scriptindex: Fix weird error cases

[xapian.git] / xapian-applications / omega / index_file.cc

/** @file
 * @brief Handle indexing a document from a file
 */
/* Copyright 1999,2000,2001 BrightStation PLC
 * Copyright 2001,2005 James Aylett
 * Copyright 2001,2002 Ananova Ltd
 * Copyright 2002,2003,2004,2005,2006,2007,2008,2009,2010,2011,2012,2013,2014,2015,2016,2017,2018 Olly Betts
 * Copyright 2009 Frank J Bruzzaniti
 * Copyright 2012 Mihai Bivol
 *
 * 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 "index_file.h"

#include <algorithm>
#include <iostream>
#include <limits>
#include <string>
#include <map>
#include <vector>

#include <sys/types.h>
#include "safeunistd.h"
#include <cerrno>
#include <cstdio>
#include <cstdlib>
#include <cstring>
#include "safefcntl.h"
#include <ctime>

#include <xapian.h>

#include "append_filename_arg.h"
#include "atomparse.h"
#include "diritor.h"
#include "failed.h"
#include "md5wrap.h"
#include "metaxmlparse.h"
#include "mimemap.h"
#include "msxmlparse.h"
#include "myhtmlparse.h"
#include "opendocparse.h"
#include "pkglibbindir.h"
#include "runfilter.h"
#include "sample.h"
#include "str.h"
#include "stringutils.h"
#include "svgparse.h"
#include "tmpdir.h"
#include "utf8convert.h"
#include "utils.h"
#include "values.h"
#include "xmlparse.h"
#include "xlsxparse.h"
#include "xpsxmlparse.h"

using namespace std;

static Xapian::WritableDatabase db;
static Xapian::TermGenerator indexer;

static Xapian::doccount old_docs_not_seen;
static Xapian::docid old_lastdocid;
static vector<bool> updated;

static bool verbose;
static bool retry_failed;
static bool use_ctime;
static dup_action_type dup_action;
static bool ignore_exclusions;
static bool description_as_sample;

static time_t last_altered_max;
static size_t sample_size;
static size_t title_size;
static size_t max_ext_len;

static empty_body_type empty_body;

static string root;
static string site_term, host_term;

static Failed failed;

map<string, Filter> commands;

static void
mark_as_seen(Xapian::docid did)
{
    if (usual(did < updated.size() && !updated[did])) {
        updated[did] = true;
        --old_docs_not_seen;
    }
}

void
skip(const string & urlterm, const string & context, const string & msg,
     off_t size, time_t last_mod, unsigned flags)
{
    failed.add(urlterm, last_mod, size);

    if (!verbose || (flags & SKIP_SHOW_FILENAME)) {
        if (!verbose && (flags & SKIP_VERBOSE_ONLY)) return;
        cout << context << ": ";
    }

    cout << "Skipping - " << msg << endl;
}

static void
skip_cmd_failed(const string & urlterm, const string & context, const string & cmd,
                off_t size, time_t last_mod)
{
    skip(urlterm, context, "\"" + cmd + "\" failed", size, last_mod);
}

static void
skip_meta_tag(const string & urlterm, const string & context,
              off_t size, time_t last_mod)
{
    skip(urlterm, context, "indexing disallowed by meta tag", size, last_mod);
}

static void
skip_unknown_mimetype(const string & urlterm, const string & context,
                      const string & mimetype, off_t size, time_t last_mod)
{
    skip(urlterm, context, "unknown MIME type '" + mimetype + "'", size, last_mod);
}

void
index_add_default_filters()
{
    index_command("application/msword", Filter("antiword -mUTF-8.txt", false));
    index_command("application/vnd.ms-excel",
                  Filter("xls2csv -c' ' -q0 -dutf-8", false));
    index_command("application/vnd.ms-powerpoint",
                  Filter("catppt -dutf-8", false));
    // Looking at the source of wpd2html and wpd2text I think both output
    // UTF-8, but it's hard to be sure without sample Unicode .wpd files
    // as they don't seem to be at all well documented.
    index_command("application/vnd.wordperfect", Filter("wpd2text", false));
    // wps2text produces UTF-8 output from the sample files I've tested.
    index_command("application/vnd.ms-works", Filter("wps2text", false));
    // Output is UTF-8 according to "man djvutxt".  Generally this seems to
    // be true, though some examples from djvu.org generate isolated byte
    // 0x95 in a context which suggests it might be intended to be a bullet
    // (as it is in CP1252).
    index_command("image/vnd.djvu", Filter("djvutxt", false));
    index_command("text/markdown", Filter("markdown", "text/html", false));
    // The --text option unhelpfully converts all non-ASCII characters to "?"
    // so we use --html instead, which produces HTML entities.  The --nopict
    // option suppresses exporting picture files as pictNNNN.wmf in the current
    // directory.  Note that this option was ignored in some older versions,
    // but it was fixed in unrtf 0.20.4.
    index_command("application/rtf",
                  Filter("unrtf --nopict --html 2>/dev/null", "text/html",
                         false));
    index_command("text/rtf",
                  Filter("unrtf --nopict --html 2>/dev/null", "text/html",
                         false));
    index_command("text/x-rst", Filter("rst2html", "text/html", false));
    index_command("application/x-mspublisher",
                  Filter("pub2xhtml", "text/html", false));
    index_command("application/vnd.ms-outlook",
                  Filter(get_pkglibbindir() + "/outlookmsg2html", "text/html",
                         false));
    index_command("application/vnd.ms-visio.drawing",
                  Filter("vsd2xhtml", "image/svg+xml", false));
    index_command("application/vnd.ms-visio.stencil",
                  Filter("vsd2xhtml", "image/svg+xml", false));
    index_command("application/vnd.ms-visio.template",
                  Filter("vsd2xhtml", "image/svg+xml", false));
    index_command("application/vnd.visio",
                  Filter("vsd2xhtml", "image/svg+xml", false));
    // pod2text's output character set doesn't seem to be documented, but from
    // inspecting the source it looks like it's probably iso-8859-1.  We need
    // to pass "--errors=stderr" or else minor POD formatting errors cause a
    // file not to be indexed.
    index_command("text/x-perl",
                  Filter("pod2text --errors=stderr",
                         "text/plain", "iso-8859-1", false));
    // FIXME: -e0 means "UTF-8", but that results in "fi", "ff", "ffi", etc
    // appearing as single ligatures.  For European languages, it's actually
    // better to use -e2 (ISO-8859-1) and then convert, so let's do that for
    // now until we handle Unicode "compatibility decompositions".
    index_command("application/x-dvi",
                  Filter("catdvi -e2 -s", "text/plain", "iso-8859-1", false));
    // Simplistic - ought to look in index.rdf files for filename and character
    // set.
    index_command("application/x-maff",
                  Filter("unzip -p %f '*/*.*htm*'", "text/html", "iso-8859-1",
                         false));
    index_command("application/x-mimearchive",
                  Filter(get_pkglibbindir() + "/mhtml2html", "text/html",
                         false));
    index_command("message/news",
                  Filter(get_pkglibbindir() + "/rfc822tohtml", "text/html",
                         false));
    index_command("message/rfc822",
                  Filter(get_pkglibbindir() + "/rfc822tohtml", "text/html",
                         false));
    index_command("text/vcard",
                  Filter(get_pkglibbindir() + "/vcard2text", false));
    index_command("application/vnd.apple.keynote",
                  Filter("key2text", false));
    index_command("application/vnd.apple.numbers",
                  Filter("numbers2text", false));
    index_command("application/vnd.apple.pages",
                  Filter("pages2text", false));
}

void
index_init(const string & dbpath, const Xapian::Stem & stemmer,
           const string & root_, const string & site_term_,
           const string & host_term_,
           empty_body_type empty_body_, dup_action_type dup_action_,
           size_t sample_size_, size_t title_size_, size_t max_ext_len_,
           bool overwrite, bool retry_failed_,
           bool delete_removed_documents, bool verbose_, bool use_ctime_,
           bool spelling, bool ignore_exclusions_, bool description_as_sample_)
{
    root = root_;
    site_term = site_term_;
    host_term = host_term_;
    empty_body = empty_body_;
    dup_action = dup_action_;
    sample_size = sample_size_;
    title_size = title_size_;
    max_ext_len = max_ext_len_;
    verbose = verbose_;
    use_ctime = use_ctime_;
    ignore_exclusions = ignore_exclusions_;
    description_as_sample = description_as_sample_;

    if (!overwrite) {
        db = Xapian::WritableDatabase(dbpath, Xapian::DB_CREATE_OR_OPEN);
        old_docs_not_seen = db.get_doccount();
        // Handle an initially empty database exactly the same way as when
        // overwrite is true.
        if (old_docs_not_seen != 0) {
            old_lastdocid = db.get_lastdocid();
            if (delete_removed_documents) {
                // + 1 so that old_lastdocid is a valid subscript.
                updated.resize(old_lastdocid + 1);
            }
            Xapian::valueno slot = use_ctime ? VALUE_CTIME : VALUE_LASTMOD;
            string ubound = db.get_value_upper_bound(slot);
            if (!ubound.empty())
                last_altered_max = binary_string_to_int(ubound);
        }
    } else {
        db = Xapian::WritableDatabase(dbpath, Xapian::DB_CREATE_OR_OVERWRITE);
    }

    if (spelling) {
        indexer.set_database(db);
        indexer.set_flags(indexer.FLAG_SPELLING);
    }
    indexer.set_stemmer(stemmer);

    runfilter_init();

    failed.init(db);

    if (overwrite) {
        // There are no failures to retry, so setting this flag doesn't
        // change the outcome, but does mean we avoid the overhead of
        // checking for a previous failure.
        retry_failed = true;
    } else if (retry_failed_) {
        failed.clear();
        retry_failed = true;
    } else {
        // If there are no existing failures, setting this flag doesn't
        // change the outcome, but does mean we avoid the overhead of
        // checking for a previous failure.
        retry_failed = failed.empty();
    }
}

static void
parse_pdfinfo_field(const char * p, const char * end, string & out, const char * field, size_t len)
{
    if (size_t(end - p) > len && memcmp(p, field, len) == 0) {
        p += len;
        while (p != end && *p == ' ')
            ++p;
        if (p != end && (end[-1] != '\r' || --end != p))
            out.assign(p, end - p);
    }
}

#define PARSE_PDFINFO_FIELD(P, END, OUT, FIELD) \
    parse_pdfinfo_field((P), (END), (OUT), FIELD":", CONST_STRLEN(FIELD) + 1)

static void
parse_pdf_metainfo(const string& pdfinfo, string &author, string &title,
                   string &keywords, string &topic, int& pages)
{
    const char * p = pdfinfo.data();
    const char * end = p + pdfinfo.size();
    while (p != end) {
        const char * start = p;
        p = static_cast<const char *>(memchr(p, '\n', end - p));
        const char * eol;
        if (p) {
            eol = p;
            ++p;
        } else {
            p = eol = end;
        }
        switch (*start) {
            case 'A':
                PARSE_PDFINFO_FIELD(start, eol, author, "Author");
                break;
            case 'K':
                PARSE_PDFINFO_FIELD(start, eol, keywords, "Keywords");
                break;
            case 'P': {
                string s;
                PARSE_PDFINFO_FIELD(start, eol, s, "Pages");
                if (!s.empty())
                    pages = atoi(s.c_str());
                break;
            }
            case 'S':
                PARSE_PDFINFO_FIELD(start, eol, topic, "Subject");
                break;
            case 'T':
                PARSE_PDFINFO_FIELD(start, eol, title, "Title");
                break;
        }
    }
}

static void
get_pdf_metainfo(int fd, string &author, string &title,
                 string &keywords, string &topic, int& pages)
{
    try {
        string pdfinfo;
        run_filter(fd, "pdfinfo -enc UTF-8 -", false, &pdfinfo);
        parse_pdf_metainfo(pdfinfo, author, title, keywords, topic, pages);
    } catch (const ReadError&) {
        // It's probably best to index the document even if pdfinfo fails.
    }
}

static void
get_pdf_metainfo(const string& file, string &author, string &title,
                 string &keywords, string &topic, int& pages)
{
    try {
        string cmd = "pdfinfo -enc UTF-8";
        append_filename_argument(cmd, file);
        parse_pdf_metainfo(stdout_to_string(cmd, false),
                           author, title, keywords, topic, pages);
    } catch (const ReadError&) {
        // It's probably best to index the document even if pdfinfo fails.
    }
}

static void
generate_sample_from_csv(const string & csv_data, string & sample)
{
    // Add 3 to allow for a 4 byte utf-8 sequence being appended when
    // output is sample_size - 1 bytes long.  Use csv_data.size() if smaller
    // since the user might reasonably set sample_size really high.
    sample.reserve(min(sample_size + 3, csv_data.size()));
    size_t last_word_end = 0;
    bool in_space = true;
    bool in_quotes = false;
    for (Xapian::Utf8Iterator i(csv_data); i != Xapian::Utf8Iterator(); ++i) {
        unsigned ch = *i;

        if (!in_quotes) {
            // If not already in double quotes, '"' starts quoting and
            // ',' starts a new field.
            if (ch == '"') {
                in_quotes = true;
                continue;
            }
            if (ch == ',')
                ch = ' ';
        } else if (ch == '"') {
            // In double quotes, '"' either ends double quotes, or
            // if followed by another '"', means a literal '"'.
            if (++i == Xapian::Utf8Iterator())
                break;
            ch = *i;
            if (ch != '"') {
                in_quotes = false;
                if (ch == ',')
                    ch = ' ';
            }
        }

        if (ch <= ' ' || ch == 0xa0) {
            // FIXME: if all the whitespace characters between two
            // words are 0xa0 (non-breaking space) then perhaps we
            // should output 0xa0.
            if (in_space)
                continue;
            last_word_end = sample.size();
            sample += ' ';
            in_space = true;
        } else {
            Xapian::Unicode::append_utf8(sample, ch);
            in_space = false;
        }

        if (sample.size() >= sample_size) {
            // Need to truncate sample.
            if (last_word_end <= sample_size / 2) {
                // Monster word!  We'll have to just split it.
                sample.replace(sample_size - 3, string::npos, "...", 3);
            } else {
                sample.replace(last_word_end, string::npos, " ...", 4);
            }
            break;
        }
    }
}

static bool
index_check_existing(const string & urlterm, time_t last_altered,
                     Xapian::docid & did)
{
    switch (dup_action) {
        case DUP_SKIP: {
            Xapian::PostingIterator p = db.postlist_begin(urlterm);
            if (p != db.postlist_end(urlterm)) {
                if (verbose)
                    cout << "already indexed, not updating" << endl;
                did = *p;
                mark_as_seen(did);
                return true;
            }
            break;
        }
        case DUP_CHECK_LAZILY: {
            // If last_altered > last_altered_max, we know for sure that the
            // file is new or updated.
            if (last_altered > last_altered_max) {
                return false;
            }

            Xapian::PostingIterator p = db.postlist_begin(urlterm);
            if (p != db.postlist_end(urlterm)) {
                did = *p;
                Xapian::Document doc = db.get_document(did);
                Xapian::valueno slot = use_ctime ? VALUE_CTIME : VALUE_LASTMOD;
                string value = doc.get_value(slot);
                time_t old_last_altered = binary_string_to_int(value);
                if (last_altered <= old_last_altered) {
                    if (verbose)
                        cout << "already indexed" << endl;
                    // The docid should be in updated - the only valid
                    // exception is if the URL was long and hashed to the
                    // same URL as an existing document indexed in the same
                    // batch.
                    mark_as_seen(did);
                    return true;
                }
            }
            break;
        }
    }
    return false;
}

void
index_remove_failed_entry(const string& urlterm)
{
    failed.del(urlterm);
}

void
index_add_document(const string & urlterm, time_t last_altered,
                   Xapian::docid did, const Xapian::Document & doc)
{
    if (dup_action != DUP_SKIP) {
        // If this document has already been indexed, update the existing
        // entry.
        if (did) {
            // We already found out the document id above.
            db.replace_document(did, doc);
        } else if (last_altered <= last_altered_max) {
            // We checked for the UID term and didn't find it.
            did = db.add_document(doc);
        } else {
            did = db.replace_document(urlterm, doc);
        }
        mark_as_seen(did);
        if (verbose) {
            if (did <= old_lastdocid) {
                cout << "updated" << endl;
            } else {
                cout << "added" << endl;
            }
        }
    } else {
        // If this were a duplicate, we'd have skipped it above.
        db.add_document(doc);
        if (verbose)
            cout << "added" << endl;
    }
}

void
index_mimetype(const string & file, const string & urlterm, const string & url,
               const string & ext,
               const string &mimetype, DirectoryIterator &d,
               Xapian::Document & newdocument,
               string record)
{
    string context(file, root.size(), string::npos);

    // FIXME: We could be cleverer here and check mtime too when use_ctime is
    // set - if the ctime has changed but the mtime is unchanged, we can just
    // update the existing Document and avoid having to re-extract text, etc.
    time_t last_altered = use_ctime ? d.get_ctime() : d.get_mtime();

    Xapian::docid did = 0;
    if (index_check_existing(urlterm, last_altered, did))
        return;

    if (!retry_failed) {
        // We only store and check the mtime (last modified) - a change to the
        // metadata won't generally cause a previous failure to now work
        // (FIXME: except permissions).
        time_t failed_last_mod;
        off_t failed_size;
        if (failed.contains(urlterm, failed_last_mod, failed_size)) {
            if (d.get_mtime() <= failed_last_mod &&
                d.get_size() == failed_size) {
                if (verbose)
                    cout << "failed to extract text on earlier run" << endl;
                return;
            }
            // The file has changed, so remove the entry for it.  If it fails
            // again on this attempt, we'll add a new one.
            failed.del(urlterm);
        }
    }

    if (verbose) cout << flush;

    string author, title, sample, keywords, topic, dump;
    string md5;
    time_t created = time_t(-1);
    int pages = -1;

    map<string, Filter>::const_iterator cmd_it = commands.find(mimetype);
    if (cmd_it == commands.end()) {
        size_t slash = mimetype.find('/');
        if (slash != string::npos) {
            string wildtype(mimetype, 0, slash + 2);
            wildtype[slash + 1] = '*';
            cmd_it = commands.find(wildtype);
            if (cmd_it == commands.end()) {
                cmd_it = commands.find("*/*");
            }
        }
        if (cmd_it == commands.end()) {
            cmd_it = commands.find("*");
        }
    }
    try {
        if (cmd_it != commands.end()) {
            // Easy "run a command and read text or HTML from stdout or a
            // temporary file" cases.
            auto& filter = cmd_it->second;
            string cmd = filter.cmd;
            if (cmd.empty()) {
                skip(urlterm, context, "required filter not installed",
                     d.get_size(), d.get_mtime(), SKIP_VERBOSE_ONLY);
                return;
            }
            if (cmd == "false") {
                // Allow setting 'false' as a filter to mean that a MIME type
                // should be quietly ignored.
                string m = "ignoring MIME type '";
                m += cmd_it->first;
                m += "'";
                skip(urlterm, context, m, d.get_size(), d.get_mtime(),
                     SKIP_VERBOSE_ONLY);
                return;
            }
            bool use_shell = filter.use_shell();
            bool substituted = false;
            string tmpout;
            size_t pcent = 0;
            while (true) {
                pcent = cmd.find('%', pcent);
                if (pcent >= cmd.size() - 1)
                    break;
                switch (cmd[pcent + 1]) {
                    case '%': // %% -> %.
                        cmd.erase(++pcent, 1);
                        break;
                    case 'f': { // %f -> escaped filename.
                        substituted = true;
                        string tail(cmd, pcent + 2);
                        cmd.resize(pcent);
                        // Suppress the space append_filename_argument()
                        // usually adds before the argument - the command
                        // string either includes one, or won't expect one
                        // (e.g. --input=%f).
                        append_filename_argument(cmd, file, false);
                        pcent = cmd.size();
                        cmd += tail;
                        break;
                    }
                    case 't': { // %t -> temporary output file.
                        if (tmpout.empty()) {
                            // Use a temporary file with a suitable extension
                            // in case the command cares, and for more helpful
                            // error messages from the command.
                            if (filter.output_type == "text/html") {
                                tmpout = get_tmpfile("tmp.html");
                            } else if (filter.output_type == "image/svg+xml") {
                                tmpout = get_tmpfile("tmp.svg");
                            } else {
                                tmpout = get_tmpfile("tmp.txt");
                            }
                        }
                        substituted = true;
                        string tail(cmd, pcent + 2);
                        cmd.resize(pcent);
                        // Suppress the space append_filename_argument()
                        // usually adds before the argument - the command
                        // string either includes one, or won't expect one
                        // (e.g. --output=%t).
                        append_filename_argument(cmd, tmpout, false);
                        pcent = cmd.size();
                        cmd += tail;
                        break;
                    }
                    default:
                        // Leave anything else alone for now.
                        pcent += 2;
                        break;
                }
            }
            if (!substituted && cmd != "true") {
                // If no %f, append the filename to the command.
                append_filename_argument(cmd, file);
            }
            try {
                if (!tmpout.empty()) {
                    // Output in temporary file.
                    run_filter(cmd, use_shell);
                    if (!load_file(tmpout, dump, NOCACHE)) {
                        throw ReadError("Couldn't read output file");
                    }
                    unlink(tmpout.c_str());
                } else if (cmd == "true") {
                    // Ignore the file's contents, just index metadata from the
                    // filing system.
                } else {
                    // Output on stdout.
                    run_filter(cmd, use_shell, &dump);
                }
                const string & charset = filter.output_charset;
                if (filter.output_type == "text/html") {
                    MyHtmlParser p;
                    p.ignore_metarobots();
                    p.description_as_sample = description_as_sample;
                    try {
                        p.parse_html(dump, charset, false);
                    } catch (const string & newcharset) {
                        p.reset();
                        p.ignore_metarobots();
                        p.description_as_sample = description_as_sample;
                        p.parse_html(dump, newcharset, true);
                    } catch (const ReadError&) {
                        skip_cmd_failed(urlterm, context, cmd,
                                        d.get_size(), d.get_mtime());
                        return;
                    }
                    dump = p.dump;
                    title = p.title;
                    keywords = p.keywords;
                    topic = p.topic;
                    sample = p.sample;
                    author = p.author;
                    created = p.created;
                } else if (filter.output_type == "image/svg+xml") {
                    SvgParser svgparser;
                    svgparser.parse(dump);
                    dump = svgparser.dump;
                    title = svgparser.title;
                    keywords = svgparser.keywords;
                    // FIXME: topic = svgparser.topic;
                    author = svgparser.author;
                } else if (!charset.empty()) {
                    convert_to_utf8(dump, charset);
                }
            } catch (const ReadError&) {
                skip_cmd_failed(urlterm, context, cmd,
                                d.get_size(), d.get_mtime());
                return;
            }
        } else if (mimetype == "text/html" || mimetype == "text/x-php") {
            const string & text = d.file_to_string();
            MyHtmlParser p;
            if (ignore_exclusions) p.ignore_metarobots();
            p.description_as_sample = description_as_sample;
            try {
                // Default HTML character set is latin 1, though not specifying
                // one is deprecated these days.
                p.parse_html(text, "iso-8859-1", false);
            } catch (const string & newcharset) {
                p.reset();
                if (ignore_exclusions) p.ignore_metarobots();
                p.description_as_sample = description_as_sample;
                p.parse_html(text, newcharset, true);
            }
            if (!p.indexing_allowed) {
                skip_meta_tag(urlterm, context,
                              d.get_size(), d.get_mtime());
                return;
            }
            dump = p.dump;
            title = p.title;
            keywords = p.keywords;
            topic = p.topic;
            sample = p.sample;
            author = p.author;
            created = p.created;
            md5_string(text, md5);
        } else if (mimetype == "text/plain") {
            // Currently we assume that text files are UTF-8 unless they have a
            // byte-order mark.
            dump = d.file_to_string();
            md5_string(dump, md5);

            // Look for Byte-Order Mark (BOM).
            if (startswith(dump, "\xfe\xff") || startswith(dump, "\xff\xfe")) {
                // UTF-16 in big-endian/little-endian order - we just convert
                // it as "UTF-16" and let the conversion handle the BOM as that
                // way we avoid the copying overhead of erasing 2 bytes from
                // the start of dump.
                convert_to_utf8(dump, "UTF-16");
            } else if (startswith(dump, "\xef\xbb\xbf")) {
                // UTF-8 with stupid Windows not-the-byte-order mark.
                dump.erase(0, 3);
            } else {
                // FIXME: What charset is the file?  Look at contents?
            }
        } else if (mimetype == "application/pdf") {
            const char* cmd = "pdftotext -enc UTF-8 - -";
            try {
                run_filter(d.get_fd(), cmd, false, &dump);
            } catch (const ReadError&) {
                skip_cmd_failed(urlterm, context, cmd,
                                d.get_size(), d.get_mtime());
                return;
            }
            get_pdf_metainfo(d.get_fd(), author, title, keywords, topic, pages);
        } else if (mimetype == "application/postscript") {
            // There simply doesn't seem to be a Unicode capable PostScript to
            // text converter (e.g. pstotext always outputs ISO-8859-1).  The
            // only solution seems to be to convert via PDF using ps2pdf and
            // then pdftotext.  This gives plausible looking UTF-8 output for
            // some Chinese PostScript files I found using Google.  It also has
            // the benefit of allowing us to extract meta information from
            // PostScript files.
            string tmpfile = get_tmpfile("tmp.pdf");
            if (tmpfile.empty()) {
                // FIXME: should this be fatal?  Or disable indexing postscript?
                string msg = "Couldn't create temporary directory (";
                msg += strerror(errno);
                msg += ")";
                skip(urlterm, context, msg,
                     d.get_size(), d.get_mtime());
                return;
            }
            string cmd = "ps2pdf -";
            append_filename_argument(cmd, tmpfile);
            try {
                run_filter(d.get_fd(), cmd, false);
                cmd = "pdftotext -enc UTF-8";
                append_filename_argument(cmd, tmpfile);
                cmd += " -";
                run_filter(cmd, false, &dump);
            } catch (const ReadError&) {
                skip_cmd_failed(urlterm, context, cmd,
                                d.get_size(), d.get_mtime());
                unlink(tmpfile.c_str());
                return;
            } catch (...) {
                unlink(tmpfile.c_str());
                throw;
            }
            try {
                get_pdf_metainfo(tmpfile, author, title, keywords, topic,
                                 pages);
            } catch (...) {
                unlink(tmpfile.c_str());
                throw;
            }
            unlink(tmpfile.c_str());
        } else if (startswith(mimetype, "application/vnd.sun.xml.") ||
                   startswith(mimetype, "application/vnd.oasis.opendocument."))
        {
            // Inspired by http://mjr.towers.org.uk/comp/sxw2text
            string cmd = "unzip -p";
            append_filename_argument(cmd, file);
            cmd += " content.xml ; unzip -p";
            append_filename_argument(cmd, file);
            cmd += " styles.xml";
            try {
                OpenDocParser parser;
                parser.parse(stdout_to_string(cmd, true));
                dump = parser.dump;
            } catch (const ReadError&) {
                skip_cmd_failed(urlterm, context, cmd,
                                d.get_size(), d.get_mtime());
                return;
            }

            cmd = "unzip -p";
            append_filename_argument(cmd, file);
            cmd += " meta.xml";
            try {
                MetaXmlParser metaxmlparser;
                metaxmlparser.parse(stdout_to_string(cmd, false));
                title = metaxmlparser.title;
                keywords = metaxmlparser.keywords;
                // FIXME: topic = metaxmlparser.topic;
                sample = metaxmlparser.sample;
                author = metaxmlparser.author;
            } catch (const ReadError&) {
                // It's probably best to index the document even if this fails.
            }
        } else if (startswith(mimetype, "application/vnd.openxmlformats-officedocument.")) {
            const char * args = NULL;
            string tail(mimetype, 46);
            if (startswith(tail, "wordprocessingml.")) {
                // unzip returns exit code 11 if a file to extract wasn't found
                // which we want to ignore, because there may be no headers or
                // no footers.
                args = " word/document.xml 'word/header*.xml' 'word/footer*.xml' 2>/dev/null";
            } else if (startswith(tail, "spreadsheetml.")) {
                // Extract the shared string table first, so our parser can
                // grab those ready for parsing the sheets which will reference
                // the shared strings.
                string cmd = "unzip -p";
                append_filename_argument(cmd, file);
                cmd += " xl/styles.xml xl/workbook.xml xl/sharedStrings.xml ; unzip -p";
                append_filename_argument(cmd, file);
                cmd += " xl/worksheets/sheet\\*.xml";
                try {
                    XlsxParser parser;
                    parser.parse(stdout_to_string(cmd, true));
                    dump = parser.dump;
                } catch (const ReadError&) {
                    skip_cmd_failed(urlterm, context, cmd,
                                    d.get_size(), d.get_mtime());
                    return;
                }
            } else if (startswith(tail, "presentationml.")) {
                // unzip returns exit code 11 if a file to extract wasn't found
                // which we want to ignore, because there may be no notesSlides
                // or comments.
                args = " 'ppt/slides/slide*.xml' 'ppt/notesSlides/notesSlide*.xml' 'ppt/comments/comment*.xml' 2>/dev/null";
            } else {
                // Don't know how to index this type.
                skip_unknown_mimetype(urlterm, context, mimetype,
                                      d.get_size(), d.get_mtime());
                return;
            }

            if (args) {
                string cmd = "unzip -p";
                append_filename_argument(cmd, file);
                cmd += args;
                try {
                    MSXmlParser xmlparser;
                    // Treat exit status 11 from unzip as success - this is
                    // what we get if one of the listed filenames to extract
                    // doesn't match anything in the zip file.
                    xmlparser.parse_xml(stdout_to_string(cmd, false, 11));
                    dump = xmlparser.dump;
                } catch (const ReadError&) {
                    skip_cmd_failed(urlterm, context, cmd,
                                    d.get_size(), d.get_mtime());
                    return;
                }
            }

            string cmd = "unzip -p";
            append_filename_argument(cmd, file);
            cmd += " docProps/core.xml";
            try {
                MetaXmlParser metaxmlparser;
                metaxmlparser.parse(stdout_to_string(cmd, false));
                title = metaxmlparser.title;
                keywords = metaxmlparser.keywords;
                // FIXME: topic = metaxmlparser.topic;
                sample = metaxmlparser.sample;
                author = metaxmlparser.author;
            } catch (const ReadError&) {
                // It's probably best to index the document even if this fails.
            }
        } else if (mimetype == "application/x-abiword") {
            // FIXME: Implement support for metadata.
            XmlParser xmlparser;
            const string & text = d.file_to_string();
            xmlparser.parse_xml(text);
            dump = xmlparser.dump;
            md5_string(text, md5);
        } else if (mimetype == "application/x-abiword-compressed") {
            // FIXME: Implement support for metadata.
            XmlParser xmlparser;
            xmlparser.parse_xml(d.gzfile_to_string());
            dump = xmlparser.dump;
        } else if (mimetype == "application/oxps" ||
                   mimetype == "application/vnd.ms-xpsdocument") {
            string cmd = "unzip -p";
            append_filename_argument(cmd, file);
            cmd += " 'Documents/1/Pages/*.fpage'";
            try {
                XpsXmlParser xpsparser;
                run_filter(cmd, false, &dump);
                xpsparser.parse(dump);
                dump = xpsparser.dump;
            } catch (const ReadError&) {
                skip_cmd_failed(urlterm, context, cmd,
                                d.get_size(), d.get_mtime());
                return;
            }

            cmd = "unzip -p";
            append_filename_argument(cmd, file);
            cmd += " docProps/core.xml";
            try {
                MetaXmlParser metaparser;
                metaparser.parse(stdout_to_string(cmd, false));
                title = metaparser.title;
                keywords = metaparser.keywords;
                // FIXME: topic = metaparser.topic;
                sample = metaparser.sample;
                author = metaparser.author;
            } catch (const ReadError&) {
                // Ignore errors as not all XPS files contain this file.
            }
        } else if (mimetype == "text/csv") {
            // Currently we assume that text files are UTF-8 unless they have a
            // byte-order mark.
            dump = d.file_to_string();
            md5_string(dump, md5);

            // Look for Byte-Order Mark (BOM).
            if (startswith(dump, "\xfe\xff") || startswith(dump, "\xff\xfe")) {
                // UTF-16 in big-endian/little-endian order - we just convert
                // it as "UTF-16" and let the conversion handle the BOM as that
                // way we avoid the copying overhead of erasing 2 bytes from
                // the start of dump.
                convert_to_utf8(dump, "UTF-16");
            } else if (startswith(dump, "\xef\xbb\xbf")) {
                // UTF-8 with stupid Windows not-the-byte-order mark.
                dump.erase(0, 3);
            } else {
                // FIXME: What charset is the file?  Look at contents?
            }

            generate_sample_from_csv(dump, sample);
        } else if (mimetype == "image/svg+xml") {
            SvgParser svgparser;
            const string & text = d.file_to_string();
            md5_string(text, md5);
            svgparser.parse(text);
            dump = svgparser.dump;
            title = svgparser.title;
            keywords = svgparser.keywords;
            // FIXME: topic = svgparser.topic;
            author = svgparser.author;
        } else if (mimetype == "application/vnd.debian.binary-package" ||
                   mimetype == "application/x-debian-package") {
            const char* cmd = "dpkg-deb -f - Description";
            string desc;
            run_filter(d.get_fd(), cmd, false, &desc);
            // First line is short description, which we use as the title.
            string::size_type idx = desc.find('\n');
            title.assign(desc, 0, idx);
            if (idx != string::npos) {
                dump.assign(desc, idx + 1, string::npos);
            }
        } else if (mimetype == "application/x-redhat-package-manager" ||
                   mimetype == "application/x-rpm") {
            string cmd("rpm -q --qf '%{SUMMARY}\\n%{DESCRIPTION}' -p");
            append_filename_argument(cmd, file);
            string desc;
            run_filter(cmd, false, &desc);
            // First line is summary, which we use as the title.
            string::size_type idx = desc.find('\n');
            title.assign(desc, 0, idx);
            if (idx != string::npos) {
                dump.assign(desc, idx + 1, string::npos);
            }
        } else if (mimetype == "application/atom+xml") {
            AtomParser atomparser;
            const string & text = d.file_to_string();
            md5_string(text, md5);
            atomparser.parse(text);
            dump = atomparser.dump;
            title = atomparser.title;
            keywords = atomparser.keywords;
            // FIXME: topic = atomparser.topic;
            author = atomparser.author;
        } else {
            // Don't know how to index this type.
            skip_unknown_mimetype(urlterm, context, mimetype,
                                  d.get_size(), d.get_mtime());
            return;
        }

        // Compute the MD5 of the file if we haven't already.
        if (md5.empty() && !d.md5(md5)) {
            if (errno == ENOENT || errno == ENOTDIR) {
                skip(urlterm, context, "File removed during indexing",
                     d.get_size(), d.get_mtime(),
                     SKIP_VERBOSE_ONLY | SKIP_SHOW_FILENAME);
            } else {
                skip(urlterm, context, "failed to read file to calculate MD5 checksum",
                     d.get_size(), d.get_mtime());
            }
            return;
        }

        // Remove any trailing formfeeds, so we don't consider them when
        // considering if we extracted any text (e.g. pdftotext outputs a
        // formfeed between each page, even for blank pages).
        //
        // If dump contain only formfeeds, then trim_end will be string::npos
        // and ++trim_end will be 0, which is the correct new size.
        string::size_type trim_end = dump.find_last_not_of('\f');
        if (++trim_end != dump.size())
            dump.resize(trim_end);

        if (dump.empty()) {
            switch (empty_body) {
                case EMPTY_BODY_INDEX:
                    break;
                case EMPTY_BODY_WARN:
                    cout << "no text extracted from document body, "
                            "but indexing metadata anyway" << endl;
                    break;
                case EMPTY_BODY_SKIP:
                    skip(urlterm, context, "no text extracted from document body",
                         d.get_size(), d.get_mtime());
                    return;
            }
        }

        // Produce a sample
        if (sample.empty()) {
            sample = generate_sample(dump, sample_size, "...", " ...");
        } else {
            sample = generate_sample(sample, sample_size, "...", " ...");
        }

        // Put the data in the document
        if (record.empty()) {
            record = "url=";
        } else {
            record += "\nurl=";
        }
        record += url;
        record += "\nsample=";
        record += sample;
        if (!title.empty()) {
            record += "\ncaption=";
            record += generate_sample(title, title_size, "...", " ...");
        }
        if (!author.empty()) {
            record += "\nauthor=";
            record += author;
        }
        record += "\ntype=";
        record += mimetype;
        time_t mtime = d.get_mtime();
        if (mtime != static_cast<time_t>(-1)) {
            record += "\nmodtime=";
            record += str(mtime);
        }
        if (created != static_cast<time_t>(-1)) {
            record += "\ncreated=";
            record += str(created);
        }
        if (pages >= 0) {
            record += "\npages=";
            record += str(pages);
        }
        off_t size = d.get_size();
        record += "\nsize=";
        record += str(size);
        newdocument.set_data(record);

        // Index the title, document text, keywords and topic.
        indexer.set_document(newdocument);
        if (!title.empty()) {
            indexer.index_text(title, 5, "S");
            indexer.increase_termpos(100);
        }
        if (!dump.empty()) {
            indexer.index_text(dump);
        }
        if (!keywords.empty()) {
            indexer.increase_termpos(100);
            indexer.index_text(keywords);
        }
        if (!topic.empty()) {
            indexer.increase_termpos(100);
            indexer.index_text(topic, 1, "B");
        }
        // Index the leafname of the file.
        {
            indexer.increase_termpos(100);
            string leaf = d.leafname();
            string::size_type dot = leaf.find_last_of('.');
            if (dot != string::npos && leaf.size() - dot - 1 <= max_ext_len)
                leaf.resize(dot);
            indexer.index_text(leaf, 1, "F");

            // Also index with underscores and ampersands replaced by spaces.
            bool modified = false;
            string::size_type rep = 0;
            while ((rep = leaf.find_first_of("_&", rep)) != string::npos) {
                leaf[rep++] = ' ';
                modified = true;
            }
            if (modified) {
                indexer.increase_termpos(100);
                indexer.index_text(leaf, 1, "F");
            }
        }

        if (!author.empty()) {
            indexer.increase_termpos(100);
            indexer.index_text(author, 1, "A");
        }

        // mimeType:
        newdocument.add_boolean_term("T" + mimetype);

        newdocument.add_boolean_term(site_term);

        if (!host_term.empty())
            newdocument.add_boolean_term(host_term);

        struct tm *tm = localtime(&mtime);
        string date_term = "D" + date_to_string(tm->tm_year + 1900, tm->tm_mon + 1, tm->tm_mday);
        newdocument.add_boolean_term(date_term); // Date (YYYYMMDD)
        date_term.resize(7);
        date_term[0] = 'M';
        newdocument.add_boolean_term(date_term); // Month (YYYYMM)
        date_term.resize(5);
        date_term[0] = 'Y';
        newdocument.add_boolean_term(date_term); // Year (YYYY)

        newdocument.add_boolean_term(urlterm); // Url

        // Add mtime as a value to allow "sort by date".
        newdocument.add_value(VALUE_LASTMOD,
                              int_to_binary_string(uint32_t(mtime)));
        if (use_ctime) {
            // Add ctime as a value to track modifications.
            time_t ctime = d.get_ctime();
            newdocument.add_value(VALUE_CTIME,
                                  int_to_binary_string(uint32_t(ctime)));
        }

        // Add MD5 as a value to allow duplicate documents to be collapsed
        // together.
        newdocument.add_value(VALUE_MD5, md5);

        // Add the file size as a value to allow "sort by size" and size ranges.
        newdocument.add_value(VALUE_SIZE,
                              Xapian::sortable_serialise(size));

        bool inc_tag_added = false;
        if (d.is_other_readable()) {
            inc_tag_added = true;
            newdocument.add_boolean_term("I*");
        } else if (d.is_group_readable()) {
            const char * group = d.get_group();
            if (group) {
                newdocument.add_boolean_term(string("I#") + group);
            }
        }
        const char * owner = d.get_owner();
        if (owner) {
            newdocument.add_boolean_term(string("O") + owner);
            if (!inc_tag_added && d.is_owner_readable())
                newdocument.add_boolean_term(string("I@") + owner);
        }

        string ext_term("E");
        for (string::const_iterator i = ext.begin(); i != ext.end(); ++i) {
            char ch = *i;
            if (ch >= 'A' && ch <= 'Z')
                ch |= 32;
            ext_term += ch;
        }
        newdocument.add_boolean_term(ext_term);

        index_add_document(urlterm, last_altered, did, newdocument);
    } catch (const ReadError&) {
        skip(urlterm, context, string("can't read file: ") + strerror(errno),
             d.get_size(), d.get_mtime());
    } catch (const NoSuchFilter&) {
        string filter_entry;
        if (cmd_it != commands.end()) {
            filter_entry = cmd_it->first;
        } else {
            filter_entry = mimetype;
        }
        string m = "Filter for \"";
        m += filter_entry;
        m += "\" not installed";
        skip(urlterm, context, m, d.get_size(), d.get_mtime());
        commands[filter_entry] = Filter();
    } catch (const FileNotFound&) {
        skip(urlterm, context, "File removed during indexing",
             d.get_size(), d.get_mtime(),
             SKIP_VERBOSE_ONLY | SKIP_SHOW_FILENAME);
    } catch (const std::string & error) {
        skip(urlterm, context, error, d.get_size(), d.get_mtime());
    } catch (const std::bad_alloc&) {
        // Attempt to flag the file as failed and commit changes, though that
        // might fail too if we're low on memory rather than being asked to
        // allocate a ludicrous amount.
        skip(urlterm, context, "Out of memory trying to extract text from file",
             d.get_size(), d.get_mtime(),
             SKIP_SHOW_FILENAME);
        throw CommitAndExit("Caught std::bad_alloc", "");
    }
}

void
index_handle_deletion()
{
    if (updated.empty() || old_docs_not_seen == 0) return;

    if (verbose) {
        cout << "Deleting " << old_docs_not_seen << " old documents which weren't found" << endl;
    }
    Xapian::PostingIterator alldocs = db.postlist_begin(string());
    Xapian::docid did = *alldocs;
    while (did < updated.size()) {
        if (!updated[did]) {
            alldocs.skip_to(did);
            if (alldocs == db.postlist_end(string()))
                break;
            if (*alldocs != did) {
                // Document #did didn't exist before we started.
                did = *alldocs;
                continue;
            }
            db.delete_document(did);
            if (--old_docs_not_seen == 0)
                break;
        }
        ++did;
    }
}

void
index_commit()
{
    db.commit();
}

void
index_done()
{
    // If we created a temporary directory then delete it.
    remove_tmpdir();
}