NON_BOOLEAN, prefix)));
    } else {
        // Check that this is the same type of filter as the existing one(s).
        if (p->second.type != NON_BOOLEAN) {
            throw Xapian::InvalidOperationError("Can't use add_prefix() and add_boolean_prefix() on the same field name, or add_boolean_prefix() with different values of the 'exclusive' parameter");
        }
        if (!p->second.procs.empty())
            throw Xapian::FeatureUnavailableError("Mixing FieldProcessor objects and string prefixes currently not supported");
        p->second.prefixes.push_back(prefix);
   }
}

void
QueryParser::Internal::add_prefix(const string &field, FieldProcessor *proc)
{
    map<string, FieldInfo>::iterator p = field_map.find(field);
    if (p == field_map.end()) {
        field_map.insert(make_pair(field, FieldInfo(NON_BOOLEAN, proc)));
    } else {
        // Check that this is the same type of filter as the existing one(s).
        if (p->second.type != NON_BOOLEAN) {
            throw Xapian::InvalidOperationError("Can't use add_prefix() and add_boolean_prefix() on the same field name, or add_boolean_prefix() with different values of the 'exclusive' parameter");
        }
        if (!p->second.prefixes.empty())
            throw Xapian::FeatureUnavailableError("Mixing FieldProcessor objects and string prefixes currently not supported");
        throw Xapian::FeatureUnavailableError("Multiple FieldProcessor objects for the same prefix currently not supported");
        // p->second.procs.push_back(proc);
   }
}

void
QueryParser::Internal::add_boolean_prefix(const string &field,
                                          const string &prefix,
                                          const string* grouping)
{
    // Don't allow the empty prefix to be set as boolean as it doesn't
    // really make sense.
    if (field.empty())
        throw Xapian::UnimplementedError("Can't set the empty prefix to be a boolean filter");
    if (!grouping) grouping = &field;
    filter_type type = grouping->empty() ? BOOLEAN : BOOLEAN_EXCLUSIVE;
    map<string, FieldInfo>::iterator p = field_map.find(field);
    if (p == field_map.end()) {
        field_map.insert(make_pair(field, FieldInfo(type, prefix, *grouping)));
    } else {
        // Check that this is the same type of filter as the existing one(s).
        if (p->second.type != type) {
            throw Xapian::InvalidOperationError("Can't use add_prefix() and add_boolean_prefix() on the same field name, or add_boolean_prefix() with different values of the 'exclusive' parameter"); // FIXME
        }
        if (!p->second.procs.empty())
            throw Xapian::FeatureUnavailableError("Mixing FieldProcessor objects and string prefixes currently not supported");
        p->second.prefixes.push_back(prefix); // FIXME grouping
   }
}

void
QueryParser::Internal::add_boolean_prefix(const string &field,
                                          FieldProcessor *proc,
                                          const string* grouping)
{
    // Don't allow the empty prefix to be set as boolean as it doesn't
    // really make sense.
    if (field.empty())
        throw Xapian::UnimplementedError("Can't set the empty prefix to be a boolean filter");
    if (!grouping) grouping = &field;
    filter_type type = grouping->empty() ? BOOLEAN : BOOLEAN_EXCLUSIVE;
    map<string, FieldInfo>::iterator p = field_map.find(field);
    if (p == field_map.end()) {
        field_map.insert(make_pair(field, FieldInfo(type, proc, *grouping)));
    } else {
        // Check that this is the same type of filter as the existing one(s).
        if (p->second.type != type) {
            throw Xapian::InvalidOperationError("Can't use add_prefix() and add_boolean_prefix() on the same field name, or add_boolean_prefix() with different values of the 'exclusive' parameter"); // FIXME
        }
        if (!p->second.prefixes.empty())
            throw Xapian::FeatureUnavailableError("Mixing FieldProcessor objects and string prefixes currently not supported");
        throw Xapian::FeatureUnavailableError("Multiple FieldProcessor objects for the same prefix currently not supported");
        // p->second.procs.push_back(proc);
   }
}

string
QueryParser::Internal::parse_term(Utf8Iterator &it, const Utf8Iterator &end,
                                  bool cjk_ngram, bool & is_cjk_term,
                                  bool &was_acronym)
{
    string term;
    // Look for initials separated by '.' (e.g. P.T.O., U.N.C.L.E).
    // Don't worry if there's a trailing '.' or not.
    if (U_isupper(*it)) {
        string t;
        Utf8Iterator p = it;
        do {
            Unicode::append_utf8(t, *p++);
        } while (p != end && *p == '.' && ++p != end && U_isupper(*p));
        // One letter does not make an acronym!  If we handled a single
        // uppercase letter here, we wouldn't catch M&S below.
        if (t.length() > 1) {
            // Check there's not a (lower case) letter or digit
            // immediately after it.
            // FIXME: should I.B.M..P.T.O be a range search?
            if (p == end || !is_wordchar(*p)) {
                it = p;
                swap(term, t);
            }
        }
    }
    was_acronym = !term.empty();

    if (cjk_ngram && term.empty() && CJK::codepoint_is_cjk(*it)) {
        term = CJK::get_cjk(it);
        is_cjk_term = true;
    }

    if (term.empty()) {
        unsigned prevch = *it;
        Unicode::append_utf8(term, prevch);
        while (++it != end) {
            if (cjk_ngram && CJK::codepoint_is_cjk(*it)) break;
            unsigned ch = *it;
            if (!is_wordchar(ch)) {
                // Treat a single embedded '&' or "'" or similar as a word
                // character (e.g. AT&T, Fred's).  Also, normalise
                // apostrophes to ASCII apostrophe.
                Utf8Iterator p = it;
                ++p;
                if (p == end || !is_wordchar(*p)) break;
                unsigned nextch = *p;
                if (is_digit(prevch) && is_digit(nextch)) {
                    ch = check_infix_digit(ch);
                } else {
                    ch = check_infix(ch);
                }
                if (!ch) break;
                if (ch == UNICODE_IGNORE)
                    continue;
            }
            Unicode::append_utf8(term, ch);
            prevch = ch;
        }
        if (it != end && is_suffix(*it)) {
            string suff_term = term;
            Utf8Iterator p = it;
            // Keep trailing + (e.g. C++, Na+) or # (e.g. C#).
            do {
                if (suff_term.size() - term.size() == 3) {
                    suff_term.resize(0);
                    break;
                }
                suff_term += *p;
            } while (is_suffix(*++p));
            if (!suff_term.empty() && (p == end || !is_wordchar(*p))) {
                // If the suffixed term doesn't exist, check that the
                // non-suffixed term does.  This also takes care of
                // the case when QueryParser::set_database() hasn't
                // been called.
                bool use_suff_term = false;
                string lc = Unicode::tolower(suff_term);
                if (db.term_exists(lc)) {
                    use_suff_term = true;
                } else {
                    lc = Unicode::tolower(term);
                    if (!db.term_exists(lc)) use_suff_term = true;
                }
                if (use_suff_term) {
                    term = suff_term;
                    it = p;
                }
            }
        }
    }
    return term;
}

class ParserHandler {
    yyParser * parser;

  public:
    explicit ParserHandler(yyParser * parser_) : parser(parser_) { }
    operator yyParser*() const { return parser; }
    ~ParserHandler() { ParseFree(parser); }
};

Query
QueryParser::Internal::parse_query(const string &qs, unsigned flags,
                                   const string &default_prefix)
{
    bool cjk_ngram = (flags & FLAG_CJK_NGRAM) || CJK::is_cjk_enabled();

    // Set ranges if we may have to handle ranges in the query.
    bool ranges = !rangeprocs.empty() && (qs.find("..") != string::npos);

    termpos term_pos = 1;
    Utf8Iterator it(qs), end;

    State state(this, flags);

    // To successfully apply more than one spelling correction to a query
    // string, we must keep track of the offset due to previous corrections.
    int correction_offset = 0;
    corrected_query.resize(0);

    // Stack of prefixes, used for phrases and subexpressions.
    list<const FieldInfo *> prefix_stack;

    // If default_prefix is specified, use it.  Otherwise, use any list
    // that has been set for the empty prefix.
    const FieldInfo def_pfx(NON_BOOLEAN, default_prefix);
    {
        const FieldInfo * default_field_info = &def_pfx;
        if (default_prefix.empty()) {
            auto f = field_map.find(string());
            if (f != field_map.end()) default_field_info = &(f->second);
        }

        // We always have the current prefix on the top of the stack.
        prefix_stack.push_back(default_field_info);
    }

    ParserHandler pParser(ParseAlloc());

    unsigned newprev = ' ';
main_lex_loop:
    enum {
        DEFAULT, IN_QUOTES, IN_PREFIXED_QUOTES, IN_PHRASED_TERM, IN_GROUP,
        IN_GROUP2, EXPLICIT_SYNONYM
    } mode = DEFAULT;
    while (it != end && !state.error) {
        bool last_was_operator = false;
        bool last_was_operator_needing_term = false;
        if (mode == EXPLICIT_SYNONYM) mode = DEFAULT;
        if (false) {
just_had_operator:
            if (it == end) break;
            mode = DEFAULT;
            last_was_operator_needing_term = false;
            last_was_operator = true;
        }
        if (false) {
just_had_operator_needing_term:
            last_was_operator_needing_term = true;
            last_was_operator = true;
        }
        if (mode == IN_PHRASED_TERM) mode = DEFAULT;
        if (is_whitespace(*it)) {
            newprev = ' ';
            ++it;
            it = find_if(it, end, is_not_whitespace);
            if (it == end) break;
        }

        if (ranges &&
            (mode == DEFAULT || mode == IN_GROUP || mode == IN_GROUP2)) {
            // Scan forward to see if this could be the "start of range"
            // token.  Sadly this has O(n^2) tendencies, though at least
            // "n" is the number of words in a query which is likely to
            // remain fairly small.  FIXME: can we tokenise more elegantly?
            Utf8Iterator it_initial = it;
            Utf8Iterator p = it;
            unsigned ch = 0;
            while (p != end) {
                if (ch == '.' && *p == '.') {
                    string a;
                    while (it != p) {
                        Unicode::append_utf8(a, *it++);
                    }
                    // Trim off the trailing ".".
                    a.resize(a.size() - 1);
                    ++p;
                    // Either end of the range can be empty (for an open-ended
                    // range) but both can't be empty.
                    if (!a.empty() || (p != end && *p > ' ' && *p != ')')) {
                        string b;
                        // Allow any character except whitespace and ')' in the
                        // upper bound.
                        while (p != end && *p > ' ' && *p != ')') {
                            Unicode::append_utf8(b, *p++);
                        }
                        Term * range = state.range(a, b);
                        if (!range) {
                            state.error = "Unknown range operation";
                            if (a.find(':', 1) == string::npos) {
                                goto done;
                            }
                            // Might be a boolean filter with ".." in.  Leave
                            // state.error in case it isn't.
                            it = it_initial;
                            break;
                        }
                        Parse(pParser, RANGE, range, &state);
                    }
                    it = p;
                    goto main_lex_loop;
                }
                ch = *p;
                // Allow any character except whitespace and '(' in the lower
                // bound.
                if (ch <= ' ' || ch == '(') break;
                ++p;
            }
        }

        if (!is_wordchar(*it)) {
            unsigned prev = newprev;
            unsigned ch = *it++;
            newprev = ch;
            // Drop out of IN_GROUP mode.
            if (mode == IN_GROUP || mode == IN_GROUP2)
                mode = DEFAULT;
            switch (ch) {
              case '"':
              case 0x201c: // Left curly double quote.
              case 0x201d: // Right curly double quote.
                // Quoted phrase.
                if (mode == DEFAULT) {
                    // Skip whitespace.
                    it = find_if(it, end, is_not_whitespace);
                    if (it == end) {
                        // Ignore an unmatched " at the end of the query to
                        // avoid generating an empty pair of QUOTEs which will
                        // cause a parse error.
                        goto done;
                    }
                    if (is_double_quote(*it)) {
                        // Ignore empty "" (but only if we're not already
                        // IN_QUOTES as we don't merge two adjacent quoted
                        // phrases!)
                        newprev = *it++;
                        break;
                    }
                }
                if (flags & QueryParser::FLAG_PHRASE) {
                    Parse(pParser, QUOTE, NULL, &state);
                    if (mode == DEFAULT) {
                        mode = IN_QUOTES;
                    } else {
                        // Remove the prefix we pushed for this phrase.
                        if (mode == IN_PREFIXED_QUOTES)
                            prefix_stack.pop_back();
                        mode = DEFAULT;
                    }
                }
                break;

              case '+': case '-': // Loved or hated term/phrase/subexpression.
                // Ignore + or - at the end of the query string.
                if (it == end) goto done;
                if (prev > ' ' && prev != '(') {
                    // Or if not after whitespace or an open bracket.
                    break;
                }
                if (is_whitespace(*it) || *it == '+' || *it == '-') {
                    // Ignore + or - followed by a space, or further + or -.
                    // Postfix + (such as in C++ and H+) is handled as part of
                    // the term lexing code in parse_term().
                    newprev = *it++;
                    break;
                }
                if (mode == DEFAULT && (flags & FLAG_LOVEHATE)) {
                    int token;
                    if (ch == '+') {
                        token = LOVE;
                    } else if (last_was_operator) {
                        token = HATE_AFTER_AND;
                    } else {
                        token = HATE;
                    }
                    Parse(pParser, token, NULL, &state);
                    goto just_had_operator_needing_term;
                }
                // Need to prevent the term after a LOVE or HATE starting a
                // term group...
                break;

              case '(': // Bracketed subexpression.
                // Skip whitespace.
                it = find_if(it, end, is_not_whitespace);
                // Ignore ( at the end of the query string.
                if (it == end) goto done;
                if (prev > ' ' && strchr("()+-", prev) == NULL) {
                    // Or if not after whitespace or a bracket or '+' or '-'.
                    break;
                }
                if (*it == ')') {
                    // Ignore empty ().
                    newprev = *it++;
                    break;
                }
                if (mode == DEFAULT && (flags & FLAG_BOOLEAN)) {
                    prefix_stack.push_back(prefix_stack.back());
                    Parse(pParser, BRA, NULL, &state);
                }
                break;

              case ')': // End of bracketed subexpression.
                if (mode == DEFAULT && (flags & FLAG_BOOLEAN)) {
                    // Remove the prefix we pushed for the corresponding BRA.
                    // If brackets are unmatched, it's a syntax error, but
                    // that's no excuse to SEGV!
                    if (prefix_stack.size() > 1) prefix_stack.pop_back();
                    Parse(pParser, KET, NULL, &state);
                }
                break;

              case '~': // Synonym expansion.
                // Ignore at the end of the query string.
                if (it == end) goto done;
                if (mode == DEFAULT && (flags & FLAG_SYNONYM)) {
                    if (prev > ' ' && strchr("+-(", prev) == NULL) {
                        // Or if not after whitespace, +, -, or an open bracket.
                        break;
                    }
                    if (!is_wordchar(*it)) {
                        // Ignore if not followed by a word character.
                        break;
                    }
                    Parse(pParser, SYNONYM, NULL, &state);
                    mode = EXPLICIT_SYNONYM;
                    goto just_had_operator_needing_term;
                }
                break;
            }
            // Skip any other characters.
            continue;
        }

        Assert(is_wordchar(*it));

        size_t term_start_index = it.raw() - qs.data();

        newprev = 'A'; // Any letter will do...

        // A term, a prefix, or a boolean operator.
        const FieldInfo * field_info = NULL;
        if ((mode == DEFAULT || mode == IN_GROUP || mode == IN_GROUP2 || mode == EXPLICIT_SYNONYM) &&
            !field_map.empty()) {
            // Check for a fieldname prefix (e.g. title:historical).
            Utf8Iterator p = find_if(it, end, is_not_wordchar);
            if (p != end && *p == ':' && ++p != end && *p > ' ' && *p != ')') {
                string field;
                p = it;
                while (*p != ':')
                    Unicode::append_utf8(field, *p++);
                map<string, FieldInfo>::const_iterator f;
                f = field_map.find(field);
                if (f != field_map.end()) {
                    // Special handling for prefixed fields, depending on the
                    // type of the prefix.
                    unsigned ch = *++p;
                    field_info = &(f->second);

                    if (field_info->type != NON_BOOLEAN) {
                        // Drop out of IN_GROUP if we're in it.
                        if (mode == IN_GROUP || mode == IN_GROUP2)
                            mode = DEFAULT;
                        it = p;
                        string name;
                        if (it != end && is_double_quote(*it)) {
                            // Quoted boolean term (can contain any character).
                            bool fancy = (*it != '"');
                            ++it;
                            while (it != end) {
                                if (*it == '"') {
                                    // Interpret "" as an escaped ".
                                    if (++it == end || *it != '"')
                                        break;
                                } else if (fancy && is_double_quote(*it)) {
                                    // If the opening quote was ASCII, then the
                                    // closing one must be too - otherwise
                                    // the user can't protect non-ASCII double
                                    // quote characters by quoting or escaping.
                                    ++it;
                                    break;
                                }
                                Unicode::append_utf8(name, *it++);
                            }
                        } else {
                            // Can't boolean filter prefix a subexpression, so
                            // just use anything following the prefix until the
                            // next space or ')' as part of the boolean filter
                            // term.
                            while (it != end && *it > ' ' && *it != ')')
                                Unicode::append_utf8(name, *it++);
                        }
                        // Build the unstemmed form in field.
                        field += ':';
                        field += name;
                        // Clear any pending range error.
                        state.error = NULL;
                        Term * token = new Term(&state, name, field_info, field);
                        Parse(pParser, BOOLEAN_FILTER, token, &state);
                        continue;
                    }

                    if ((flags & FLAG_PHRASE) && is_double_quote(ch)) {
                        // Prefixed phrase, e.g.: subject:"space flight"
                        mode = IN_PREFIXED_QUOTES;
                        Parse(pParser, QUOTE, NULL, &state);
                        it = p;
                        newprev = ch;
                        ++it;
                        prefix_stack.push_back(field_info);
                        continue;
                    }

                    if (ch == '(' && (flags & FLAG_BOOLEAN)) {
                        // Prefixed subexpression, e.g.: title:(fast NEAR food)
                        mode = DEFAULT;
                        Parse(pParser, BRA, NULL, &state);
                        it = p;
                        newprev = ch;
                        ++it;
                        prefix_stack.push_back(field_info);
                        continue;
                    }

                    if (ch != ':') {
                        // Allow 'path:/usr/local' but not 'foo::bar::baz'.
                        while (is_phrase_generator(ch)) {
                            if (++p == end)
                                goto not_prefix;
                            ch = *p;
                        }
                    }

                    if (is_wordchar(ch)) {
                        // Prefixed term.
                        it = p;
                    } else {
not_prefix:
                        // It looks like a prefix but isn't, so parse it as
                        // text instead.
                        field_info = NULL;
                    }
                }
            }
        }

phrased_term:
        bool was_acronym;
        bool is_cjk_term = false;
        string term = parse_term(it, end, cjk_ngram, is_cjk_term, was_acronym);

        // Boolean operators.
        if ((mode == DEFAULT || mode == IN_GROUP || mode == IN_GROUP2) &&
            (flags & FLAG_BOOLEAN) &&
            // Don't want to interpret A.N.D. as an AND operator.
            !was_acronym &&
            !field_info &&
            term.size() >= 2 && term.size() <= 4 && U_isalpha(term[0])) {

            string op = term;
            if (flags & FLAG_BOOLEAN_ANY_CASE) {
                for (string::iterator i = op.begin(); i != op.end(); ++i) {
                    *i = C_toupper(*i);
                }
            }
            if (op.size() == 3) {
                if (op == "AND") {
                    Parse(pParser, AND, NULL, &state);
                    goto just_had_operator;
                }
                if (op == "NOT") {
                    Parse(pParser, NOT, NULL, &state);
                    goto just_had_operator;
                }
                if (op == "XOR") {
                    Parse(pParser, XOR, NULL, &state);
                    goto just_had_operator;
                }
                if (op == "ADJ") {
                    if (it != end && *it == '/') {
                        size_t width = 0;
                        Utf8Iterator p = it;
                        while (++p != end && U_isdigit(*p)) {
                            width = (width * 10) + (*p - '0');
                        }
                        if (width && (p == end || is_whitespace(*p))) {
                            it = p;
                            Parse(pParser, ADJ, new Term(width), &state);
                            goto just_had_operator;
                        }
                    } else {
                        Parse(pParser, ADJ, NULL, &state);
                        goto just_had_operator;
                    }
                }
            } else if (op.size() == 2) {
                if (op == "OR") {
                    Parse(pParser, OR, NULL, &state);
                    goto just_had_operator;
                }
            } else if (op.size() == 4) {
                if (op == "NEAR") {
                    if (it != end && *it == '/') {
                        size_t width = 0;
                        Utf8Iterator p = it;
                        while (++p != end && U_isdigit(*p)) {
                            width = (width * 10) + (*p - '0');
                        }
                        if (width && (p == end || is_whitespace(*p))) {
                            it = p;
                            Parse(pParser, NEAR, new Term(width), &state);
                            goto just_had_operator;
                        }
                    } else {
                        Parse(pParser, NEAR, NULL, &state);
                        goto just_had_operator;
                    }
                }
            }
        }

        // If no prefix is set, use the default one.
        if (!field_info) field_info = prefix_stack.back();

        Assert(field_info->type == NON_BOOLEAN);

        {
            string unstemmed_term(term);
            term = Unicode::tolower(term);

            // Reuse stem_strategy - STEM_SOME here means "stem terms except
            // when used with positional operators".
            stem_strategy stem_term = stem_action;
            if (stem_term != STEM_NONE) {
                if (!stemmer.internal.get()) {
                    // No stemmer is set.
                    stem_term = STEM_NONE;
                } else if (stem_term == STEM_SOME) {
                    if (!should_stem(unstemmed_term) ||
                        (it != end && is_stem_preventer(*it))) {
                        // Don't stem this particular term.
                        stem_term = STEM_NONE;
                    }
                }
            }

            Term * term_obj = new Term(&state, term, field_info,
                                       unstemmed_term, stem_term, term_pos++);

            if (is_cjk_term) {
                Parse(pParser, CJKTERM, term_obj, &state);
                if (it == end) break;
                continue;
            }

            if (mode == DEFAULT || mode == IN_GROUP || mode == IN_GROUP2) {
                if (it != end) {
                    if ((flags & FLAG_WILDCARD) && *it == '*') {
                        Utf8Iterator p(it);
                        ++p;
                        if (p == end || !is_wordchar(*p)) {
                            it = p;
                            if (mode == IN_GROUP || mode == IN_GROUP2) {
                                // Drop out of IN_GROUP and flag that the group
                                // can be empty if all members are stopwords.
                                if (mode == IN_GROUP2)
                                    Parse(pParser, EMPTY_GROUP_OK, NULL, &state);
                                mode = DEFAULT;
                            }
                            // Wildcard at end of term (also known as
                            // "right truncation").
                            Parse(pParser, WILD_TERM, term_obj, &state);
                            continue;
                        }
                    }
                } else {
                    if (flags & FLAG_PARTIAL) {
                        if (mode == IN_GROUP || mode == IN_GROUP2) {
                            // Drop out of IN_GROUP and flag that the group
                            // can be empty if all members are stopwords.
                            if (mode == IN_GROUP2)
                                Parse(pParser, EMPTY_GROUP_OK, NULL, &state);
                            mode = DEFAULT;
                        }
                        // Final term of a partial match query, with no
                        // following characters - treat as a wildcard.
                        Parse(pParser, PARTIAL_TERM, term_obj, &state);
                        continue;
                    }
                }
            }

            // Check spelling, if we're a normal term, and any of the prefixes
            // are empty.
            if ((flags & FLAG_SPELLING_CORRECTION) && !was_acronym) {
                const list<string> & pfxes = field_info->prefixes;
                list<string>::const_iterator pfx_it;
                for (pfx_it = pfxes.begin(); pfx_it != pfxes.end(); ++pfx_it) {
                    if (!pfx_it->empty())
                        continue;
                    const string & suggest = db.get_spelling_suggestion(term);
                    if (!suggest.empty()) {
                        if (corrected_query.empty()) corrected_query = qs;
                        size_t term_end_index = it.raw() - qs.data();
                        size_t n = term_end_index - term_start_index;
                        size_t pos = term_start_index + correction_offset;
                        corrected_query.replace(pos, n, suggest);
                        correction_offset += suggest.size();
                        correction_offset -= n;
                    }
                    break;
                }
            }

            if (mode == IN_PHRASED_TERM) {
                Parse(pParser, PHR_TERM, term_obj, &state);
            } else {
                // See if the next token will be PHR_TERM - if so, this one
                // needs to be TERM not GROUP_TERM.
                if ((mode == IN_GROUP || mode == IN_GROUP2) &&
                    is_phrase_generator(*it)) {
                    // FIXME: can we clean this up?
                    Utf8Iterator p = it;
                    do {
                        ++p;
                    } while (p != end && is_phrase_generator(*p));
                    // Don't generate a phrase unless the phrase generators are
                    // immediately followed by another term.
                    if (p != end && is_wordchar(*p)) {
                        mode = DEFAULT;
                    }
                }

                int token = TERM;
                if (mode == IN_GROUP || mode == IN_GROUP2) {
                    mode = IN_GROUP2;
                    token = GROUP_TERM;
                }
                Parse(pParser, token, term_obj, &state);
                if (token == TERM && mode != DEFAULT)
                    continue;
            }
        }

        if (it == end) break;

        if (is_phrase_generator(*it)) {
            // Skip multiple phrase generators.
            do {
                ++it;
            } while (it != end && is_phrase_generator(*it));
            // Don't generate a phrase unless the phrase generators are
            // immediately followed by another term.
            if (it != end && is_wordchar(*it)) {
                mode = IN_PHRASED_TERM;
                term_start_index = it.raw() - qs.data();
                goto phrased_term;
            }
        } else if (mode == DEFAULT || mode == IN_GROUP || mode == IN_GROUP2) {
            int old_mode = mode;
            mode = DEFAULT;
            if (!last_was_operator_needing_term && is_whitespace(*it)) {
                newprev = ' ';
                // Skip multiple whitespace.
                do {
                    ++it;
                } while (it != end && is_whitespace(*it));
                // Don't generate a group unless the terms are only separated
                // by whitespace.
                if (it != end && is_wordchar(*it)) {
                    if (old_mode == IN_GROUP || old_mode == IN_GROUP2) {
                        mode = IN_GROUP2;
                    } else {
                        mode = IN_GROUP;
                    }
                }
            }
        }
    }
done:
    if (!state.error) {
        // Implicitly close any unclosed quotes.
        if (mode == IN_QUOTES || mode == IN_PREFIXED_QUOTES)
            Parse(pParser, QUOTE, NULL, &state);

        // Implicitly close all unclosed brackets.
        while (prefix_stack.size() > 1) {
            Parse(pParser, KET, NULL, &state);
            prefix_stack.pop_back();
        }
        Parse(pParser, 0, NULL, &state);
    }

    errmsg = state.error;
    return state.query;
}

struct ProbQuery {
    Query * query;
    Query * love;
    Query * hate;
    // filter is a map from prefix to a query for that prefix.  Queries with
    // the same prefix are combined with OR, and the results of this are
    // combined with AND to get the full filter.
    map<string, Query> filter;

    ProbQuery() : query(0), love(0), hate(0) { }
    ~ProbQuery() {
        delete query;
        delete love;
        delete hate;
    }

    void add_filter(const string& grouping, const Query & q) {
        filter[grouping] = q;
    }

    void append_filter(const string& grouping, const Query & qnew) {
        auto it = filter.find(grouping);
        if (it == filter.end()) {
            filter.insert(make_pair(grouping, qnew));
        } else {
            Query & q = it->second;
            // We OR multiple filters with the same prefix if they're
            // exclusive, otherwise we AND them.
            bool exclusive = !grouping.empty();
            Query::op op = exclusive ? Query::OP_OR : Query::OP_AND;
            q = Query(op, q, qnew);
        }
    }

    void add_filter_range(const string& grouping, const Query & range) {
        filter[grouping] = range;
    }

    void append_filter_range(const string& grouping, const Query & range) {
        Query & q = filter[grouping];
        q = Query(Query::OP_OR, q, range);
    }

    Query merge_filters() const {
        auto i = filter.begin();
        Assert(i != filter.end());
        Query q = i->second;
        while (++i != filter.end()) {
            q = Query(Query::OP_AND, q, i->second);
        }
        return q;
    }
};

/// A group of terms separated only by whitespace.
class TermGroup {
    vector<Term *> terms;

    /** Controls how to handle a group where all terms are stopwords.
     *
     *  If true, then as_group() returns NULL.  If false, then the
     *  stopword status of the terms is ignored.
     */
    bool empty_ok;

    TermGroup(Term* t1, Term* t2) : empty_ok(false) {
        add_term(t1);
        add_term(t2);
    }

  public:
    /// Factory function - ensures heap allocation.
    static TermGroup* create(Term* t1, Term* t2) {
        return new TermGroup(t1, t2);
    }

    ~TermGroup() {
        for (auto&& t : terms) {
            delete t;
        }
    }

    /// Add a Term object to this TermGroup object.
    void add_term(Term * term) {
        terms.push_back(term);
    }

    /// Set the empty_ok flag.
    void set_empty_ok() { empty_ok = true; }

    /// Convert to a Xapian::Query * using default_op.
    Query * as_group(State *state) const;
};

Query *
TermGroup::as_group(State *state) const
{
    const Xapian::Stopper * stopper = state->get_stopper();
    size_t stoplist_size = state->stoplist_size();
    bool default_op_is_positional = is_positional(state->default_op());
reprocess:
    Query::op default_op = state->default_op();
    vector<Query> subqs;
    subqs.reserve(terms.size());
    if (state->flags & QueryParser::FLAG_AUTO_MULTIWORD_SYNONYMS) {
        // Check for multi-word synonyms.
        Database db = state->get_database();

        string key;
        vector<Term*>::const_iterator begin = terms.begin();
        vector<Term*>::const_iterator i = begin;
        while (i != terms.end()) {
            TermIterator synkey(db.synonym_keys_begin((*i)->name));
            TermIterator synend(db.synonym_keys_end((*i)->name));
            if (synkey == synend) {
                // No multi-synonym matches.
                if (stopper && (*stopper)((*i)->name)) {
                    state->add_to_stoplist(*i);
                } else {
                    if (default_op_is_positional)
                        (*i)->need_positions();
                    subqs.push_back((*i)->get_query_with_auto_synonyms());
                }
                begin = ++i;
                continue;
            }
            key.resize(0);
            while (i != terms.end()) {
                if (!key.empty()) key += ' ';
                key += (*i)->name;
                ++i;
                synkey.skip_to(key);
                if (synkey == synend || !startswith(*synkey, key)) break;
            }
            // Greedily try to match as many consecutive words as possible.
            TermIterator syn, end;
            while (true) {
                syn = db.synonyms_begin(key);
                end = db.synonyms_end(key);
                if (syn != end) break;
                if (--i == begin) break;
                key.resize(key.size() - (*i)->name.size() - 1);
            }
            if (i == begin) {
                // No multi-synonym matches.
                if (stopper && (*stopper)((*i)->name)) {
                    state->add_to_stoplist(*i);
                } else {
                    if (default_op_is_positional)
                        (*i)->need_positions();
                    subqs.push_back((*i)->get_query_with_auto_synonyms());
                }
                begin = ++i;
                continue;
            }

            vector<Query> subqs2;
            vector<Term*>::const_iterator j;
            for (j = begin; j != i; ++j) {
                if (stopper && (*stopper)((*j)->name)) {
                    state->add_to_stoplist(*j);
                } else {
                    if (default_op_is_positional)
                        (*i)->need_positions();
                    subqs2.push_back((*j)->get_query());
                }
            }
            Query q_original_terms;
            if (default_op_is_positional) {
                q_original_terms = Query(default_op,
                                         subqs2.begin(), subqs2.end(),
                                         subqs2.size() + 9);
            } else {
                q_original_terms = Query(default_op,
                                         subqs2.begin(), subqs2.end());
            }
            subqs2.clear();

            // Use the position of the first term for the synonyms.
            Query q(Query::OP_SYNONYM,
                    SynonymIterator(syn, (*begin)->pos, &q_original_terms),
                    SynonymIterator(end));
            subqs.push_back(q);

            begin = i;
        }
    } else {
        vector<Term*>::const_iterator i;
        for (i = terms.begin(); i != terms.end(); ++i) {
            if (stopper && (*stopper)((*i)->name)) {
                state->add_to_stoplist(*i);
            } else {
                if (default_op_is_positional)
                    (*i)->need_positions();
                subqs.push_back((*i)->get_query_with_auto_synonyms());
            }
        }
    }

    if (!empty_ok && stopper && subqs.empty() &&
        stoplist_size < state->stoplist_size()) {
        // This group is all stopwords, so roll-back, disable stopper
        // temporarily, and reprocess this group.
        state->stoplist_resize(stoplist_size);
        stopper = NULL;
        goto reprocess;
    }

    Query * q = NULL;
    if (!subqs.empty()) {
        if (default_op_is_positional) {
            q = new Query(default_op, subqs.begin(), subqs.end(),
                             subqs.size() + 9);
        } else {
            q = new Query(default_op, subqs.begin(), subqs.end());
        }
    }
    delete this;
    return q;
}

/// Some terms which form a positional sub-query.
class Terms {
    vector<Term *> terms;
    size_t window;

    /** Keep track of whether the terms added all have the same list of
     *  prefixes.  If so, we'll build a set of phrases, one using each prefix.
     *  This works around the limitation that a phrase cannot have multiple
     *  components which are "OR" combinations of terms, but is also probably
     *  what users expect: i.e., if a user specifies a phrase in a field, and
     *  that field maps to multiple prefixes, the user probably wants a phrase
     *  returned with all terms having one of those prefixes, rather than a
     *  phrase comprised of terms with differing prefixes.
     */
    bool uniform_prefixes;

    /** The list of prefixes of the terms added.
     *  This will be NULL if the terms have different prefixes.
     */
    const list<string> * prefixes;

    /// Convert to a query using the given operator and window size.
    Query * as_opwindow_query(Query::op op, Xapian::termcount w_delta) const {
        Query * q = NULL;
        size_t n_terms = terms.size();
        Xapian::termcount w = w_delta + terms.size();
        if (uniform_prefixes) {
            if (prefixes) {
                list<string>::const_iterator piter;
                for (piter = prefixes->begin(); piter != prefixes->end(); ++piter) {
                    vector<Query> subqs;
                    subqs.reserve(n_terms);
                    vector<Term *>::const_iterator titer;
                    for (titer = terms.begin(); titer != terms.end(); ++titer) {
                        Term * t = *titer;
                        subqs.push_back(Query(t->make_term(*piter), 1, t->pos));
                    }
                    add_to_query(q, Query::OP_OR,
                                 Query(op, subqs.begin(), subqs.end(), w));
                }
            }
        } else {
            vector<Query> subqs;
            subqs.reserve(n_terms);
            vector<Term *>::const_iterator titer;
            for (titer = terms.begin(); titer != terms.end(); ++titer) {
                subqs.push_back((*titer)->get_query());
            }
            q = new Query(op, subqs.begin(), subqs.end(), w);
        }

        delete this;
        return q;
    }

    Terms() : window(0), uniform_prefixes(true), prefixes(NULL) { }

  public:
    /// Factory function - ensures heap allocation.
    static Terms* create() {
        return new Terms();
    }

    ~Terms() {
        for (auto&& t : terms) {
            delete t;
        }
    }

    /// Add an unstemmed Term object to this Terms object.
    void add_positional_term(Term * term) {
        const list<string> & term_prefixes = term->field_info->prefixes;
        if (terms.empty()) {
            prefixes = &term_prefixes;
        } else if (uniform_prefixes && prefixes != &term_prefixes) {
            if (*prefixes != term_prefixes)  {
                prefixes = NULL;
                uniform_prefixes = false;
            }
        }
        term->need_positions();
        terms.push_back(term);
    }

    void adjust_window(size_t alternative_window) {
        if (alternative_window > window) window = alternative_window;
    }

    /// Convert to a Xapian::Query * using adjacent OP_PHRASE.
    Query * as_phrase_query() const {
        return as_opwindow_query(Query::OP_PHRASE, 0);
    }

    /// Convert to a Xapian::Query * using OP_NEAR.
    Query * as_near_query() const {
        // The common meaning of 'a NEAR b' is "a within 10 terms of b", which
        // means a window size of 11.  For more than 2 terms, we just add one
        // to the window size for each extra term.
        size_t w = window;
        if (w == 0) w = 10;
        return as_opwindow_query(Query::OP_NEAR, w - 1);
    }

    /// Convert to a Xapian::Query * using OP_PHRASE to implement ADJ.
    Query * as_adj_query() const {
        // The common meaning of 'a ADJ b' is "a at most 10 terms before b",
        // which means a window size of 11.  For more than 2 terms, we just add
        // one to the window size for each extra term.
        size_t w = window;
        if (w == 0) w = 10;
        return as_opwindow_query(Query::OP_PHRASE, w - 1);
    }
};

void
Term::as_positional_cjk_term(Terms * terms) const
{
    // Add each individual CJK character to the phrase.
    string t;
    for (Utf8Iterator it(name); it != Utf8Iterator(); ++it) {
        Unicode::append_utf8(t, *it);
        Term * c = new Term(state, t, field_info, unstemmed, stem, pos);
        terms->add_positional_term(c);
        t.resize(0);
    }

    // FIXME: we want to add the n-grams as filters too for efficiency.

    delete this;
}

// Helper macro for converting a boolean operation into a Xapian::Query.
#define BOOL_OP_TO_QUERY(E, A, OP, B, OP_TXT) \
    do {\
        if (!A || !B) {\
            state->error = "Syntax: <expression> " OP_TXT " <expression>";\
            yy_parse_failed(yypParser);\
            return;\
        }\
        E = new Query(OP, *A, *B);\
        delete A;\
        delete B;\
    } while (0)

}

%token_type {Term *}
%token_destructor {delete $$;}

%extra_argument {State * state}

%parse_failure {
    // If we've not already set an error message, set a default one.
    if (!state->error) state->error = "parse error";
}

%syntax_error {
    yy_parse_failed(yypParser);
}

// Operators, grouped in order of increasing precedence:
%nonassoc ERROR.
%left OR.
%left XOR.
%left AND NOT.
%left NEAR ADJ.
%left LOVE HATE HATE_AFTER_AND SYNONYM.

// Destructors for terminal symbols:

// TERM is a query term, including prefix (if any).
%destructor TERM {delete $$;}

// GROUP_TERM is a query term which follows a TERM or another GROUP_TERM and
// is only separated by whitespace characters.
%destructor GROUP_TERM {delete $$;}

// PHR_TERM is a query term which follows a TERM or another PHR_TERM and is
// separated only by one or more phrase generator characters (hyphen and
// apostrophe are common examples - see is_phrase_generator() for the list
// of all punctuation which does this).
%destructor PHR_TERM {delete $$;}

// WILD_TERM is like a TERM, but has a trailing wildcard which needs to be
// expanded.
%destructor WILD_TERM {delete $$;}

// PARTIAL_TERM is like a TERM, but it's at the end of the query string and
// we're doing "search as you type".  It expands to something like WILD_TERM
// OR stemmed_form.
%destructor PARTIAL_TERM {delete $$;}

// BOOLEAN_FILTER is a query term with a prefix registered using
// add_boolean_prefix().  It's added to the query using an OP_FILTER operator,
// (or OP_AND_NOT if it's negated) e.g. site:xapian.org or -site:xapian.org
%destructor BOOLEAN_FILTER {delete $$;}

// Grammar rules:

// query - The whole query - just an expr or nothing.

// query non-terminal doesn't need a type, so just give a dummy one.
%type query {int}

query ::= expr(E). {
    // Save the parsed query in the State structure so we can return it.
    if (E) {
        state->query = *E;
        delete E;
    } else {
        state->query = Query();
    }
}

query ::= . {
    // Handle a query string with no terms in.
    state->query = Query();
}

// expr - A query expression.

%type expr {Query *}
%destructor expr {delete $$;}

expr(E) ::= prob_expr(P).
        { E = P; }

expr(E) ::= bool_arg(A) AND bool_arg(B).
        { BOOL_OP_TO_QUERY(E, A, Query::OP_AND, B, "AND"); }

expr(E) ::= bool_arg(A) NOT bool_arg(B). {
    // 'NOT foo' -> '<alldocuments> NOT foo'
    if (!A && (state->flags & QueryParser::FLAG_PURE_NOT)) {
        A = new Query("", 1, 0);
    }
    BOOL_OP_TO_QUERY(E, A, Query::OP_AND_NOT, B, "NOT");
}

expr(E) ::= bool_arg(A) AND NOT bool_arg(B). [NOT]
        { BOOL_OP_TO_QUERY(E, A, Query::OP_AND_NOT, B, "AND NOT"); }

expr(E) ::= bool_arg(A) AND HATE_AFTER_AND bool_arg(B). [AND]
        { BOOL_OP_TO_QUERY(E, A, Query::OP_AND_NOT, B, "AND"); }

expr(E) ::= bool_arg(A) OR bool_arg(B).
        { BOOL_OP_TO_QUERY(E, A, Query::OP_OR, B, "OR"); }

expr(E) ::= bool_arg(A) XOR bool_arg(B).
        { BOOL_OP_TO_QUERY(E, A, Query::OP_XOR, B, "XOR"); }

// bool_arg - an argument to a boolean operator such as AND or OR.

%type bool_arg {Query *}
%destructor bool_arg {delete $$;}

bool_arg(A) ::= expr(E). { A = E; }

bool_arg(A) ::= . [ERROR] {
    // Set the argument to NULL, which enables the bool_arg-using rules in
    // expr above to report uses of AND, OR, etc which don't have two
    // arguments.
    A = NULL;
}

// prob_expr - a single compound term, or a prob.

%type prob_expr {Query *}
%destructor prob_expr {delete $$;}

prob_expr(E) ::= prob(P). {
    E = P->query;
    P->query = NULL;
    // Handle any "+ terms".
    if (P->love) {
        if (P->love->empty()) {
            // +<nothing>.
            delete E;
            E = P->love;
        } else if (E) {
            swap(E, P->love);
            add_to_query(E, Query::OP_AND_MAYBE, P->love);
        } else {
            E = P->love;
        }
        P->love = NULL;
    }
    // Handle any boolean filters.
    if (!P->filter.empty()) {
        if (E) {
            add_to_query(E, Query::OP_FILTER, P->merge_filters());
        } else {
            // Make the query a boolean one.
            E = new Query(Query::OP_SCALE_WEIGHT, P->merge_filters(), 0.0);
        }
    }
    // Handle any "- terms".
    if (P->hate && !P->hate->empty()) {
        if (!E) {
            // Can't just hate!
            yy_parse_failed(yypParser);
            return;
        }
        *E = Query(Query::OP_AND_NOT, *E, *P->hate);
    }
    delete P;
}

prob_expr(E) ::= term(T). {
    E = T;
}

// prob - a probabilistic sub-expression consisting of stop_terms, "+" terms,
// "-" terms, boolean filters, and/or ranges.
//
// Note: stop_term can also be several other things other than a simple term!

%type prob {ProbQuery *}
%destructor prob {delete $$;}

prob(P) ::= RANGE(R). {
    string grouping = R->name;
    const Query & range = R->as_range_query();
    P = new ProbQuery;
    P->add_filter_range(grouping, range);
}

prob(P) ::= stop_prob(Q) RANGE(R). {
    string grouping = R->name;
    const Query & range = R->as_range_query();
    P = Q;
    P->append_filter_range(grouping, range);
}

prob(P) ::= stop_term(T) stop_term(U). {
    P = new ProbQuery;
    P->query = T;
    if (U) {
        Query::op op = state->default_op();
        if (P->query && is_positional(op)) {
            // If default_op is OP_NEAR or OP_PHRASE, set the window size to
            // 11 for the first pair of terms and it will automatically grow
            // by one for each subsequent term.
            Query * subqs[2] = { P->query, U };
            *(P->query) = Query(op, subqs, subqs + 2, 11);
            delete U;
        } else {
            add_to_query(P->query, op, U);
        }
    }
}

prob(P) ::= prob(Q) stop_term(T). {
    P = Q;
    // If T is a stopword, there's nothing to do here.
    if (T) add_to_query(P->query, state->default_op(), T);
}

prob(P) ::= LOVE term(T). {
    P = new ProbQuery;
    if (state->default_op() == Query::OP_AND) {
        P->query = T;
    } else {
        P->love = T;
    }
}

prob(P) ::= stop_prob(Q) LOVE term(T). {
    P = Q;
    if (state->default_op() == Query::OP_AND) {
        /* The default op is AND, so we just put loved terms into the query
         * (in this case the only effect of love is to ignore the stopword
         * list). */
        add_to_query(P->query, Query::OP_AND, T);
    } else {
        add_to_query(P->love, Query::OP_AND, T);
    }
}

prob(P) ::= HATE term(T). {
    P = new ProbQuery;
    P->hate = T;
}

prob(P) ::= stop_prob(Q) HATE term(T). {
    P = Q;
    add_to_query(P->hate, Query::OP_OR, T);
}

prob(P) ::= HATE BOOLEAN_FILTER(T). {
    P = new ProbQuery;
    P->hate = new Query(T->get_query());
    delete T;
}

prob(P) ::= stop_prob(Q) HATE BOOLEAN_FILTER(T). {
    P = Q;
    add_to_query(P->hate, Query::OP_OR, T->get_query());
    delete T;
}

prob(P) ::= BOOLEAN_FILTER(T). {
    P = new ProbQuery;
    P->add_filter(T->get_grouping(), T->get_query());
    delete T;
}

prob(P) ::= stop_prob(Q) BOOLEAN_FILTER(T). {
    P = Q;
    P->append_filter(T->get_grouping(), T->get_query());
    delete T;
}

prob(P) ::= LOVE BOOLEAN_FILTER(T). {
    // LOVE BOOLEAN_FILTER(T) is just the same as BOOLEAN_FILTER
    P = new ProbQuery;
    P->filter[T->get_grouping()] = T->get_query();
    delete T;
}

prob(P) ::= stop_prob(Q) LOVE BOOLEAN_FILTER(T). {
    // LOVE BOOLEAN_FILTER(T) is just the same as BOOLEAN_FILTER
    P = Q;
    // We OR filters with the same prefix...
    Query & q = P->filter[T->get_grouping()];
    q = Query(Query::OP_OR, q, T->get_query());
    delete T;
}

// stop_prob - A prob or a stop_term.

%type stop_prob {ProbQuery *}
%destructor stop_prob {delete $$;}

stop_prob(P) ::= prob(Q).
    { P = Q; }

stop_prob(P) ::= stop_term(T). {
    P = new ProbQuery;
    P->query = T;
}

// stop_term - A term which should be checked against the stopword list,
// or a compound_term.
//
// If a term is loved, hated, or in a phrase, we don't want to consult the
// stopword list, so stop_term isn't used there (instead term is).

%type stop_term {Query *}
%destructor stop_term {delete $$;}

stop_term(T) ::= TERM(U). {
    if (state->is_stopword(U)) {
        T = NULL;
        state->add_to_stoplist(U);
    } else {
        T = new Query(U->get_query_with_auto_synonyms());
    }
    delete U;
}

stop_term(T) ::= compound_term(U). {
    T = U;
}

// term - A term or a compound_term.

%type term {Query *}
%destructor term {delete $$;}

term(T) ::= TERM(U). {
    T = new Query(U->get_query_with_auto_synonyms());
    delete U;
}

term(T) ::= compound_term(U). {
    T = U;
}

// compound_term - A WILD_TERM, a quoted phrase (with or without prefix), a
// phrased_term, group, near_expr, adj_expr, or a bracketed subexpression (with
// or without prefix).

%type compound_term {Query *}
%destructor compound_term {delete $$;}

compound_term(T) ::= WILD_TERM(U).
        { T = U->as_wildcarded_query(state); }

compound_term(T) ::= PARTIAL_TERM(U).
        { T = U->as_partial_query(state); }

compound_term(T) ::= QUOTE phrase(P) QUOTE.
        { T = P->as_phrase_query(); }

compound_term(T) ::= phrased_term(P).
        { T = P->as_phrase_query(); }

compound_term(T) ::= group(P).
        { T = P->as_group(state); }

compound_term(T) ::= near_expr(P).
        { T = P->as_near_query(); }

compound_term(T) ::= adj_expr(P).
        { T = P->as_adj_query(); }

compound_term(T) ::= BRA expr(E) KET.
        { T = E; }

compound_term(T) ::= SYNONYM TERM(U). {
    T = new Query(U->get_query_with_synonyms());
    delete U;
}

compound_term(T) ::= CJKTERM(U). {
    { T = U->as_cjk_query(); }
}

// phrase - The "inside the quotes" part of a double-quoted phrase.

%type phrase {Terms *}

%destructor phrase {delete $$;}

phrase(P) ::= TERM(T). {
    P = Terms::create();
    P->add_positional_term(T);
}

phrase(P) ::= CJKTERM(T). {
    P = Terms::create();
    T->as_positional_cjk_term(P);
}

phrase(P) ::= phrase(Q) TERM(T). {
    P = Q;
    P->add_positional_term(T);
}

phrase(P) ::= phrase(Q) CJKTERM(T). {
    P = Q;
    T->as_positional_cjk_term(P);
}

// phrased_term - A phrased term works like a single term, but is actually
// 2 or more terms linked together into a phrase by punctuation.  There must be
// at least 2 terms in order to be able to have punctuation between the terms!

%type phrased_term {Terms *}
%destructor phrased_term {delete $$;}

phrased_term(P) ::= TERM(T) PHR_TERM(U). {
    P = Terms::create();
    P->add_positional_term(T);
    P->add_positional_term(U);
}

phrased_term(P) ::= phrased_term(Q) PHR_TERM(T). {
    P = Q;
    P->add_positional_term(T);
}

// group - A group of terms separated only by whitespace - candidates for
// multi-term synonyms.

%type group {TermGroup *}
%destructor group {delete $$;}

group(P) ::= TERM(T) GROUP_TERM(U). {
    P = TermGroup::create(T, U);
}

group(P) ::= group(Q) GROUP_TERM(T). {
    P = Q;
    P->add_term(T);
}

group(P) ::= group(Q) EMPTY_GROUP_OK. {
    P = Q;
    P->set_empty_ok();
}

// near_expr - 2 or more terms with NEAR in between.  There must be at least 2
// terms in order for there to be any NEAR operators!

%type near_expr {Terms *}
%destructor near_expr {delete $$;}

near_expr(P) ::= TERM(T) NEAR(N) TERM(U). {
    P = Terms::create();
    P->add_positional_term(T);
    P->add_positional_term(U);
    if (N) {
        P->adjust_window(N->get_termpos());
        delete N;
    }
}

near_expr(P) ::= near_expr(Q) NEAR(N) TERM(T). {
    P = Q;
    P->add_positional_term(T);
    if (N) {
        P->adjust_window(N->get_termpos());
        delete N;
    }
}

// adj_expr - 2 or more terms with ADJ in between.  There must be at least 2
// terms in order for there to be any ADJ operators!

%type adj_expr {Terms *}
%destructor adj_expr {delete $$;}

adj_expr(P) ::= TERM(T) ADJ(N) TERM(U). {
    P = Terms::create();
    P->add_positional_term(T);
    P->add_positional_term(U);
    if (N) {
        P->adjust_window(N->get_termpos());
        delete N;
    }
}

adj_expr(P) ::= adj_expr(Q) ADJ(N) TERM(T). {
    P = Q;
    P->add_positional_term(T);
    if (N) {
        P->adjust_window(N->get_termpos());
        delete N;
    }
}

// Select yacc syntax highlighting in vim editor: vim: syntax=yacc
// (lemon syntax colouring isn't supplied by default; yacc does an OK job).