mirror of
https://github.com/SickGear/SickGear.git
synced 2024-11-30 00:13:38 +00:00
ae4173e8c0
Change use specific logger functions for debug, warning, error.
2090 lines
98 KiB
Python
2090 lines
98 KiB
Python
#
|
|
# This file is part of SickGear.
|
|
#
|
|
# SickGear is free software: you can redistribute it and/or modify
|
|
# it under the terms of the GNU General Public License as published by
|
|
# the Free Software Foundation, either version 3 of the License, or
|
|
# (at your option) any later version.
|
|
#
|
|
# SickGear is distributed in the hope that it will be useful,
|
|
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
# GNU General Public License for more details.
|
|
#
|
|
# You should have received a copy of the GNU General Public License
|
|
# along with SickGear. If not, see <http://www.gnu.org/licenses/>.
|
|
|
|
import datetime
|
|
import os.path
|
|
import re
|
|
|
|
from .. import db, common, logger
|
|
from ..name_parser.parser import NameParser, InvalidNameException, InvalidShowException
|
|
import sickgear
|
|
|
|
from six import iteritems
|
|
|
|
# noinspection PyUnreachableCode
|
|
if False:
|
|
from _23 import DirEntry
|
|
|
|
MIN_DB_VERSION = 9 # oldest db version we support migrating from
|
|
MAX_DB_VERSION = 20016
|
|
TEST_BASE_VERSION = None # the base production db version, only needed for TEST db versions (>=100000)
|
|
|
|
|
|
class MainSanityCheck(db.DBSanityCheck):
|
|
def check(self):
|
|
self.fix_missing_table_indexes()
|
|
self.fix_duplicate_shows()
|
|
self.fix_duplicate_episodes()
|
|
self.fix_orphan_episodes()
|
|
self.fix_unaired_episodes()
|
|
self.fix_scene_exceptions()
|
|
self.fix_orphan_not_found_show()
|
|
self.fix_fallback_mapping()
|
|
self.fix_indexer_mapping_tvdb()
|
|
self.fix_episode_subtitles()
|
|
self.fix_genre_separator()
|
|
|
|
def fix_episode_subtitles(self):
|
|
if not self.connection.has_flag('fix_episode_subtitles'):
|
|
cleaned = False
|
|
cl = []
|
|
|
|
ep_result = self.connection.select(
|
|
'SELECT episode_id'
|
|
' FROM tv_episodes'
|
|
' WHERE subtitles LIKE "%,%"')
|
|
|
|
ep_len, cur_p = len(ep_result), 0
|
|
ep_step = ep_len / 100.0
|
|
fix_msg = 'Fixing subtitles: %s'
|
|
|
|
if ep_len:
|
|
self.connection.upgrade_log(fix_msg % ('%s%%' % 0))
|
|
|
|
for _cur_count, cur_ep in enumerate(ep_result):
|
|
if cur_p < int(_cur_count / ep_step):
|
|
cur_p = int(_cur_count / ep_step)
|
|
self.connection.upgrade_log(fix_msg % ('%s%%' % cur_p))
|
|
if not cleaned:
|
|
logger.log('Removing duplicate subtitles data in TV Episodes table, this WILL take some time')
|
|
cleaned = True
|
|
|
|
sql_result = self.connection.select(
|
|
'SELECT SUBSTR(REPLACE(subtitles, ",,", ""), -2000) AS truncated_langs'
|
|
' FROM tv_episodes'
|
|
' WHERE episode_id = ? LIMIT 1', [cur_ep['episode_id']])
|
|
|
|
for cur_result in sql_result:
|
|
raw_langs = re.sub(r',+', '', cur_result['truncated_langs'])
|
|
subt_value = ','.join(re.findall('[a-z]{2}', raw_langs))
|
|
cl.append(['UPDATE tv_episodes SET subtitles = ? WHERE episode_id = ?',
|
|
[(subt_value, '')[bool(len(raw_langs) % 2)], cur_ep['episode_id']]])
|
|
|
|
if 0 < len(cl):
|
|
self.connection.mass_action(cl)
|
|
|
|
logger.debug('Performing a vacuum on the database.')
|
|
self.connection.upgrade_log(fix_msg % 'VACUUM')
|
|
self.connection.action('VACUUM')
|
|
self.connection.upgrade_log(fix_msg % 'finished')
|
|
|
|
self.connection.set_flag('fix_episode_subtitles')
|
|
|
|
def fix_indexer_mapping_tvdb(self):
|
|
if not self.connection.has_flag('fix_indexer_mapping_tvdb'):
|
|
self.connection.action('DELETE FROM indexer_mapping WHERE mindexer = ?', [10001])
|
|
self.connection.set_flag('fix_indexer_mapping_tvdb')
|
|
|
|
def fix_duplicate_shows(self, column='indexer_id'):
|
|
|
|
# This func would break with multi tv info sources and without tvid, so added check min db version to mitigate
|
|
# Also, tv_show table had a unique index added at some time to prevent further dupes,
|
|
# therefore, this func is kept to cleanse legacy data given that it's redundant for new row insertions
|
|
if self.connection.check_db_version() < 20004:
|
|
|
|
sql_result = self.connection.select(
|
|
'SELECT show_id, %(col)s, COUNT(%(col)s) AS count FROM tv_shows GROUP BY %(col)s HAVING count > 1'
|
|
% {'col': column})
|
|
|
|
for cur_result in sql_result:
|
|
|
|
logger.debug(f'Duplicate show detected! {column}: {cur_result[column]} count: {cur_result["count"]}')
|
|
|
|
cur_dupe_results = self.connection.select(
|
|
'SELECT show_id, ' + column + ' FROM tv_shows WHERE ' + column + ' = ? LIMIT ?',
|
|
[cur_result[column], int(cur_result['count']) - 1]
|
|
)
|
|
|
|
cl = []
|
|
for cur_dupe_id in cur_dupe_results:
|
|
logger.log(f'Deleting duplicate show with {column}: {cur_dupe_id[column]}'
|
|
f' show_id: {cur_dupe_id["show_id"]}')
|
|
cl.append(['DELETE FROM tv_shows WHERE show_id = ?', [cur_dupe_id['show_id']]])
|
|
|
|
if 0 < len(cl):
|
|
self.connection.mass_action(cl)
|
|
|
|
else:
|
|
logger.log('No duplicate show, check passed')
|
|
|
|
def fix_duplicate_episodes(self):
|
|
|
|
# This func would break with multi tv info sources and without tvid, so added check min db version to mitigate
|
|
# Also, tv_show table had a unique index added at some time to prevent further dupes,
|
|
# therefore, this func is kept to cleanse legacy data given that it's redundant for new row insertions
|
|
if self.connection.check_db_version() < 20007:
|
|
|
|
sql_result = self.connection.select(
|
|
'SELECT indexer AS tv_id, showid AS prod_id, season, episode, COUNT(showid) as count'
|
|
' FROM tv_episodes'
|
|
' GROUP BY tv_id, prod_id, season, episode'
|
|
' HAVING count > 1')
|
|
|
|
for cur_result in sql_result:
|
|
|
|
logger.debug(f'Duplicate episode detected! prod_id: {cur_result["prod_id"]}'
|
|
f' season: {cur_result["season"]} episode: {cur_result["episode"]}'
|
|
f' count: {cur_result["count"]}')
|
|
|
|
cur_dupe_results = self.connection.select(
|
|
'SELECT episode_id'
|
|
' FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' AND season = ? AND episode = ?'
|
|
' ORDER BY episode_id DESC LIMIT ?',
|
|
[cur_result['tv_id'], cur_result['prod_id'],
|
|
cur_result['season'], cur_result['episode'],
|
|
int(cur_result['count']) - 1]
|
|
)
|
|
|
|
cl = []
|
|
for cur_dupe_id in cur_dupe_results:
|
|
logger.log(f'Deleting duplicate episode with episode_id: {cur_dupe_id["episode_id"]}')
|
|
cl.append(['DELETE FROM tv_episodes WHERE episode_id = ?', [cur_dupe_id["episode_id"]]])
|
|
|
|
if 0 < len(cl):
|
|
self.connection.mass_action(cl)
|
|
|
|
else:
|
|
logger.log('No duplicate episode, check passed')
|
|
|
|
def fix_orphan_episodes(self):
|
|
|
|
sql_result = self.connection.select(
|
|
'SELECT episode_id, showid, tv_shows.indexer_id'
|
|
' FROM tv_episodes'
|
|
' LEFT JOIN tv_shows ON tv_episodes.showid=tv_shows.indexer_id AND tv_episodes.indexer=tv_shows.indexer '
|
|
' WHERE tv_shows.indexer_id is NULL')
|
|
|
|
cl = []
|
|
for cur_result in sql_result:
|
|
logger.debug(f'Orphan episode detected! episode_id: {cur_result["episode_id"]}'
|
|
f' showid: {cur_result["showid"]}')
|
|
logger.log(f'Deleting orphan episode with episode_id: {cur_result["episode_id"]}')
|
|
cl.append(['DELETE FROM tv_episodes WHERE episode_id = ?', [cur_result['episode_id']]])
|
|
|
|
if 0 < len(cl):
|
|
self.connection.mass_action(cl)
|
|
|
|
else:
|
|
logger.log('No orphan episodes, check passed')
|
|
|
|
def fix_missing_table_indexes(self):
|
|
if not self.connection.select('PRAGMA index_info("idx_indexer_id")'):
|
|
logger.log('Updating TV Shows table with index idx_indexer_id')
|
|
self.connection.action('CREATE UNIQUE INDEX idx_indexer_id ON tv_shows (indexer, indexer_id);')
|
|
|
|
if not self.connection.select('PRAGMA index_info("idx_tv_episodes_showid_airdate")'):
|
|
logger.log('Updating TV Episode table with index idx_tv_episodes_showid_airdate')
|
|
self.connection.action('CREATE INDEX idx_tv_episodes_showid_airdate'
|
|
' ON tv_episodes(indexer,showid,airdate);')
|
|
|
|
if not self.connection.select('PRAGMA index_info("idx_status")'):
|
|
logger.log('Updating TV Episode table with index idx_status')
|
|
self.connection.action('CREATE INDEX idx_status ON tv_episodes (status, season, episode, airdate)')
|
|
|
|
if not self.connection.select('PRAGMA index_info("idx_sta_epi_air")'):
|
|
logger.log('Updating TV Episode table with index idx_sta_epi_air')
|
|
self.connection.action('CREATE INDEX idx_sta_epi_air ON tv_episodes (status, episode, airdate)')
|
|
|
|
if not self.connection.select('PRAGMA index_info("idx_sta_epi_sta_air")'):
|
|
logger.log('Updating TV Episode table with index idx_sta_epi_sta_air')
|
|
self.connection.action('CREATE INDEX idx_sta_epi_sta_air ON tv_episodes (season, episode, status, airdate)')
|
|
|
|
if not self.connection.has_index('tv_episodes', 'idx_tv_ep_ids'):
|
|
logger.log('Updating TV Episode table with index idx_tv_ep_ids')
|
|
self.connection.action('CREATE INDEX idx_tv_ep_ids ON tv_episodes (indexer, showid)')
|
|
|
|
if not self.connection.has_index('tv_episodes', 'idx_tv_episodes_unique'):
|
|
self.connection.action('CREATE UNIQUE INDEX idx_tv_episodes_unique ON '
|
|
'tv_episodes(indexer,showid,season,episode)')
|
|
|
|
allowtbl, blocktbl = (('allow', 'block'), ('white', 'black'))[not self.connection.has_table('blocklist')]
|
|
for t in [('%slist' % allowtbl, 'show_id'), ('%slist' % blocktbl, 'show_id'),
|
|
('history', 'showid'), ('scene_exceptions', 'indexer_id')]:
|
|
if not self.connection.has_index('%s' % t[0], 'idx_id_indexer_%s' % t[0]):
|
|
# noinspection SqlResolve
|
|
self.connection.action('CREATE INDEX idx_id_indexer_%s ON %s (indexer, %s)' % (t[0], t[0], t[1]))
|
|
|
|
def fix_unaired_episodes(self):
|
|
|
|
cur_date = datetime.date.today() + datetime.timedelta(days=1)
|
|
|
|
sql_result = self.connection.select(
|
|
'SELECT episode_id, showid FROM tv_episodes WHERE status = ? or ( airdate > ? AND status in (?,?) ) or '
|
|
'( airdate <= 1 AND status = ? )', ['', cur_date.toordinal(), common.SKIPPED, common.WANTED, common.WANTED])
|
|
|
|
cl = []
|
|
for cur_result in sql_result:
|
|
logger.debug(f'UNAIRED episode detected! episode_id: {cur_result["episode_id"]}'
|
|
f' showid: {cur_result["showid"]}')
|
|
logger.log(f'Fixing unaired episode status with episode_id: {cur_result["episode_id"]}')
|
|
cl.append(['UPDATE tv_episodes SET status = ? WHERE episode_id = ?',
|
|
[common.UNAIRED, cur_result['episode_id']]])
|
|
|
|
if 0 < len(cl):
|
|
self.connection.mass_action(cl)
|
|
|
|
else:
|
|
logger.log('No UNAIRED episodes, check passed')
|
|
|
|
def fix_scene_exceptions(self):
|
|
|
|
# noinspection SqlResolve
|
|
sql_result = self.connection.select(
|
|
'SELECT exception_id FROM scene_exceptions WHERE season = "null"')
|
|
|
|
if 0 < len(sql_result):
|
|
logger.log('Fixing invalid scene exceptions')
|
|
# noinspection SqlResolve
|
|
self.connection.action('UPDATE scene_exceptions SET season = -1 WHERE season = "null"')
|
|
|
|
def fix_orphan_not_found_show(self):
|
|
sql_result = self.connection.action(
|
|
'DELETE FROM tv_shows_not_found'
|
|
' WHERE NOT EXISTS (SELECT NULL FROM tv_shows WHERE tv_shows_not_found.indexer == tv_shows.indexer AND'
|
|
' tv_shows_not_found.indexer_id == tv_shows.indexer_id)')
|
|
if sql_result.rowcount:
|
|
logger.log('Fixed orphaned not found shows')
|
|
|
|
def fix_fallback_mapping(self):
|
|
fallback_indexer = [i for i in sickgear.TVInfoAPI().fallback_sources]
|
|
if fallback_indexer:
|
|
sql_result = self.connection.action(
|
|
'DELETE FROM indexer_mapping WHERE mindexer IN (%s) OR indexer in (%s)' %
|
|
(','.join(['?'] * len(fallback_indexer)), ','.join(['?'] * len(fallback_indexer))),
|
|
fallback_indexer + fallback_indexer)
|
|
if sql_result.rowcount:
|
|
logger.log('Fixed fallback indexer mappings')
|
|
|
|
def fix_genre_separator(self):
|
|
try:
|
|
sql_result = self.connection.select('SELECT indexer, indexer_id, genre FROM tv_shows')
|
|
c_l = []
|
|
sep_chk = re.compile('(^[|]|[|]$)')
|
|
for cur_show in sql_result:
|
|
if sep_chk.search(cur_show['genre']):
|
|
genres = '|'.join([_g for _g in (cur_show['genre'] or '').split('|') if _g])
|
|
c_l.append(['UPDATE tv_shows SET genre = ? WHERE indexer = ? AND indexer_id = ?',
|
|
[genres, cur_show['indexer'], cur_show['indexer_id']]])
|
|
if c_l:
|
|
progress_msg = 'Fixing genres separator'
|
|
logger.log(progress_msg)
|
|
sickgear.classes.loading_msg.message = progress_msg
|
|
self.connection.mass_action(c_l)
|
|
except(BaseException, Exception):
|
|
import traceback
|
|
logger.error('Error fixing genres separator')
|
|
logger.debug('%s' % traceback.format_exc())
|
|
|
|
|
|
class InitialSchema(db.SchemaUpgrade):
|
|
# ======================
|
|
# = Main DB Migrations =
|
|
# ======================
|
|
# Add new migrations at the bottom of the list; subclass the previous migration.
|
|
# 0 -> 20009
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
if not self.has_table('tv_shows') and not self.has_table('db_version'):
|
|
queries = [
|
|
# anime allow and block list
|
|
'CREATE TABLE allowlist (show_id INTEGER, range TEXT, keyword TEXT, indexer NUMERIC)',
|
|
'CREATE TABLE blocklist (show_id INTEGER, range TEXT, keyword TEXT, indexer NUMERIC)',
|
|
# db_version
|
|
'CREATE TABLE db_version (db_version INTEGER)',
|
|
'INSERT INTO db_version (db_version) VALUES (20008)',
|
|
# flags
|
|
'CREATE TABLE flags (flag PRIMARY KEY NOT NULL)',
|
|
# history
|
|
'CREATE TABLE history (action NUMERIC, date NUMERIC, showid NUMERIC, season NUMERIC, episode NUMERIC,'
|
|
' quality NUMERIC, resource TEXT, provider TEXT, version NUMERIC, hide NUMERIC DEFAULT 0)',
|
|
# imdb_info
|
|
'CREATE TABLE imdb_info (indexer_id INTEGER PRIMARY KEY, imdb_id TEXT, title TEXT, year NUMERIC,'
|
|
' akas TEXT, runtimes NUMERIC, genres TEXT, countries TEXT, country_codes TEXT, certificates TEXT,'
|
|
' rating TEXT, votes INTEGER, last_update NUMERIC)',
|
|
# indexer_mapping
|
|
'CREATE TABLE indexer_mapping (indexer_id INTEGER, indexer NUMERIC, mindexer_id INTEGER NOT NULL,'
|
|
' mindexer NUMERIC, date NUMERIC NOT NULL DEFAULT 0, status INTEGER NOT NULL DEFAULT 0,'
|
|
' PRIMARY KEY (indexer_id, indexer, mindexer))',
|
|
'CREATE INDEX idx_mapping ON indexer_mapping (indexer_id, indexer)',
|
|
# info
|
|
'CREATE TABLE info (last_backlog NUMERIC, last_indexer NUMERIC, last_proper_search NUMERIC,'
|
|
' last_run_backlog NUMERIC NOT NULL DEFAULT 1)',
|
|
# scene_exceptions
|
|
'CREATE TABLE scene_exceptions (exception_id INTEGER PRIMARY KEY, indexer_id INTEGER KEY,'
|
|
' show_name TEXT, season NUMERIC, custom NUMERIC)',
|
|
# scene_exceptions_refresh
|
|
'CREATE TABLE scene_exceptions_refresh (list TEXT PRIMARY KEY, last_refreshed INTEGER)',
|
|
# scene_numbering
|
|
'CREATE TABLE scene_numbering (indexer TEXT, indexer_id INTEGER, season INTEGER, episode INTEGER,'
|
|
' scene_season INTEGER, scene_episode INTEGER, absolute_number NUMERIC, scene_absolute_number NUMERIC,'
|
|
' PRIMARY KEY (indexer_id, season, episode))',
|
|
# tv_episodes
|
|
'CREATE TABLE tv_episodes (episode_id INTEGER PRIMARY KEY, showid NUMERIC, indexerid NUMERIC,'
|
|
' indexer NUMERIC, name TEXT, season NUMERIC, episode NUMERIC, description TEXT, airdate NUMERIC,'
|
|
' hasnfo NUMERIC, hastbn NUMERIC, status NUMERIC, location TEXT, file_size NUMERIC, release_name TEXT,'
|
|
' subtitles TEXT, subtitles_searchcount NUMERIC, subtitles_lastsearch TIMESTAMP, is_proper NUMERIC,'
|
|
' scene_season NUMERIC, scene_episode NUMERIC, absolute_number NUMERIC, scene_absolute_number NUMERIC,'
|
|
' version NUMERIC, release_group TEXT)',
|
|
'CREATE INDEX idx_showid ON tv_episodes (showid)',
|
|
'CREATE INDEX idx_tv_episodes_showid_airdate ON tv_episodes (showid,airdate)',
|
|
'CREATE INDEX idx_sta_epi_air ON tv_episodes (status,episode, airdate)',
|
|
'CREATE INDEX idx_sta_epi_sta_air ON tv_episodes (season,episode, status, airdate)',
|
|
'CREATE INDEX idx_status ON tv_episodes (status,season,episode,airdate)',
|
|
# tv_episodes_watched
|
|
'CREATE TABLE tv_episodes_watched (tvep_id NUMERIC NOT NULL, clientep_id TEXT, label TEXT,'
|
|
' played NUMERIC NOT NULL DEFAULT 0, date_watched NUMERIC NOT NULL, date_added NUMERIC,'
|
|
' status NUMERIC, location TEXT, file_size NUMERIC, hide INT NOT NULL DEFAULT 0)',
|
|
# tv_shows
|
|
'CREATE TABLE tv_shows (show_id INTEGER PRIMARY KEY, indexer_id NUMERIC, indexer NUMERIC,'
|
|
' show_name TEXT, location TEXT, network TEXT, genre TEXT, classification TEXT, runtime NUMERIC,'
|
|
' quality NUMERIC, airs TEXT, status TEXT, flatten_folders NUMERIC, paused NUMERIC, startyear NUMERIC,'
|
|
' air_by_date NUMERIC, lang TEXT, subtitles NUMERIC, notify_list TEXT, imdb_id TEXT,'
|
|
' last_update_indexer NUMERIC, dvdorder NUMERIC, archive_firstmatch NUMERIC, rls_require_words TEXT,'
|
|
' rls_ignore_words TEXT, sports NUMERIC, anime NUMERIC, scene NUMERIC, overview TEXT, tag TEXT,'
|
|
' prune INT DEFAULT 0)',
|
|
'CREATE UNIQUE INDEX idx_indexer_id ON tv_shows (indexer_id)',
|
|
# tv_shows_not_found
|
|
'CREATE TABLE tv_shows_not_found (indexer NUMERIC NOT NULL, indexer_id NUMERIC NOT NULL,'
|
|
' fail_count NUMERIC NOT NULL DEFAULT 0, last_check NUMERIC NOT NULL, last_success NUMERIC,'
|
|
' PRIMARY KEY (indexer_id, indexer))',
|
|
# webdl_types
|
|
'CREATE TABLE webdl_types (dname TEXT NOT NULL, regex TEXT NOT NULL)',
|
|
# xem_refresh
|
|
'CREATE TABLE xem_refresh (indexer TEXT, indexer_id INTEGER PRIMARY KEY, last_refreshed INTEGER)',
|
|
]
|
|
for query in queries:
|
|
self.connection.action(query)
|
|
|
|
else:
|
|
cur_db_version = self.call_check_db_version()
|
|
|
|
if cur_db_version < MIN_DB_VERSION:
|
|
logger.log_error_and_exit(
|
|
f'Your database version ({cur_db_version}) is too old to migrate from'
|
|
f' what this version of SickGear supports ({MIN_DB_VERSION}).\n'
|
|
+ 'Upgrade using a previous version (tag) build 496 to build 501 of SickGear'
|
|
' first or remove database file to begin fresh.')
|
|
|
|
if cur_db_version > MAX_DB_VERSION:
|
|
logger.log_error_and_exit(
|
|
f'Your database version ({cur_db_version}) has been incremented past'
|
|
f' what this version of SickGear supports ({MAX_DB_VERSION}).\n'
|
|
+ 'If you have used other forks of SickGear,'
|
|
' your database may be unusable due to their modifications.')
|
|
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 9 -> 10
|
|
class AddSizeAndSceneNameFields(db.SchemaUpgrade):
|
|
def execute(self):
|
|
"""
|
|
This func is only for 9->10 where older db columns exist,
|
|
those columns have since changed
|
|
"""
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
if not self.has_column('tv_episodes', 'file_size'):
|
|
self.add_column('tv_episodes', 'file_size')
|
|
|
|
if not self.has_column('tv_episodes', 'release_name'):
|
|
self.add_column('tv_episodes', 'release_name', 'TEXT', '')
|
|
|
|
sql_result = self.connection.select('SELECT episode_id, location, file_size FROM tv_episodes')
|
|
|
|
self.upgrade_log('Adding file size to all episodes in DB, please be patient')
|
|
for cur_result in sql_result:
|
|
if not cur_result['location']:
|
|
continue
|
|
|
|
# if there is no size yet then populate it for us
|
|
if (not cur_result['file_size'] or not int(cur_result['file_size'])) \
|
|
and os.path.isfile(cur_result['location']):
|
|
cur_size = os.path.getsize(cur_result['location'])
|
|
self.connection.action('UPDATE tv_episodes SET file_size = ? WHERE episode_id = ?',
|
|
[cur_size, int(cur_result['episode_id'])])
|
|
|
|
# check each snatch to see if we can use it to get a release name from
|
|
# noinspection SqlRedundantOrderingDirection
|
|
history_sql_result = self.connection.select('SELECT * FROM history WHERE provider != -1 ORDER BY date ASC')
|
|
|
|
self.upgrade_log('Adding release name to all episodes still in history')
|
|
for cur_result in history_sql_result:
|
|
# find the associated download, if there isn't one then ignore it
|
|
# noinspection SqlResolve
|
|
download_sql_result = self.connection.select(
|
|
'SELECT resource'
|
|
' FROM history'
|
|
' WHERE provider = -1 AND showid = ? AND season = ? AND episode = ? AND date > ?',
|
|
[cur_result['showid'], cur_result['season'], cur_result['episode'], cur_result['date']])
|
|
if not download_sql_result:
|
|
self.upgrade_log(f'Found a snatch in the history for {cur_result["resource"]}'
|
|
f' but couldn\'t find the associated download, skipping it', logger.DEBUG)
|
|
continue
|
|
|
|
nzb_name = cur_result['resource']
|
|
file_name = os.path.basename(download_sql_result[0]['resource'])
|
|
|
|
# take the extension off the filename, it's not needed
|
|
if '.' in file_name:
|
|
file_name = file_name.rpartition('.')[0]
|
|
|
|
# find the associated episode on disk
|
|
# noinspection SqlResolve
|
|
sql_result = self.connection.select(
|
|
'SELECT episode_id, status'
|
|
' FROM tv_episodes'
|
|
' WHERE showid = ? AND season = ? AND episode = ? AND location != ""',
|
|
[cur_result['showid'], cur_result['season'], cur_result['episode']])
|
|
if not sql_result:
|
|
logger.debug(f'The episode {nzb_name} was found in history but doesn\'t exist on disk anymore,'
|
|
f' skipping')
|
|
continue
|
|
|
|
# get the status/quality of the existing ep and make sure it's what we expect
|
|
ep_status, ep_quality = common.Quality.split_composite_status(int(sql_result[0]['status']))
|
|
if ep_status != common.DOWNLOADED:
|
|
continue
|
|
|
|
if ep_quality != int(cur_result['quality']):
|
|
continue
|
|
|
|
# make sure this is actually a real release name and not a season pack or something
|
|
for cur_name in (nzb_name, file_name):
|
|
logger.debug(f'Checking if {cur_name} is actually a good release name')
|
|
try:
|
|
np = NameParser(False)
|
|
parse_result = np.parse(cur_name)
|
|
except (InvalidNameException, InvalidShowException):
|
|
continue
|
|
|
|
if parse_result.series_name and parse_result.season_number is not None\
|
|
and parse_result.episode_numbers and parse_result.release_group:
|
|
# if all is well by this point we'll just put the release name into the database
|
|
self.connection.action('UPDATE tv_episodes SET release_name = ? WHERE episode_id = ?',
|
|
[cur_name, sql_result[0]['episode_id']])
|
|
break
|
|
|
|
# check each snatch to see if we can use it to get a release name from
|
|
# noinspection SqlResolve
|
|
empty_sql_result = self.connection.select('SELECT episode_id, location'
|
|
' FROM tv_episodes'
|
|
' WHERE release_name = ""')
|
|
|
|
self.upgrade_log('Adding release name to all episodes with obvious scene filenames')
|
|
for cur_result in empty_sql_result:
|
|
|
|
ep_file_name = os.path.basename(cur_result['location'])
|
|
ep_file_name = os.path.splitext(ep_file_name)[0]
|
|
|
|
# only want to find real scene names here so anything with a space in it is out
|
|
if ' ' in ep_file_name:
|
|
continue
|
|
|
|
try:
|
|
np = NameParser(False)
|
|
parse_result = np.parse(ep_file_name)
|
|
except (InvalidNameException, InvalidShowException):
|
|
continue
|
|
|
|
if not parse_result.release_group:
|
|
continue
|
|
|
|
logger.debug(f'Name {ep_file_name} gave release group of {parse_result.release_group}, seems valid')
|
|
self.connection.action('UPDATE tv_episodes SET release_name = ? WHERE episode_id = ?',
|
|
[ep_file_name, cur_result['episode_id']])
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 10 -> 11
|
|
class RenameSeasonFolders(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
# rename the column
|
|
self.connection.action('ALTER TABLE tv_shows RENAME TO tmp_tv_shows')
|
|
self.connection.action(
|
|
'CREATE TABLE tv_shows (show_id INTEGER PRIMARY KEY, location TEXT, show_name TEXT, tvdb_id NUMERIC,'
|
|
' network TEXT, genre TEXT, runtime NUMERIC, quality NUMERIC, airs TEXT, status TEXT,'
|
|
' flatten_folders NUMERIC, paused NUMERIC, startyear NUMERIC, tvr_id NUMERIC, tvr_name TEXT,'
|
|
' air_by_date NUMERIC, lang TEXT)')
|
|
# noinspection SqlResolve
|
|
sql = 'INSERT INTO tv_shows(show_id, location, show_name, tvdb_id, network, genre, runtime,' \
|
|
' quality, airs, status, flatten_folders, paused, startyear, tvr_id, tvr_name, air_by_date, lang)' \
|
|
' SELECT show_id, location, show_name, tvdb_id, network, genre, runtime, quality, airs, status,' \
|
|
' seasonfolders, paused, startyear, tvr_id, tvr_name, air_by_date, lang FROM tmp_tv_shows'
|
|
self.connection.action(sql)
|
|
|
|
# flip the values to be opposite of what they were before
|
|
self.connection.action('UPDATE tv_shows SET flatten_folders = 2 WHERE flatten_folders = 1')
|
|
self.connection.action('UPDATE tv_shows SET flatten_folders = 1 WHERE flatten_folders = 0')
|
|
self.connection.action('UPDATE tv_shows SET flatten_folders = 0 WHERE flatten_folders = 2')
|
|
# noinspection SqlResolve
|
|
self.connection.action('DROP TABLE tmp_tv_shows')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 11 -> 12
|
|
class Add1080pAndRawHDQualities(db.SchemaUpgrade):
|
|
"""
|
|
Add support for 1080p related qualities along with RawHD
|
|
|
|
Quick overview of what the upgrade needs to do:
|
|
|
|
quality | old | new
|
|
--------------------------
|
|
hdwebdl | 1<<3 | 1<<5
|
|
hdbluray | 1<<4 | 1<<7
|
|
fullhdbluray | 1<<5 | 1<<8
|
|
--------------------------
|
|
rawhdtv | | 1<<3
|
|
fullhdtv | | 1<<4
|
|
fullhdwebdl | | 1<<6
|
|
"""
|
|
|
|
def _update_status(self, old_status):
|
|
(status, quality) = common.Quality.split_composite_status(old_status)
|
|
return common.Quality.composite_status(status, self._update_quality(quality))
|
|
|
|
@staticmethod
|
|
def _update_quality(old_quality):
|
|
"""
|
|
Update bitwise flags to reflect new quality values
|
|
|
|
Check flag bits (clear old then set their new locations) starting
|
|
with the highest bits so we dont overwrite data we need later on
|
|
"""
|
|
|
|
result = old_quality
|
|
# move fullhdbluray from 1<<5 to 1<<8 if set
|
|
if result & (1 << 5):
|
|
result &= ~(1 << 5)
|
|
result |= 1 << 8
|
|
# move hdbluray from 1<<4 to 1<<7 if set
|
|
if result & (1 << 4):
|
|
result &= ~(1 << 4)
|
|
result |= 1 << 7
|
|
# move hdwebdl from 1<<3 to 1<<5 if set
|
|
if result & (1 << 3):
|
|
result &= ~(1 << 3)
|
|
result |= 1 << 5
|
|
|
|
return result
|
|
|
|
def _update_composite_qualities(self, status):
|
|
"""
|
|
Unpack, Update, Return new quality values
|
|
|
|
Unpack the composite archive/initial values.
|
|
Update either qualities if needed.
|
|
Then return the new compsite quality value.
|
|
"""
|
|
|
|
best = (status & (0xffff << 16)) >> 16
|
|
initial = status & 0xffff
|
|
|
|
best = self._update_quality(best)
|
|
initial = self._update_quality(initial)
|
|
|
|
result = ((best << 16) | initial)
|
|
return result
|
|
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
# update the default quality so we dont grab the wrong qualities after migration
|
|
sickgear.QUALITY_DEFAULT = self._update_composite_qualities(sickgear.QUALITY_DEFAULT)
|
|
sickgear.save_config()
|
|
|
|
# upgrade previous HD to HD720p -- shift previous qualities to new placevalues
|
|
old_hd = common.Quality.combine_qualities(
|
|
[common.Quality.HDTV, common.Quality.HDWEBDL >> 2, common.Quality.HDBLURAY >> 3], [])
|
|
new_hd = common.Quality.combine_qualities([common.Quality.HDTV, common.Quality.HDWEBDL,
|
|
common.Quality.HDBLURAY], [])
|
|
|
|
# update ANY -- shift existing qualities and add new 1080p qualities,
|
|
# note that rawHD was not added to the ANY template
|
|
old_any = common.Quality.combine_qualities(
|
|
[common.Quality.SDTV, common.Quality.SDDVD, common.Quality.HDTV, common.Quality.HDWEBDL >> 2,
|
|
common.Quality.HDBLURAY >> 3, common.Quality.UNKNOWN], [])
|
|
new_any = common.Quality.combine_qualities(
|
|
[common.Quality.SDTV, common.Quality.SDDVD, common.Quality.HDTV, common.Quality.FULLHDTV,
|
|
common.Quality.HDWEBDL, common.Quality.FULLHDWEBDL, common.Quality.HDBLURAY, common.Quality.FULLHDBLURAY,
|
|
common.Quality.UNKNOWN], [])
|
|
|
|
# update qualities (including templates)
|
|
self.upgrade_log('[1/4] Updating pre-defined templates and the quality for each show...')
|
|
cl = []
|
|
shows = self.connection.select('SELECT * FROM tv_shows')
|
|
for cur_show in shows:
|
|
if old_hd == cur_show['quality']:
|
|
new_quality = new_hd
|
|
elif old_any == cur_show['quality']:
|
|
new_quality = new_any
|
|
else:
|
|
new_quality = self._update_composite_qualities(cur_show['quality'])
|
|
cl.append(['UPDATE tv_shows SET quality = ? WHERE show_id = ?', [new_quality, cur_show['show_id']]])
|
|
self.connection.mass_action(cl)
|
|
|
|
# update status that are are within the old hdwebdl
|
|
# (1<<3 which is 8) and better -- exclude unknown (1<<15 which is 32768)
|
|
self.upgrade_log('[2/4] Updating the status for the episodes within each show...')
|
|
cl = []
|
|
sql_result = self.connection.select('SELECT * FROM tv_episodes WHERE status < 3276800 AND status >= 800')
|
|
for cur_result in sql_result:
|
|
cl.append(['UPDATE tv_episodes SET status = ? WHERE episode_id = ?',
|
|
[self._update_status(cur_result['status']), cur_result['episode_id']]])
|
|
self.connection.mass_action(cl)
|
|
|
|
# make two seperate passes through the history since snatched and downloaded (action & quality)
|
|
# may not always coordinate together
|
|
|
|
# update previous history so it shows the correct action
|
|
self.upgrade_log('[3/4] Updating history to reflect the correct action...')
|
|
cl = []
|
|
# noinspection SqlResolve
|
|
history_action = self.connection.select('SELECT * FROM history WHERE action < 3276800 AND action >= 800')
|
|
for cur_entry in history_action:
|
|
cl.append(['UPDATE history SET action = ? WHERE showid = ? AND date = ?',
|
|
[self._update_status(cur_entry['action']), cur_entry['showid'], cur_entry['date']]])
|
|
self.connection.mass_action(cl)
|
|
|
|
# update previous history so it shows the correct quality
|
|
self.upgrade_log('[4/4] Updating history to reflect the correct quality...')
|
|
cl = []
|
|
# noinspection SqlResolve
|
|
history_quality = self.connection.select('SELECT * FROM history WHERE quality < 32768 AND quality >= 8')
|
|
for cur_entry in history_quality:
|
|
cl.append(['UPDATE history SET quality = ? WHERE showid = ? AND date = ?',
|
|
[self._update_quality(cur_entry['quality']), cur_entry['showid'], cur_entry['date']]])
|
|
self.connection.mass_action(cl)
|
|
|
|
self.inc_db_version()
|
|
|
|
# cleanup and reduce db if any previous data was removed
|
|
self.upgrade_log('Performing a vacuum on the database.', logger.DEBUG)
|
|
self.connection.action('VACUUM')
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 12 -> 13
|
|
class AddShowidTvdbidIndex(db.SchemaUpgrade):
|
|
# Adding index on tvdb_id (tv_shows) and showid (tv_episodes) to speed up searches/queries
|
|
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Checking for duplicate shows before adding unique index.')
|
|
MainSanityCheck(self.connection).fix_duplicate_shows('tvdb_id')
|
|
|
|
self.upgrade_log('Adding index on tvdb_id (tv_shows) and showid (tv_episodes) to speed up searches/queries.')
|
|
if not self.has_table('idx_showid'):
|
|
self.connection.action('CREATE INDEX idx_showid ON tv_episodes (showid);')
|
|
if not self.has_table('idx_tvdb_id'):
|
|
# noinspection SqlResolve
|
|
self.connection.action('CREATE UNIQUE INDEX idx_tvdb_id ON tv_shows (tvdb_id);')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 13 -> 14
|
|
class AddLastUpdateTVDB(db.SchemaUpgrade):
|
|
# Adding column last_update_tvdb to tv_shows for controlling nightly updates
|
|
def execute(self):
|
|
|
|
if not self.has_column('tv_shows', 'last_update_tvdb'):
|
|
self.upgrade_log('Adding column last_update_tvdb to tv_shows')
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.add_column('tv_shows', 'last_update_tvdb', default=1)
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 14 -> 15
|
|
class AddDBIncreaseTo15(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log(f'Bumping database version to v{self.call_check_db_version()}')
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 15 -> 16
|
|
class AddIMDbInfo(db.SchemaUpgrade):
|
|
def execute(self):
|
|
|
|
db_backed_up = False
|
|
if not self.has_table('imdb_info'):
|
|
self.upgrade_log('Creating IMDb table imdb_info')
|
|
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
db_backed_up = True
|
|
self.connection.action(
|
|
'CREATE TABLE imdb_info (tvdb_id INTEGER PRIMARY KEY, imdb_id TEXT, title TEXT, year NUMERIC,'
|
|
' akas TEXT, runtimes NUMERIC, genres TEXT, countries TEXT, country_codes TEXT, certificates TEXT,'
|
|
' rating TEXT, votes INTEGER, last_update NUMERIC)')
|
|
|
|
if not self.has_column('tv_shows', 'imdb_id'):
|
|
self.upgrade_log('Adding IMDb column imdb_id to tv_shows')
|
|
|
|
if not db_backed_up:
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.add_column('tv_shows', 'imdb_id')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 16 -> 17
|
|
class AddProperNamingSupport(db.SchemaUpgrade):
|
|
def execute(self):
|
|
|
|
if not self.has_column('tv_shows', 'imdb_id')\
|
|
and self.has_column('tv_shows', 'rls_require_words')\
|
|
and self.has_column('tv_shows', 'rls_ignore_words'):
|
|
return self.set_db_version(5816)
|
|
|
|
if not self.has_column('tv_episodes', 'is_proper'):
|
|
self.upgrade_log('Adding column is_proper to tv_episodes')
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.add_column('tv_episodes', 'is_proper')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 17 -> 18
|
|
class AddEmailSubscriptionTable(db.SchemaUpgrade):
|
|
def execute(self):
|
|
|
|
if not self.has_column('tv_episodes', 'is_proper')\
|
|
and self.has_column('tv_shows', 'rls_require_words')\
|
|
and self.has_column('tv_shows', 'rls_ignore_words')\
|
|
and self.has_column('tv_shows', 'skip_notices'):
|
|
return self.set_db_version(5817)
|
|
|
|
if not self.has_column('tv_shows', 'notify_list'):
|
|
self.upgrade_log('Adding column notify_list to tv_shows')
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.add_column('tv_shows', 'notify_list', 'TEXT', None)
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 18 -> 19
|
|
class AddProperSearch(db.SchemaUpgrade):
|
|
def execute(self):
|
|
if not self.has_column('tv_episodes', 'is_proper'):
|
|
return self.set_db_version(12)
|
|
|
|
if not self.has_column('tv_shows', 'notify_list')\
|
|
and self.has_column('tv_shows', 'rls_require_words')\
|
|
and self.has_column('tv_shows', 'rls_ignore_words')\
|
|
and self.has_column('tv_shows', 'skip_notices')\
|
|
and self.has_column('history', 'source'):
|
|
return self.set_db_version(5818)
|
|
|
|
if not self.has_column('info', 'last_proper_search'):
|
|
self.upgrade_log('Adding column last_proper_search to info')
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.add_column('info', 'last_proper_search', default=1)
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 19 -> 20
|
|
class AddDvdOrderOption(db.SchemaUpgrade):
|
|
def execute(self):
|
|
if not self.has_column('tv_shows', 'dvdorder'):
|
|
self.upgrade_log('Adding column dvdorder to tv_shows')
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.add_column('tv_shows', 'dvdorder', 'NUMERIC', '0')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 20 -> 21
|
|
class AddSubtitlesSupport(db.SchemaUpgrade):
|
|
def execute(self):
|
|
if not self.has_column('tv_shows', 'subtitles'):
|
|
self.upgrade_log('Adding subtitles to tv_shows and tv_episodes')
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.add_column('tv_shows', 'subtitles')
|
|
self.add_column('tv_episodes', 'subtitles', 'TEXT', '')
|
|
self.add_column('tv_episodes', 'subtitles_searchcount')
|
|
self.add_column('tv_episodes', 'subtitles_lastsearch', 'TIMESTAMP', str(datetime.datetime.min))
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 21 -> 22
|
|
class ConvertTVShowsToIndexerScheme(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Converting TV Shows table to Indexer Scheme...')
|
|
|
|
if self.has_table('tmp_tv_shows'):
|
|
self.upgrade_log('Removing temp tv show tables left behind from previous updates...')
|
|
# noinspection SqlResolve
|
|
self.connection.action('DROP TABLE tmp_tv_shows')
|
|
|
|
self.connection.action('ALTER TABLE tv_shows RENAME TO tmp_tv_shows')
|
|
self.connection.action(
|
|
'CREATE TABLE tv_shows (show_id INTEGER PRIMARY KEY, indexer_id NUMERIC, indexer NUMERIC, show_name TEXT,'
|
|
' location TEXT, network TEXT, genre TEXT, classification TEXT, runtime NUMERIC, quality NUMERIC,'
|
|
' airs TEXT, status TEXT, flatten_folders NUMERIC, paused NUMERIC, startyear NUMERIC, air_by_date NUMERIC,'
|
|
' lang TEXT, subtitles NUMERIC, notify_list TEXT, imdb_id TEXT,'
|
|
' last_update_indexer NUMERIC, dvdorder NUMERIC)')
|
|
# noinspection SqlResolve
|
|
self.connection.action(
|
|
'INSERT INTO tv_shows(show_id, indexer_id, show_name, location, network, genre, runtime, quality, airs,'
|
|
' status, flatten_folders, paused, startyear, air_by_date, lang, subtitles, notify_list, imdb_id,'
|
|
' last_update_indexer, dvdorder)'
|
|
' SELECT show_id, tvdb_id, show_name, location, network, genre, runtime,'
|
|
' quality, airs, status, flatten_folders, paused, startyear, air_by_date, lang, subtitles, notify_list,'
|
|
' imdb_id, last_update_tvdb, dvdorder FROM tmp_tv_shows')
|
|
# noinspection SqlResolve
|
|
self.connection.action('DROP TABLE tmp_tv_shows')
|
|
|
|
self.connection.action('CREATE UNIQUE INDEX idx_indexer_id ON tv_shows (indexer_id);')
|
|
|
|
# noinspection SqlResolve,SqlConstantCondition
|
|
self.connection.action('UPDATE tv_shows SET classification = "Scripted" WHERE 1=1')
|
|
# noinspection SqlConstantCondition
|
|
self.connection.action('UPDATE tv_shows SET indexer = 1 WHERE 1=1')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 22 -> 23
|
|
class ConvertTVEpisodesToIndexerScheme(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Converting TV Episodes table to Indexer Scheme...')
|
|
|
|
if self.has_table('tmp_tv_episodes'):
|
|
self.upgrade_log('Removing temp tv episode tables left behind from previous updates...')
|
|
# noinspection SqlResolve
|
|
self.connection.action('DROP TABLE tmp_tv_episodes')
|
|
|
|
self.connection.action('ALTER TABLE tv_episodes RENAME TO tmp_tv_episodes')
|
|
self.connection.action(
|
|
'CREATE TABLE tv_episodes (episode_id INTEGER PRIMARY KEY, showid NUMERIC, indexerid NUMERIC,'
|
|
' indexer NUMERIC, name TEXT, season NUMERIC, episode NUMERIC, description TEXT, airdate NUMERIC,'
|
|
' hasnfo NUMERIC, hastbn NUMERIC, status NUMERIC, location TEXT, file_size NUMERIC, release_name TEXT,'
|
|
' subtitles TEXT, subtitles_searchcount NUMERIC, subtitles_lastsearch TIMESTAMP, is_proper NUMERIC)')
|
|
# noinspection SqlResolve
|
|
self.connection.action(
|
|
'INSERT INTO tv_episodes(episode_id, showid, indexerid, name, season, episode, description, airdate,'
|
|
' hasnfo, hastbn, status, location, file_size, release_name, subtitles, subtitles_searchcount,'
|
|
' subtitles_lastsearch, is_proper) SELECT episode_id, showid, tvdbid, name, season, episode, description,'
|
|
' airdate, hasnfo, hastbn, status, location, file_size, release_name, subtitles, subtitles_searchcount,'
|
|
' subtitles_lastsearch, is_proper FROM tmp_tv_episodes')
|
|
# noinspection SqlResolve
|
|
self.connection.action('DROP TABLE tmp_tv_episodes')
|
|
|
|
self.connection.action('CREATE INDEX idx_tv_episodes_showid_airdate ON tv_episodes(showid,airdate);')
|
|
self.connection.action('CREATE INDEX idx_showid ON tv_episodes (showid);')
|
|
self.connection.action('CREATE INDEX idx_status ON tv_episodes (status,season,episode,airdate)')
|
|
self.connection.action('CREATE INDEX idx_sta_epi_air ON tv_episodes (status,episode, airdate)')
|
|
self.connection.action('CREATE INDEX idx_sta_epi_sta_air ON tv_episodes (season,episode, status, airdate)')
|
|
|
|
# noinspection SqlConstantCondition
|
|
self.connection.action('UPDATE tv_episodes SET indexer = 1 WHERE 1=1')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 23 -> 24
|
|
class ConvertIMDBInfoToIndexerScheme(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Converting IMDb Info table to Indexer Scheme...')
|
|
|
|
if self.has_table('tmp_imdb_info'):
|
|
self.upgrade_log('Removing temp imdb info tables left behind from previous updates...')
|
|
# noinspection SqlResolve
|
|
self.connection.action('DROP TABLE tmp_imdb_info')
|
|
|
|
self.connection.action('ALTER TABLE imdb_info RENAME TO tmp_imdb_info')
|
|
self.connection.action(
|
|
'CREATE TABLE imdb_info (indexer_id INTEGER PRIMARY KEY, imdb_id TEXT, title TEXT, year NUMERIC, akas TEXT,'
|
|
' runtimes NUMERIC, genres TEXT, countries TEXT, country_codes TEXT, certificates TEXT, rating TEXT,'
|
|
' votes INTEGER, last_update NUMERIC)')
|
|
# noinspection SqlResolve
|
|
self.connection.action(
|
|
'INSERT INTO imdb_info(indexer_id, imdb_id, title, year, akas, runtimes, genres, countries, country_codes,'
|
|
' certificates, rating, votes, last_update) SELECT tvdb_id, imdb_id, title, year, akas, runtimes, genres,'
|
|
' countries, country_codes, certificates, rating, votes, last_update FROM tmp_imdb_info')
|
|
# noinspection SqlResolve
|
|
self.connection.action('DROP TABLE tmp_imdb_info')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 24 -> 25
|
|
class ConvertInfoToIndexerScheme(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Converting Info table to Indexer Scheme...')
|
|
|
|
if self.has_table('tmp_info'):
|
|
self.upgrade_log('Removing temp info tables left behind from previous updates...')
|
|
# noinspection SqlResolve
|
|
self.connection.action('DROP TABLE tmp_info')
|
|
|
|
self.connection.action('ALTER TABLE info RENAME TO tmp_info')
|
|
self.connection.action(
|
|
'CREATE TABLE info (last_backlog NUMERIC, last_indexer NUMERIC, last_proper_search NUMERIC)')
|
|
# noinspection SqlResolve
|
|
self.connection.action(
|
|
'INSERT INTO info(last_backlog, last_indexer, last_proper_search)'
|
|
' SELECT last_backlog, last_tvdb, last_proper_search FROM tmp_info')
|
|
# noinspection SqlResolve
|
|
self.connection.action('DROP TABLE tmp_info')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 25 -> 26
|
|
class AddArchiveFirstMatchOption(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
if not self.has_column('tv_shows', 'archive_firstmatch'):
|
|
self.upgrade_log('Adding column archive_firstmatch to tv_shows')
|
|
self.add_column('tv_shows', 'archive_firstmatch', 'NUMERIC', '0')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 26 -> 27
|
|
class AddSceneNumbering(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
if self.has_table('scene_numbering'):
|
|
self.connection.action('DROP TABLE scene_numbering')
|
|
|
|
self.upgrade_log('Upgrading table scene_numbering ...')
|
|
self.connection.action(
|
|
'CREATE TABLE scene_numbering (indexer TEXT, indexer_id INTEGER, season INTEGER, episode INTEGER,'
|
|
' scene_season INTEGER, scene_episode INTEGER,'
|
|
' PRIMARY KEY (indexer_id,season,episode))')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 27 -> 28
|
|
class ConvertIndexerToInteger(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
cl = []
|
|
self.upgrade_log('Converting Indexer to Integer ...')
|
|
cl.append(['UPDATE tv_shows SET indexer = ? WHERE LOWER(indexer) = ?', ['1', 'tvdb']])
|
|
cl.append(['UPDATE tv_shows SET indexer = ? WHERE LOWER(indexer) = ?', ['2', 'tvrage']])
|
|
cl.append(['UPDATE tv_episodes SET indexer = ? WHERE LOWER(indexer) = ?', ['1', 'tvdb']])
|
|
cl.append(['UPDATE tv_episodes SET indexer = ? WHERE LOWER(indexer) = ?', ['2', 'tvrage']])
|
|
cl.append(['UPDATE scene_numbering SET indexer = ? WHERE LOWER(indexer) = ?', ['1', 'tvdb']])
|
|
cl.append(['UPDATE scene_numbering SET indexer = ? WHERE LOWER(indexer) = ?', ['2', 'tvrage']])
|
|
|
|
self.connection.mass_action(cl)
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 28 -> 29
|
|
class AddRequireAndIgnoreWords(db.SchemaUpgrade):
|
|
# Adding column rls_require_words and rls_ignore_words to tv_shows
|
|
def execute(self):
|
|
if self.has_column('tv_shows', 'rls_require_words') and self.has_column('tv_shows', 'rls_ignore_words'):
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
db_backed_up = False
|
|
if not self.has_column('tv_shows', 'rls_require_words'):
|
|
self.upgrade_log('Adding column rls_require_words to tv_shows')
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
db_backed_up = True
|
|
self.add_column('tv_shows', 'rls_require_words', 'TEXT', '')
|
|
|
|
if not self.has_column('tv_shows', 'rls_ignore_words'):
|
|
self.upgrade_log('Adding column rls_ignore_words to tv_shows')
|
|
if not db_backed_up:
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.add_column('tv_shows', 'rls_ignore_words', 'TEXT', '')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 29 -> 30
|
|
class AddSportsOption(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db_backed_up = False
|
|
if not self.has_column('tv_shows', 'sports'):
|
|
self.upgrade_log('Adding column sports to tv_shows')
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
db_backed_up = True
|
|
self.add_column('tv_shows', 'sports', 'NUMERIC', '0')
|
|
|
|
if self.has_column('tv_shows', 'air_by_date') and self.has_column('tv_shows', 'sports'):
|
|
# update sports column
|
|
self.upgrade_log('[4/4] Updating tv_shows to reflect the correct sports value...')
|
|
if not db_backed_up:
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
cl = []
|
|
history_quality = self.connection.select(
|
|
'SELECT * FROM tv_shows WHERE LOWER(classification) = "sports" AND air_by_date = 1 AND sports = 0')
|
|
for cur_entry in history_quality:
|
|
cl.append(['UPDATE tv_shows SET sports = ? WHERE show_id = ?',
|
|
[cur_entry['air_by_date'], cur_entry['show_id']]])
|
|
cl.append(['UPDATE tv_shows SET air_by_date = 0 WHERE show_id = ?', [cur_entry['show_id']]])
|
|
self.connection.mass_action(cl)
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 30 -> 31
|
|
class AddSceneNumberingToTvEpisodes(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Adding columns scene_season and scene_episode to tvepisodes')
|
|
self.add_column('tv_episodes', 'scene_season', 'NUMERIC', 'NULL')
|
|
self.add_column('tv_episodes', 'scene_episode', 'NUMERIC', 'NULL')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 31 -> 32
|
|
class AddAnimeTVShow(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Adding column anime to tv_episodes')
|
|
self.add_column('tv_shows', 'anime', 'NUMERIC', '0')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 32 -> 33
|
|
class AddAbsoluteNumbering(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Adding column absolute_number to tv_episodes')
|
|
self.add_column('tv_episodes', 'absolute_number', 'NUMERIC', '0')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 33 -> 34
|
|
class AddSceneAbsoluteNumbering(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Adding columns absolute_number and scene_absolute_number to scene_numbering')
|
|
self.add_column('scene_numbering', 'absolute_number', 'NUMERIC', '0')
|
|
self.add_column('scene_numbering', 'scene_absolute_number', 'NUMERIC', '0')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 34 -> 35
|
|
class AddAnimeAllowlistBlocklist(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
cl = [['CREATE TABLE allowlist (show_id INTEGER, range TEXT, keyword TEXT, indexer NUMERIC)'],
|
|
['CREATE TABLE blocklist (show_id INTEGER, range TEXT, keyword TEXT, indexer NUMERIC)']]
|
|
self.upgrade_log('Creating tables for anime allow and block lists')
|
|
self.connection.mass_action(cl)
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 35 -> 36
|
|
class AddSceneAbsoluteNumbering2(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Adding column scene_absolute_number to tv_episodes')
|
|
self.add_column('tv_episodes', 'scene_absolute_number', 'NUMERIC', '0')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 36 -> 37
|
|
class AddXemRefresh(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Creating table xem_refresh')
|
|
self.connection.action(
|
|
'CREATE TABLE xem_refresh (indexer TEXT, indexer_id INTEGER PRIMARY KEY, last_refreshed INTEGER)')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 37 -> 38
|
|
class AddSceneToTvShows(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Adding column scene to tv_shows')
|
|
self.add_column('tv_shows', 'scene', 'NUMERIC', '0')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 38 -> 39
|
|
class AddIndexerMapping(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
if self.has_table('indexer_mapping'):
|
|
self.connection.action('DROP TABLE indexer_mapping')
|
|
|
|
self.upgrade_log('Adding table indexer_mapping')
|
|
self.connection.action(
|
|
'CREATE TABLE indexer_mapping (indexer_id INTEGER, indexer NUMERIC, mindexer_id INTEGER, mindexer NUMERIC,'
|
|
' PRIMARY KEY (indexer_id, indexer))')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 39 -> 40
|
|
class AddVersionToTvEpisodes(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Adding columns release_group and version to tv_episodes')
|
|
self.add_column('tv_episodes', 'release_group', 'TEXT', '')
|
|
self.add_column('tv_episodes', 'version', 'NUMERIC', '-1')
|
|
|
|
self.upgrade_log('Adding column version to history')
|
|
self.add_column('history', 'version', 'NUMERIC', '-1')
|
|
|
|
self.inc_db_version()
|
|
return self.call_check_db_version()
|
|
|
|
|
|
# 40 -> 10000
|
|
class BumpDatabaseVersion(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Bumping database version')
|
|
|
|
return self.set_db_version(10000)
|
|
|
|
|
|
# 41,42 -> 10001
|
|
class Migrate41(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Bumping database version')
|
|
|
|
return self.set_db_version(10001)
|
|
|
|
|
|
# 43,44 -> 10001
|
|
class Migrate43(db.SchemaUpgrade):
|
|
def execute(self):
|
|
|
|
db_backed_up = False
|
|
db_chg = None
|
|
table = 'tmdb_info'
|
|
if self.has_table(table):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
db_backed_up = True
|
|
self.upgrade_log('Dropping redundant table tmdb_info')
|
|
# noinspection SqlResolve
|
|
self.connection.action('DROP TABLE [%s]' % table)
|
|
db_chg = True
|
|
|
|
if self.has_column('tv_shows', 'tmdb_id'):
|
|
if not db_backed_up:
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
db_backed_up = True
|
|
self.upgrade_log('Dropping redundant tmdb_info refs')
|
|
self.drop_columns('tv_shows', 'tmdb_id')
|
|
db_chg = True
|
|
|
|
if not self.has_table('db_version'):
|
|
if not db_backed_up:
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.connection.action('PRAGMA user_version = 0')
|
|
self.connection.action('CREATE TABLE db_version (db_version INTEGER);')
|
|
self.connection.action('INSERT INTO db_version (db_version) VALUES (0);')
|
|
|
|
if not db_chg:
|
|
self.upgrade_log('Bumping database version')
|
|
|
|
return self.set_db_version(10001)
|
|
|
|
|
|
# 4301 -> 10002
|
|
class Migrate4301(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Bumping database version')
|
|
|
|
return self.set_db_version(10002)
|
|
|
|
|
|
# 4302,4400 -> 10003
|
|
class Migrate4302(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Bumping database version')
|
|
|
|
return self.set_db_version(10003)
|
|
|
|
|
|
# 5816 - 5818 -> 15
|
|
class MigrateUpstream(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log(f'Migrate SickBeard db v{self.call_check_db_version().replace("58", "")} into v15')
|
|
|
|
return self.set_db_version(15)
|
|
|
|
|
|
# 10000 -> 20000
|
|
class SickGearDatabaseVersion(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Bumping database version to new SickGear standards')
|
|
|
|
return self.set_db_version(20000)
|
|
|
|
|
|
# 10001 -> 10000
|
|
class RemoveDefaultEpStatusFromTvShows(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Dropping redundant column default_ep_status from tv_shows')
|
|
self.drop_columns('tv_shows', 'default_ep_status')
|
|
|
|
return self.set_db_version(10000)
|
|
|
|
|
|
# 10002 -> 10001
|
|
class RemoveMinorDBVersion(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Dropping redundant column db_minor_version from db_version')
|
|
self.drop_columns('db_version', 'db_minor_version')
|
|
|
|
return self.set_db_version(10001)
|
|
|
|
|
|
# 10003 -> 10002
|
|
class RemoveMetadataSub(db.SchemaUpgrade):
|
|
def execute(self):
|
|
if self.has_column('tv_shows', 'sub_use_sr_metadata'):
|
|
self.upgrade_log('Dropping redundant column metadata sub')
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.drop_columns('tv_shows', 'sub_use_sr_metadata')
|
|
|
|
return self.set_db_version(10002)
|
|
|
|
|
|
# 20000 -> 20001
|
|
class DBIncreaseTo20001(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Bumping database version to force a backup before new database code')
|
|
|
|
self.connection.action('VACUUM')
|
|
self.upgrade_log('Performed a vacuum on the database', logger.DEBUG)
|
|
|
|
return self.set_db_version(20001)
|
|
|
|
|
|
# 20001 -> 20002
|
|
class AddTvShowOverview(db.SchemaUpgrade):
|
|
def execute(self):
|
|
if not self.has_column('tv_shows', 'overview'):
|
|
self.upgrade_log('Adding column overview to tv_shows')
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.add_column('tv_shows', 'overview', 'TEXT', '')
|
|
|
|
return self.set_db_version(20002)
|
|
|
|
|
|
# 20002 -> 20003
|
|
class AddTvShowTags(db.SchemaUpgrade):
|
|
def execute(self):
|
|
if not self.has_column('tv_shows', 'tag'):
|
|
self.upgrade_log('Adding tag to tv_shows')
|
|
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.add_column('tv_shows', 'tag', 'TEXT', 'Show List')
|
|
|
|
return self.set_db_version(20003)
|
|
|
|
|
|
# 20003 -> 20004
|
|
class ChangeMapIndexer(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
if self.has_table('indexer_mapping'):
|
|
self.connection.action('DROP TABLE indexer_mapping')
|
|
|
|
self.upgrade_log('Changing table indexer_mapping')
|
|
self.connection.action(
|
|
'CREATE TABLE indexer_mapping (indexer_id INTEGER, indexer NUMERIC, mindexer_id INTEGER NOT NULL,'
|
|
' mindexer NUMERIC, date NUMERIC NOT NULL DEFAULT 0, status INTEGER NOT NULL DEFAULT 0,'
|
|
' PRIMARY KEY (indexer_id, indexer, mindexer))')
|
|
|
|
self.connection.action('CREATE INDEX IF NOT EXISTS idx_mapping ON indexer_mapping (indexer_id, indexer)')
|
|
|
|
if not self.has_column('info', 'last_run_backlog'):
|
|
self.upgrade_log('Adding last_run_backlog to info')
|
|
self.add_column('info', 'last_run_backlog', 'NUMERIC', 1)
|
|
|
|
self.upgrade_log('Moving table scene_exceptions from cache.db to sickbeard.db')
|
|
if self.has_table('scene_exceptions_refresh'):
|
|
self.connection.action('DROP TABLE scene_exceptions_refresh')
|
|
self.connection.action('CREATE TABLE scene_exceptions_refresh (list TEXT PRIMARY KEY, last_refreshed INTEGER)')
|
|
if self.has_table('scene_exceptions'):
|
|
self.connection.action('DROP TABLE scene_exceptions')
|
|
self.connection.action('CREATE TABLE scene_exceptions (exception_id INTEGER PRIMARY KEY,'
|
|
' indexer_id INTEGER KEY, show_name TEXT, season NUMERIC, custom NUMERIC)')
|
|
|
|
try:
|
|
cachedb = db.DBConnection(filename='cache.db')
|
|
if cachedb.has_table('scene_exceptions'):
|
|
sql_result = cachedb.action('SELECT * FROM scene_exceptions')
|
|
cs = []
|
|
for cur_result in sql_result:
|
|
cs.append(
|
|
['INSERT OR REPLACE INTO scene_exceptions (exception_id, indexer_id, show_name, season, custom)'
|
|
' VALUES (?,?,?,?,?)',
|
|
[cur_result['exception_id'], cur_result['indexer_id'],
|
|
cur_result['show_name'], cur_result['season'], cur_result['custom']]])
|
|
|
|
if 0 < len(cs):
|
|
self.connection.mass_action(cs)
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
keep_tables = {'allowlist', 'blocklist', 'whitelist', 'blacklist',
|
|
'scene_exceptions', 'scene_exceptions_refresh', 'info', 'indexer_mapping',
|
|
'db_version', 'history', 'imdb_info', 'lastUpdate', 'scene_numbering', 'tv_episodes', 'tv_shows',
|
|
'xem_refresh'}
|
|
current_tables = set(self.list_tables())
|
|
remove_tables = list(current_tables - keep_tables)
|
|
for table in remove_tables:
|
|
# noinspection SqlResolve
|
|
self.connection.action('DROP TABLE [%s]' % table)
|
|
|
|
self.connection.action('VACUUM')
|
|
|
|
return self.set_db_version(20004)
|
|
|
|
|
|
# 20004 -> 20005
|
|
class AddShowNotFoundCounter(db.SchemaUpgrade):
|
|
def execute(self):
|
|
if not self.has_table('tv_shows_not_found'):
|
|
self.upgrade_log('Adding table tv_shows_not_found')
|
|
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.connection.action(
|
|
'CREATE TABLE tv_shows_not_found (indexer NUMERIC NOT NULL, indexer_id NUMERIC NOT NULL,'
|
|
' fail_count NUMERIC NOT NULL DEFAULT 0, last_check NUMERIC NOT NULL, last_success NUMERIC,'
|
|
' PRIMARY KEY (indexer_id, indexer))')
|
|
|
|
return self.set_db_version(20005)
|
|
|
|
|
|
# 20005 -> 20006
|
|
class AddFlagTable(db.SchemaUpgrade):
|
|
def execute(self):
|
|
if not self.has_table('flags'):
|
|
self.upgrade_log('Adding table flags')
|
|
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.connection.action('CREATE TABLE flags (flag PRIMARY KEY NOT NULL )')
|
|
|
|
return self.set_db_version(20006)
|
|
|
|
|
|
# 20006 -> 20007
|
|
class DBIncreaseTo20007(db.SchemaUpgrade):
|
|
def execute(self):
|
|
|
|
self.upgrade_log('Bumping database version')
|
|
|
|
return self.set_db_version(20007)
|
|
|
|
|
|
# 20007 -> 20008
|
|
class AddWebdlTypesTable(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.connection.action('CREATE TABLE webdl_types (dname TEXT NOT NULL , regex TEXT NOT NULL )')
|
|
|
|
return self.set_db_version(20008)
|
|
|
|
|
|
# 20008 -> 20009
|
|
class AddWatched(db.SchemaUpgrade):
|
|
def execute(self):
|
|
# remove old table from version 20007
|
|
if self.has_table('tv_episodes_watched') and not self.has_column('tv_episodes_watched', 'clientep_id'):
|
|
self.connection.action('DROP TABLE tv_episodes_watched')
|
|
self.connection.action('VACUUM')
|
|
|
|
if not self.has_table('tv_episodes_watched'):
|
|
self.upgrade_log('Adding table tv_episodes_watched')
|
|
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.connection.action(
|
|
'CREATE TABLE tv_episodes_watched (tvep_id NUMERIC NOT NULL, clientep_id TEXT, label TEXT,'
|
|
' played NUMERIC DEFAULT 0 NOT NULL, date_watched NUMERIC NOT NULL, date_added NUMERIC,'
|
|
' status NUMERIC, location TEXT, file_size NUMERIC, hide INT default 0 not null)'
|
|
)
|
|
|
|
return self.set_db_version(20009)
|
|
|
|
|
|
# 20009 -> 20010
|
|
class AddPrune(db.SchemaUpgrade):
|
|
def execute(self):
|
|
|
|
if not self.has_column('tv_shows', 'prune'):
|
|
self.upgrade_log('Adding prune to tv_shows')
|
|
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.add_column('tv_shows', 'prune', 'INT', 0)
|
|
|
|
return self.set_db_version(20010)
|
|
|
|
|
|
# 20010 -> 20011
|
|
class AddIndexerToTables(db.SchemaUpgrade):
|
|
def execute(self):
|
|
sickgear.helpers.upgrade_new_naming()
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
show_ids = {s['prod_id']: s['tv_id'] for s in
|
|
self.connection.select('SELECT indexer AS tv_id, indexer_id AS prod_id FROM tv_shows')}
|
|
|
|
allowtbl, blocktbl = (('allow', 'block'), ('white', 'black'))[not self.connection.has_table('blocklist')]
|
|
allowtbl, blocktbl = '%slist' % allowtbl, '%slist' % blocktbl
|
|
columns = {allowtbl: 'show_id, range, keyword, indexer',
|
|
blocktbl: 'show_id, range, keyword, indexer',
|
|
'history': 'action, date, showid, season, episode, quality, resource, provider, version, indexer',
|
|
'scene_exceptions': 'exception_id , indexer_id, show_name, season, custom, indexer'}
|
|
|
|
# add missing indexer column
|
|
for t in [(allowtbl, 'show_id'), (blocktbl, 'show_id'),
|
|
('history', 'showid'), ('scene_exceptions', 'indexer_id')]:
|
|
if not self.has_column(t[0], 'indexer'):
|
|
self.upgrade_log(f'Adding TV info support to {t[0]} table')
|
|
self.add_column(t[0], 'indexer')
|
|
cl = []
|
|
for s_id, i in iteritems(show_ids):
|
|
# noinspection SqlResolve
|
|
cl.append(['UPDATE %s SET indexer = ? WHERE %s = ?' % (t[0], t[1]), [i, s_id]])
|
|
self.connection.mass_action(cl)
|
|
# noinspection SqlResolve
|
|
self.connection.action('CREATE INDEX idx_id_indexer_%s ON %s (indexer, %s)' % (t[0], t[0], t[1]))
|
|
|
|
if 'history' != t[0]:
|
|
# remove any unknown ids (exception history table)
|
|
# noinspection SqlResolve
|
|
self.connection.action('DELETE FROM %s WHERE indexer = ?' % t[0], [0])
|
|
if 0 < self.connection.connection.total_changes:
|
|
self.upgrade_log('Removed orphaned data from %s' % t[0])
|
|
|
|
if self.connection.has_table('backup_%s' % t[0]):
|
|
self.upgrade_log('Adding backup data to %s' % t[0])
|
|
self.connection.action('REPLACE INTO %s SELECT %s FROM %s' % ('%s (%s)' % (t[0], columns[t[0]]),
|
|
columns[t[0]], 'backup_%s' % t[0]))
|
|
self.connection.remove_table('backup_%s' % t[0])
|
|
|
|
# recreate tables that have wrong primary key = indexer_id without indexer
|
|
self.upgrade_log('Adding TV info support to scene_numbering')
|
|
# noinspection SqlResolve
|
|
self.connection.mass_action([['ALTER TABLE scene_numbering RENAME TO tmp_scene_numbering'],
|
|
['CREATE TABLE scene_numbering (indexer TEXT, indexer_id INTEGER, season INTEGER, '
|
|
'episode INTEGER, scene_season INTEGER, scene_episode INTEGER, '
|
|
'absolute_number NUMERIC, scene_absolute_number NUMERIC, '
|
|
'PRIMARY KEY (indexer,indexer_id,season,episode))'],
|
|
['REPLACE INTO scene_numbering (indexer, indexer_id, '
|
|
'season, episode, scene_season, scene_episode, absolute_number, '
|
|
'scene_absolute_number) SELECT "0" AS indexer, indexer_id, '
|
|
'season, episode, scene_season, scene_episode, absolute_number, '
|
|
'scene_absolute_number FROM tmp_scene_numbering'],
|
|
['DROP TABLE tmp_scene_numbering']])
|
|
|
|
cl = []
|
|
for s_id, i in iteritems(show_ids):
|
|
cl.append(['UPDATE scene_numbering SET indexer = ? WHERE indexer_id = ?', [i, s_id]])
|
|
cl.append(['DELETE FROM scene_numbering WHERE indexer = ?', [0]])
|
|
self.connection.mass_action(cl)
|
|
|
|
self.upgrade_log('Adding TV info support to imdb_info')
|
|
# noinspection SqlResolve
|
|
self.connection.mass_action([['ALTER TABLE imdb_info RENAME TO tmp_imdb_info'],
|
|
['CREATE TABLE imdb_info (indexer NUMERIC, indexer_id NUMERIC, imdb_id TEXT, '
|
|
'title TEXT, year NUMERIC, akas TEXT, runtimes NUMERIC, genres TEXT, '
|
|
'countries TEXT, country_codes TEXT, certificates TEXT, rating TEXT, '
|
|
'votes INTEGER, last_update NUMERIC, PRIMARY KEY (indexer,indexer_id))'],
|
|
['REPLACE INTO imdb_info (indexer, indexer_id, imdb_id, '
|
|
'title, year, akas, runtimes, genres, countries, country_codes, certificates, '
|
|
'rating, votes, last_update) SELECT "0" AS indexer, indexer_id, imdb_id, '
|
|
'title, year, akas, runtimes, genres, countries, country_codes, certificates, '
|
|
'rating, votes, last_update FROM tmp_imdb_info'],
|
|
['DROP TABLE tmp_imdb_info']])
|
|
|
|
cl = []
|
|
for s_id, i in iteritems(show_ids):
|
|
cl.append(['UPDATE imdb_info SET indexer = ? WHERE indexer_id = ?', [i, s_id]])
|
|
cl.append(['DELETE FROM imdb_info WHERE indexer = ?', [0]])
|
|
self.connection.mass_action(cl)
|
|
self.connection.action('CREATE INDEX idx_id_indexer_imdb_info ON imdb_info (indexer,indexer_id)')
|
|
|
|
if self.connection.has_table('backup_imdb_info'):
|
|
self.upgrade_log('Adding backup data to imdb_info')
|
|
# noinspection SqlResolve
|
|
self.connection.action('REPLACE INTO imdb_info (indexer, indexer_id, imdb_id, title, year, akas, '
|
|
'runtimes, genres, countries, country_codes, certificates, rating, votes, '
|
|
'last_update) SELECT indexer, indexer_id, imdb_id, title, year, akas, runtimes, '
|
|
'genres, countries, country_codes, certificates, rating, votes, last_update '
|
|
'FROM backup_imdb_info')
|
|
self.connection.remove_table('backup_imdb_info')
|
|
|
|
# remove an index of an no longer existing column
|
|
self.upgrade_log('Changing/Re-Creating Indexes')
|
|
if self.connection.has_index('tv_shows', 'idx_tvdb_id'):
|
|
self.connection.remove_index('tv_shows', 'idx_tvdb_id')
|
|
|
|
if self.connection.has_index('tv_shows', 'idx_indexer_id'):
|
|
self.connection.remove_index('tv_shows', 'idx_indexer_id')
|
|
self.connection.action('CREATE UNIQUE INDEX idx_indexer_id ON tv_shows (indexer,indexer_id)')
|
|
|
|
if self.connection.has_index('tv_episodes', 'idx_showid'):
|
|
self.connection.remove_index('tv_episodes', 'idx_showid')
|
|
|
|
if self.connection.has_index('tv_episodes', 'idx_tv_episodes_showid_airdate'):
|
|
self.connection.remove_index('tv_episodes', 'idx_tv_episodes_showid_airdate')
|
|
self.connection.action('CREATE INDEX idx_tv_episodes_showid_airdate ON tv_episodes(indexer,showid,airdate)')
|
|
|
|
if not self.connection.has_index('tv_episodes', 'idx_tv_episodes_unique'):
|
|
self.connection.action('CREATE UNIQUE INDEX idx_tv_episodes_unique ON '
|
|
'tv_episodes(indexer,showid,season,episode)')
|
|
|
|
if self.connection.has_table('backup_tv_episodes'):
|
|
self.upgrade_log('Adding backup data to tv_episodes')
|
|
# noinspection SqlResolve
|
|
self.connection.action('REPLACE INTO tv_episodes (episode_id, showid, indexerid, indexer, name, season, '
|
|
'episode, description, airdate, hasnfo, hastbn, status, location, file_size, '
|
|
'release_name, subtitles, subtitles_searchcount, subtitles_lastsearch, is_proper, '
|
|
'scene_season, scene_episode, absolute_number, scene_absolute_number, '
|
|
'release_group, version) SELECT episode_id, showid, indexerid, indexer, name, '
|
|
'season, episode, description, airdate, hasnfo, hastbn, status, location, '
|
|
'file_size, release_name, subtitles, subtitles_searchcount, subtitles_lastsearch, '
|
|
'is_proper, scene_season, scene_episode, absolute_number, scene_absolute_number, '
|
|
'release_group, version FROM backup_tv_episodes')
|
|
self.connection.remove_table('backup_tv_episodes')
|
|
|
|
if self.connection.has_table('backup_tv_shows'):
|
|
self.upgrade_log('Adding backup data to tv_shows')
|
|
# noinspection SqlResolve
|
|
self.connection.action('REPLACE INTO tv_shows (show_id, indexer_id, indexer, show_name, location, '
|
|
'network, genre, classification, runtime, quality, airs, status, flatten_folders, '
|
|
'paused, startyear, air_by_date, lang, subtitles, notify_list, imdb_id, '
|
|
'last_update_indexer, dvdorder, archive_firstmatch, rls_require_words, '
|
|
'rls_ignore_words, sports, anime, scene, overview, tag, prune) '
|
|
'SELECT show_id, indexer_id, '
|
|
'indexer, show_name, location, network, genre, classification, runtime, quality, '
|
|
'airs, status, flatten_folders, paused, startyear, air_by_date, lang, subtitles, '
|
|
'notify_list, imdb_id, last_update_indexer, dvdorder, archive_firstmatch, '
|
|
'rls_require_words, rls_ignore_words, sports, anime, scene, overview, tag, prune '
|
|
'FROM backup_tv_shows')
|
|
self.connection.remove_table('backup_tv_shows')
|
|
|
|
self.connection.action('VACUUM')
|
|
|
|
return self.set_db_version(20011)
|
|
|
|
|
|
# 20011 -> 20012
|
|
class AddShowExludeGlobals(db.SchemaUpgrade):
|
|
def execute(self):
|
|
|
|
if not self.has_column('tv_shows', 'rls_global_exclude_ignore'):
|
|
self.upgrade_log('Adding rls_global_exclude_ignore, rls_global_exclude_require to tv_shows')
|
|
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
self.add_column('tv_shows', 'rls_global_exclude_ignore', data_type='TEXT', default='')
|
|
self.add_column('tv_shows', 'rls_global_exclude_require', data_type='TEXT', default='')
|
|
|
|
if self.has_table('tv_shows_exclude_backup'):
|
|
self.upgrade_log('Adding rls_global_exclude_ignore, rls_global_exclude_require from backup to tv_shows')
|
|
# noinspection SqlResolve
|
|
self.connection.mass_action([['UPDATE tv_shows SET rls_global_exclude_ignore = '
|
|
'(SELECT te.rls_global_exclude_ignore FROM tv_shows_exclude_backup te WHERE '
|
|
'te.show_id = tv_shows.show_id AND te.indexer = tv_shows.indexer), '
|
|
'rls_global_exclude_require = (SELECT te.rls_global_exclude_require FROM '
|
|
'tv_shows_exclude_backup te WHERE te.show_id = tv_shows.show_id AND '
|
|
'te.indexer = tv_shows.indexer) WHERE EXISTS (SELECT 1 FROM '
|
|
'tv_shows_exclude_backup WHERE tv_shows.show_id = '
|
|
'tv_shows_exclude_backup.show_id AND '
|
|
'tv_shows.indexer = tv_shows_exclude_backup.indexer)'],
|
|
['DROP TABLE tv_shows_exclude_backup']
|
|
])
|
|
|
|
return self.set_db_version(20012)
|
|
|
|
|
|
# 20012 -> 20013
|
|
class RenameAllowBlockListTables(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
if not self.connection.has_table('blocklist'):
|
|
self.upgrade_log('Renaming allow/block list tables')
|
|
|
|
for old, new in (('black', 'block'), ('white', 'allow')):
|
|
# noinspection SqlResolve
|
|
self.connection.mass_action([
|
|
['ALTER TABLE %slist RENAME TO tmp_%slist' % (old, new)],
|
|
['CREATE TABLE %slist (show_id INTEGER, range TEXT, keyword TEXT, indexer NUMERIC)' % new],
|
|
['INSERT INTO %slist(show_id, range, keyword, indexer)'
|
|
' SELECT show_id, range, keyword, indexer FROM tmp_%slist' % (new, new)],
|
|
['DROP TABLE tmp_%slist' % new]
|
|
])
|
|
|
|
return self.set_db_version(20013)
|
|
|
|
|
|
# 20013 -> 20014
|
|
class AddHistoryHideColumn(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
if not self.has_column('history', 'hide'):
|
|
self.upgrade_log('Adding hide column to history')
|
|
self.add_column('history', 'hide', default=0, set_default=True)
|
|
|
|
if self.has_table('history_hide_backup'):
|
|
self.upgrade_log('Restoring hide status in history from backup')
|
|
# noinspection SqlResolve
|
|
self.connection.mass_action([
|
|
[
|
|
"""
|
|
UPDATE history SET hide = (SELECT hide FROM history_hide_backup as hh WHERE
|
|
hh.ROWID = history.ROWID AND hh.showid = history.showid AND hh.indexer = history.indexer AND
|
|
hh.season = history.season AND hh.episode = history.episode AND hh.action = history.action AND
|
|
hh.date = history.date)
|
|
"""
|
|
],
|
|
['DROP TABLE history_hide_backup']
|
|
])
|
|
|
|
return self.set_db_version(20014)
|
|
|
|
|
|
# 20014 -> 20015
|
|
class ChangeShowData(db.SchemaUpgrade):
|
|
def execute(self):
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
|
|
self.upgrade_log('Adding new data columns to tv_shows')
|
|
self.add_columns('tv_shows', [('timezone', 'TEXT', ''), ('airtime', 'NUMERIC'),
|
|
('network_country', 'TEXT', ''), ('network_country_code', 'TEXT', ''),
|
|
('network_id', 'NUMERIC'), ('network_is_stream', 'INTEGER'),
|
|
('src_update_timestamp', 'INTEGER')])
|
|
|
|
self.upgrade_log('Adding new data columns to tv_episodes')
|
|
self.add_columns('tv_episodes', [('timezone', 'TEXT', ''), ('airtime', 'NUMERIC'),
|
|
('runtime', 'NUMERIC', 0), ('timestamp', 'NUMERIC'),
|
|
('network', 'TEXT', ''), ('network_country', 'TEXT', ''),
|
|
('network_country_code', 'TEXT', ''), ('network_id', 'NUMERIC'),
|
|
('network_is_stream', 'INTEGER')])
|
|
|
|
if not self.has_column('imdb_info', 'is_mini_series'):
|
|
self.upgrade_log('Adding new data columns to imdb_info')
|
|
self.add_columns('imdb_info', [('is_mini_series', 'INTEGER', 0), ('episode_count', 'NUMERIC')])
|
|
|
|
self.upgrade_log('Adding Character and Persons tables')
|
|
|
|
table_create_sql = {
|
|
'castlist': [
|
|
[
|
|
"""
|
|
CREATE TABLE castlist
|
|
(
|
|
id INTEGER PRIMARY KEY AUTOINCREMENT,
|
|
indexer NUMERIC NOT NULL,
|
|
indexer_id NUMERIC NOT NULL,
|
|
character_id NUMERIC NOT NULL,
|
|
sort_order NUMERIC DEFAULT 0 NOT NULL,
|
|
updated NUMERIC
|
|
);
|
|
"""
|
|
],
|
|
],
|
|
'idx_castlist': [
|
|
['CREATE INDEX idx_castlist ON castlist (indexer, indexer_Id);'],
|
|
['CREATE UNIQUE INDEX idx_unique_castlist ON castlist (indexer, indexer_id, character_id);']
|
|
],
|
|
'characters': [
|
|
[
|
|
"""
|
|
CREATE TABLE characters
|
|
(
|
|
id INTEGER NOT NULL PRIMARY KEY AUTOINCREMENT,
|
|
name TEXT,
|
|
bio TEXT,
|
|
thumb_url TEXT,
|
|
image_url TEXT,
|
|
updated NUMERIC
|
|
);
|
|
"""
|
|
]
|
|
],
|
|
'character_ids': [
|
|
[
|
|
"""
|
|
CREATE TABLE character_ids
|
|
(
|
|
id INTEGER PRIMARY KEY AUTOINCREMENT,
|
|
src NUMERIC NOT NULL,
|
|
src_id NUMERIC NOT NULL,
|
|
character_id NUMERIC NOT NULL
|
|
);
|
|
"""
|
|
],
|
|
],
|
|
'idx_character_ids': [
|
|
['CREATE UNIQUE INDEX idx_unique_character_ids ON character_ids (src, character_id);'],
|
|
['CREATE INDEX idx_character_ids ON character_ids (character_id);']
|
|
],
|
|
'character_person_map': [
|
|
[
|
|
"""
|
|
CREATE TABLE character_person_map
|
|
(
|
|
id INTEGER PRIMARY KEY AUTOINCREMENT,
|
|
character_id NUMERIC NOT NULL,
|
|
person_id NUMERIC NOT NULL
|
|
);
|
|
"""
|
|
],
|
|
],
|
|
'idx_character_person_map': [
|
|
['CREATE INDEX idx_character_person_map_character ON character_person_map (character_id);'],
|
|
['CREATE INDEX idx_character_person_map_person ON character_person_map (person_id);'],
|
|
['CREATE UNIQUE INDEX idx_unique_character_person ON character_person_map (character_id, person_id);']
|
|
],
|
|
'character_person_years': [
|
|
[
|
|
"""CREATE TABLE character_person_years
|
|
(
|
|
character_id NUMERIC NOT NULL,
|
|
person_id NUMERIC NOT NULL,
|
|
start_year NUMERIC,
|
|
end_year NUMERIC
|
|
);
|
|
"""
|
|
],
|
|
],
|
|
'idx_character_person_years': [
|
|
['CREATE UNIQUE INDEX idx_unique_character_person_years '
|
|
'ON character_person_years (character_id, person_id)'],
|
|
['CREATE INDEX idx_character_person_years ON character_person_years (character_id)'],
|
|
],
|
|
'persons': [
|
|
[
|
|
"""
|
|
CREATE TABLE persons
|
|
(
|
|
id INTEGER PRIMARY KEY AUTOINCREMENT,
|
|
name TEXT,
|
|
gender INTEGER,
|
|
birthdate NUMERIC,
|
|
deathdate NUMERIC,
|
|
birthplace TEXT,
|
|
deathplace TEXT,
|
|
height NUMERIC,
|
|
realname TEXT,
|
|
nicknames TEXT,
|
|
akas TEXT,
|
|
homepage TEXT,
|
|
bio TEXT,
|
|
thumb_url TEXT,
|
|
image_url TEXT,
|
|
updated NUMERIC
|
|
);
|
|
"""
|
|
]
|
|
],
|
|
'person_ids': [
|
|
[
|
|
"""
|
|
CREATE TABLE person_ids
|
|
(
|
|
id INTEGER PRIMARY KEY AUTOINCREMENT,
|
|
src INTEGER NOT NULL,
|
|
src_id TEXT NOT NULL,
|
|
person_id NUMERIC NOT NULL
|
|
);
|
|
"""
|
|
],
|
|
],
|
|
'idx_person_ids': [
|
|
['CREATE UNIQUE INDEX idx_unique_person_ids ON person_ids (src, person_id);'],
|
|
['CREATE INDEX idx_person_ids ON person_ids (person_id);']
|
|
],
|
|
'tv_src_switch': [
|
|
[
|
|
"""
|
|
CREATE TABLE tv_src_switch
|
|
(
|
|
old_indexer NUMERIC NOT NULL,
|
|
old_indexer_id NUMERIC NOT NULL,
|
|
new_indexer NUMERIC NOT NULL,
|
|
new_indexer_id NUMERIC,
|
|
force_id NUMERIC DEFAULT 0 NOT NULL,
|
|
set_pause INTEGER DEFAULT 0 NOT NULL,
|
|
mark_wanted INTEGER DEFAULT 0 NOT NULL,
|
|
status NUMERIC DEFAULT 0 NOT NULL,
|
|
action_id INTEGER NOT NULL,
|
|
uid NUMERIC NOT NULL
|
|
);
|
|
"""
|
|
],
|
|
],
|
|
'idx_tv_src_switch': [
|
|
['CREATE UNIQUE INDEX idx_unique_tv_src_switch ON tv_src_switch (old_indexer, old_indexer_id);']
|
|
],
|
|
'switch_ep_result': [
|
|
[
|
|
"""
|
|
CREATE TABLE switch_ep_result
|
|
(
|
|
old_indexer NUMERIC NOT NULL,
|
|
old_indexer_id NUMERIC NOT NULL,
|
|
new_indexer NUMERIC NOT NULL,
|
|
new_indexer_id NUMERIC NOT NULL,
|
|
season NUMERIC NOT NULL,
|
|
episode NUMERIC NOT NULL,
|
|
reason NUMERIC DEFAULT 0
|
|
);
|
|
"""
|
|
],
|
|
],
|
|
'idx_switch_ep_result': [
|
|
['CREATE INDEX idx_switch_ep_result ON switch_ep_result (new_indexer, new_indexer_id);'],
|
|
["""CREATE INDEX idx_unique_switch_ep_result
|
|
ON switch_ep_result (new_indexer, new_indexer_id, season, episode, reason);"""],
|
|
['CREATE INDEX idx_switch_ep_result_old ON switch_ep_result (old_indexer, old_indexer_id);'],
|
|
],
|
|
}
|
|
|
|
cl = []
|
|
tables = self.list_tables()
|
|
for t in ('castlist', 'characters', 'character_ids', 'persons', 'person_ids', 'character_person_map',
|
|
'character_person_years', 'tv_src_switch', 'switch_ep_result'):
|
|
if 'backup_%s' % t in tables:
|
|
# noinspection SqlResolve
|
|
cl.append(['ALTER TABLE backup_%s RENAME TO %s' % (t, t)])
|
|
elif t not in tables:
|
|
cl.extend(table_create_sql[t])
|
|
if 'idx_%s' % t in table_create_sql:
|
|
cl.extend(table_create_sql['idx_%s' % t])
|
|
|
|
cl.extend(sickgear.tv.TVShow.orphaned_cast_sql())
|
|
|
|
if cl:
|
|
self.connection.mass_action(cl)
|
|
self.connection.action('VACUUM')
|
|
|
|
return self.set_db_version(20015)
|
|
|
|
|
|
# 20015 -> 20016
|
|
class ChangeTmdbID(db.SchemaUpgrade):
|
|
def execute(self):
|
|
from lib.tvinfo_base import TVINFO_TMDB, TVINFO_TMDB_OLD, TVINFO_TVDB, TVINFO_TVMAZE
|
|
from sg_helpers import move_file, scantree
|
|
from ..image_cache import ImageCache, cache_img_src
|
|
import re
|
|
self.upgrade_log('Renaming tmdb images')
|
|
# noinspection PyProtectedMember
|
|
for _dir in (ImageCache._persons_dir(), ImageCache._characters_dir()):
|
|
for _f in scantree(_dir): # type: DirEntry
|
|
if not _f.is_file(follow_symlinks=False):
|
|
continue
|
|
try:
|
|
img_src = int(re.search(r'^(\d+)-', _f.name).group(1))
|
|
except (BaseException, Exception):
|
|
continue
|
|
if img_src != TVINFO_TMDB_OLD and img_src not in cache_img_src:
|
|
continue
|
|
try:
|
|
move_file(_f.path,
|
|
os.path.join(os.path.dirname(_f.path),
|
|
re.sub('^%s-' % img_src, '%s-' %
|
|
cache_img_src[(img_src, TVINFO_TMDB)[TVINFO_TMDB_OLD == img_src]], _f.name)))
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
db.backup_database(self.connection, 'sickbeard.db', self.call_check_db_version())
|
|
has_tmdb_backups = all(self.has_table(_r) for _r in
|
|
('backup_tmdb_tv_shows', 'backup_tmdb_tv_episodes', 'backup_tmdb_indexer_mapping'))
|
|
if has_tmdb_backups:
|
|
self.upgrade_log('Checking for dupe shows in backup tables')
|
|
# get all ids from downgraded backup table = new tmdb id is used
|
|
backup_tmdb_show_ids = {_r['show_id'] for _r in
|
|
self.connection.select('SELECT show_id FROM backup_tmdb_tv_shows WHERE indexer = ?',
|
|
[TVINFO_TMDB])}
|
|
if backup_tmdb_show_ids:
|
|
# get all mapped tmdb ids from current db = old tmdb id is used
|
|
current_mapped_tmdb_ids = {
|
|
_r['mindexer_id'] for _r in
|
|
self.connection.select('SELECT mindexer_id FROM indexer_mapping WHERE mindexer = ?',
|
|
[TVINFO_TMDB_OLD])}
|
|
# set of all dupe ids = these shows exists with other source already and can't be restored
|
|
dupe_ids = backup_tmdb_show_ids.intersection(current_mapped_tmdb_ids)
|
|
backup_mapped_tvdb_ids = {
|
|
_r['mindexer_id']: _r['indexer_id'] for _r in
|
|
self.connection.select('SELECT mindexer_id, indexer_id FROM backup_tmdb_indexer_mapping'
|
|
' WHERE mindexer = ? AND indexer = ?', [TVINFO_TVDB, TVINFO_TMDB])}
|
|
current_tvdb_ids = {
|
|
_r['show_id'] for _r in
|
|
self.connection.select('SELECT show_id FROM tv_shows WHERE indexer = ?', [TVINFO_TVDB])}
|
|
dupe_tvdb_ids = current_tvdb_ids.intersection(backup_mapped_tvdb_ids.keys())
|
|
if dupe_tvdb_ids:
|
|
dupe_ids += {_v for _k, _v in iteritems(backup_mapped_tvdb_ids) if _k in dupe_tvdb_ids}
|
|
backup_mapped_tvmaze_ids = {
|
|
_r['mindexer_id']: _r['indexer_id'] for _r in
|
|
self.connection.select('SELECT mindexer_id, indexer_id FROM backup_tmdb_indexer_mapping'
|
|
' WHERE mindexer = ? AND indexer = ?', [TVINFO_TVMAZE, TVINFO_TVDB])}
|
|
current_tvmaze_ids = {
|
|
_r['show_id'] for _r in
|
|
self.connection.select('SELECT show_id FROM tv_shows WHERE indexer = ?', [TVINFO_TVMAZE])}
|
|
dupe_tvmaze_ids = current_tvmaze_ids.intersection(backup_mapped_tvmaze_ids.keys())
|
|
if dupe_tvmaze_ids:
|
|
dupe_ids += {_v for _k, _v in iteritems(backup_mapped_tvdb_ids) if _k in dupe_tvmaze_ids}
|
|
|
|
if dupe_ids:
|
|
self.upgrade_log('Dupe tmdb id detected, removing from backup')
|
|
self.connection.action(
|
|
'DELETE FROM tv_shows WHERE show_id IN (%s)' % ','.join(['?'] * len(dupe_ids)), list(dupe_ids))
|
|
|
|
self.upgrade_log('Changing tmdb id')
|
|
self.connection.mass_action([
|
|
['UPDATE indexer_mapping SET indexer = ? WHERE indexer = ?', [TVINFO_TMDB, TVINFO_TMDB_OLD]],
|
|
['UPDATE indexer_mapping SET mindexer = ? WHERE mindexer = ?', [TVINFO_TMDB, TVINFO_TMDB_OLD]],
|
|
['UPDATE person_ids SET src = ? WHERE src = ?', [TVINFO_TMDB, TVINFO_TMDB_OLD]],
|
|
['UPDATE character_ids SET src = ? WHERE src = ?', [TVINFO_TMDB, TVINFO_TMDB_OLD]],
|
|
] + ([],
|
|
[
|
|
['REPLACE INTO tv_shows '
|
|
'(show_id, indexer_id, indexer, show_name, location, network, genre, classification, runtime, quality,'
|
|
' airs, status, flatten_folders, paused, startyear, air_by_date, lang, subtitles, notify_list, imdb_id,'
|
|
' last_update_indexer, dvdorder, archive_firstmatch, rls_require_words, rls_ignore_words, sports, anime,'
|
|
' scene, overview, tag, prune, rls_global_exclude_ignore, rls_global_exclude_require, timezone, airtime,'
|
|
' network_country, network_country_code, network_id, network_is_stream, src_update_timestamp)'
|
|
' SELECT show_id, indexer_id, indexer, show_name, location, network, genre, classification, runtime,'
|
|
' quality, airs, status, flatten_folders, paused, startyear, air_by_date, lang, subtitles, notify_list,'
|
|
' imdb_id, last_update_indexer, dvdorder, archive_firstmatch, rls_require_words, rls_ignore_words,'
|
|
' sports, anime, scene, overview, tag, prune, rls_global_exclude_ignore, rls_global_exclude_require,'
|
|
' timezone, airtime, network_country, network_country_code, network_id, network_is_stream,'
|
|
' src_update_timestamp FROM backup_tmdb_tv_shows'],
|
|
['REPLACE INTO tv_episodes '
|
|
'(episode_id, showid, indexerid, indexer, name, season, episode, description, airdate, hasnfo, hastbn,'
|
|
' status, location, file_size, release_name, subtitles, subtitles_searchcount, subtitles_lastsearch,'
|
|
' is_proper, scene_season, scene_episode, absolute_number, scene_absolute_number, release_group,'
|
|
' version, timezone, airtime, runtime, timestamp, network, network_country, network_country_code,'
|
|
' network_id, network_is_stream)'
|
|
' SELECT episode_id, showid, indexerid, indexer, name, season, episode, description, airdate, hasnfo,'
|
|
' hastbn, status, location, file_size, release_name, subtitles, subtitles_searchcount,'
|
|
' subtitles_lastsearch, is_proper, scene_season, scene_episode, absolute_number, scene_absolute_number,'
|
|
' release_group, version, timezone, airtime, runtime, timestamp, network, network_country,'
|
|
' network_country_code, network_id, network_is_stream FROM backup_tmdb_tv_episodes'],
|
|
['REPLACE INTO indexer_mapping (indexer_id, indexer, mindexer_id, mindexer, date, status)'
|
|
' SELECT indexer_id, indexer, mindexer_id, mindexer, date, status FROM backup_tmdb_indexer_mapping'],
|
|
])[has_tmdb_backups])
|
|
[self.connection.remove_table(_t) for _t in ('backup_tmdb_tv_shows', 'backup_tmdb_tv_episodes',
|
|
'backup_tmdb_indexer_mapping')]
|
|
|
|
return self.set_db_version(20016)
|
|
|