calibre/db/cache.py

#!/usr/local/bin/python3.8
# vim:fileencoding=UTF-8:ts=4:sw=4:sta:et:sts=4:ai


__license__   = 'GPL v3'
__copyright__ = '2011, Kovid Goyal <kovid@kovidgoyal.net>'
__docformat__ = 'restructuredtext en'

import operator
import os
import random
import shutil
import sys
import traceback
from collections import defaultdict
from collections.abc import MutableSet, Set
from functools import partial, wraps
from io import BytesIO
from threading import Lock
from time import time

from calibre import as_unicode, isbytestring
from calibre.constants import iswindows, preferred_encoding
from calibre.customize.ui import (
    run_plugins_on_import, run_plugins_on_postadd, run_plugins_on_postimport
)
from calibre.db import SPOOL_SIZE, _get_next_series_num_for_list
from calibre.db.annotations import merge_annotations
from calibre.db.categories import get_categories
from calibre.db.errors import NoSuchBook, NoSuchFormat
from calibre.db.fields import IDENTITY, InvalidLinkTable, create_field
from calibre.db.lazy import FormatMetadata, FormatsList, ProxyMetadata
from calibre.db.listeners import EventDispatcher, EventType
from calibre.db.locking import (
    DowngradeLockError, LockingError, SafeReadLock, create_locks, try_lock
)
from calibre.db.search import Search
from calibre.db.tables import VirtualTable
from calibre.db.utils import type_safe_sort_key_function
from calibre.db.write import get_series_values, uniq
from calibre.ebooks import check_ebook_format
from calibre.ebooks.metadata import author_to_author_sort, string_to_authors
from calibre.ebooks.metadata.book.base import Metadata
from calibre.ebooks.metadata.opf2 import metadata_to_opf
from calibre.ptempfile import PersistentTemporaryFile, SpooledTemporaryFile, base_dir
from calibre.utils.config import prefs, tweaks
from calibre.utils.date import UNDEFINED_DATE, now as nowf, utcnow
from calibre.utils.icu import sort_key
from calibre.utils.localization import canonicalize_lang
from polyglot.builtins import cmp, iteritems, itervalues, string_or_bytes


def api(f):
    f.is_cache_api = True
    return f


def read_api(f):
    f = api(f)
    f.is_read_api = True
    return f


def write_api(f):
    f = api(f)
    f.is_read_api = False
    return f


def wrap_simple(lock, func):
    @wraps(func)
    def call_func_with_lock(*args, **kwargs):
        try:
            with lock:
                return func(*args, **kwargs)
        except DowngradeLockError:
            # We already have an exclusive lock, no need to acquire a shared
            # lock. See the safe_read_lock properties' documentation for why
            # this is necessary.
            return func(*args, **kwargs)
    return call_func_with_lock


def run_import_plugins(path_or_stream, fmt):
    fmt = fmt.lower()
    if hasattr(path_or_stream, 'seek'):
        path_or_stream.seek(0)
        pt = PersistentTemporaryFile('_import_plugin.'+fmt)
        shutil.copyfileobj(path_or_stream, pt, 1024**2)
        pt.close()
        path = pt.name
    else:
        path = path_or_stream
    return run_plugins_on_import(path, fmt)


def _add_newbook_tag(mi):
    tags = prefs['new_book_tags']
    if tags:
        for tag in [t.strip() for t in tags]:
            if tag:
                if not mi.tags:
                    mi.tags = [tag]
                elif tag not in mi.tags:
                    mi.tags.append(tag)


def _add_default_custom_column_values(mi, fm):
    cols = fm.custom_field_metadata(include_composites=False)
    for cc,col in iteritems(cols):
        dv = col['display'].get('default_value', None)
        try:
            if dv is not None:
                if not mi.get_user_metadata(cc, make_copy=False):
                    mi.set_user_metadata(cc, col)
                dt = col['datatype']
                if dt == 'datetime' and icu_lower(dv) == 'now':
                    dv = nowf()
                mi.set(cc, dv)
        except:
            traceback.print_exc()


dynamic_category_preferences = frozenset({'grouped_search_make_user_categories', 'grouped_search_terms', 'user_categories'})


class Cache:

    '''
    An in-memory cache of the metadata.db file from a calibre library.
    This class also serves as a threadsafe API for accessing the database.
    The in-memory cache is maintained in normal form for maximum performance.

    SQLITE is simply used as a way to read and write from metadata.db robustly.
    All table reading/sorting/searching/caching logic is re-implemented. This
    was necessary for maximum performance and flexibility.
    '''
    EventType = EventType

    def __init__(self, backend):
        self.backend = backend
        self.event_dispatcher = EventDispatcher()
        self.fields = {}
        self.composites = {}
        self.read_lock, self.write_lock = create_locks()
        self.format_metadata_cache = defaultdict(dict)
        self.formatter_template_cache = {}
        self.dirtied_cache = {}
        self.vls_for_books_cache = None
        self.vls_cache_lock = Lock()
        self.dirtied_sequence = 0
        self.cover_caches = set()
        self.clear_search_cache_count = 0

        # Implement locking for all simple read/write API methods
        # An unlocked version of the method is stored with the name starting
        # with a leading underscore. Use the unlocked versions when the lock
        # has already been acquired.
        for name in dir(self):
            func = getattr(self, name)
            ira = getattr(func, 'is_read_api', None)
            if ira is not None:
                # Save original function
                setattr(self, '_'+name, func)
                # Wrap it in a lock
                lock = self.read_lock if ira else self.write_lock
                setattr(self, name, wrap_simple(lock, func))

        self._search_api = Search(self, 'saved_searches', self.field_metadata.get_search_terms())
        self.initialize_dynamic()

    @property
    def new_api(self):
        return self

    @property
    def library_id(self):
        return self.backend.library_id

    @property
    def dbpath(self):
        return self.backend.dbpath

    @property
    def safe_read_lock(self):
        ''' A safe read lock is a lock that does nothing if the thread already
        has a write lock, otherwise it acquires a read lock. This is necessary
        to prevent DowngradeLockErrors, which can happen when updating the
        search cache in the presence of composite columns. Updating the search
        cache holds an exclusive lock, but searching a composite column
        involves reading field values via ProxyMetadata which tries to get a
        shared lock. There may be other scenarios that trigger this as well.

        This property returns a new lock object on every access. This lock
        object is not recursive (for performance) and must only be used in a
        with statement as ``with cache.safe_read_lock:`` otherwise bad things
        will happen.'''
        return SafeReadLock(self.read_lock)

    @write_api
    def ensure_has_search_category(self, fail_on_existing=True):
        if len(self._search_api.saved_searches.names()) > 0:
            self.field_metadata.add_search_category(label='search', name=_('Saved searches'), fail_on_existing=fail_on_existing)

    def _initialize_dynamic_categories(self):
        # Reconstruct the user categories, putting them into field_metadata
        fm = self.field_metadata
        fm.remove_dynamic_categories()
        for user_cat in sorted(self._pref('user_categories', {}), key=sort_key):
            cat_name = '@' + user_cat  # add the '@' to avoid name collision
            while cat_name:
                try:
                    fm.add_user_category(label=cat_name, name=user_cat)
                except ValueError:
                    break  # Can happen since we are removing dots and adding parent categories ourselves
                cat_name = cat_name.rpartition('.')[0]

        # add grouped search term user categories
        muc = frozenset(self._pref('grouped_search_make_user_categories', []))
        for cat in sorted(self._pref('grouped_search_terms', {}), key=sort_key):
            if cat in muc:
                # There is a chance that these can be duplicates of an existing
                # user category. Print the exception and continue.
                try:
                    self.field_metadata.add_user_category(label='@' + cat, name=cat)
                except ValueError:
                    traceback.print_exc()
        self._ensure_has_search_category()

        self.field_metadata.add_grouped_search_terms(
                                    self._pref('grouped_search_terms', {}))
        self._refresh_search_locations()

    @write_api
    def initialize_dynamic(self):
        self.backend.dirty_books_with_dirtied_annotations()
        self.dirtied_cache = {x:i for i, x in enumerate(self.backend.dirtied_books())}
        if self.dirtied_cache:
            self.dirtied_sequence = max(itervalues(self.dirtied_cache))+1
        self._initialize_dynamic_categories()

    @write_api
    def initialize_template_cache(self):
        self.formatter_template_cache = {}

    @write_api
    def set_user_template_functions(self, user_template_functions):
        self.backend.set_user_template_functions(user_template_functions)

    @write_api
    def clear_composite_caches(self, book_ids=None):
        for field in itervalues(self.composites):
            field.clear_caches(book_ids=book_ids)

    @write_api
    def clear_search_caches(self, book_ids=None):
        self.clear_search_cache_count += 1
        self._search_api.update_or_clear(self, book_ids)
        self.vls_for_books_cache = None

    @read_api
    def last_modified(self):
        return self.backend.last_modified()

    @write_api
    def clear_caches(self, book_ids=None, template_cache=True, search_cache=True):
        if template_cache:
            self._initialize_template_cache()  # Clear the formatter template cache
        for field in itervalues(self.fields):
            if hasattr(field, 'clear_caches'):
                field.clear_caches(book_ids=book_ids)  # Clear the composite cache and ondevice caches
        if book_ids:
            for book_id in book_ids:
                self.format_metadata_cache.pop(book_id, None)
        else:
            self.format_metadata_cache.clear()
        if search_cache:
            self._clear_search_caches(book_ids)

    @write_api
    def reload_from_db(self, clear_caches=True):
        if clear_caches:
            self._clear_caches()
        with self.backend.conn:  # Prevent other processes, such as calibredb from interrupting the reload by locking the db
            self.backend.prefs.load_from_db()
            self._search_api.saved_searches.load_from_db()
            for field in itervalues(self.fields):
                if hasattr(field, 'table'):
                    field.table.read(self.backend)  # Reread data from metadata.db

    @property
    def field_metadata(self):
        return self.backend.field_metadata

    def _get_metadata(self, book_id, get_user_categories=True):  # {{{
        mi = Metadata(None, template_cache=self.formatter_template_cache)

        mi._proxy_metadata = ProxyMetadata(self, book_id, formatter=mi.formatter)

        author_ids = self._field_ids_for('authors', book_id)
        adata = self._author_data(author_ids)
        aut_list = [adata[i] for i in author_ids]
        aum = []
        aus = {}
        aul = {}
        for rec in aut_list:
            aut = rec['name']
            aum.append(aut)
            aus[aut] = rec['sort']
            aul[aut] = rec['link']
        mi.title       = self._field_for('title', book_id,
                default_value=_('Unknown'))
        mi.authors     = aum
        mi.author_sort = self._field_for('author_sort', book_id,
                default_value=_('Unknown'))
        mi.author_sort_map = aus
        mi.author_link_map = aul
        mi.comments    = self._field_for('comments', book_id)
        mi.publisher   = self._field_for('publisher', book_id)
        n = utcnow()
        mi.timestamp   = self._field_for('timestamp', book_id, default_value=n)
        mi.pubdate     = self._field_for('pubdate', book_id, default_value=n)
        mi.uuid        = self._field_for('uuid', book_id,
                default_value='dummy')
        mi.title_sort  = self._field_for('sort', book_id,
                default_value=_('Unknown'))
        mi.last_modified = self._field_for('last_modified', book_id,
                default_value=n)
        formats = self._field_for('formats', book_id)
        mi.format_metadata = {}
        mi.languages = list(self._field_for('languages', book_id))
        if not formats:
            good_formats = None
        else:
            mi.format_metadata = FormatMetadata(self, book_id, formats)
            good_formats = FormatsList(sorted(formats), mi.format_metadata)
        # These three attributes are returned by the db2 get_metadata(),
        # however, we dont actually use them anywhere other than templates, so
        # they have been removed, to avoid unnecessary overhead. The templates
        # all use _proxy_metadata.
        # mi.book_size   = self._field_for('size', book_id, default_value=0)
        # mi.ondevice_col = self._field_for('ondevice', book_id, default_value='')
        # mi.db_approx_formats = formats
        mi.formats = good_formats
        mi.has_cover = _('Yes') if self._field_for('cover', book_id,
                default_value=False) else ''
        mi.tags = list(self._field_for('tags', book_id, default_value=()))
        mi.series = self._field_for('series', book_id)
        if mi.series:
            mi.series_index = self._field_for('series_index', book_id,
                    default_value=1.0)
        mi.rating = self._field_for('rating', book_id)
        mi.set_identifiers(self._field_for('identifiers', book_id,
            default_value={}))
        mi.application_id = book_id
        mi.id = book_id
        composites = []
        for key, meta in self.field_metadata.custom_iteritems():
            mi.set_user_metadata(key, meta)
            if meta['datatype'] == 'composite':
                composites.append(key)
            else:
                val = self._field_for(key, book_id)
                if isinstance(val, tuple):
                    val = list(val)
                extra = self._field_for(key+'_index', book_id)
                mi.set(key, val=val, extra=extra)
        for key in composites:
            mi.set(key, val=self._composite_for(key, book_id, mi))

        user_cat_vals = {}
        if get_user_categories:
            user_cats = self._pref('user_categories', {})
            for ucat in user_cats:
                res = []
                for name,cat,ign in user_cats[ucat]:
                    v = mi.get(cat, None)
                    if isinstance(v, list):
                        if name in v:
                            res.append([name,cat])
                    elif name == v:
                        res.append([name,cat])
                user_cat_vals[ucat] = res
        mi.user_categories = user_cat_vals

        return mi
    # }}}

    @api
    def init(self):
        '''
        Initialize this cache with data from the backend.
        '''
        with self.write_lock:
            self.backend.read_tables()
            bools_are_tristate = self.backend.prefs['bools_are_tristate']

            for field, table in iteritems(self.backend.tables):
                self.fields[field] = create_field(field, table, bools_are_tristate,
                                          self.backend.get_template_functions)
                if table.metadata['datatype'] == 'composite':
                    self.composites[field] = self.fields[field]

            self.fields['ondevice'] = create_field('ondevice',
                    VirtualTable('ondevice'), bools_are_tristate,
                    self.backend.get_template_functions)

            for name, field in iteritems(self.fields):
                if name[0] == '#' and name.endswith('_index'):
                    field.series_field = self.fields[name[:-len('_index')]]
                    self.fields[name[:-len('_index')]].index_field = field
                elif name == 'series_index':
                    field.series_field = self.fields['series']
                    self.fields['series'].index_field = field
                elif name == 'authors':
                    field.author_sort_field = self.fields['author_sort']
                elif name == 'title':
                    field.title_sort_field = self.fields['sort']
        if self.backend.prefs['update_all_last_mod_dates_on_start']:
            self.update_last_modified(self.all_book_ids())
            self.backend.prefs.set('update_all_last_mod_dates_on_start', False)

    # Cache Layer API {{{

    @write_api
    def add_listener(self, event_callback_function):
        '''
        Register a callback function that will be called after certain actions are
        taken on this database. The function must take three arguments:
        (:class:`EventType`, library_id, event_type_specific_data)
        '''
        self.event_dispatcher.library_id = getattr(self, 'server_library_id', self.library_id)
        self.event_dispatcher.add_listener(event_callback_function)

    @write_api
    def remove_listener(self, event_callback_function):
        self.event_dispatcher.remove_listener(event_callback_function)

    @read_api
    def field_for(self, name, book_id, default_value=None):
        '''
        Return the value of the field ``name`` for the book identified
        by ``book_id``. If no such book exists or it has no defined
        value for the field ``name`` or no such field exists, then
        ``default_value`` is returned.

        ``default_value`` is not used for title, title_sort, authors, author_sort
        and series_index. This is because these always have values in the db.
        ``default_value`` is used for all custom columns.

        The returned value for is_multiple fields are always tuples, even when
        no values are found (in other words, default_value is ignored). The
        exception is identifiers for which the returned value is always a dict.
        The returned tuples are always in link order, that is, the order in
        which they were created.
        '''
        if self.composites and name in self.composites:
            return self.composite_for(name, book_id,
                    default_value=default_value)
        try:
            field = self.fields[name]
        except KeyError:
            return default_value
        if field.is_multiple:
            default_value = field.default_value
        try:
            return field.for_book(book_id, default_value=default_value)
        except (KeyError, IndexError):
            return default_value

    @read_api
    def fast_field_for(self, field_obj, book_id, default_value=None):
        ' Same as field_for, except that it avoids the extra lookup to get the field object '
        if field_obj.is_composite:
            return field_obj.get_value_with_cache(book_id, self._get_proxy_metadata)
        if field_obj.is_multiple:
            default_value = field_obj.default_value
        try:
            return field_obj.for_book(book_id, default_value=default_value)
        except (KeyError, IndexError):
            return default_value

    @read_api
    def all_field_for(self, field, book_ids, default_value=None):
        ' Same as field_for, except that it operates on multiple books at once '
        field_obj = self.fields[field]
        return {book_id:self._fast_field_for(field_obj, book_id, default_value=default_value) for book_id in book_ids}

    @read_api
    def composite_for(self, name, book_id, mi=None, default_value=''):
        try:
            f = self.fields[name]
        except KeyError:
            return default_value

        if mi is None:
            return f.get_value_with_cache(book_id, self._get_proxy_metadata)
        else:
            return f._render_composite_with_cache(book_id, mi, mi.formatter, mi.template_cache)

    @read_api
    def field_ids_for(self, name, book_id):
        '''
        Return the ids (as a tuple) for the values that the field ``name`` has on the book
        identified by ``book_id``. If there are no values, or no such book, or
        no such field, an empty tuple is returned.
        '''
        try:
            return self.fields[name].ids_for_book(book_id)
        except (KeyError, IndexError):
            return ()

    @read_api
    def books_for_field(self, name, item_id):
        '''
        Return all the books associated with the item identified by
        ``item_id``, where the item belongs to the field ``name``.

        Returned value is a set of book ids, or the empty set if the item
        or the field does not exist.
        '''
        try:
            return self.fields[name].books_for(item_id)
        except (KeyError, IndexError):
            return set()

    @read_api
    def all_book_ids(self, type=frozenset):
        '''
        Frozen set of all known book ids.
        '''
        return type(self.fields['uuid'].table.book_col_map)

    @read_api
    def all_field_ids(self, name):
        '''
        Frozen set of ids for all values in the field ``name``.
        '''
        return frozenset(iter(self.fields[name]))

    @read_api
    def all_field_names(self, field):
        ''' Frozen set of all fields names (should only be used for many-one and many-many fields) '''
        if field == 'formats':
            return frozenset(self.fields[field].table.col_book_map)

        try:
            return frozenset(self.fields[field].table.id_map.values())
        except AttributeError:
            raise ValueError('%s is not a many-one or many-many field' % field)

    @read_api
    def get_usage_count_by_id(self, field):
        ''' Return a mapping of id to usage count for all values of the specified
        field, which must be a many-one or many-many field. '''
        try:
            return {k:len(v) for k, v in iteritems(self.fields[field].table.col_book_map)}
        except AttributeError:
            raise ValueError('%s is not a many-one or many-many field' % field)

    @read_api
    def get_id_map(self, field):
        ''' Return a mapping of id numbers to values for the specified field.
        The field must be a many-one or many-many field, otherwise a ValueError
        is raised. '''
        try:
            return self.fields[field].table.id_map.copy()
        except AttributeError:
            if field == 'title':
                return self.fields[field].table.book_col_map.copy()
            raise ValueError('%s is not a many-one or many-many field' % field)

    @read_api
    def get_item_name(self, field, item_id):
        ''' Return the item name for the item specified by item_id in the
        specified field. See also :meth:`get_id_map`.'''
        return self.fields[field].table.id_map[item_id]

    @read_api
    def get_item_id(self, field, item_name):
        ' Return the item id for item_name (case-insensitive) '
        rmap = {icu_lower(v) if isinstance(v, str) else v:k for k, v in iteritems(self.fields[field].table.id_map)}
        return rmap.get(icu_lower(item_name) if isinstance(item_name, str) else item_name, None)

    @read_api
    def get_item_ids(self, field, item_names):
        ' Return the item id for item_name (case-insensitive) '
        rmap = {icu_lower(v) if isinstance(v, str) else v:k for k, v in iteritems(self.fields[field].table.id_map)}
        return {name:rmap.get(icu_lower(name) if isinstance(name, str) else name, None) for name in item_names}

    @read_api
    def author_data(self, author_ids=None):
        '''
        Return author data as a dictionary with keys: name, sort, link

        If no authors with the specified ids are found an empty dictionary is
        returned. If author_ids is None, data for all authors is returned.
        '''
        af = self.fields['authors']
        if author_ids is None:
            return {aid:af.author_data(aid) for aid in af.table.id_map}
        return {aid:af.author_data(aid) for aid in author_ids if aid in af.table.id_map}

    @read_api
    def format_hash(self, book_id, fmt):
        ''' Return the hash of the specified format for the specified book. The
        kind of hash is backend dependent, but is usually SHA-256. '''
        try:
            name = self.fields['formats'].format_fname(book_id, fmt)
            path = self._field_for('path', book_id).replace('/', os.sep)
        except:
            raise NoSuchFormat('Record %d has no fmt: %s'%(book_id, fmt))
        return self.backend.format_hash(book_id, fmt, name, path)

    @api
    def format_metadata(self, book_id, fmt, allow_cache=True, update_db=False):
        '''
        Return the path, size and mtime for the specified format for the specified book.
        You should not use path unless you absolutely have to,
        since accessing it directly breaks the threadsafe guarantees of this API. Instead use
        the :meth:`copy_format_to` method.

        :param allow_cache: If ``True`` cached values are used, otherwise a
            slow filesystem access is done. The cache values could be out of date
            if access was performed to the filesystem outside of this API.

        :param update_db: If ``True`` The max_size field of the database is updated for this book.
        '''
        if not fmt:
            return {}
        fmt = fmt.upper()
        # allow_cache and update_db are mutually exclusive. Give priority to update_db
        if allow_cache and not update_db:
            x = self.format_metadata_cache[book_id].get(fmt, None)
            if x is not None:
                return x
        with self.safe_read_lock:
            try:
                name = self.fields['formats'].format_fname(book_id, fmt)
                path = self._field_for('path', book_id).replace('/', os.sep)
            except:
                return {}

            ans = {}
            if path and name:
                ans = self.backend.format_metadata(book_id, fmt, name, path)
                self.format_metadata_cache[book_id][fmt] = ans
        if update_db and 'size' in ans:
            with self.write_lock:
                max_size = self.fields['formats'].table.update_fmt(book_id, fmt, name, ans['size'], self.backend)
                self.fields['size'].table.update_sizes({book_id: max_size})

        return ans

    @read_api
    def format_files(self, book_id):
        field = self.fields['formats']
        fmts = field.table.book_col_map.get(book_id, ())
        return {fmt:field.format_fname(book_id, fmt) for fmt in fmts}

    @read_api
    def format_db_size(self, book_id, fmt):
        field = self.fields['formats']
        return field.format_size(book_id, fmt)

    @read_api
    def pref(self, name, default=None, namespace=None):
        ' Return the value for the specified preference or the value specified as ``default`` if the preference is not set. '
        if namespace is not None:
            return self.backend.prefs.get_namespaced(namespace, name, default)
        return self.backend.prefs.get(name, default)

    @write_api
    def set_pref(self, name, val, namespace=None):
        ' Set the specified preference to the specified value. See also :meth:`pref`. '
        if namespace is not None:
            self.backend.prefs.set_namespaced(namespace, name, val)
            return
        self.backend.prefs.set(name, val)
        if name in ('grouped_search_terms', 'virtual_libraries'):
            self._clear_search_caches()
        if name in dynamic_category_preferences:
            self._initialize_dynamic_categories()

    @api
    def get_metadata(self, book_id,
            get_cover=False, get_user_categories=True, cover_as_data=False):
        '''
        Return metadata for the book identified by book_id as a :class:`calibre.ebooks.metadata.book.base.Metadata` object.
        Note that the list of formats is not verified. If get_cover is True,
        the cover is returned, either a path to temp file as mi.cover or if
        cover_as_data is True then as mi.cover_data.
        '''

        # Check if virtual_libraries_for_books rebuilt its cache. If it did then
        # we must clear the composite caches so the new data can be taken into
        # account. Clearing the caches requires getting a write lock, so it must
        # be done outside of the closure of _get_metadata().
        composite_cache_needs_to_be_cleared = False
        with self.safe_read_lock:
            vl_cache_was_none = self.vls_for_books_cache is None
            mi = self._get_metadata(book_id, get_user_categories=get_user_categories)
            if vl_cache_was_none and self.vls_for_books_cache is not None:
                composite_cache_needs_to_be_cleared = True
        if composite_cache_needs_to_be_cleared:
            try:
                self.clear_composite_caches()
            except LockingError:
                # We can't clear the composite caches because a read lock is set.
                # As a consequence the value of a composite column that calls
                # virtual_libraries() might be wrong. Oh well. Log and keep running.
                print("Couldn't get write lock after vls_for_books_cache was loaded", file=sys.stderr)
                traceback.print_exc()

        if get_cover:
            if cover_as_data:
                cdata = self.cover(book_id)
                if cdata:
                    mi.cover_data = ('jpeg', cdata)
            else:
                mi.cover = self.cover(book_id, as_path=True)

        return mi

    @read_api
    def get_proxy_metadata(self, book_id):
        ''' Like :meth:`get_metadata` except that it returns a ProxyMetadata
        object that only reads values from the database on demand. This is much
        faster than get_metadata when only a small number of fields need to be
        accessed from the returned metadata object. '''
        return ProxyMetadata(self, book_id)

    @api
    def cover(self, book_id,
            as_file=False, as_image=False, as_path=False):
        '''
        Return the cover image or None. By default, returns the cover as a
        bytestring.

        WARNING: Using as_path will copy the cover to a temp file and return
        the path to the temp file. You should delete the temp file when you are
        done with it.

        :param as_file: If True return the image as an open file object (a SpooledTemporaryFile)
        :param as_image: If True return the image as a QImage object
        :param as_path: If True return the image as a path pointing to a
                        temporary file
        '''
        if as_file:
            ret = SpooledTemporaryFile(SPOOL_SIZE)
            if not self.copy_cover_to(book_id, ret):
                return
            ret.seek(0)
        elif as_path:
            pt = PersistentTemporaryFile('_dbcover.jpg')
            with pt:
                if not self.copy_cover_to(book_id, pt):
                    return
            ret = pt.name
        else:
            buf = BytesIO()
            if not self.copy_cover_to(book_id, buf):
                return
            ret = buf.getvalue()
            if as_image:
                from qt.core import QImage
                i = QImage()
                i.loadFromData(ret)
                ret = i
        return ret

    @read_api
    def cover_or_cache(self, book_id, timestamp):
        try:
            path = self._field_for('path', book_id).replace('/', os.sep)
        except AttributeError:
            return False, None, None
        return self.backend.cover_or_cache(path, timestamp)

    @read_api
    def cover_last_modified(self, book_id):
        try:
            path = self._field_for('path', book_id).replace('/', os.sep)
        except AttributeError:
            return
        return self.backend.cover_last_modified(path)

    @read_api
    def copy_cover_to(self, book_id, dest, use_hardlink=False, report_file_size=None):
        '''
        Copy the cover to the file like object ``dest``. Returns False
        if no cover exists or dest is the same file as the current cover.
        dest can also be a path in which case the cover is
        copied to it if and only if the path is different from the current path (taking
        case sensitivity into account).
        '''
        try:
            path = self._field_for('path', book_id).replace('/', os.sep)
        except AttributeError:
            return False

        return self.backend.copy_cover_to(path, dest, use_hardlink=use_hardlink,
                                          report_file_size=report_file_size)

    @write_api
    def compress_covers(self, book_ids, jpeg_quality=100, progress_callback=None):
        '''
        Compress the cover images for the specified books. A compression quality of 100
        will perform lossless compression, otherwise lossy compression.

        The progress callback will be called with the book_id and the old and new sizes
        for each book that has been processed. If an error occurs, the new size will
        be a string with the error details.
        '''
        jpeg_quality = max(10, min(jpeg_quality, 100))
        path_map = {}
        for book_id in book_ids:
            try:
                path_map[book_id] = self._field_for('path', book_id).replace('/', os.sep)
            except AttributeError:
                continue
        self.backend.compress_covers(path_map, jpeg_quality, progress_callback)

    @read_api
    def copy_format_to(self, book_id, fmt, dest, use_hardlink=False, report_file_size=None):
        '''
        Copy the format ``fmt`` to the file like object ``dest``. If the
        specified format does not exist, raises :class:`NoSuchFormat` error.
        dest can also be a path (to a file), in which case the format is copied to it, iff
        the path is different from the current path (taking case sensitivity
        into account).
        '''
        fmt = (fmt or '').upper()
        try:
            name = self.fields['formats'].format_fname(book_id, fmt)
            path = self._field_for('path', book_id).replace('/', os.sep)
        except (KeyError, AttributeError):
            raise NoSuchFormat('Record %d has no %s file'%(book_id, fmt))

        return self.backend.copy_format_to(book_id, fmt, name, path, dest,
                                               use_hardlink=use_hardlink, report_file_size=report_file_size)

    @read_api
    def format_abspath(self, book_id, fmt):
        '''
        Return absolute path to the e-book file of format `format`. You should
        almost never use this, as it breaks the threadsafe promise of this API.
        Instead use, :meth:`copy_format_to`.

        Currently used only in calibredb list, the viewer, edit book,
        compare_format to original format, open with, bulk metadata edit and
        the catalogs (via get_data_as_dict()).

        Apart from the viewer, open with and edit book, I don't believe any of
        the others do any file write I/O with the results of this call.
        '''
        fmt = (fmt or '').upper()
        try:
            path = self._field_for('path', book_id).replace('/', os.sep)
        except:
            return None
        if path:
            if fmt == '__COVER_INTERNAL__':
                return self.backend.cover_abspath(book_id, path)
            else:
                try:
                    name = self.fields['formats'].format_fname(book_id, fmt)
                except:
                    return None
                if name:
                    return self.backend.format_abspath(book_id, fmt, name, path)

    @read_api
    def has_format(self, book_id, fmt):
        'Return True iff the format exists on disk'
        fmt = (fmt or '').upper()
        try:
            name = self.fields['formats'].format_fname(book_id, fmt)
            path = self._field_for('path', book_id).replace('/', os.sep)
        except:
            return False
        return self.backend.has_format(book_id, fmt, name, path)

    @api
    def save_original_format(self, book_id, fmt):
        ' Save a copy of the specified format as ORIGINAL_FORMAT, overwriting any existing ORIGINAL_FORMAT. '
        fmt = fmt.upper()
        if 'ORIGINAL' in fmt:
            raise ValueError('Cannot save original of an original fmt')
        fmtfile = self.format(book_id, fmt, as_file=True)
        if fmtfile is None:
            return False
        with fmtfile:
            nfmt = 'ORIGINAL_'+fmt
            return self.add_format(book_id, nfmt, fmtfile, run_hooks=False)

    @write_api
    def restore_original_format(self, book_id, original_fmt):
        ''' Restore the specified format from the previously saved
        ORIGINAL_FORMAT, if any. Return True on success. The ORIGINAL_FORMAT is
        deleted after a successful restore. '''
        original_fmt = original_fmt.upper()
        fmt = original_fmt.partition('_')[2]
        try:
            ofmt_name = self.fields['formats'].format_fname(book_id, original_fmt)
            path = self._field_for('path', book_id).replace('/', os.sep)
        except Exception:
            return False
        if self.backend.is_format_accessible(book_id, original_fmt, ofmt_name, path):
            self.add_format(book_id, fmt, BytesIO(), run_hooks=False)
            fmt_name = self.fields['formats'].format_fname(book_id, fmt)
            file_size = self.backend.rename_format_file(book_id, ofmt_name, original_fmt, fmt_name, fmt, path)
            self.fields['formats'].table.update_fmt(book_id, fmt, fmt_name, file_size, self.backend)
            self._remove_formats({book_id:(original_fmt,)})
            return True
        return False

    @read_api
    def formats(self, book_id, verify_formats=True):
        '''
        Return tuple of all formats for the specified book. If verify_formats
        is True, verifies that the files exist on disk.
        '''
        ans = self.field_for('formats', book_id)
        if verify_formats and ans:
            try:
                path = self._field_for('path', book_id).replace('/', os.sep)
            except:
                return ()

            def verify(fmt):
                try:
                    name = self.fields['formats'].format_fname(book_id, fmt)
                except:
                    return False
                return self.backend.has_format(book_id, fmt, name, path)

            ans = tuple(x for x in ans if verify(x))
        return ans

    @api
    def format(self, book_id, fmt, as_file=False, as_path=False, preserve_filename=False):
        '''
        Return the e-book format as a bytestring or `None` if the format doesn't exist,
        or we don't have permission to write to the e-book file.

        :param as_file: If True the e-book format is returned as a file object. Note
                        that the file object is a SpooledTemporaryFile, so if what you want to
                        do is copy the format to another file, use :meth:`copy_format_to`
                        instead for performance.
        :param as_path: Copies the format file to a temp file and returns the
                        path to the temp file
        :param preserve_filename: If True and returning a path the filename is
                                  the same as that used in the library. Note that using
                                  this means that repeated calls yield the same
                                  temp file (which is re-created each time)
        '''
        fmt = (fmt or '').upper()
        ext = ('.'+fmt.lower()) if fmt else ''
        if as_path:
            if preserve_filename:
                with self.safe_read_lock:
                    try:
                        fname = self.fields['formats'].format_fname(book_id, fmt)
                    except:
                        return None
                    fname += ext

                bd = base_dir()
                d = os.path.join(bd, 'format_abspath')
                try:
                    os.makedirs(d)
                except:
                    pass
                ret = os.path.join(d, fname)
                try:
                    self.copy_format_to(book_id, fmt, ret)
                except NoSuchFormat:
                    return None
            else:
                with PersistentTemporaryFile(ext) as pt:
                    try:
                        self.copy_format_to(book_id, fmt, pt)
                    except NoSuchFormat:
                        return None
                    ret = pt.name
        elif as_file:
            with self.safe_read_lock:
                try:
                    fname = self.fields['formats'].format_fname(book_id, fmt)
                except:
                    return None
                fname += ext

            ret = SpooledTemporaryFile(SPOOL_SIZE)
            try:
                self.copy_format_to(book_id, fmt, ret)
            except NoSuchFormat:
                return None
            ret.seek(0)
            # Various bits of code try to use the name as the default
            # title when reading metadata, so set it
            ret.name = fname
        else:
            buf = BytesIO()
            try:
                self.copy_format_to(book_id, fmt, buf)
            except NoSuchFormat:
                return None

            ret = buf.getvalue()

        return ret

    @read_api
    def multisort(self, fields, ids_to_sort=None, virtual_fields=None):
        '''
        Return a list of sorted book ids. If ids_to_sort is None, all book ids
        are returned.

        fields must be a list of 2-tuples of the form (field_name,
        ascending=True or False). The most significant field is the first
        2-tuple.
        '''
        ids_to_sort = self._all_book_ids() if ids_to_sort is None else ids_to_sort
        get_metadata = self._get_proxy_metadata
        lang_map = self.fields['languages'].book_value_map
        virtual_fields = virtual_fields or {}

        fm = {'title':'sort', 'authors':'author_sort'}

        def sort_key_func(field):
            'Handle series type fields, virtual fields and the id field'
            idx = field + '_index'
            is_series = idx in self.fields
            try:
                func = self.fields[fm.get(field, field)].sort_keys_for_books(get_metadata, lang_map)
            except KeyError:
                if field == 'id':
                    return IDENTITY
                else:
                    return virtual_fields[fm.get(field, field)].sort_keys_for_books(get_metadata, lang_map)
            if is_series:
                idx_func = self.fields[idx].sort_keys_for_books(get_metadata, lang_map)

                def skf(book_id):
                    return (func(book_id), idx_func(book_id))
                return skf
            return func

        # Sort only once on any given field
        fields = uniq(fields, operator.itemgetter(0))

        if len(fields) == 1:
            keyfunc = sort_key_func(fields[0][0])
            reverse = not fields[0][1]
            try:
                return sorted(ids_to_sort, key=keyfunc, reverse=reverse)
            except Exception as err:
                print('Failed to sort database on field:', fields[0][0], 'with error:', err, file=sys.stderr)
                try:
                    return sorted(ids_to_sort, key=type_safe_sort_key_function(keyfunc), reverse=reverse)
                except Exception as err:
                    print('Failed to type-safe sort database on field:', fields[0][0], 'with error:', err, file=sys.stderr)
                    return sorted(ids_to_sort, reverse=reverse)
        sort_key_funcs = tuple(sort_key_func(field) for field, order in fields)
        orders = tuple(1 if order else -1 for _, order in fields)
        Lazy = object()  # Lazy load the sort keys for sub-sort fields

        class SortKey:

            __slots__ = 'book_id', 'sort_key'

            def __init__(self, book_id):
                self.book_id = book_id
                # Calculate only the first sub-sort key since that will always be used
                self.sort_key = [key(book_id) if i == 0 else Lazy for i, key in enumerate(sort_key_funcs)]

            def compare_to_other(self, other):
                for i, (order, self_key, other_key) in enumerate(zip(orders, self.sort_key, other.sort_key)):
                    if self_key is Lazy:
                        self_key = self.sort_key[i] = sort_key_funcs[i](self.book_id)
                    if other_key is Lazy:
                        other_key = other.sort_key[i] = sort_key_funcs[i](other.book_id)
                    ans = cmp(self_key, other_key)
                    if ans != 0:
                        return ans * order
                return 0

            def __eq__(self, other):
                return self.compare_to_other(other) == 0

            def __ne__(self, other):
                return self.compare_to_other(other) != 0

            def __lt__(self, other):
                return self.compare_to_other(other) < 0

            def __le__(self, other):
                return self.compare_to_other(other) <= 0

            def __gt__(self, other):
                return self.compare_to_other(other) > 0

            def __ge__(self, other):
                return self.compare_to_other(other) >= 0

        return sorted(ids_to_sort, key=SortKey)

    @read_api
    def search(self, query, restriction='', virtual_fields=None, book_ids=None):
        '''
        Search the database for the specified query, returning a set of matched book ids.

        :param restriction: A restriction that is ANDed to the specified query. Note that
            restrictions are cached, therefore the search for a AND b will be slower than a with restriction b.

        :param virtual_fields: Used internally (virtual fields such as on_device to search over).

        :param book_ids: If not None, a set of book ids for which books will
            be searched instead of searching all books.
        '''
        return self._search_api(self, query, restriction, virtual_fields=virtual_fields, book_ids=book_ids)

    @read_api
    def books_in_virtual_library(self, vl, search_restriction=None, virtual_fields=None):
        ' Return the set of books in the specified virtual library '
        vl = self._pref('virtual_libraries', {}).get(vl) if vl else None
        if not vl and not search_restriction:
            return self.all_book_ids()
        # We utilize the search restriction cache to speed this up
        srch = partial(self._search, virtual_fields=virtual_fields)
        if vl:
            if search_restriction:
                return frozenset(srch('', vl) & srch('', search_restriction))
            return frozenset(srch('', vl))
        return frozenset(srch('', search_restriction))

    @read_api
    def number_of_books_in_virtual_library(self, vl=None, search_restriction=None):
        if not vl and not search_restriction:
            return len(self.fields['uuid'].table.book_col_map)
        return len(self.books_in_virtual_library(vl, search_restriction))

    @api
    def get_categories(self, sort='name', book_ids=None, already_fixed=None,
                       first_letter_sort=False):
        ' Used internally to implement the Tag Browser '
        try:
            with self.safe_read_lock:
                return get_categories(self, sort=sort, book_ids=book_ids,
                                      first_letter_sort=first_letter_sort)
        except InvalidLinkTable as err:
            bad_field = err.field_name
            if bad_field == already_fixed:
                raise
            with self.write_lock:
                self.fields[bad_field].table.fix_link_table(self.backend)
            return self.get_categories(sort=sort, book_ids=book_ids, already_fixed=bad_field)

    @write_api
    def update_last_modified(self, book_ids, now=None):
        if book_ids:
            if now is None:
                now = nowf()
            f = self.fields['last_modified']
            f.writer.set_books({book_id:now for book_id in book_ids}, self.backend)
            if self.composites:
                self._clear_composite_caches(book_ids)
            self._clear_search_caches(book_ids)

    @write_api
    def mark_as_dirty(self, book_ids):
        self._update_last_modified(book_ids)
        already_dirtied = set(self.dirtied_cache).intersection(book_ids)
        new_dirtied = book_ids - already_dirtied
        already_dirtied = {book_id:self.dirtied_sequence+i for i, book_id in enumerate(already_dirtied)}
        if already_dirtied:
            self.dirtied_sequence = max(itervalues(already_dirtied)) + 1
        self.dirtied_cache.update(already_dirtied)
        if new_dirtied:
            self.backend.dirty_books(new_dirtied)
            new_dirtied = {book_id:self.dirtied_sequence+i for i, book_id in enumerate(new_dirtied)}
            self.dirtied_sequence = max(itervalues(new_dirtied)) + 1
            self.dirtied_cache.update(new_dirtied)

    @write_api
    def commit_dirty_cache(self):
        if self.dirtied_cache:
            self.backend.dirty_books(self.dirtied_cache)

    @write_api
    def check_dirtied_annotations(self):
        if not self.backend.dirty_books_with_dirtied_annotations():
            return
        book_ids = set(self.backend.dirtied_books())
        new_dirtied = book_ids - set(self.dirtied_cache)
        if new_dirtied:
            new_dirtied = {book_id:self.dirtied_sequence+i for i, book_id in enumerate(new_dirtied)}
            self.dirtied_sequence = max(itervalues(new_dirtied)) + 1
            self.dirtied_cache.update(new_dirtied)

    @write_api
    def set_field(self, name, book_id_to_val_map, allow_case_change=True, do_path_update=True):
        '''
        Set the values of the field specified by ``name``. Returns the set of all book ids that were affected by the change.

        :param book_id_to_val_map: Mapping of book_ids to values that should be applied.
        :param allow_case_change: If True, the case of many-one or many-many fields will be changed.
            For example, if a  book has the tag ``tag1`` and you set the tag for another book to ``Tag1``
            then the both books will have the tag ``Tag1`` if allow_case_change is True, otherwise they will
            both have the tag ``tag1``.
        :param do_path_update: Used internally, you should never change it.
        '''
        f = self.fields[name]
        is_series = f.metadata['datatype'] == 'series'
        update_path = name in {'title', 'authors'}
        if update_path and iswindows:
            paths = (x for x in (self._field_for('path', book_id) for book_id in book_id_to_val_map) if x)
            self.backend.windows_check_if_files_in_use(paths)

        if is_series:
            bimap, simap = {}, {}
            sfield = self.fields[name + '_index']
            for k, v in iteritems(book_id_to_val_map):
                if isinstance(v, string_or_bytes):
                    v, sid = get_series_values(v)
                else:
                    v = sid = None
                if sid is None and name.startswith('#'):
                    sid = self._fast_field_for(sfield, k)
                    sid = 1.0 if sid is None else sid  # The value to be set the db link table
                bimap[k] = v
                if sid is not None:
                    simap[k] = sid
            book_id_to_val_map = bimap

        dirtied = f.writer.set_books(
            book_id_to_val_map, self.backend, allow_case_change=allow_case_change)

        if is_series and simap:
            sf = self.fields[f.name+'_index']
            dirtied |= sf.writer.set_books(simap, self.backend, allow_case_change=False)

        if dirtied:
            if update_path and do_path_update:
                self._update_path(dirtied, mark_as_dirtied=False)
            self._mark_as_dirty(dirtied)
            self.event_dispatcher(EventType.metadata_changed, name, dirtied)
        return dirtied

    @write_api
    def update_path(self, book_ids, mark_as_dirtied=True):
        for book_id in book_ids:
            title = self._field_for('title', book_id, default_value=_('Unknown'))
            try:
                author = self._field_for('authors', book_id, default_value=(_('Unknown'),))[0]
            except IndexError:
                author = _('Unknown')
            self.backend.update_path(book_id, title, author, self.fields['path'], self.fields['formats'])
            self.format_metadata_cache.pop(book_id, None)
            if mark_as_dirtied:
                self._mark_as_dirty(book_ids)

    @read_api
    def get_a_dirtied_book(self):
        if self.dirtied_cache:
            return random.choice(tuple(self.dirtied_cache))
        return None

    @read_api
    def get_metadata_for_dump(self, book_id):
        mi = None
        # get the current sequence number for this book to pass back to the
        # backup thread. This will avoid double calls in the case where the
        # thread has not done the work between the put and the get_metadata
        sequence = self.dirtied_cache.get(book_id, None)
        if sequence is not None:
            try:
                # While a book is being created, the path is empty. Don't bother to
                # try to write the opf, because it will go to the wrong folder.
                if self._field_for('path', book_id):
                    mi = self._get_metadata(book_id)
                    # Always set cover to cover.jpg. Even if cover doesn't exist,
                    # no harm done. This way no need to call dirtied when
                    # cover is set/removed
                    mi.cover = 'cover.jpg'
                    mi.all_annotations = self._all_annotations_for_book(book_id)
            except:
                # This almost certainly means that the book has been deleted while
                # the backup operation sat in the queue.
                import traceback
                traceback.print_exc()
        return mi, sequence

    @write_api
    def clear_dirtied(self, book_id, sequence):
        # Clear the dirtied indicator for the books. This is used when fetching
        # metadata, creating an OPF, and writing a file are separated into steps.
        # The last step is clearing the indicator
        dc_sequence = self.dirtied_cache.get(book_id, None)
        if dc_sequence is None or sequence is None or dc_sequence == sequence:
            self.backend.mark_book_as_clean(book_id)
            self.dirtied_cache.pop(book_id, None)

    @write_api
    def write_backup(self, book_id, raw):
        try:
            path = self._field_for('path', book_id).replace('/', os.sep)
        except:
            return

        self.backend.write_backup(path, raw)

    @read_api
    def dirty_queue_length(self):
        return len(self.dirtied_cache)

    @read_api
    def read_backup(self, book_id):
        ''' Return the OPF metadata backup for the book as a bytestring or None
        if no such backup exists.  '''
        try:
            path = self._field_for('path', book_id).replace('/', os.sep)
        except:
            return

        try:
            return self.backend.read_backup(path)
        except OSError:
            return None

    @write_api
    def dump_metadata(self, book_ids=None, remove_from_dirtied=True,
            callback=None):
        # Write metadata for each record to an individual OPF file. If callback
        # is not None, it is called once at the start with the number of book_ids
        # being processed. And once for every book_id, with arguments (book_id,
        # mi, ok).
        if book_ids is None:
            book_ids = set(self.dirtied_cache)

        if callback is not None:
            callback(len(book_ids), True, False)

        for book_id in book_ids:
            if self._field_for('path', book_id) is None:
                if callback is not None:
                    callback(book_id, None, False)
                continue
            mi, sequence = self._get_metadata_for_dump(book_id)
            if mi is None:
                if callback is not None:
                    callback(book_id, mi, False)
                continue
            try:
                raw = metadata_to_opf(mi)
                self._write_backup(book_id, raw)
                if remove_from_dirtied:
                    self._clear_dirtied(book_id, sequence)
            except:
                pass
            if callback is not None:
                callback(book_id, mi, True)

    @write_api
    def set_cover(self, book_id_data_map):
        ''' Set the cover for this book. The data can be either a QImage,
        QPixmap, file object or bytestring. It can also be None, in which
        case any existing cover is removed. '''

        for book_id, data in iteritems(book_id_data_map):
            try:
                path = self._field_for('path', book_id).replace('/', os.sep)
            except AttributeError:
                self._update_path((book_id,))
                path = self._field_for('path', book_id).replace('/', os.sep)

            self.backend.set_cover(book_id, path, data)
        for cc in self.cover_caches:
            cc.invalidate(book_id_data_map)
        return self._set_field('cover', {
            book_id:(0 if data is None else 1) for book_id, data in iteritems(book_id_data_map)})

    @write_api
    def add_cover_cache(self, cover_cache):
        if not callable(cover_cache.invalidate):
            raise ValueError('Cover caches must have an invalidate method')
        self.cover_caches.add(cover_cache)

    @write_api
    def remove_cover_cache(self, cover_cache):
        self.cover_caches.discard(cover_cache)

    @write_api
    def set_metadata(self, book_id, mi, ignore_errors=False, force_changes=False,
                     set_title=True, set_authors=True, allow_case_change=False):
        '''
        Set metadata for the book `id` from the `Metadata` object `mi`

        Setting force_changes=True will force set_metadata to update fields even
        if mi contains empty values. In this case, 'None' is distinguished from
        'empty'. If mi.XXX is None, the XXX is not replaced, otherwise it is.
        The tags, identifiers, and cover attributes are special cases. Tags and
        identifiers cannot be set to None so they will always be replaced if
        force_changes is true. You must ensure that mi contains the values you
        want the book to have. Covers are always changed if a new cover is
        provided, but are never deleted. Also note that force_changes has no
        effect on setting title or authors.
        '''
        dirtied = set()

        try:
            # Handle code passing in an OPF object instead of a Metadata object
            mi = mi.to_book_metadata()
        except (AttributeError, TypeError):
            pass

        def set_field(name, val):
            dirtied.update(self._set_field(name, {book_id:val}, do_path_update=False, allow_case_change=allow_case_change))

        path_changed = False
        if set_title and mi.title:
            path_changed = True
            set_field('title', mi.title)
        authors_changed = False
        if set_authors:
            path_changed = True
            if not mi.authors:
                mi.authors = [_('Unknown')]
            authors = []
            for a in mi.authors:
                authors += string_to_authors(a)
            set_field('authors', authors)
            authors_changed = True

        if path_changed:
            self._update_path({book_id})

        def protected_set_field(name, val):
            try:
                set_field(name, val)
            except:
                if ignore_errors:
                    traceback.print_exc()
                else:
                    raise

        # force_changes has no effect on cover manipulation
        try:
            cdata = mi.cover_data[1]
            if cdata is None and isinstance(mi.cover, string_or_bytes) and mi.cover and os.access(mi.cover, os.R_OK):
                with lopen(mi.cover, 'rb') as f:
                    cdata = f.read() or None
            if cdata is not None:
                self._set_cover({book_id: cdata})
        except:
            if ignore_errors:
                traceback.print_exc()
            else:
                raise

        try:
            with self.backend.conn:  # Speed up set_metadata by not operating in autocommit mode
                for field in ('rating', 'series_index', 'timestamp'):
                    val = getattr(mi, field)
                    if val is not None:
                        protected_set_field(field, val)

                val = mi.get('author_sort', None)
                if authors_changed and (not val or mi.is_null('author_sort')):
                    val = self._author_sort_from_authors(mi.authors)
                if authors_changed or (force_changes and val is not None) or not mi.is_null('author_sort'):
                    protected_set_field('author_sort', val)

                for field in ('publisher', 'series', 'tags', 'comments',
                    'languages', 'pubdate'):
                    val = mi.get(field, None)
                    if (force_changes and val is not None) or not mi.is_null(field):
                        protected_set_field(field, val)

                val = mi.get('title_sort', None)
                if (force_changes and val is not None) or not mi.is_null('title_sort'):
                    protected_set_field('sort', val)

                # identifiers will always be replaced if force_changes is True
                mi_idents = mi.get_identifiers()
                if force_changes:
                    protected_set_field('identifiers', mi_idents)
                elif mi_idents:
                    identifiers = self._field_for('identifiers', book_id, default_value={})
                    for key, val in iteritems(mi_idents):
                        if val and val.strip():  # Don't delete an existing identifier
                            identifiers[icu_lower(key)] = val
                    protected_set_field('identifiers', identifiers)

                user_mi = mi.get_all_user_metadata(make_copy=False)
                fm = self.field_metadata
                for key in user_mi:
                    if (key in fm and user_mi[key]['datatype'] == fm[key]['datatype'] and (
                        user_mi[key]['datatype'] != 'text' or (
                            user_mi[key]['is_multiple'] == fm[key]['is_multiple']))):
                        val = mi.get(key, None)
                        if force_changes or val is not None:
                            protected_set_field(key, val)
                            idx = key + '_index'
                            if idx in self.fields:
                                extra = mi.get_extra(key)
                                if extra is not None or force_changes:
                                    protected_set_field(idx, extra)
        except:
            # sqlite will rollback the entire transaction, thanks to the with
            # statement, so we have to re-read everything form the db to ensure
            # the db and Cache are in sync
            self._reload_from_db()
            raise
        return dirtied

    def _do_add_format(self, book_id, fmt, stream, name=None, mtime=None):
        path = self._field_for('path', book_id)
        if path is None:
            # Theoretically, this should never happen, but apparently it
            # does: https://www.mobileread.com/forums/showthread.php?t=233353
            self._update_path({book_id}, mark_as_dirtied=False)
            path = self._field_for('path', book_id)

        path = path.replace('/', os.sep)
        title = self._field_for('title', book_id, default_value=_('Unknown'))
        try:
            author = self._field_for('authors', book_id, default_value=(_('Unknown'),))[0]
        except IndexError:
            author = _('Unknown')

        size, fname = self.backend.add_format(book_id, fmt, stream, title, author, path, name, mtime=mtime)
        return size, fname

    @api
    def add_format(self, book_id, fmt, stream_or_path, replace=True, run_hooks=True, dbapi=None):
        '''
        Add a format to the specified book. Return True if the format was added successfully.

        :param replace: If True replace existing format, otherwise if the format already exists, return False.
        :param run_hooks: If True, file type plugins are run on the format before and after being added.
        :param dbapi: Internal use only.
        '''
        needs_close = False
        if run_hooks:
            # Run import plugins, the write lock is not held to cater for
            # broken plugins that might spin the event loop by popping up a
            # message in the GUI during the processing.
            npath = run_import_plugins(stream_or_path, fmt)
            fmt = os.path.splitext(npath)[-1].lower().replace('.', '').upper()
            stream_or_path = lopen(npath, 'rb')
            needs_close = True
            fmt = check_ebook_format(stream_or_path, fmt)

        with self.write_lock:
            if not self._has_id(book_id):
                raise NoSuchBook(book_id)
            fmt = (fmt or '').upper()
            self.format_metadata_cache[book_id].pop(fmt, None)
            try:
                name = self.fields['formats'].format_fname(book_id, fmt)
            except Exception:
                name = None

            if name and not replace:
                return False

            if hasattr(stream_or_path, 'read'):
                stream = stream_or_path
            else:
                stream = lopen(stream_or_path, 'rb')
                needs_close = True
            try:
                stream = stream_or_path if hasattr(stream_or_path, 'read') else lopen(stream_or_path, 'rb')
                size, fname = self._do_add_format(book_id, fmt, stream, name)
            finally:
                if needs_close:
                    stream.close()
            del stream

            max_size = self.fields['formats'].table.update_fmt(book_id, fmt, fname, size, self.backend)
            self.fields['size'].table.update_sizes({book_id: max_size})
            self._update_last_modified((book_id,))
            self.event_dispatcher(EventType.format_added, book_id, fmt)

        if run_hooks:
            # Run post import plugins, the write lock is released so the plugin
            # can call api without a locking violation.
            run_plugins_on_postimport(dbapi or self, book_id, fmt)
            stream_or_path.close()

        return True

    @write_api
    def remove_formats(self, formats_map, db_only=False):
        '''
        Remove the specified formats from the specified books.

        :param formats_map: A mapping of book_id to a list of formats to be removed from the book.
        :param db_only: If True, only remove the record for the format from the db, do not delete the actual format file from the filesystem.
        '''
        table = self.fields['formats'].table
        formats_map = {book_id:frozenset((f or '').upper() for f in fmts) for book_id, fmts in iteritems(formats_map)}

        for book_id, fmts in iteritems(formats_map):
            for fmt in fmts:
                self.format_metadata_cache[book_id].pop(fmt, None)

        if not db_only:
            removes = defaultdict(set)
            for book_id, fmts in iteritems(formats_map):
                try:
                    path = self._field_for('path', book_id).replace('/', os.sep)
                except:
                    continue
                for fmt in fmts:
                    try:
                        name = self.fields['formats'].format_fname(book_id, fmt)
                    except:
                        continue
                    if name and path:
                        removes[book_id].add((fmt, name, path))
            if removes:
                self.backend.remove_formats(removes)

        size_map = table.remove_formats(formats_map, self.backend)
        self.fields['size'].table.update_sizes(size_map)
        self._update_last_modified(tuple(formats_map))
        self.event_dispatcher(EventType.formats_removed, formats_map)

    @read_api
    def get_next_series_num_for(self, series, field='series', current_indices=False):
        '''
        Return the next series index for the specified series, taking into account the various preferences that
        control next series number generation.

        :param field: The series-like field (defaults to the builtin series column)
        :param current_indices: If True, returns a mapping of book_id to current series_index value instead.
        '''
        books = ()
        sf = self.fields[field]
        if series:
            q = icu_lower(series)
            for val, book_ids in sf.iter_searchable_values(self._get_proxy_metadata, frozenset(self._all_book_ids())):
                if q == icu_lower(val):
                    books = book_ids
                    break
        idf = sf.index_field
        index_map = {book_id:self._fast_field_for(idf, book_id, default_value=1.0) for book_id in books}
        if current_indices:
            return index_map
        series_indices = sorted(itervalues(index_map))
        return _get_next_series_num_for_list(tuple(series_indices), unwrap=False)

    @read_api
    def author_sort_from_authors(self, authors, key_func=icu_lower):
        '''Given a list of authors, return the author_sort string for the authors,
        preferring the author sort associated with the author over the computed
        string. '''
        table = self.fields['authors'].table
        result = []
        rmap = {key_func(v):k for k, v in iteritems(table.id_map)}
        for aut in authors:
            aid = rmap.get(key_func(aut), None)
            result.append(author_to_author_sort(aut) if aid is None else table.asort_map[aid])
        return ' & '.join(_f for _f in result if _f)

    @read_api
    def data_for_has_book(self):
        ''' Return data suitable for use in :meth:`has_book`. This can be used for an
        implementation of :meth:`has_book` in a worker process without access to the
        db. '''
        try:
            return {icu_lower(title) for title in itervalues(self.fields['title'].table.book_col_map)}
        except TypeError:
            # Some non-unicode titles in the db
            return {icu_lower(as_unicode(title)) for title in itervalues(self.fields['title'].table.book_col_map)}

    @read_api
    def has_book(self, mi):
        ''' Return True iff the database contains an entry with the same title
        as the passed in Metadata object. The comparison is case-insensitive.
        See also :meth:`data_for_has_book`.  '''
        title = mi.title
        if title:
            if isbytestring(title):
                title = title.decode(preferred_encoding, 'replace')
            q = icu_lower(title).strip()
            for title in itervalues(self.fields['title'].table.book_col_map):
                if q == icu_lower(title):
                    return True
        return False

    @read_api
    def has_id(self, book_id):
        ' Return True iff the specified book_id exists in the db '''
        return book_id in self.fields['title'].table.book_col_map

    @write_api
    def create_book_entry(self, mi, cover=None, add_duplicates=True, force_id=None, apply_import_tags=True, preserve_uuid=False):
        if mi.tags:
            mi.tags = list(mi.tags)
        if apply_import_tags:
            _add_newbook_tag(mi)
            _add_default_custom_column_values(mi, self.field_metadata)
        if not add_duplicates and self._has_book(mi):
            return
        series_index = (self._get_next_series_num_for(mi.series) if mi.series_index is None else mi.series_index)
        try:
            series_index = float(series_index)
        except Exception:
            try:
                series_index = float(self._get_next_series_num_for(mi.series))
            except Exception:
                series_index = 1.0
        if not mi.authors:
            mi.authors = (_('Unknown'),)
        aus = mi.author_sort if not mi.is_null('author_sort') else self._author_sort_from_authors(mi.authors)
        mi.title = mi.title or _('Unknown')
        if isbytestring(aus):
            aus = aus.decode(preferred_encoding, 'replace')
        if isbytestring(mi.title):
            mi.title = mi.title.decode(preferred_encoding, 'replace')
        if force_id is None:
            self.backend.execute('INSERT INTO books(title, series_index, author_sort) VALUES (?, ?, ?)',
                         (mi.title, series_index, aus))
        else:
            self.backend.execute('INSERT INTO books(id, title, series_index, author_sort) VALUES (?, ?, ?, ?)',
                         (force_id, mi.title, series_index, aus))
        book_id = self.backend.last_insert_rowid()
        self.event_dispatcher(EventType.book_created, book_id)

        mi.timestamp = utcnow() if mi.timestamp is None else mi.timestamp
        mi.pubdate = UNDEFINED_DATE if mi.pubdate is None else mi.pubdate
        if cover is not None:
            mi.cover, mi.cover_data = None, (None, cover)
        self._set_metadata(book_id, mi, ignore_errors=True)
        if preserve_uuid and mi.uuid:
            self._set_field('uuid', {book_id:mi.uuid})
        # Update the caches for fields from the books table
        self.fields['size'].table.book_col_map[book_id] = 0
        row = next(self.backend.execute('SELECT sort, series_index, author_sort, uuid, has_cover FROM books WHERE id=?', (book_id,)))
        for field, val in zip(('sort', 'series_index', 'author_sort', 'uuid', 'cover'), row):
            if field == 'cover':
                val = bool(val)
            elif field == 'uuid':
                self.fields[field].table.uuid_to_id_map[val] = book_id
            self.fields[field].table.book_col_map[book_id] = val

        return book_id

    @api
    def add_books(self, books, add_duplicates=True, apply_import_tags=True, preserve_uuid=False, run_hooks=True, dbapi=None):
        '''
        Add the specified books to the library. Books should be an iterable of
        2-tuples, each 2-tuple of the form :code:`(mi, format_map)` where mi is a
        Metadata object and format_map is a dictionary of the form :code:`{fmt: path_or_stream}`,
        for example: :code:`{'EPUB': '/path/to/file.epub'}`.

        Returns a pair of lists: :code:`ids, duplicates`. ``ids`` contains the book ids for all newly created books in the
        database. ``duplicates`` contains the :code:`(mi, format_map)` for all books that already exist in the database
        as per the simple duplicate detection heuristic used by :meth:`has_book`.
        '''
        duplicates, ids = [], []
        fmt_map = {}
        for mi, format_map in books:
            book_id = self.create_book_entry(mi, add_duplicates=add_duplicates, apply_import_tags=apply_import_tags, preserve_uuid=preserve_uuid)
            if book_id is None:
                duplicates.append((mi, format_map))
            else:
                ids.append(book_id)
                for fmt, stream_or_path in iteritems(format_map):
                    if self.add_format(book_id, fmt, stream_or_path, dbapi=dbapi, run_hooks=run_hooks):
                        fmt_map[fmt.lower()] = getattr(stream_or_path, 'name', stream_or_path) or '<stream>'
            run_plugins_on_postadd(dbapi or self, book_id, fmt_map)
        return ids, duplicates

    @write_api
    def remove_books(self, book_ids, permanent=False):
        ''' Remove the books specified by the book_ids from the database and delete
        their format files. If ``permanent`` is False, then the format files
        are placed in the recycle bin. '''
        path_map = {}
        for book_id in book_ids:
            try:
                path = self._field_for('path', book_id).replace('/', os.sep)
            except:
                path = None
            path_map[book_id] = path
        if iswindows:
            paths = (x.replace(os.sep, '/') for x in itervalues(path_map) if x)
            self.backend.windows_check_if_files_in_use(paths)

        self.backend.remove_books(path_map, permanent=permanent)
        for field in itervalues(self.fields):
            try:
                table = field.table
            except AttributeError:
                continue  # Some fields like ondevice do not have tables
            else:
                table.remove_books(book_ids, self.backend)
        self._search_api.discard_books(book_ids)
        self._clear_caches(book_ids=book_ids, template_cache=False, search_cache=False)
        for cc in self.cover_caches:
            cc.invalidate(book_ids)
        self.event_dispatcher(EventType.books_removed, book_ids)

    @read_api
    def author_sort_strings_for_books(self, book_ids):
        val_map = {}
        for book_id in book_ids:
            authors = self._field_ids_for('authors', book_id)
            adata = self._author_data(authors)
            val_map[book_id] = tuple(adata[aid]['sort'] for aid in authors)
        return val_map

    @write_api
    def rename_items(self, field, item_id_to_new_name_map, change_index=True, restrict_to_book_ids=None):
        '''
        Rename items from a many-one or many-many field such as tags or series.

        :param change_index: When renaming in a series-like field also change the series_index values.
        :param restrict_to_book_ids: An optional set of book ids for which the rename is to be performed, defaults to all books.
        '''

        f = self.fields[field]
        affected_books = set()
        try:
            sv = f.metadata['is_multiple']['ui_to_list']
        except (TypeError, KeyError, AttributeError):
            sv = None

        if restrict_to_book_ids is not None:
            # We have a VL. Only change the item name for those books
            if not isinstance(restrict_to_book_ids, (Set, MutableSet)):
                restrict_to_book_ids = frozenset(restrict_to_book_ids)
            id_map = {}
            default_process_map = {}
            for old_id, new_name in iteritems(item_id_to_new_name_map):
                new_names = tuple(x.strip() for x in new_name.split(sv)) if sv else (new_name,)
                # Get a list of books in the VL with the item
                books_with_id = f.books_for(old_id)
                books_to_process = books_with_id & restrict_to_book_ids
                if len(books_with_id) == len(books_to_process):
                    # All the books with the ID are in the VL, so we can use
                    # the normal processing
                    default_process_map[old_id] = new_name
                elif books_to_process:
                    affected_books.update(books_to_process)
                    newvals = {}
                    for book_id in books_to_process:
                        # Get the current values, remove the one being renamed, then add
                        # the new value(s) back.
                        vals = self._field_for(field, book_id)
                        # Check for is_multiple
                        if isinstance(vals, tuple):
                            # We must preserve order.
                            vals = list(vals)
                            # Don't need to worry about case here because we
                            # are fetching its one-true spelling. But lets be
                            # careful anyway
                            try:
                                dex = vals.index(self._get_item_name(field, old_id))
                                # This can put the name back with a different case
                                vals[dex] = new_names[0]
                                # now add any other items if they aren't already there
                                if len(new_names) > 1:
                                    set_vals = {icu_lower(x) for x in vals}
                                    for v in new_names[1:]:
                                        lv = icu_lower(v)
                                        if lv not in set_vals:
                                            vals.append(v)
                                            set_vals.add(lv)
                                newvals[book_id] = vals
                            except Exception:
                                traceback.print_exc()
                        else:
                            newvals[book_id] = new_names[0]
                    # Allow case changes
                    self._set_field(field, newvals)
                    id_map[old_id] = self._get_item_id(field, new_names[0])
            if default_process_map:
                ab, idm = self._rename_items(field, default_process_map, change_index=change_index)
                affected_books.update(ab)
                id_map.update(idm)
            self.event_dispatcher(EventType.items_renamed, field, affected_books, id_map)
            return affected_books, id_map

        try:
            func = f.table.rename_item
        except AttributeError:
            raise ValueError('Cannot rename items for one-one fields: %s' % field)
        moved_books = set()
        id_map = {}
        for item_id, new_name in iteritems(item_id_to_new_name_map):
            new_names = tuple(x.strip() for x in new_name.split(sv)) if sv else (new_name,)
            books, new_id = func(item_id, new_names[0], self.backend)
            affected_books.update(books)
            id_map[item_id] = new_id
            if new_id != item_id:
                moved_books.update(books)
            if len(new_names) > 1:
                # Add the extra items to the books
                extra = new_names[1:]
                self._set_field(field, {book_id:self._fast_field_for(f, book_id) + extra for book_id in books})

        if affected_books:
            if field == 'authors':
                self._set_field('author_sort',
                                {k:' & '.join(v) for k, v in iteritems(self._author_sort_strings_for_books(affected_books))})
                self._update_path(affected_books, mark_as_dirtied=False)
            elif change_index and hasattr(f, 'index_field') and tweaks['series_index_auto_increment'] != 'no_change':
                for book_id in moved_books:
                    self._set_field(f.index_field.name, {book_id:self._get_next_series_num_for(self._fast_field_for(f, book_id), field=field)})
            self._mark_as_dirty(affected_books)
        self.event_dispatcher(EventType.items_renamed, field, affected_books, id_map)
        return affected_books, id_map

    @write_api
    def remove_items(self, field, item_ids, restrict_to_book_ids=None):
        ''' Delete all items in the specified field with the specified ids.
        Returns the set of affected book ids. ``restrict_to_book_ids`` is an
        optional set of books ids. If specified the items will only be removed
        from those books. '''
        field = self.fields[field]
        if restrict_to_book_ids is not None and not isinstance(restrict_to_book_ids, (MutableSet, Set)):
            restrict_to_book_ids = frozenset(restrict_to_book_ids)
        affected_books = field.table.remove_items(item_ids, self.backend,
                                                  restrict_to_book_ids=restrict_to_book_ids)
        if affected_books:
            if hasattr(field, 'index_field'):
                self._set_field(field.index_field.name, {bid:1.0 for bid in affected_books})
            else:
                self._mark_as_dirty(affected_books)
        self.event_dispatcher(EventType.items_removed, field, affected_books, item_ids)
        return affected_books

    @write_api
    def add_custom_book_data(self, name, val_map, delete_first=False):
        ''' Add data for name where val_map is a map of book_ids to values. If
        delete_first is True, all previously stored data for name will be
        removed. '''
        missing = frozenset(val_map) - self._all_book_ids()
        if missing:
            raise ValueError('add_custom_book_data: no such book_ids: %d'%missing)
        self.backend.add_custom_data(name, val_map, delete_first)

    @read_api
    def get_custom_book_data(self, name, book_ids=(), default=None):
        ''' Get data for name. By default returns data for all book_ids, pass
        in a list of book ids if you only want some data. Returns a map of
        book_id to values. If a particular value could not be decoded, uses
        default for it. '''
        return self.backend.get_custom_book_data(name, book_ids, default)

    @write_api
    def delete_custom_book_data(self, name, book_ids=()):
        ''' Delete data for name. By default deletes all data, if you only want
        to delete data for some book ids, pass in a list of book ids. '''
        self.backend.delete_custom_book_data(name, book_ids)

    @read_api
    def get_ids_for_custom_book_data(self, name):
        ''' Return the set of book ids for which name has data. '''
        return self.backend.get_ids_for_custom_book_data(name)

    @read_api
    def conversion_options(self, book_id, fmt='PIPE'):
        return self.backend.conversion_options(book_id, fmt)

    @read_api
    def has_conversion_options(self, ids, fmt='PIPE'):
        return self.backend.has_conversion_options(ids, fmt)

    @write_api
    def delete_conversion_options(self, book_ids, fmt='PIPE'):
        return self.backend.delete_conversion_options(book_ids, fmt)

    @write_api
    def set_conversion_options(self, options, fmt='PIPE'):
        ''' options must be a map of the form {book_id:conversion_options} '''
        return self.backend.set_conversion_options(options, fmt)

    @write_api
    def refresh_format_cache(self):
        self.fields['formats'].table.read(self.backend)
        self.format_metadata_cache.clear()

    @write_api
    def refresh_ondevice(self):
        self.fields['ondevice'].clear_caches()
        self.clear_search_caches()
        self.clear_composite_caches()

    @read_api
    def books_matching_device_book(self, lpath):
        ans = set()
        for book_id, (_, _, _, _, lpaths) in self.fields['ondevice'].cache.items():
            if lpath in lpaths:
                ans.add(book_id)
        return ans

    @read_api
    def tags_older_than(self, tag, delta=None, must_have_tag=None, must_have_authors=None):
        '''
        Return the ids of all books having the tag ``tag`` that are older than
        the specified time. tag comparison is case insensitive.

        :param delta: A timedelta object or None. If None, then all ids with
            the tag are returned.

        :param must_have_tag: If not None the list of matches will be
            restricted to books that have this tag

        :param must_have_authors: A list of authors. If not None the list of
            matches will be restricted to books that have these authors (case
            insensitive).

        '''
        tag_map = {icu_lower(v):k for k, v in iteritems(self._get_id_map('tags'))}
        tag = icu_lower(tag.strip())
        mht = icu_lower(must_have_tag.strip()) if must_have_tag else None
        tag_id, mht_id = tag_map.get(tag, None), tag_map.get(mht, None)
        ans = set()
        if mht_id is None and mht:
            return ans
        if tag_id is not None:
            tagged_books = self._books_for_field('tags', tag_id)
            if mht_id is not None and tagged_books:
                tagged_books = tagged_books.intersection(self._books_for_field('tags', mht_id))
            if tagged_books:
                if must_have_authors is not None:
                    amap = {icu_lower(v):k for k, v in iteritems(self._get_id_map('authors'))}
                    books = None
                    for author in must_have_authors:
                        abooks = self._books_for_field('authors', amap.get(icu_lower(author), None))
                        books = abooks if books is None else books.intersection(abooks)
                        if not books:
                            break
                    tagged_books = tagged_books.intersection(books or set())
                if delta is None:
                    ans = tagged_books
                else:
                    now = nowf()
                    for book_id in tagged_books:
                        ts = self._field_for('timestamp', book_id)
                        if (now - ts) > delta:
                            ans.add(book_id)
        return ans

    @write_api
    def set_sort_for_authors(self, author_id_to_sort_map, update_books=True):
        sort_map = self.fields['authors'].table.set_sort_names(author_id_to_sort_map, self.backend)
        changed_books = set()
        if update_books:
            val_map = {}
            for author_id in sort_map:
                books = self._books_for_field('authors', author_id)
                changed_books |= books
                for book_id in books:
                    authors = self._field_ids_for('authors', book_id)
                    adata = self._author_data(authors)
                    sorts = [adata[x]['sort'] for x in authors]
                    val_map[book_id] = ' & '.join(sorts)
            if val_map:
                self._set_field('author_sort', val_map)
        if changed_books:
            self._mark_as_dirty(changed_books)
        return changed_books

    @write_api
    def set_link_for_authors(self, author_id_to_link_map):
        link_map = self.fields['authors'].table.set_links(author_id_to_link_map, self.backend)
        changed_books = set()
        for author_id in link_map:
            changed_books |= self._books_for_field('authors', author_id)
        if changed_books:
            self._mark_as_dirty(changed_books)
        return changed_books

    @read_api
    def lookup_by_uuid(self, uuid):
        return self.fields['uuid'].table.lookup_by_uuid(uuid)

    @write_api
    def delete_custom_column(self, label=None, num=None):
        self.backend.delete_custom_column(label, num)

    @write_api
    def create_custom_column(self, label, name, datatype, is_multiple, editable=True, display={}):
        return self.backend.create_custom_column(label, name, datatype, is_multiple, editable=editable, display=display)

    @write_api
    def set_custom_column_metadata(self, num, name=None, label=None, is_editable=None,
                                   display=None, update_last_modified=False):
        changed = self.backend.set_custom_column_metadata(num, name=name, label=label, is_editable=is_editable, display=display)
        if changed:
            if update_last_modified:
                self._update_last_modified(self._all_book_ids())
            else:
                self.backend.prefs.set('update_all_last_mod_dates_on_start', True)
        return changed

    @read_api
    def get_books_for_category(self, category, item_id_or_composite_value):
        f = self.fields[category]
        if hasattr(f, 'get_books_for_val'):
            # Composite field
            return f.get_books_for_val(item_id_or_composite_value, self._get_proxy_metadata, self._all_book_ids())
        return self._books_for_field(f.name, int(item_id_or_composite_value))

    @read_api
    def data_for_find_identical_books(self):
        ''' Return data that can be used to implement
        :meth:`find_identical_books` in a worker process without access to the
        db. See db.utils for an implementation. '''
        at = self.fields['authors'].table
        author_map = defaultdict(set)
        for aid, author in iteritems(at.id_map):
            author_map[icu_lower(author)].add(aid)
        return (author_map, at.col_book_map.copy(), self.fields['title'].table.book_col_map.copy(), self.fields['languages'].book_value_map.copy())

    @read_api
    def update_data_for_find_identical_books(self, book_id, data):
        author_map, author_book_map, title_map, lang_map = data
        title_map[book_id] = self._field_for('title', book_id)
        lang_map[book_id] = self._field_for('languages', book_id)
        at = self.fields['authors'].table
        for aid in at.book_col_map.get(book_id, ()):
            author_map[icu_lower(at.id_map[aid])].add(aid)
            try:
                author_book_map[aid].add(book_id)
            except KeyError:
                author_book_map[aid] = {book_id}

    @read_api
    def find_identical_books(self, mi, search_restriction='', book_ids=None):
        ''' Finds books that have a superset of the authors in mi and the same
        title (title is fuzzy matched). See also :meth:`data_for_find_identical_books`. '''
        from calibre.db.utils import fuzzy_title
        identical_book_ids = set()
        langq = tuple(x for x in map(canonicalize_lang, mi.languages or ()) if x and x != 'und')
        if mi.authors:
            try:
                quathors = mi.authors[:20]  # Too many authors causes parsing of the search expression to fail
                query = ' and '.join('authors:"=%s"'%(a.replace('"', '')) for a in quathors)
                qauthors = mi.authors[20:]
            except ValueError:
                return identical_book_ids
            try:
                book_ids = self._search(query, restriction=search_restriction, book_ids=book_ids)
            except:
                traceback.print_exc()
                return identical_book_ids
            if qauthors and book_ids:
                matches = set()
                qauthors = {icu_lower(x) for x in qauthors}
                for book_id in book_ids:
                    aut = self._field_for('authors', book_id)
                    if aut:
                        aut = {icu_lower(x) for x in aut}
                        if aut.issuperset(qauthors):
                            matches.add(book_id)
                book_ids = matches

            for book_id in book_ids:
                fbook_title = self._field_for('title', book_id)
                fbook_title = fuzzy_title(fbook_title)
                mbook_title = fuzzy_title(mi.title)
                if fbook_title == mbook_title:
                    bl = self._field_for('languages', book_id)
                    if not langq or not bl or bl == langq:
                        identical_book_ids.add(book_id)
        return identical_book_ids

    @read_api
    def get_top_level_move_items(self):
        all_paths = {self._field_for('path', book_id).partition('/')[0] for book_id in self._all_book_ids()}
        return self.backend.get_top_level_move_items(all_paths)

    @write_api
    def move_library_to(self, newloc, progress=None, abort=None):
        def progress_callback(item_name, item_count, total):
            try:
                if progress is not None:
                    progress(item_name, item_count, total)
            except Exception:
                import traceback
                traceback.print_exc()

        all_paths = {self._field_for('path', book_id).partition('/')[0] for book_id in self._all_book_ids()}
        self.backend.move_library_to(all_paths, newloc, progress=progress_callback, abort=abort)

    @read_api
    def saved_search_names(self):
        return self._search_api.saved_searches.names()

    @read_api
    def saved_search_lookup(self, name):
        return self._search_api.saved_searches.lookup(name)

    @write_api
    def saved_search_set_all(self, smap):
        self._search_api.saved_searches.set_all(smap)
        self._clear_search_caches()

    @write_api
    def saved_search_delete(self, name):
        self._search_api.saved_searches.delete(name)
        self._clear_search_caches()

    @write_api
    def saved_search_add(self, name, val):
        self._search_api.saved_searches.add(name, val)

    @write_api
    def saved_search_rename(self, old_name, new_name):
        self._search_api.saved_searches.rename(old_name, new_name)
        self._clear_search_caches()

    @write_api
    def change_search_locations(self, newlocs):
        self._search_api.change_locations(newlocs)

    @write_api
    def refresh_search_locations(self):
        self._search_api.change_locations(self.field_metadata.get_search_terms())

    @write_api
    def dump_and_restore(self, callback=None, sql=None):
        return self.backend.dump_and_restore(callback=callback, sql=sql)

    @write_api
    def vacuum(self):
        self.backend.vacuum()

    @write_api
    def close(self):
        self.event_dispatcher.close()
        from calibre.customize.ui import available_library_closed_plugins
        for plugin in available_library_closed_plugins():
            try:
                plugin.run(self)
            except Exception:
                import traceback
                traceback.print_exc()
        self.backend.close()

    @property
    def is_closed(self):
        return self.backend.is_closed

    @write_api
    def restore_book(self, book_id, mi, last_modified, path, formats, annotations=()):
        ''' Restore the book entry in the database for a book that already exists on the filesystem '''
        cover = mi.cover
        mi.cover = None
        self._create_book_entry(mi, add_duplicates=True,
                force_id=book_id, apply_import_tags=False, preserve_uuid=True)
        self._update_last_modified((book_id,), last_modified)
        if cover and os.path.exists(cover):
            self._set_field('cover', {book_id:1})
        self.backend.restore_book(book_id, path, formats)
        if annotations:
            self._restore_annotations(book_id, annotations)

    @read_api
    def virtual_libraries_for_books(self, book_ids, virtual_fields=None):
        # use a primitive lock to ensure that only one thread is updating
        # the cache and that recursive calls don't do the update. This
        # method can recurse via self._search()
        with try_lock(self.vls_cache_lock) as got_lock:
            # Using a list is slightly faster than a set.
            c = defaultdict(list)
            if not got_lock:
                # We get here if resolving the books in a VL triggers another VL
                # calculation. This can be 'real' recursion, in which case the
                # eventual answer will be wrong. It can also be a  search using
                # a location of 'all' that causes evaluation of a composite that
                # references virtual_libraries(). If the composite isn't used in a
                # VL then the eventual answer will be correct because get_metadata
                # will clear the caches.
                return c
            if self.vls_for_books_cache is None:
                self.vls_for_books_cache_is_loading = True
                libraries = self._pref('virtual_libraries', {})
                for lib, expr in libraries.items():
                    book = None
                    try:
                        for book in self._search(expr, virtual_fields=virtual_fields):
                            c[book].append(lib)
                    except Exception as e:
                        if book:
                            c[book].append(_('[Error in Virtual library {0}: {1}]').format(lib, str(e)))
                self.vls_for_books_cache = {b:tuple(sorted(libs, key=sort_key)) for b, libs in c.items()}
        if not book_ids:
            book_ids = self._all_book_ids()
        # book_ids is usually 1 long. The loop will be faster than a comprehension
        r = {}
        default = ()
        for b in book_ids:
            r[b] = self.vls_for_books_cache.get(b, default)
        return r

    @read_api
    def user_categories_for_books(self, book_ids, proxy_metadata_map=None):
        ''' Return the user categories for the specified books.
        proxy_metadata_map is optional and is useful for a performance boost,
        in contexts where a ProxyMetadata object for the books already exists.
        It should be a mapping of book_ids to their corresponding ProxyMetadata
        objects.
        '''
        user_cats = self._pref('user_categories', {})
        pmm = proxy_metadata_map or {}
        ans = {}

        for book_id in book_ids:
            proxy_metadata = pmm.get(book_id) or self._get_proxy_metadata(book_id)
            user_cat_vals = ans[book_id] = {}
            for ucat, categories in iteritems(user_cats):
                user_cat_vals[ucat] = res = []
                for name, cat, ign in categories:
                    try:
                        field_obj = self.fields[cat]
                    except KeyError:
                        continue

                    if field_obj.is_composite:
                        v = field_obj.get_value_with_cache(book_id, lambda x:proxy_metadata)
                    else:
                        v = self._fast_field_for(field_obj, book_id)

                    if isinstance(v, (list, tuple)):
                        if name in v:
                            res.append([name, cat])
                    elif name == v:
                        res.append([name, cat])
        return ans

    @write_api
    def embed_metadata(self, book_ids, only_fmts=None, report_error=None, report_progress=None):
        ''' Update metadata in all formats of the specified book_ids to current metadata in the database. '''
        field = self.fields['formats']
        from calibre.customize.ui import apply_null_metadata
        from calibre.ebooks.metadata.meta import set_metadata
        from calibre.ebooks.metadata.opf2 import pretty_print
        if only_fmts:
            only_fmts = {f.lower() for f in only_fmts}

        def doit(fmt, mi, stream):
            with apply_null_metadata, pretty_print:
                set_metadata(stream, mi, stream_type=fmt, report_error=report_error)
            stream.seek(0, os.SEEK_END)
            return stream.tell()

        for i, book_id in enumerate(book_ids):
            fmts = field.table.book_col_map.get(book_id, ())
            if not fmts:
                continue
            mi = self.get_metadata(book_id, get_cover=True, cover_as_data=True)
            try:
                path = self._field_for('path', book_id).replace('/', os.sep)
            except:
                continue
            for fmt in fmts:
                if only_fmts is not None and fmt.lower() not in only_fmts:
                    continue
                try:
                    name = self.fields['formats'].format_fname(book_id, fmt)
                except:
                    continue
                if name and path:
                    new_size = self.backend.apply_to_format(book_id, path, name, fmt, partial(doit, fmt, mi))
                    if new_size is not None:
                        self.format_metadata_cache[book_id].get(fmt, {})['size'] = new_size
                        max_size = self.fields['formats'].table.update_fmt(book_id, fmt, name, new_size, self.backend)
                        self.fields['size'].table.update_sizes({book_id: max_size})
            if report_progress is not None:
                report_progress(i+1, len(book_ids), mi)

    @read_api
    def get_last_read_positions(self, book_id, fmt, user):
        fmt = fmt.upper()
        ans = []
        for device, cfi, epoch, pos_frac in self.backend.execute(
                'SELECT device,cfi,epoch,pos_frac FROM last_read_positions WHERE book=? AND format=? AND user=?',
                (book_id, fmt, user)):
            ans.append({'device':device, 'cfi': cfi, 'epoch':epoch, 'pos_frac':pos_frac})
        return ans

    @write_api
    def set_last_read_position(self, book_id, fmt, user='_', device='_', cfi=None, epoch=None, pos_frac=0):
        fmt = fmt.upper()
        device = device or '_'
        user = user or '_'
        if not cfi:
            self.backend.execute(
                'DELETE FROM last_read_positions WHERE book=? AND format=? AND user=? AND device=?',
                (book_id, fmt, user, device))
        else:
            self.backend.execute(
                'INSERT OR REPLACE INTO last_read_positions(book,format,user,device,cfi,epoch,pos_frac) VALUES (?,?,?,?,?,?,?)',
                (book_id, fmt, user, device, cfi, epoch or time(), pos_frac))

    @read_api
    def export_library(self, library_key, exporter, progress=None, abort=None):
        from polyglot.binary import as_hex_unicode
        key_prefix = as_hex_unicode(library_key)
        book_ids = self._all_book_ids()
        total = len(book_ids) + 1
        format_metadata = {}
        if progress is not None:
            progress('metadata.db', 0, total)
        pt = PersistentTemporaryFile('-export.db')
        pt.close()
        self.backend.backup_database(pt.name)
        dbkey = key_prefix + ':::' + 'metadata.db'
        with lopen(pt.name, 'rb') as f:
            exporter.add_file(f, dbkey)
        os.remove(pt.name)
        metadata = {'format_data':format_metadata, 'metadata.db':dbkey, 'total':total}
        for i, book_id in enumerate(book_ids):
            if abort is not None and abort.is_set():
                return
            if progress is not None:
                progress(self._field_for('title', book_id), i + 1, total)
            format_metadata[book_id] = {}
            for fmt in self._formats(book_id):
                mdata = self.format_metadata(book_id, fmt)
                key = '%s:%s:%s' % (key_prefix, book_id, fmt)
                format_metadata[book_id][fmt] = key
                with exporter.start_file(key, mtime=mdata.get('mtime')) as dest:
                    self._copy_format_to(book_id, fmt, dest, report_file_size=dest.ensure_space)
            cover_key = '%s:%s:%s' % (key_prefix, book_id, '.cover')
            with exporter.start_file(cover_key) as dest:
                if not self.copy_cover_to(book_id, dest, report_file_size=dest.ensure_space):
                    dest.discard()
                else:
                    format_metadata[book_id]['.cover'] = cover_key
        exporter.set_metadata(library_key, metadata)
        if progress is not None:
            progress(_('Completed'), total, total)

    @read_api
    def annotations_map_for_book(self, book_id, fmt, user_type='local', user='viewer'):
        ans = {}
        for annot in self.backend.annotations_for_book(book_id, fmt, user_type, user):
            ans.setdefault(annot['type'], []).append(annot)
        return ans

    @read_api
    def all_annotations_for_book(self, book_id):
        return tuple(self.backend.all_annotations_for_book(book_id))

    @read_api
    def annotation_count_for_book(self, book_id):
        return self.backend.annotation_count_for_book(book_id)

    @read_api
    def all_annotation_users(self):
        return tuple(self.backend.all_annotation_users())

    @read_api
    def all_annotation_types(self):
        return tuple(self.backend.all_annotation_types())

    @read_api
    def all_annotations(self, restrict_to_user=None, limit=None, annotation_type=None, ignore_removed=False, restrict_to_book_ids=None):
        return tuple(self.backend.all_annotations(restrict_to_user, limit, annotation_type, ignore_removed, restrict_to_book_ids))

    @read_api
    def search_annotations(
        self,
        fts_engine_query,
        use_stemming=True,
        highlight_start=None,
        highlight_end=None,
        snippet_size=None,
        annotation_type=None,
        restrict_to_book_ids=None,
        restrict_to_user=None,
        ignore_removed=False
    ):
        return tuple(self.backend.search_annotations(
            fts_engine_query, use_stemming, highlight_start, highlight_end,
            snippet_size, annotation_type, restrict_to_book_ids, restrict_to_user,
            ignore_removed
        ))

    @write_api
    def delete_annotations(self, annot_ids):
        self.backend.delete_annotations(annot_ids)

    @write_api
    def update_annotations(self, annot_id_map):
        self.backend.update_annotations(annot_id_map)

    @write_api
    def restore_annotations(self, book_id, annotations):
        from calibre.utils.date import EPOCH
        from calibre.utils.iso8601 import parse_iso8601
        umap = defaultdict(list)
        for adata in annotations:
            key = adata['user_type'], adata['user'], adata['format']
            a = adata['annotation']
            ts = (parse_iso8601(a['timestamp']) - EPOCH).total_seconds()
            umap[key].append((a, ts))
        for (user_type, user, fmt), annots_list in iteritems(umap):
            self._set_annotations_for_book(book_id, fmt, annots_list, user_type=user_type, user=user)

    @write_api
    def set_annotations_for_book(self, book_id, fmt, annots_list, user_type='local', user='viewer'):
        self.backend.set_annotations_for_book(book_id, fmt, annots_list, user_type, user)

    @write_api
    def merge_annotations_for_book(self, book_id, fmt, annots_list, user_type='local', user='viewer'):
        from calibre.utils.date import EPOCH
        from calibre.utils.iso8601 import parse_iso8601
        amap = self._annotations_map_for_book(book_id, fmt, user_type=user_type, user=user)
        merge_annotations(annots_list, amap)
        alist = []
        for val in itervalues(amap):
            for annot in val:
                ts = (parse_iso8601(annot['timestamp']) - EPOCH).total_seconds()
                alist.append((annot, ts))
        self._set_annotations_for_book(book_id, fmt, alist, user_type=user_type, user=user)

    @write_api
    def reindex_annotations(self):
        self.backend.reindex_annotations()


def import_library(library_key, importer, library_path, progress=None, abort=None):
    from calibre.db.backend import DB
    metadata = importer.metadata[library_key]
    total = metadata['total']
    if progress is not None:
        progress('metadata.db', 0, total)
    if abort is not None and abort.is_set():
        return
    with open(os.path.join(library_path, 'metadata.db'), 'wb') as f:
        src = importer.start_file(metadata['metadata.db'], 'metadata.db for ' + library_path)
        shutil.copyfileobj(src, f)
        src.close()
    cache = Cache(DB(library_path, load_user_formatter_functions=False))
    cache.init()
    format_data = {int(book_id):data for book_id, data in iteritems(metadata['format_data'])}
    for i, (book_id, fmt_key_map) in enumerate(iteritems(format_data)):
        if abort is not None and abort.is_set():
            return
        title = cache._field_for('title', book_id)
        if progress is not None:
            progress(title, i + 1, total)
        cache._update_path((book_id,), mark_as_dirtied=False)
        for fmt, fmtkey in iteritems(fmt_key_map):
            if fmt == '.cover':
                stream = importer.start_file(fmtkey, _('Cover for %s') % title)
                path = cache._field_for('path', book_id).replace('/', os.sep)
                cache.backend.set_cover(book_id, path, stream, no_processing=True)
            else:
                stream = importer.start_file(fmtkey, _('{0} format for {1}').format(fmt.upper(), title))
                size, fname = cache._do_add_format(book_id, fmt, stream, mtime=stream.mtime)
                cache.fields['formats'].table.update_fmt(book_id, fmt, fname, size, cache.backend)
            stream.close()
        cache.dump_metadata({book_id})
    if progress is not None:
        progress(_('Completed'), total, total)
    return cache
# }}}