/xapian-applications/omega/query.cc

Large files files are truncated, but you can click here to view the full file

/* query.cc: query executor for omega
 *
 * Copyright 1999,2000,2001 BrightStation PLC
 * Copyright 2001 James Aylett
 * Copyright 2001,2002 Ananova Ltd
 * Copyright 2002 Intercede 1749 Ltd
 * Copyright 2002,2003,2004,2005,2006,2007,2008,2009,2010,2011,2013,2014 Olly Betts
 * Copyright 2008 Thomas Viehmann
 *
 * 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>

// If we're building against git after the expand API changed but before the
// version gets bumped to 1.3.2, we'll get a deprecation warning from
// get_eset() unless we suppress such warnings here.
#define XAPIAN_DEPRECATED(D) D

#include <algorithm>
#include <iostream>
#include <map>
#include <set>
#include <vector>

#include <cassert>
#include <cctype>
#include "safeerrno.h"
#include <stdio.h>
#include <cstdlib>
#include <cstring>
#include "strcasecmp.h"
#include <ctime>

#include "safeunistd.h"
#include <sys/types.h>
#include "safesysstat.h"
#include "safefcntl.h"

#include "realtime.h"

#include <cdb.h>

#include "date.h"
#include "datematchdecider.h"
#include "jsonescape.h"
#include "utils.h"
#include "omega.h"
#include "query.h"
#include "cgiparam.h"
#include "loadfile.h"
#include "sample.h"
#include "str.h"
#include "stringutils.h"
#include "transform.h"
#include "urldecode.h"
#include "urlencode.h"
#include "unixperm.h"
#include "values.h"
#include "weight.h"
#include "expand.h"

#include <xapian.h>

using namespace std;

using Xapian::Utf8Iterator;

using Xapian::Unicode::is_wordchar;

#ifndef SNPRINTF
#include <cstdarg>

static int my_snprintf(char *str, size_t size, const char *format, ...)
{
    int res;
    va_list ap;
    va_start(ap, format);
    str[size - 1] = '\0';
    res = vsprintf(str, format, ap);
    if (str[size - 1] || res < 0 || size_t(res) >= size)
	abort(); /* Overflowed! */
    va_end(ap);
    return res;
}
#else
#define my_snprintf SNPRINTF
#endif

static bool query_parsed = false;
static bool done_query = false;
static Xapian::docid last = 0;

static Xapian::MSet mset;

static map<Xapian::docid, bool> ticked;

static void ensure_query_parsed();
static void ensure_match();

static Xapian::Query query;
//static string url_query_string;
Xapian::Query::op default_op = Xapian::Query::OP_AND; // default matching mode

static Xapian::QueryParser qp;
static Xapian::NumberValueRangeProcessor * size_vrp = NULL;
static Xapian::Stem *stemmer = NULL;

static string eval_file(const string &fmtfile);

static set<string> termset;

// Holds mapping from term prefix to user prefix (e.g. 'S' -> 'subject:').
static map<string, string> termprefix_to_userprefix;

static string queryterms;

static string error_msg;

static double secs = -1;

static const char DEFAULT_LOG_ENTRY[] =
	"$or{$env{REMOTE_HOST},$env{REMOTE_ADDR},-}\t"
	"[$date{$now,%d/%b/%Y:%H:%M:%S} +0000]\t"
	"$if{$cgi{X},add,$if{$cgi{MORELIKE},morelike,query}}\t"
	"$dbname\t"
	"$query\t"
	"$msize$if{$env{HTTP_REFERER},\t$env{HTTP_REFERER}}";

class MyStopper : public Xapian::Stopper {
  public:
    bool operator()(const string &t) const {
	switch (t[0]) {
	    case 'a':
		return (t == "a" || t == "about" || t == "an" || t == "and" ||
			t == "are" || t == "as" || t == "at");
	    case 'b':
		return (t == "be" || t == "by");
	    case 'e':
		return (t == "en");
	    case 'f':
		return (t == "for" || t == "from");
	    case 'h':
		return (t == "how");
	    case 'i':
		return (t == "i" || t == "in" || t == "is" || t == "it");
	    case 'o':
		return (t == "of" || t == "on" || t == "or");
	    case 't':
		return (t == "that" || t == "the" || t == "this" || t == "to");
	    case 'w':
		return (t == "was" || t == "what" || t == "when" ||
			t == "where" || t == "which" || t == "who" ||
			t == "why" || t == "will" || t == "with");
	    case 'y':
		return (t == "you" || t == "your");
	    default:
		return false;
	}
    }
};

static size_t
prefix_from_term(string &prefix, const string &term)
{
    if (term.empty()) {
	prefix.resize(0);
	return 0;
    }
    if (term[0] == 'X') {
	const string::const_iterator begin = term.begin();
	string::const_iterator i = begin + 1;
	while (i != term.end() && isupper(static_cast<unsigned char>(*i))) ++i;
	prefix.assign(begin, i);
	if (i != term.end() && *i == ':') ++i;
	return i - begin;
    }

    prefix = term[0];
    return 1;
}

// Don't allow ".." in format names, log file names, etc as this would allow
// people to open a format "../../etc/passwd" or similar.
// FIXME: make this check more exact ("foo..bar" is safe)
// FIXME: log when this check fails
static bool
vet_filename(const string &filename)
{
    string::size_type i = filename.find("..");
    return (i == string::npos);
}

// Heuristics:
// * If any terms have been removed, it's a "fresh query" so we discard any
//   relevance judgements
// * If all previous terms are there but more have been added then we keep
//   the relevance judgements, but return the first page of hits
//
// NEW_QUERY entirely new query
// SAME_QUERY unchanged query
// EXTENDED_QUERY new query, but based on the old one
// BAD_QUERY parse error (message in error_msg)
typedef enum { NEW_QUERY, SAME_QUERY, EXTENDED_QUERY, BAD_QUERY } querytype;

static map<string, string> probabilistic_query;

void
set_probabilistic_query(const string & prefix, const string & s)
{
    string query_string = s;
    // Strip leading and trailing whitespace from query_string.
    trim(query_string);
    if (!query_string.empty())
	probabilistic_query.insert(make_pair(prefix, query_string));
}

static unsigned
read_qp_flags(const string & opt_pfx, unsigned f)
{
    map<string, string>::const_iterator i = option.lower_bound(opt_pfx);
    for (; i != option.end() && startswith(i->first, opt_pfx); ++i) {
	unsigned mask = 0;
	const char * s = i->first.c_str() + opt_pfx.size();
	switch (s[0]) {
	    case 'a':
		if (strcmp(s, "auto_multiword_synonyms") == 0) {
		    mask = Xapian::QueryParser::FLAG_AUTO_MULTIWORD_SYNONYMS;
		    break;
		}
		if (strcmp(s, "auto_synonyms") == 0) {
		    mask = Xapian::QueryParser::FLAG_AUTO_SYNONYMS;
		    break;
		}
		break;
	    case 'b':
		if (strcmp(s, "boolean") == 0) {
		    mask = Xapian::QueryParser::FLAG_BOOLEAN;
		    break;
		}
		if (strcmp(s, "boolean_any_case") == 0) {
		    mask = Xapian::QueryParser::FLAG_BOOLEAN_ANY_CASE;
		    break;
		}
		break;
	    case 'd':
		if (strcmp(s, "default") == 0) {
		    mask = Xapian::QueryParser::FLAG_DEFAULT;
		    break;
		}
		break;
	    case 'l':
		if (strcmp(s, "lovehate") == 0) {
		    mask = Xapian::QueryParser::FLAG_LOVEHATE;
		    break;
		}
		break;
	    case 'p':
		if (strcmp(s, "partial") == 0) {
		    mask = Xapian::QueryParser::FLAG_PARTIAL;
		    break;
		}
		if (strcmp(s, "phrase") == 0) {
		    mask = Xapian::QueryParser::FLAG_PHRASE;
		    break;
		}
		if (strcmp(s, "pure_not") == 0) {
		    mask = Xapian::QueryParser::FLAG_PURE_NOT;
		    break;
		}
		break;
	    case 's':
		if (strcmp(s, "spelling_correction") == 0) {
		    mask = Xapian::QueryParser::FLAG_SPELLING_CORRECTION;
		    break;
		}
		if (strcmp(s, "synonym") == 0) {
		    mask = Xapian::QueryParser::FLAG_SYNONYM;
		    break;
		}
		break;
	    case 'w':
		if (strcmp(s, "wildcard") == 0) {
		    mask = Xapian::QueryParser::FLAG_WILDCARD;
		    break;
		}
		break;
	}

	if (i->second.empty()) {
	    f &= ~mask;
	} else {
	    f |= mask;
	}
    }
    return f;
}

static querytype
set_probabilistic(const string &oldp)
{
    // Parse the query string.
    qp.set_stemming_strategy(option["stem_all"] == "true" ? Xapian::QueryParser::STEM_ALL : Xapian::QueryParser::STEM_SOME);
    qp.set_stopper(new MyStopper());
    qp.set_default_op(default_op);
    qp.set_database(db);
    // FIXME: provide a custom VRP which handles size:10..20K, etc.
    if (!size_vrp)
	size_vrp = new Xapian::NumberValueRangeProcessor(VALUE_SIZE, "size:",
							 true);
    qp.add_valuerangeprocessor(size_vrp);
    map<string, string>::const_iterator pfx = option.lower_bound("prefix,");
    for (; pfx != option.end() && startswith(pfx->first, "prefix,"); ++pfx) {
	string user_prefix(pfx->first, 7);
	const string & term_pfx_list = pfx->second;
	string::size_type i = 0;
	do {
	    string::size_type i0 = i;
	    i = term_pfx_list.find('\t', i);
	    const string & term_pfx = term_pfx_list.substr(i0, i - i0);
	    qp.add_prefix(user_prefix, term_pfx);
	    // std::map::insert() won't overwrite an existing entry, so we'll
	    // prefer the first user_prefix for which a particular term prefix
	    // is specified.
	    termprefix_to_userprefix.insert(make_pair(term_pfx, user_prefix));
	} while (++i);
    }
    pfx = option.lower_bound("boolprefix,");
    for (; pfx != option.end() && startswith(pfx->first, "boolprefix,"); ++pfx) {
	string user_prefix = pfx->first.substr(11);
	qp.add_boolean_prefix(user_prefix, pfx->second);
	termprefix_to_userprefix.insert(make_pair(pfx->second, user_prefix));
    }

    try {
	unsigned default_flags = read_qp_flags("flag_", 0);
	if (option["spelling"] == "true")
	    default_flags |= qp.FLAG_SPELLING_CORRECTION;

	vector<Xapian::Query> queries;
	queries.reserve(probabilistic_query.size());

	map<string, string>::const_iterator j;
	for (j = probabilistic_query.begin();
	     j != probabilistic_query.end();
	     ++j) {
	    const string & prefix = j->first;

	    // Choose the stemmer to use for this input.
	    string stemlang = option[prefix + ":stemmer"];
	    if (stemlang.empty())
		stemlang = option["stemmer"];
	    qp.set_stemmer(Xapian::Stem(stemlang));

	    // Work out the flags to use for this input.
	    unsigned f = read_qp_flags(prefix + ":flag_", default_flags);

	    const string & query_string = j->second;
	    queries.push_back(qp.parse_query(query_string, f, prefix));
	}
	query = Xapian::Query(query.OP_AND, queries.begin(), queries.end());
    } catch (Xapian::QueryParserError &e) {
	error_msg = e.get_msg();
	return BAD_QUERY;
    }

    Xapian::termcount n_new_terms = 0;
    for (Xapian::TermIterator i = query.get_terms_begin();
	 i != query.get_terms_end(); ++i) {
	if (termset.find(*i) == termset.end()) {
	    termset.insert(*i);
	    if (!queryterms.empty()) queryterms += '\t';
	    queryterms += *i;
	}
	n_new_terms++;
    }

    // Check new query against the previous one
    if (oldp.empty()) {
	// FIXME: should take into account other probabilistic prefixes here...
	return probabilistic_query[string()].empty() ? SAME_QUERY : NEW_QUERY;
    }

    // The terms in oldp are separated by tabs.
    const char oldp_separator = '\t';
    size_t n_old_terms = count(oldp.begin(), oldp.end(), oldp_separator) + 1;

    // short-cut: if the new query has fewer terms, it must be a new one
    if (n_new_terms < n_old_terms) return NEW_QUERY;

    const char *term = oldp.c_str();
    const char *pend;
    while ((pend = strchr(term, oldp_separator)) != NULL) {
	if (termset.find(string(term, pend - term)) == termset.end())
	    return NEW_QUERY;
	term = pend + 1;
    }
    if (*term) {
	if (termset.find(string(term)) == termset.end())
	    return NEW_QUERY;
    }

    // Use termset.size() rather than n_new_terms so we correctly handle
    // the case when the query has repeated terms.
    // This works wrongly in the case when the user extends the query
    // by adding a term already in it, but that's unlikely and the behaviour
    // isn't too bad (we just don't reset page 1).  We also mishandle a few
    // other obscure cases e.g. adding quotes to turn a query into a phrase.
    if (termset.size() > n_old_terms) return EXTENDED_QUERY;
    return SAME_QUERY;
}

static multimap<string, string> filter_map;

typedef multimap<string, string>::const_iterator FMCI;

void add_bterm(const string &term) {
    string prefix;
    if (prefix_from_term(prefix, term) > 0)
	filter_map.insert(multimap<string, string>::value_type(prefix, term));
}

static void
run_query()
{
    bool force_boolean = false;
    if (!filter_map.empty()) {
	// OR together filters with the same prefix, then AND together
	vector<Xapian::Query> filter_vec;
	vector<string> or_vec;
	string current;
	for (FMCI i = filter_map.begin(); ; i++) {
	    bool over = (i == filter_map.end());
	    if (over || i->first != current) {
		switch (or_vec.size()) {
		    case 0:
		        break;
		    case 1:
			filter_vec.push_back(Xapian::Query(or_vec[0]));
		        break;
		    default:
			filter_vec.push_back(Xapian::Query(Xapian::Query::OP_OR,
						     or_vec.begin(),
						     or_vec.end()));
		        break;
		}
		or_vec.clear();
		if (over) break;
		current = i->first;
	    }
	    or_vec.push_back(i->second);
	}

	Xapian::Query filter(Xapian::Query::OP_AND,
			     filter_vec.begin(), filter_vec.end());

	if (query.empty()) {
	    // If no probabilistic query is provided then promote the filters
	    // to be THE query - filtering an empty query will give no
	    // matches.
	    std::swap(query, filter);
	    if (enquire) force_boolean = true;
	} else {
	    query = Xapian::Query(Xapian::Query::OP_FILTER, query, filter);
	}
    }

    Xapian::MatchDecider * mdecider = NULL;
    if (!date_start.empty() || !date_end.empty() || !date_span.empty()) {
	MCI i = cgi_params.find("DATEVALUE");
	if (i != cgi_params.end()) {
	    Xapian::valueno datevalue = string_to_int(i->second);
	    mdecider = new DateMatchDecider(datevalue, date_start, date_end, date_span);
	} else {
	    Xapian::Query date_filter(Xapian::Query::OP_OR,
				      date_range_filter(date_start, date_end,
							date_span),
				      Xapian::Query("Dlatest"));

	    // If no probabilistic query is provided then promote the daterange
	    // filter to be THE query instead of filtering an empty query.
	    if (query.empty()) {
		query = date_filter;
	    } else {
		query = Xapian::Query(Xapian::Query::OP_FILTER, query, date_filter);
	    }
	}
    }

    if (!enquire || !error_msg.empty()) return;

    set_weighting_scheme(*enquire, option, force_boolean);

    enquire->set_cutoff(threshold);

    if (sort_key != Xapian::BAD_VALUENO) {
	if (sort_after) {
	    enquire->set_sort_by_relevance_then_value(sort_key, sort_ascending);
	} else {
	    enquire->set_sort_by_value_then_relevance(sort_key, sort_ascending);
	}
    }

    enquire->set_docid_order(docid_order);

    if (collapse) {
	enquire->set_collapse_key(collapse_key);
    }

    if (!query.empty()) {
#if 0
	// FIXME: If we start doing permissions checks based on $REMOTE_USER
	// we're going to break some existing setups if users upgrade.  We
	// probably want a way to set this from OmegaScript.
	const char * remote_user = getenv("REMOTE_USER");
	if (remote_user)
	    apply_unix_permissions(query, remote_user);
#endif

	enquire->set_query(query);
	// We could use the value of topdoc as first parameter, but we
	// need to know the first few items in the mset to fake a
	// relevance set for topterms.
	//
	// If min_hits isn't set, check at least one extra result so we
	// know if we've reached the end of the matches or not - then we
	// can avoid offering a "next" button which leads to an empty page.
	mset = enquire->get_mset(0, topdoc + hits_per_page,
				 topdoc + max(hits_per_page + 1, min_hits),
				 &rset, mdecider);
    }
}

string
html_escape(const string &str)
{
    string res;
    string::size_type p = 0;
    while (p < str.size()) {
	char ch = str[p++];
	switch (ch) {
	    case '<':
	        res += "&lt;";
	        continue;
	    case '>':
	        res += "&gt;";
	        continue;
	    case '&':
	        res += "&amp;";
	        continue;
	    case '"':
	        res += "&quot;";
	        continue;
	    default:
	        res += ch;
	}
    }
    return res;
}

static string
html_strip(const string &str)
{
    string res;
    string::size_type p = 0;
    bool skip = false;
    while (p < str.size()) {
	char ch = str[p++];
	switch (ch) {
	    case '<':
	        skip = true;
	        continue;
	    case '>':
	        skip = false;
	        continue;
	    default:
	        if (! skip) res += ch;
	}
    }
    return res;
}

// FIXME split list into hash or map and use that rather than linear lookup?
static int word_in_list(const string& word, const string& list)
{
    string::size_type split = 0, split2;
    int count = 0;
    while ((split2 = list.find('\t', split)) != string::npos) {
	if (word.size() == split2 - split) {
	    if (memcmp(word.data(), list.data() + split, word.size()) == 0)
		return count;
	}
	split = split2 + 1;
	++count;
    }
    if (word.size() == list.size() - split) {
	if (memcmp(word.data(), list.data() + split, word.size()) == 0)
	    return count;
    }
    return -1;
}

// Not a character in an identifier
inline static bool
p_notid(unsigned int c)
{
    return !isalnum(static_cast<unsigned char>(c)) && c != '_';
}

// Not a character in an HTML tag name
inline static bool
p_nottag(unsigned int c)
{
    return !isalnum(static_cast<unsigned char>(c)) && c != '.' && c != '-';
}

// FIXME: shares algorithm with indextext.cc!
static string
html_highlight(const string &s, const string &list,
	       const string &bra, const string &ket)
{
    if (!stemmer) {
	stemmer = new Xapian::Stem(option["stemmer"]);
    }

    string res;

    Utf8Iterator j(s);
    const Utf8Iterator s_end;
    while (true) {
	Utf8Iterator first = j;
	while (first != s_end && !is_wordchar(*first)) ++first;
	if (first == s_end) break;
	Utf8Iterator term_end;
	string term;
	string word;
	const char *l = j.raw();
	if (*first < 128 && isupper(*first)) {
	    j = first;
	    Xapian::Unicode::append_utf8(term, *j);
	    while (++j != s_end && *j == '.' && ++j != s_end && *j < 128 && isupper(*j)) {
		Xapian::Unicode::append_utf8(term, *j);
	    }
	    if (term.length() < 2 || (j != s_end && is_wordchar(*j))) {
		term.resize(0);
	    }
	    term_end = j;
	}
	if (term.empty()) {
	    j = first;
	    while (is_wordchar(*j)) {
		Xapian::Unicode::append_utf8(term, *j);
		++j;
		if (j == s_end) break;
		if (*j == '&' || *j == '\'') {
		    Utf8Iterator next = j;
		    ++next;
		    if (next == s_end || !is_wordchar(*next)) break;
		    term += *j;
		    j = next;
		}
	    }
	    term_end = j;
	    if (j != s_end && (*j == '+' || *j == '-' || *j == '#')) {
		string::size_type len = term.length();
		if (*j == '#') {
		    term += '#';
		    do { ++j; } while (j != s_end && *j == '#');
		} else {
		    while (j != s_end && (*j == '+' || *j == '-')) {
			Xapian::Unicode::append_utf8(term, *j);
			++j;
		    }
		}
		if (term.size() - len > 3 || (j != s_end && is_wordchar(*j))) {
		    term.resize(len);
		} else {
		    term_end = j;
		}
	    }
	}
	j = term_end;
	term = Xapian::Unicode::tolower(term);
	int match = word_in_list(term, list);
	if (match == -1) {
	    string stem = "Z";
	    stem += (*stemmer)(term);
	    match = word_in_list(stem, list);
	}
	if (match >= 0) {
	    res += html_escape(string(l, first.raw() - l));
	    if (!bra.empty()) {
		res += bra;
	    } else {
		static const char * colours[] = {
		    "ffff66", "99ff99", "99ffff", "ff66ff", "ff9999",
		    "990000", "009900", "996600", "006699", "990099"
		};
		size_t idx = match % (sizeof(colours) / sizeof(colours[0]));
		const char * bg = colours[idx];
		if (strchr(bg, 'f')) {
		    res += "<b style=\"color:black;background-color:#";
		} else {
		    res += "<b style=\"color:white;background-color:#";
		}
		res += bg;
		res += "\">";
	    }
	    word = string(first.raw(), j.raw() - first.raw());
	    res += html_escape(word);
	    if (!bra.empty()) {
		res += ket;
	    } else {
		res += "</b>";
	    }
	} else {
	    res += html_escape(string(l, j.raw() - l));
	}
    }
    if (j != s_end) res += html_escape(string(j.raw(), j.left()));
    return res;
}

#if 0
static void
print_query_string(const char *after)
{
    if (after && strncmp(after, "&B=", 3) == 0) {
	char prefix = after[3];
	string::size_type start = 0, amp = 0;
	while (true) {
	    amp = url_query_string.find('&', amp);
	    if (amp == string::npos) {
		cout << url_query_string.substr(start);
		return;
	    }
	    amp++;
	    while (url_query_string[amp] == 'B' &&
		   url_query_string[amp + 1] == '=' &&
		   url_query_string[amp + 2] == prefix) {
		cout << url_query_string.substr(start, amp - start - 1);
		start = url_query_string.find('&', amp + 3);
		if (start == string::npos) return;
		amp = start + 1;
	    }
	}
    }
    cout << url_query_string;
}
#endif

class Fields {
    mutable Xapian::docid did_cached;
    mutable map<string, string> fields;

    void read_fields(Xapian::docid did) const;

  public:
    Fields() : did_cached(0) { }

    const string & get_field(Xapian::docid did, const string & field) const {
	if (did != did_cached) read_fields(did);
	return fields[field];
    }
};

void
Fields::read_fields(Xapian::docid did) const
{
    fields.clear();
    did_cached = did;
    const string & data = db.get_document(did).get_data();

    // Parse document data.
    string::size_type i = 0;
    const string & names = option["fieldnames"];
    if (!names.empty()) {
	// Each line is a field, with fieldnames taken from corresponding
	// entries in the tab-separated list specified by $opt{fieldnames}.
	string::size_type n = 0;
	do {
	    string::size_type n0 = n;
	    n = names.find('\t', n);
	    string::size_type i0 = i;
	    i = data.find('\n', i);
	    fields.insert(make_pair(names.substr(n0, n  - n0),
				    data.substr(i0, i - i0)));
	} while (++n && ++i);
    } else {
	// Each line is a field, in the format NAME=VALUE.  We assume the field
	// name doesn't contain an "=".  Lines without an "=" are currently
	// just ignored.
	do {
	    string::size_type i0 = i;
	    i = data.find('\n', i);
	    string line = data.substr(i0, i - i0);
	    string::size_type j = line.find('=');
	    if (j != string::npos) {
		string & value = fields[line.substr(0, j)];
		if (!value.empty()) value += '\t';
		value += line.substr(j + 1);
	    }
	} while (++i);
    }
}

static Fields fields;
static Xapian::docid q0;
static Xapian::doccount hit_no;
static int percent;
static double weight;
static Xapian::doccount collapsed;

static string print_caption(const string &fmt, const vector<string> &param);

enum tagval {
CMD_,
CMD_add,
CMD_addfilter,
CMD_allterms,
CMD_and,
CMD_cgi,
CMD_cgilist,
CMD_collapsed,
CMD_date,
CMD_dbname,
CMD_dbsize,
CMD_def,
CMD_defaultop,
CMD_div,
CMD_eq,
CMD_emptydocs,
CMD_env,
CMD_error,
CMD_field,
CMD_filesize,
CMD_filters,
CMD_filterterms,
CMD_find,
CMD_fmt,
CMD_freq,
CMD_ge,
CMD_gt,
CMD_highlight,
CMD_hit,
CMD_hitlist,
CMD_hitsperpage,
CMD_hostname,
CMD_html,
CMD_htmlstrip,
CMD_httpheader,
CMD_id,
CMD_if,
CMD_include,
CMD_json,
CMD_jsonarray,
CMD_last,
CMD_lastpage,
CMD_le,
CMD_length,
CMD_list,
CMD_log,
CMD_lookup,
CMD_lower,
CMD_lt,
CMD_map,
CMD_max,
CMD_min,
CMD_mod,
CMD_msize,
CMD_msizeexact,
CMD_mul,
CMD_muldiv,
CMD_ne,
CMD_nice,
CMD_not,
CMD_now,
CMD_opt,
CMD_or,
CMD_pack,
CMD_percentage,
CMD_prettyterm,
CMD_prettyurl,
CMD_query,
CMD_querydescription,
CMD_queryterms,
CMD_range,
CMD_record,
CMD_relevant,
CMD_relevants,
CMD_score,
CMD_set,
CMD_setmap,
CMD_setrelevant,
CMD_slice,
CMD_snippet,
CMD_split,
CMD_stoplist,
CMD_sub,
CMD_substr,
CMD_suggestion,
CMD_terms,
CMD_thispage,
CMD_time,
CMD_topdoc,
CMD_topterms,
CMD_transform,
CMD_truncate,
CMD_uniq,
CMD_unpack,
CMD_unstem,
CMD_upper,
CMD_url,
CMD_value,
CMD_version,
CMD_weight,
CMD_MACRO // special tag for macro evaluation
};

struct func_attrib {
    int tag;
    int minargs, maxargs, evalargs;
    char ensure;
};

#define T(F,A,B,C,D) {STRINGIZE(F),{CMD_##F,A,B,C,D}}
struct func_desc {
    const char *name;
    struct func_attrib a;
};

#define N -1
#define M 'M'
#define Q 'Q'
// NB when adding a new command which ensures M or Q, update the list in
// docs/omegascript.rst
static struct func_desc func_tab[] = {
//name minargs maxargs evalargs ensure
{"",{CMD_,	   N, N, 0, 0}},// commented out code
T(add,		   0, N, N, 0), // add a list of numbers
T(addfilter,	   1, 1, N, 0), // add filter term
T(allterms,	   0, 1, N, 0), // list of all terms matching document
T(and,		   1, N, 0, 0), // logical shortcutting and of a list of values
T(cgi,		   1, 1, N, 0), // return cgi parameter value
T(cgilist,	   1, 1, N, 0), // return list of values for cgi parameter
T(collapsed,	   0, 0, N, 0), // return number of hits collapsed into this
T(date,		   1, 2, N, 0), // convert time_t to strftime format
				// (default: YYYY-MM-DD)
T(dbname,	   0, 0, N, 0), // database name
T(dbsize,	   0, 0, N, 0), // database size (# of documents)
T(def,		   2, 2, 1, 0), // define a macro
T(defaultop,	   0, 0, N, 0), // default operator: "and" or "or"
T(div,		   2, 2, N, 0), // integer divide
T(emptydocs,	   0, 1, N, 0), // list of empty documents
T(env,		   1, 1, N, 0), // environment variable
T(error,	   0, 0, N, 0), // error message
T(eq,		   2, 2, N, 0), // test equality
T(field,	   1, 2, N, 0), // lookup field in record
T(filesize,	   1, 1, N, 0), // pretty printed filesize
T(filters,	   0, 0, N, 0), // serialisation of current filters
T(filterterms,	   1, 1, N, 0), // list of terms with a given prefix
T(find,		   2, 2, N, 0), // find entry in list
T(fmt,		   0, 0, N, 0), // name of current format
T(freq,		   1, 1, N, 0), // frequency of a term
T(ge,		   2, 2, N, 0), // test >=
T(gt,		   2, 2, N, 0), // test >
T(highlight,	   2, 4, N, 0), // html escape and highlight words from list
T(hit,		   0, 0, N, 0), // hit number of current mset entry (starting
				// from 0
T(hitlist,	   1, 1, 0, M), // display hitlist using format in argument
T(hitsperpage,	   0, 0, N, 0), // hits per page
T(hostname,	   1, 1, N, 0), // extract hostname from URL
T(html,		   1, 1, N, 0), // html escape string (<>&")
T(htmlstrip,	   1, 1, N, 0), // html strip tags string (s/<[^>]*>?//g)
T(httpheader,      2, 2, N, 0), // arbitrary HTTP header
T(id,		   0, 0, N, 0), // docid of current doc
T(if,		   2, 3, 1, 0), // conditional
T(include,	   1, 1, 1, 0), // include another file
T(json,		   1, 1, N, 0), // JSON string escaping
T(jsonarray,	   1, 1, N, 0), // Format list as a JSON array of strings
T(last,		   0, 0, N, M), // m-set number of last hit on page
T(lastpage,	   0, 0, N, M), // number of last hit page
T(le,		   2, 2, N, 0), // test <=
T(length,	   1, 1, N, 0), // length of list
T(list,		   2, 5, N, 0), // pretty print list
T(log,		   1, 2, 1, 0), // create a log entry
T(lookup,	   2, 2, N, 0), // lookup in named cdb file
T(lower,	   1, 1, N, 0), // convert string to lower case
T(lt,		   2, 2, N, 0), // test <
T(map,		   1, 2, 1, 0), // map a list into another list
T(max,		   1, N, N, 0), // maximum of a list of values
T(min,		   1, N, N, 0), // minimum of a list of values
T(mod,		   2, 2, N, 0), // integer modulus
T(msize,	   0, 0, N, M), // number of matches
T(msizeexact,	   0, 0, N, M), // is $msize exact?
T(mul,		   2, N, N, 0), // multiply a list of numbers
T(muldiv,	   3, 3, N, 0), // calculate A*B/C
T(ne,		   2, 2, N, 0), // test not equal
T(nice,		   1, 1, N, 0), // pretty print integer (with thousands sep)
T(not,		   1, 1, N, 0), // logical not
T(now,		   0, 0, N, 0), // current date/time as a time_t
T(opt,		   1, 2, N, 0), // lookup an option value
T(or,		   1, N, 0, 0), // logical shortcutting or of a list of values
T(pack,		   1, 1, N, 0), // convert a number to a 4 byte big endian binary string
T(percentage,	   0, 0, N, 0), // percentage score of current hit
T(prettyterm,	   1, 1, N, Q), // pretty print term name
T(prettyurl,	   1, 1, N, 0), // pretty version of URL
T(query,	   0, 1, N, Q), // query
T(querydescription,0, 0, N, Q), // query.get_description()
T(queryterms,	   0, 0, N, Q), // list of query terms
T(range,	   2, 2, N, 0), // return list of values between start and end
T(record,	   0, 1, N, 0), // record contents of document
T(relevant,	   0, 1, N, Q), // is document relevant?
T(relevants,	   0, 0, N, Q), // return list of relevant documents
T(score,	   0, 0, N, 0), // score (0-10) of current hit
T(set,		   2, 2, N, 0), // set option value
T(setmap,	   1, N, N, 0), // set map of option values
T(setrelevant,     0, 1, N, Q), // set rset
T(slice,	   2, 2, N, 0), // slice a list using a second list
T(snippet,	   1, 2, N, 0), // generate snippet from text
T(split,	   1, 2, N, 0), // split a string to give a list
T(stoplist,	   0, 0, N, Q), // return list of stopped terms
T(sub,		   2, 2, N, 0), // subtract
T(substr,	   2, 3, N, 0), // substring
T(suggestion,	   0, 0, N, Q), // misspelled word correction suggestion
T(terms,	   0, 0, N, M), // list of matching terms
T(thispage,	   0, 0, N, M), // page number of current page
T(time,		   0, 0, N, M), // how long the match took (in seconds)
T(topdoc,	   0, 0, N, M), // first document on current page of hit list
				// (counting from 0)
T(topterms,	   0, 1, N, M), // list of up to N top relevance feedback terms
				// (default 16)
T(transform,	   3, 3, N, 0), // transform with a regexp
T(truncate,	   2, 4, N, 0), // truncate after a word
T(uniq,		   1, 1, N, 0), // removed duplicates from a sorted list
T(unpack,	   1, 1, N, 0), // convert 4 byte big endian binary string to a number
T(unstem,	   1, 1, N, Q), // return list of probabilistic terms from
				// the query which stemmed to this term
T(upper,	   1, 1, N, 0), // convert string to upper case
T(url,		   1, 1, N, 0), // url encode argument
T(value,	   1, 2, N, 0), // return document value
T(version,	   0, 0, N, 0), // omega version string
T(weight,	   0, 0, N, 0), // weight of the current hit
{ NULL,{0,	   0, 0, 0, 0}}
};

#undef T // Leaving T defined screws up Sun's C++ compiler!

static vector<string> macros;

// Call write() repeatedly until all data is written or we get a
// non-recoverable error.
static ssize_t
write_all(int fd, const char * buf, size_t count)
{
    while (count) {
	ssize_t r = write(fd, buf, count);
	if (rare(r < 0)) {
	    if (errno == EINTR) continue;
	    return r;
	}
	buf += r;
	count -= r;
    }
    return 0;
}

static string
eval(const string &fmt, const vector<string> &param)
{
    static map<string, const struct func_attrib *> func_map;
    if (func_map.empty()) {
	struct func_desc *p;
	for (p = func_tab; p->name != NULL; p++) {
	    func_map[string(p->name)] = &(p->a);
	}
    }
    string res;
    string::size_type p = 0, q;
    while ((q = fmt.find('$', p)) != string::npos) try {
	res += fmt.substr(p, q - p);
	string::size_type code_start = q; // note down for error reporting
	q++;
	if (q >= fmt.size()) break;
	unsigned char ch = fmt[q];
	switch (ch) {
	    // Magic sequences:
	    // '$$' -> '$', '$(' -> '{', '$)' -> '}', '$.' -> ','
	    case '$':
		res += '$';
		p = q + 1;
		continue;
	    case '(':
		res += '{';
		p = q + 1;
		continue;
	    case ')':
		res += '}';
		p = q + 1;
		continue;
	    case '.':
		res += ',';
		p = q + 1;
		continue;
	    case '_':
		ch = '0';
		// FALL THRU
	    case '1': case '2': case '3': case '4': case '5':
	    case '6': case '7': case '8': case '9':
		ch -= '0';
		if (ch < param.size()) res += param[ch];
		p = q + 1;
		continue;
	    case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
	    case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
	    case 'm': case 'n': case 'o': case 'p': case 'q': case 'r':
	    case 's': case 't': case 'u': case 'v': case 'w': case 'x':
	    case 'y': case 'z':
	    case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
	    case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
	    case 'M': case 'N': case 'O': case 'P': case 'Q': case 'R':
	    case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
	    case 'Y': case 'Z':
	    case '{':
		break;
	    default:
		string msg = "Unknown $ code in: $" + fmt.substr(q);
		throw msg;
	}
	p = find_if(fmt.begin() + q, fmt.end(), p_notid) - fmt.begin();
	string var = fmt.substr(q, p - q);
	map<string, const struct func_attrib *>::const_iterator func;
	func = func_map.find(var);
	if (func == func_map.end()) {
	    throw "Unknown function '" + var + "'";
	}
	vector<string> args;
	if (fmt[p] == '{') {
	    q = p + 1;
	    int nest = 1;
	    while (true) {
		p = fmt.find_first_of(",{}", p + 1);
		if (p == string::npos)
		    throw "missing } in " + fmt.substr(code_start);
		if (fmt[p] == '{') {
		    ++nest;
		} else {
		    if (nest == 1) {
			// should we split the args
			if (func->second->minargs != N) {
			    args.push_back(fmt.substr(q, p - q));
			    q = p + 1;
			}
		    }
		    if (fmt[p] == '}' && --nest == 0) break;
		}
	    }
	    if (func->second->minargs == N)
		args.push_back(fmt.substr(q, p - q));
	    p++;
	}

	if (func->second->minargs != N) {
	    if ((int)args.size() < func->second->minargs)
		throw "too few arguments to $" + var;
	    if (func->second->maxargs != N &&
		(int)args.size() > func->second->maxargs)
		throw "too many arguments to $" + var;

	    vector<string>::size_type n;
	    if (func->second->evalargs != N)
		n = func->second->evalargs;
	    else
		n = args.size();

	    for (vector<string>::size_type j = 0; j < n; j++)
		args[j] = eval(args[j], param);
	}
	if (func->second->ensure == 'Q' || func->second->ensure == 'M')
	    ensure_query_parsed();
	if (func->second->ensure == 'M') ensure_match();
	string value;
	switch (func->second->tag) {
	    case CMD_:
	        break;
	    case CMD_add: {
		int total = 0;
		vector<string>::const_iterator i;
		for (i = args.begin(); i != args.end(); i++)
		    total += string_to_int(*i);
		value = str(total);
		break;
	    }
	    case CMD_addfilter:
		add_bterm(args[0]);
		break;
	    case CMD_allterms: {
		// list of all terms indexing document
		int id = q0;
		if (!args.empty()) id = string_to_int(args[0]);
		Xapian::TermIterator term = db.termlist_begin(id);
		for ( ; term != db.termlist_end(id); term++)
		    value = value + *term + '\t';

		if (!value.empty()) value.erase(value.size() - 1);
		break;
	    }
	    case CMD_and: {
		value = "true";
		for (vector<string>::const_iterator i = args.begin();
		     i != args.end(); i++) {
		    if (eval(*i, param).empty()) {
			value.resize(0);
			break;
		    }
	        }
		break;
	    }
	    case CMD_cgi: {
		MCI i = cgi_params.find(args[0]);
		if (i != cgi_params.end()) value = i->second;
		break;
	    }
	    case CMD_cgilist: {
		pair<MCI, MCI> g;
		g = cgi_params.equal_range(args[0]);
		for (MCI i = g.first; i != g.second; i++)
		    value = value + i->second + '\t';
		if (!value.empty()) value.erase(value.size() - 1);
		break;
	    }
	    case CMD_collapsed: {
		value = str(collapsed);
		break;
	    }
	    case CMD_date:
		value = args[0];
		if (!value.empty()) {
		    char buf[64] = "";
		    time_t date = string_to_int(value);
		    if (date != (time_t)-1) {
			struct tm *then;
			then = gmtime(&date);
			string date_fmt = "%Y-%m-%d";
			if (args.size() > 1) date_fmt = eval(args[1], param);
			strftime(buf, sizeof buf, date_fmt.c_str(), then);
		    }
		    value = buf;
		}
		break;
	    case CMD_dbname:
		value = dbname;
		break;
	    case CMD_dbsize: {
		static Xapian::doccount dbsize;
		if (!dbsize) dbsize = db.get_doccount();
		value = str(dbsize);
		break;
	    }
	    case CMD_def: {
		func_attrib *fa = new func_attrib;
		fa->tag = CMD_MACRO + macros.size();
		fa->minargs = 0;
		fa->maxargs = 9;
		fa->evalargs = N; // FIXME: or 0?
		fa->ensure = 0;

		macros.push_back(args[1]);
		func_map[args[0]] = fa;
		break;
	    }
	    case CMD_defaultop:
		if (default_op == Xapian::Query::OP_AND) {
		    value = "and";
		} else {
		    value = "or";
		}
		break;
	    case CMD_div: {
		int denom = string_to_int(args[1]);
		if (denom == 0) {
		    value = "divide by 0";
		} else {
		    value = str(string_to_int(args[0]) /
				string_to_int(args[1]));
		}
		break;
	    }
	    case CMD_eq:
		if (args[0] == args[1]) value = "true";
		break;
	    case CMD_emptydocs: {
		string t;
		if (!args.empty())
		    t = args[0];
		Xapian::PostingIterator i;
		for (i = db.postlist_begin(t); i != db.postlist_end(t); ++i) {
		    if (i.get_doclength() != 0) continue;
		    if (!value.empty()) value += '\t';
		    value += str(*i);
		}
		break;
	    }
	    case CMD_env: {
		char *env = getenv(args[0].c_str());
		if (env != NULL) value = env;
		break;
	    }
	    case CMD_error:
		if (error_msg.empty() && enquire == NULL && !dbname.empty()) {
		    error_msg = "Database '" + dbname + "' couldn't be opened";
		}
		value = error_msg;
		break;
	    case CMD_field: {
		Xapian::docid did = q0;
		if (args.size() > 1) did = string_to_int(args[1]);
		value = fields.get_field(did, args[0]);
		break;
	    }
	    case CMD_filesize: {
		// FIXME: rounding?  i18n?
		int size = string_to_int(args[0]);
		int intpart = size;
		int fraction = -1;
		const char * format = 0;
		if (size < 0) {
		    // Negative size -> empty result.
		} else if (size == 1) {
		    format = "%d byte";
		} else if (size < 1024) {
		    format = "%d bytes";
		} else {
		    if (size < 1024*1024) {
			format = "%d.%cK";
		    } else {
			size /= 1024;
			if (size < 1024*1024) {
			    format = "%d.%cM";
			} else {
			    size /= 1024;
			    format = "%d.%cG";
			}
		    }
		    intpart = unsigned(size) / 1024;
		    fraction = unsigned(size) % 1024;
		}
		if (format) {
		    char buf[200];
		    int len;
		    if (fraction == -1) {
			len = my_snprintf(buf, sizeof(buf), format, intpart);
		    } else {
			fraction = (fraction * 10 / 1024) + '0';
			len = my_snprintf(buf, sizeof(buf), format, intpart, fraction);
		    }
		    if (len < 0 || (unsigned)len > sizeof(buf)) len = sizeof(buf);
		    value.assign(buf, len);
		}
		break;
	    }
	    case CMD_filters:
		value = filters;
		break;
	    case CMD_filterterms: {
		Xapian::TermIterator term = db.allterms_begin();
		term.skip_to(args[0]);
		while (term != db.allterms_end()) {
		    string t = *term;
		    if (!startswith(t, args[0])) break;
		    value = value + t + '\t';
		    ++term;
		}

		if (!value.empty()) value.erase(value.size() - 1);
		break;
	    }
	    case CMD_find: {
		string l = args[0], s = args[1];
		string::size_type i = 0, j = 0;
		size_t count = 0;
		while (j != l.size()) {
		    j = l.find('\t', i);
		    if (j == string::npos) j = l.size();
		    if (j - i == s.length()) {
			if (memcmp(s.data(), l.data() + i, j - i) == 0) {
			    value = str(count);
			    break;
			}
		    }
		    ++count;
		    i = j + 1;
		}
		break;
	    }
	    case CMD_fmt:
		value = fmtname;
		break;
	    case CMD_freq:
		try {
		    value = str(mset.get_termfreq(args[0]));
		} catch (const Xapian::InvalidOperationError&) {
		    // An MSet will raise this error if it's empty and not
		    // associated with a search.
		    value = str(db.get_termfreq(args[0]));
		}
		break;
            case CMD_ge:
		if (string_to_int(args[0]) >= string_to_int(args[1]))
		    value = "true";
		break;
            case CMD_gt:
		if (string_to_int(args[0]) > string_to_int(args[1]))
		    value = "true";
		break;
	    case CMD_highlight: {
		string bra, ket;
		if (args.size() > 2) {
		    bra = args[2];
		    if (args.size() > 3) {
			ket = args[3];
		    } else {
			string::const_iterator i;
			i = find_if(bra.begin() + 2, bra.end(), p_nottag);
			ket = "</";
			ket += bra.substr(1, i - bra.begin() - 1);
			ket += '>';
		    }
		}

		value = html_highlight(args[0], args[1], bra, ket);
		break;
	    }
	    case CMD_hit:
		// 0-based mset index
		value = str(hit_no);
		break;
	    case CMD_hitlist:
#if 0
		const char *q;
		int ch;

		url_query_string = "?DB=";
		url_query_string += dbname;
		url_query_string += "&P=";
		q = probabilistic_query[string()].c_str();
		while ((ch = *q++) != '\0') {
		    switch (ch) {
		     case '+':
			url_query_string += "%2b";
			break;
		     case '"':
			url_query_string += "%22";
			break;
		     case ' ':
			ch = '+';
			/* fall through */
		     default:
			url_query_string += ch;
		    }
		}
	        // add any boolean terms
		for (FMCI i = filter_map.begin(); i != filter_map.end(); i++) {
		    url_query_string += "&B=";
		    url_query_string += i->second;
		}
#endif
		for (hit_no = topdoc; hit_no < last; hit_no++)
		    value += print_caption(args[0], param);
		hit_no = 0;
		break;
	    case CMD_hitsperpage:
		value = str(hits_per_page);
		break;
	    case CMD_hostname: {
	        value = args[0];
		// remove URL scheme and/or path
		string::size_type i = value.find("://");
		if (i == string::npos) i = 0; else i += 3;
		value = value.substr(i, value.find('/', i) - i);
		// remove user@ or user:password@
		i = value.find('@');
		if (i != string::npos) value.erase(0, i + 1);
		// remove :port
		i = value.find(':');
		if (i != string::npos) value.resize(i);
		break;
	    }
	    case CMD_html:
	        value = html_escape(args[0]);
		break;
	    case CMD_htmlstrip:
	        value = html_strip(args[0]);
		break;
	    case CMD_httpheader:
		if (!suppress_http_headers) {
		    cout << args[0] << ": " << args[1] << endl;
		    if (!set_content_type && args[0].length() == 12 &&
			    strcasecmp(args[0].c_str(), "Content-Type") == 0) {
			set_content_type = true;
		    }
		}
	        break;
	    case CMD_id:
		// document id
		value = str(q0);
		break;
	    case CMD_if:
		if (!args[0].empty())
		    value = eval(args[1], param);
		else if (args.size() > 2)
		    value = eval(args[2], param);
		break;
	    case CMD_include:
	        value = eval_file(args[0]);
	        break;
	    case CMD_json:
		value = args[0];
		json_escape(value);
		break;
	    case CMD_jsonarray: {
		const string & l = args[0];
		string::size_type i = 0, j;
		if (l.empty()) {
		    value = "[]";
		    break;
		}
		value = "[\"]";
		while (true) {
		    j = l.find('\t', i);
		    string elt(l, i, j - i);
		    json_escape(elt);
		    value += elt;
		    if (j == string::npos) break;
		    value += "\",\"";
		    i = j + 1;
		}
		value += "\"]";
		break;
	    }
	    case CMD_last:
		value = str(last);
		break;
	    case CMD_lastpage: {
		int l = mset.get_matches_estimated();
		if (l > 0) l = (l - 1) / hits_per_page + 1;
		value = str(l);
		break;
	    }
            case CMD_le:
		if (string_to_int(args[0]) <= string_to_int(args[1]))
		    value = "true";
		break;
            case CMD_length:
		if (args[0].empty()) {
		    value = "0";
		} else {
		    size_t length = count(args[0].begin(), args[0].end(), '\t');
		    value = str(length + 1);
		}
		break;
	    case CMD_list: {
		if (!args[0].empty()) {
		    string pre, inter, interlast, post;
		    switch (args.size()) {
		     case 2:
			inter = interlast = args[1];
			break;
		     case 3:
			inter = args[1];
			interlast = args[2];
			break;
		     case 4:
			pre = args[1];
			inter = interlast = args[2];
			post = args[3];
			break;
		     case 5:
			pre = args[1];
			inter = args[2];
			interlast = args[3];
			post = args[4];
			break;
		    }
		    value += pre;
		    string list = args[0];
		    string::size_type split = 0, split2;
		    while ((split2 = list.find('\t', split)) != string::npos) {
			if (split) value += inter;
			value += list.substr(split, split2 - split);
			split = split2 + 1;
		    }
		    if (split) value += interlast;
		    value += list.substr(split);
		    value += post;
		}
		break;
	    }
	    case CMD_log: {
		if (!vet_filename(args[0])) break;
		string logfile = log_dir + args[0];
	        int fd = open(logfile.c_str(), O_CREAT|O_APPEND|O_WRONLY, 0644);
		if (fd == -1) break;
		vector<string> noargs;
		noargs.resize(1);
		string line;
		if (args.size() > 1) {
		    line = args[1];
		} else {
		    line = DEFAULT_LOG_ENTRY;
		}
		line = eval(line, noargs);
		line += '\n';
		(void)write_all(fd, line.data(), line.length());
		close(fd);
		break;
	    }
	    case CMD_lookup: {
		if (!vet_filename(args[0])) break;
		string cdbfile = cdb_dir + args[0];
	        int fd = open(cdbfile.c_str(), O_RDONLY);
		if (fd == -1) break;

		struct cdb cdb;
		cdb_init(&cdb, fd);

		if (cdb_find(&cdb, args[1].data(), args[1].length()) > 0) {
		    size_t datalen = cdb_datalen(&cdb);
		    const void *dat = cdb_get(&cdb, datalen, cdb_datapos(&cdb));
		    if (q) {
			value = string(static_cast<const char *>(dat), datalen);
		    }
		}

		cdb_free(&cdb);
		close(fd); // FIXME: cache fds?
		break;
	    }
	    case CMD_lower:
		value = Xapian::Unicode::tolower(args[0]);
		break;
            case CMD_lt:
		if (string_to_int(args[0]) < string_to_int(args[1]))
		    value = "true";
		break;
	    case CMD_map:
		if (!args[0].empty()) {
		    string l = args[0], pat = args[1];
		    vector<string> new_args(param);
		    string::size_type i = 0, j;
		    while (true) {
			j = l.find('\t', i);
			new_args[0] = l.substr(i, j - i);
			value += eval(pat, new_args);
			if (j == string::npos) break;
			value += '\t';
			i = j + 1;
		    }
		}
	        break;
	    case CMD_max: {
		vector<string>::const_iterator i = args.begin();
		int val = string_to_int(*i++);
		for (; i != args.end(); i++) {
		    int x = string_to_int(*i);
		    if (x > val) val = x;
	        }
		value = str(val);
		break;
	    }
	    case CMD_min: {
		vector<string>::const_iterator i = args.begin();
		int val = string_to_int(*i++);
		for (; i != args.end(); i++) {
		    int x = string_to_int(*i);
		    if (x < val) val = x;
	        }
		value = str(val);
		break;
	    }
	    case CMD_msize:
		// number of matches
		value = str(mset.get_matches_estimated());
		break;
	    case CMD_msizeexact:
		// is msize exact?
		if (mset.get_matches_lower_bound()
		    == mset.get_matches_upper_bound())
		    value = "true";
		break;
	    case CMD_mod: {
		int denom = string_to_int(args[1]);
		if (denom == 0) {
		    value = "divide by 0";
		} else {
		    value = str(string_to_int(args[0]) %
				string_to_int(args[1]));
		}
		break;
	    }
	    case CMD_mul: {
		vector<string>::const_iterator i = args.begin();
		int total = string_to_int(*i++);
		while (i != args.end())
		    total *= string_to_int(*i++);
		value = str(total);
		break;
	    }
	    case CMD_muldiv: {
		int denom = string_to_int(args[2]);
		if (denom == 0) {
		    value = "divide by 0";
		} else {
		    int num = string_to_int(args[0]) * string_to_int(args[1]);
		    value = str(num / denom);
		}
		break;
	    }
            case CMD_ne:
		if (args[0] != args[1]) value = "true";
		break;
	    case CMD_nice: {
		string::const_iterator i = args[0].begin();
		int len = args[0].length();
		while (len) {
		    value += *i++;
		    if (--len && len % 3 == 0) value += option["thousand"];
		}
		break;
	    }
	    case CMD_not:
		if (args[0].empty()) value = "true";
		break;
	    case CMD_now: {
		char buf[64];
		my_snprintf(buf, sizeof(buf), "%lu", (unsigned long)time(NULL));
		// MSVC's snprintf omits the zero byte if the string if
		// sizeof(buf) long.
		buf[sizeof(buf) - 1] = '\0';
		value = buf;
		break;
	    }
	    case CMD_opt:
		if (args.size() == 2) {
		    value = option[args[0] + "," + args[1]];
		} else {
		    value = option[args[0]];
		}
		break;
	    case CMD_or: {
		for (vector<string>::const_iterator i = args.begin();
		     i != args.end(); i++) {
		    value = eval(*i, param);
		    if (!value.empty()) break;
	        }
		break;
	    }
	    case CMD_pack:
		value = int_to_binary_string(string_to_int(args[0]));
		break;
	    case CMD_percentage:
		// percentage score
		value = str(percent);
		break;
	    case CMD_prettyterm:
		value = pretty_term(args[0]);
		break;
	    case CMD_prettyurl:
		value = args[0];
		url_prettify(value);
		break;
	    case CMD_query:
		value = probabilistic_query[args.empty() ? string() : args[0]];
		break;
	    case CMD_querydescription:
		value = query.get_description();
		break;
	    case CMD_queryterms:
		value = queryterms;
		break;
	    case CMD_range: {
		int start = string_to_int(args[0]);
		int end = string_to_int(args[1]);
	        while (start <= end) {
		    value += str(start);
		    if (start < end) value += '\t';
		    start++;
		}
		break;
	    }
	    case CMD_record: {
		int id = q0;
		if (!args.empty()) id = string_to_int(args[0]);
		value = db.get_document(id).get_data();
		break;
	    }
	    case CMD_relevant: {
		// document id if relevant; empty otherwise
		int id = q0;
		if (!args.empty()) id = string_to_int(args[0]);
		map<Xapian::docid, bool>::iterator i = ticked.find(id);
		if (i != ticked.end()) {
		    i->second = false; // icky side-effect
		    value = str(id);
		}
		break;
	    }
	    case CMD_relevants:	{
		for (map <Xapian::docid, bool>::const_iterator i = ticked.begin();
		     i != ticked.end(); i++) {
		    if (i->second) {
			value += str(i->first);
			value += '\t';
		    }
		}
		if (!value.empty()) value.erase(value.size() - 1);
		break;
	    }
	    case CMD_score:
	        // Score (0 to 10)
		value = str(percent / 10);
		break;
	    case CMD_set:
		option[args[0]] = args[1];
		break;
	    case CMD_setmap: {
		string base = args[0] + ',';
		if (args.size() % 2 != 1)
		    throw string("$setmap requires an odd number of arguments");
		for (unsigned int i = 1; i + 1 < args.size(); i += 2) {
		    option[base + args[i]] = args[i + 1];
		}
		break;
	    }
	    case CMD_setrelevant: {
		string::size_type i = 0, j;
		while (true) {
		    j = args[0].find_first_not_of("0123456789", i);
		    Xapian::docid id = atoi(args[0].substr(i, j - i).c_str());
		    if (id) {
			rset.add_document(id);
			ticked[id] = true;
		    }
		    if (j == string::npos) break;
		    i = j + 1;
		}
		break;
	    }
	    case CMD_slice: {
		string list = args[0], pos = args[1];
		vector<string> items;
		string::size_type i = 0, j;
		while (true) {
		    j = list.find('\t', i);
		    items.push_back(list.substr(i, j - i));
		    if (j ==…