mirror of
https://github.com/SickGear/SickGear.git
synced 2024-11-21 20:35:05 +00:00
10264 lines
473 KiB
Python
10264 lines
473 KiB
Python
# coding=utf-8
|
|
#
|
|
# 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/>.
|
|
|
|
# noinspection PyProtectedMember
|
|
from datetime import date as dt_date, datetime, time as dt_time, timedelta, timezone
|
|
from mimetypes import MimeTypes
|
|
from urllib.parse import urljoin
|
|
|
|
import base64
|
|
import copy
|
|
import glob
|
|
import hashlib
|
|
import io
|
|
import os
|
|
import random
|
|
import re
|
|
import sys
|
|
import threading
|
|
import time
|
|
import traceback
|
|
import zipfile
|
|
|
|
from exceptions_helper import ex, MultipleShowObjectsException
|
|
import exceptions_helper
|
|
from json_helper import json_dumps, json_loads
|
|
import sg_helpers
|
|
from sg_helpers import remove_file, scantree, is_virtualenv
|
|
|
|
from sg_futures import SgThreadPoolExecutor
|
|
try:
|
|
from multiprocessing import cpu_count
|
|
except ImportError:
|
|
# some platforms don't have multiprocessing
|
|
def cpu_count():
|
|
return None
|
|
|
|
import sickgear
|
|
from . import classes, clients, config, db, helpers, history, image_cache, logger, name_cache, naming, \
|
|
network_timezones, notifiers, nzbget, processTV, sab, scene_exceptions, search_queue, subtitles, ui
|
|
from .anime import AniGroupList, pull_anidb_groups, short_group_names
|
|
from .browser import folders_at_path
|
|
from .common import ARCHIVED, DOWNLOADED, FAILED, IGNORED, SKIPPED, SNATCHED, SNATCHED_ANY, UNAIRED, UNKNOWN, WANTED, \
|
|
SD, HD720p, HD1080p, UHD2160p, Overview, Quality, qualityPresetStrings, statusStrings
|
|
from .helpers import get_media_stats, has_image_ext, real_path, remove_article, remove_file_perm, starify
|
|
from .indexermapper import MapStatus, map_indexers_to_show, save_mapping
|
|
from .indexers.indexer_config import TVINFO_IMDB, TVINFO_TMDB, TVINFO_TRAKT, TVINFO_TVDB, TVINFO_TVMAZE, \
|
|
TVINFO_TRAKT_SLUG, TVINFO_TVDB_SLUG
|
|
from .name_parser.parser import InvalidNameException, InvalidShowException, NameParser
|
|
from .providers import newznab, rsstorrent
|
|
from .scene_numbering import get_scene_absolute_numbering_for_show, get_scene_numbering_for_show, \
|
|
get_xem_absolute_numbering_for_show, get_xem_numbering_for_show, set_scene_numbering_helper
|
|
from .scheduler import Scheduler
|
|
from .search_backlog import FORCED_BACKLOG
|
|
from .sgdatetime import SGDatetime
|
|
from .show_name_helpers import abbr_showname
|
|
|
|
from .show_updater import clean_ignore_require_words
|
|
from .trakt_helpers import build_config, trakt_collection_remove_account
|
|
from .tv import TVidProdid, Person as TVPerson, Character as TVCharacter, TVSWITCH_NORMAL, tvswitch_names, \
|
|
TVSWITCH_EP_DELETED, tvswitch_ep_names, usable_id
|
|
|
|
from bs4_parser import BS4Parser
|
|
# noinspection PyPackageRequirements
|
|
from Cheetah.Template import Template
|
|
from unidecode import unidecode
|
|
import dateutil.parser
|
|
|
|
from tornado import gen, iostream
|
|
from tornado.escape import utf8
|
|
from tornado.web import RequestHandler, StaticFileHandler, authenticated
|
|
from tornado.concurrent import run_on_executor
|
|
|
|
from lib import subliminal
|
|
from lib.cfscrape import CloudflareScraper
|
|
from lib.dateutil import tz, zoneinfo
|
|
from lib.dateutil.relativedelta import relativedelta
|
|
try:
|
|
from lib.thefuzz import fuzz
|
|
except ImportError as e:
|
|
from lib.fuzzywuzzy import fuzz
|
|
from lib.api_trakt import TraktAPI
|
|
from lib.api_trakt.exceptions import TraktException, TraktAuthException
|
|
from lib.tvinfo_base import TVInfoEpisode, RoleTypes
|
|
from lib.tvinfo_base.base import tv_src_names
|
|
|
|
import lib.rarfile.rarfile as rarfile
|
|
|
|
from _23 import decode_bytes, decode_str, getargspec, \
|
|
map_consume, map_none, quote_plus, unquote_plus, urlparse
|
|
from six import binary_type, integer_types, iteritems, iterkeys, itervalues, moves, string_types
|
|
|
|
# noinspection PyUnreachableCode
|
|
if False:
|
|
from typing import Any, AnyStr, Dict, List, Optional, Set, Tuple, Union
|
|
from sickgear.providers.generic import TorrentProvider
|
|
# prevent pyc TVInfoBase resolution by typing the derived used class to TVInfoAPI instantiation
|
|
from lib.tvinfo_base import TVInfoBase, TVInfoCharacter, TVInfoPerson, TVInfoShow
|
|
# from api_imdb.imdb_api import IMDbIndexer
|
|
from api_tmdb.tmdb_api import TmdbIndexer
|
|
from api_trakt.indexerapiinterface import TraktIndexer
|
|
from api_tvmaze.tvmaze_api import TvMaze as TvmazeIndexer
|
|
|
|
|
|
# noinspection PyAbstractClass
|
|
class PageTemplate(Template):
|
|
|
|
def __init__(self, web_handler, *args, **kwargs):
|
|
|
|
headers = web_handler.request.headers
|
|
self.xsrf_form_html = re.sub(r'\s*/>$', '>', web_handler.xsrf_form_html())
|
|
self.sbHost = headers.get('X-Forwarded-Host')
|
|
if None is self.sbHost:
|
|
sb_host = headers.get('Host') or 'localhost'
|
|
self.sbHost = re.match('(?msx)^' + (('[^:]+', r'\[.*\]')['[' == sb_host[0]]), sb_host).group(0)
|
|
self.sbHttpPort = sickgear.WEB_PORT
|
|
self.sbHttpsPort = headers.get('X-Forwarded-Port') or self.sbHttpPort
|
|
self.sbRoot = sickgear.WEB_ROOT
|
|
self.sbHttpsEnabled = 'https' == headers.get('X-Forwarded-Proto') or sickgear.ENABLE_HTTPS
|
|
self.sbHandleReverseProxy = sickgear.HANDLE_REVERSE_PROXY
|
|
self.sbThemeName = sickgear.THEME_NAME
|
|
|
|
self.log_num_errors = len(classes.ErrorViewer.errors)
|
|
if None is not sickgear.showList:
|
|
self.log_num_not_found_shows = len([cur_so for cur_so in sickgear.showList
|
|
if 0 < cur_so.not_found_count])
|
|
self.log_num_not_found_shows_all = len([cur_so for cur_so in sickgear.showList
|
|
if 0 != cur_so.not_found_count])
|
|
self.sbPID = str(sickgear.PID)
|
|
self.menu = [
|
|
{'title': 'Home', 'key': 'home'},
|
|
{'title': 'Episodes', 'key': 'daily-schedule'},
|
|
{'title': 'History', 'key': 'history'},
|
|
{'title': 'Manage', 'key': 'manage'},
|
|
{'title': 'Config', 'key': 'config'},
|
|
]
|
|
|
|
kwargs['file'] = os.path.join(sickgear.PROG_DIR, 'gui/%s/interfaces/default/' %
|
|
sickgear.GUI_NAME, kwargs['file'])
|
|
|
|
self.addtab_limit = sickgear.MEMCACHE.get('history_tab_limit', 0)
|
|
if not web_handler.application.is_loading_handler:
|
|
self.history_compact = sickgear.MEMCACHE.get('history_tab')
|
|
self.tvinfo_switch_running = sickgear.show_queue_scheduler.action.is_switch_running()
|
|
|
|
super(PageTemplate, self).__init__(*args, **kwargs)
|
|
|
|
def compile(self, *args, **kwargs):
|
|
if not os.path.exists(os.path.join(sickgear.CACHE_DIR, 'cheetah')):
|
|
os.mkdir(os.path.join(sickgear.CACHE_DIR, 'cheetah'))
|
|
|
|
kwargs['cacheModuleFilesForTracebacks'] = True
|
|
kwargs['cacheDirForModuleFiles'] = os.path.join(sickgear.CACHE_DIR, 'cheetah')
|
|
return super(PageTemplate, self).compile(*args, **kwargs)
|
|
|
|
|
|
class BaseStaticFileHandler(StaticFileHandler):
|
|
|
|
def write_error(self, status_code, **kwargs):
|
|
body = ''
|
|
try:
|
|
if self.request.body:
|
|
body = '\nRequest body: %s' % decode_str(self.request.body)
|
|
except (BaseException, Exception):
|
|
pass
|
|
logger.warning(f'Sent {status_code} error response to a `{self.request.method}`'
|
|
f' request for `{self.request.path}` with headers:\n'
|
|
f'{self.request.headers}{body}')
|
|
# suppress traceback by removing 'exc_info' kwarg
|
|
if 'exc_info' in kwargs:
|
|
logger.debug('Gracefully handled exception text:\n%s' % traceback.format_exception(*kwargs["exc_info"]))
|
|
del kwargs['exc_info']
|
|
return super(BaseStaticFileHandler, self).write_error(status_code, **kwargs)
|
|
|
|
def validate_absolute_path(self, root, absolute_path):
|
|
if '\\images\\flags\\' in absolute_path and not os.path.isfile(absolute_path):
|
|
absolute_path = re.sub(r'\\[^\\]+\.png$', '\\\\unknown.png', absolute_path)
|
|
return super(BaseStaticFileHandler, self).validate_absolute_path(root, absolute_path)
|
|
|
|
def data_received(self, *args):
|
|
pass
|
|
|
|
def set_extra_headers(self, path):
|
|
self.set_header('X-Robots-Tag', 'noindex, nofollow, noarchive, nocache, noodp, noydir, noimageindex, nosnippet')
|
|
self.set_header('Cache-Control', 'no-cache, max-age=0')
|
|
self.set_header('Pragma', 'no-cache')
|
|
self.set_header('Expires', '0')
|
|
if sickgear.SEND_SECURITY_HEADERS:
|
|
self.set_header('X-Frame-Options', 'SAMEORIGIN')
|
|
|
|
|
|
class RouteHandler(RequestHandler):
|
|
|
|
executor = SgThreadPoolExecutor(thread_name_prefix='WEBSERVER', max_workers=min(32, (cpu_count() or 1) + 4))
|
|
|
|
def redirect(self, url, permanent=False, status=None):
|
|
"""Send a redirect to the given (optionally relative) URL.
|
|
|
|
----->>>>> NOTE: Removed self.finish <<<<<-----
|
|
|
|
If the ``status`` argument is specified, that value is used as the
|
|
HTTP status code; otherwise either 301 (permanent) or 302
|
|
(temporary) is chosen based on the ``permanent`` argument.
|
|
The default is 302 (temporary).
|
|
"""
|
|
if not url.startswith(sickgear.WEB_ROOT):
|
|
url = sickgear.WEB_ROOT + url
|
|
|
|
# noinspection PyUnresolvedReferences
|
|
if self._headers_written:
|
|
raise Exception('Cannot redirect after headers have been written')
|
|
if status is None:
|
|
status = 301 if permanent else 302
|
|
else:
|
|
assert isinstance(status, int)
|
|
assert 300 <= status <= 399
|
|
self.set_status(status)
|
|
self.set_header('Location', urljoin(utf8(self.request.uri), utf8(url)))
|
|
|
|
def write_error(self, status_code, **kwargs):
|
|
body = ''
|
|
try:
|
|
if self.request.body:
|
|
body = '\nRequest body: %s' % decode_str(self.request.body)
|
|
except (BaseException, Exception):
|
|
pass
|
|
logger.warning(f'Sent {status_code} error response to a `{self.request.method}`'
|
|
f' request for `{self.request.path}` with headers:\n{self.request.headers}{body}')
|
|
# suppress traceback by removing 'exc_info' kwarg
|
|
if 'exc_info' in kwargs:
|
|
logger.debug('Gracefully handled exception text:\n%s' % traceback.format_exception(*kwargs["exc_info"]))
|
|
del kwargs['exc_info']
|
|
return super(RouteHandler, self).write_error(status_code, **kwargs)
|
|
|
|
def data_received(self, *args):
|
|
pass
|
|
|
|
def decode_data(self, data):
|
|
if isinstance(data, binary_type):
|
|
return decode_str(data)
|
|
if isinstance(data, list):
|
|
return [self.decode_data(d) for d in data]
|
|
if not isinstance(data, string_types):
|
|
return data
|
|
return data.encode('latin1').decode('utf-8')
|
|
|
|
@gen.coroutine
|
|
def route_method(self, route, use_404=False, limit_route=None, xsrf_filter=True):
|
|
|
|
route = route.strip('/')
|
|
if not route and None is limit_route:
|
|
route = 'index'
|
|
if limit_route:
|
|
route = limit_route(route)
|
|
if '-' in route:
|
|
parts = re.split(r'([/?])', route)
|
|
route = '%s%s' % (parts[0].replace('-', '_'), '' if not len(parts) else ''.join(parts[1:]))
|
|
|
|
try:
|
|
method = getattr(self, route)
|
|
except (BaseException, Exception):
|
|
self.finish(use_404 and self.page_not_found() or None)
|
|
else:
|
|
request_kwargs = {k: self.decode_data(v if not (isinstance(v, list) and 1 == len(v)) else v[0])
|
|
for k, v in iteritems(self.request.arguments) if not xsrf_filter or ('_xsrf' != k)}
|
|
if 'tvid_prodid' in request_kwargs and request_kwargs['tvid_prodid'] in sickgear.switched_shows:
|
|
# in case show has been switched, redirect to new id
|
|
url = self.request.uri.replace('tvid_prodid=%s' % request_kwargs['tvid_prodid'],
|
|
'tvid_prodid=%s' %
|
|
sickgear.switched_shows[request_kwargs['tvid_prodid']])
|
|
self.redirect(url, permanent=True)
|
|
self.finish()
|
|
return
|
|
# filter method specified arguments so *args and **kwargs are not required and unused vars safely dropped
|
|
method_args = []
|
|
# noinspection PyDeprecation
|
|
for arg in list(getargspec(method)):
|
|
if not isinstance(arg, list):
|
|
arg = [arg]
|
|
method_args += [item for item in arg if None is not item]
|
|
if 'kwargs' in method_args or re.search('[A-Z]', route):
|
|
# no filtering for legacy and routes that depend on *args and **kwargs
|
|
result = yield self.async_call(method, request_kwargs) # method(**request_kwargs)
|
|
else:
|
|
filter_kwargs = dict(filter(lambda kv: kv[0] in method_args, iteritems(request_kwargs)))
|
|
result = yield self.async_call(method, filter_kwargs) # method(**filter_kwargs)
|
|
self.finish(result)
|
|
|
|
@run_on_executor
|
|
def async_call(self, function, kw):
|
|
try:
|
|
return function(**kw)
|
|
except (BaseException, Exception) as e:
|
|
raise e
|
|
|
|
def page_not_found(self):
|
|
self.set_status(404)
|
|
t = PageTemplate(web_handler=self, file='404.tmpl')
|
|
return t.respond()
|
|
|
|
|
|
class BaseHandler(RouteHandler):
|
|
|
|
def set_default_headers(self):
|
|
self.set_header('Cache-Control', 'no-store, no-cache, must-revalidate, max-age=0')
|
|
self.set_header('X-Robots-Tag', 'noindex, nofollow, noarchive, nocache, noodp, noydir, noimageindex, nosnippet')
|
|
if sickgear.SEND_SECURITY_HEADERS:
|
|
self.set_header('X-Frame-Options', 'SAMEORIGIN')
|
|
|
|
def redirect(self, url, permanent=False, status=None):
|
|
if not url.startswith(sickgear.WEB_ROOT):
|
|
url = sickgear.WEB_ROOT + url
|
|
|
|
super(BaseHandler, self).redirect(url, permanent, status)
|
|
|
|
def get_current_user(self):
|
|
if sickgear.WEB_USERNAME or sickgear.WEB_PASSWORD:
|
|
return self.get_signed_cookie('sickgear-session-%s' % helpers.md5_for_text(sickgear.WEB_PORT))
|
|
return True
|
|
|
|
def get_image(self, image):
|
|
if os.path.isfile(image):
|
|
mime_type, encoding = MimeTypes().guess_type(image)
|
|
self.set_header('Content-Type', mime_type)
|
|
with open(image, 'rb') as img:
|
|
return img.read()
|
|
|
|
def show_poster(self, tvid_prodid=None, which=None, api=None):
|
|
# Redirect initial poster/banner thumb to default images
|
|
if 'poster' == which[0:6]:
|
|
default_image_name = 'poster.png'
|
|
elif 'banner' == which[0:6]:
|
|
default_image_name = 'banner.png'
|
|
else:
|
|
default_image_name = 'backart.png'
|
|
|
|
static_image_path = os.path.join('/images', default_image_name)
|
|
if helpers.find_show_by_id(tvid_prodid):
|
|
cache_obj = image_cache.ImageCache()
|
|
tvid_prodid_obj = tvid_prodid and TVidProdid(tvid_prodid)
|
|
|
|
image_file_name = []
|
|
if 'poster' == which[0:6]:
|
|
if '_thumb' == which[6:]:
|
|
image_file_name = [cache_obj.poster_thumb_path(*tvid_prodid_obj.tuple)]
|
|
image_file_name += [cache_obj.poster_path(*tvid_prodid_obj.tuple)]
|
|
elif 'banner' == which[0:6]:
|
|
if '_thumb' == which[6:]:
|
|
image_file_name = [cache_obj.banner_thumb_path(*tvid_prodid_obj.tuple)]
|
|
image_file_name += [cache_obj.banner_path(*tvid_prodid_obj.tuple)]
|
|
elif 'fanart' == which[0:6]:
|
|
image_file_name = [cache_obj.fanart_path(
|
|
*tvid_prodid_obj.tuple +
|
|
('%s' % (re.sub(r'.*?fanart_(\d+(?:\.\w{1,20})?\.\w{5,8}).*', r'\1.', which, 0, re.I)),))]
|
|
|
|
for cur_name in image_file_name:
|
|
if os.path.isfile(cur_name):
|
|
static_image_path = cur_name
|
|
break
|
|
|
|
if api:
|
|
used_file = os.path.basename(static_image_path)
|
|
if static_image_path.startswith('/images'):
|
|
used_file = 'default'
|
|
static_image_path = os.path.join(sickgear.PROG_DIR, 'gui', 'slick', static_image_path[1:])
|
|
mime_type, encoding = MimeTypes().guess_type(static_image_path)
|
|
self.set_header('Content-Type', mime_type)
|
|
self.set_header('X-Filename', used_file)
|
|
with open(static_image_path, 'rb') as img:
|
|
return img.read()
|
|
else:
|
|
static_image_path = os.path.normpath(static_image_path.replace(sickgear.CACHE_DIR, '/cache'))
|
|
static_image_path = static_image_path.replace('\\', '/')
|
|
self.redirect(static_image_path)
|
|
|
|
|
|
class LoginHandler(BaseHandler):
|
|
|
|
# noinspection PyUnusedLocal
|
|
def get(self, *args, **kwargs):
|
|
if self.get_current_user():
|
|
self.redirect(self.get_argument('next', '/home/'))
|
|
else:
|
|
t = PageTemplate(web_handler=self, file='login.tmpl')
|
|
t.resp = self.get_argument('resp', '')
|
|
self.set_status(401)
|
|
self.finish(t.respond())
|
|
|
|
# noinspection PyUnusedLocal
|
|
def post(self, *args, **kwargs):
|
|
username = sickgear.WEB_USERNAME
|
|
password = sickgear.WEB_PASSWORD
|
|
|
|
if (self.get_argument('username') == username) and (self.get_argument('password') == password):
|
|
params = dict(expires_days=(None, 30)[0 < int(self.get_argument('remember_me', default='0') or 0)],
|
|
httponly=True)
|
|
if sickgear.ENABLE_HTTPS:
|
|
params.update(dict(secure=True))
|
|
self.set_signed_cookie('sickgear-session-%s' % helpers.md5_for_text(sickgear.WEB_PORT),
|
|
sickgear.COOKIE_SECRET, **params)
|
|
self.redirect(self.get_argument('next', '/home/'))
|
|
else:
|
|
next_arg = '&next=' + self.get_argument('next', '/home/')
|
|
self.redirect('/login?resp=authfailed' + next_arg)
|
|
|
|
|
|
class LogoutHandler(BaseHandler):
|
|
|
|
# noinspection PyUnusedLocal
|
|
def get(self, *args, **kwargs):
|
|
self.clear_cookie('sickgear-session-%s' % helpers.md5_for_text(sickgear.WEB_PORT))
|
|
self.redirect('/login/')
|
|
|
|
|
|
class CalendarHandler(BaseHandler):
|
|
|
|
# noinspection PyUnusedLocal
|
|
def get(self, *args, **kwargs):
|
|
if sickgear.CALENDAR_UNPROTECTED or self.get_current_user():
|
|
self.write(self.calendar())
|
|
else:
|
|
self.set_status(401)
|
|
self.write('User authentication required')
|
|
|
|
def calendar(self):
|
|
""" iCalendar (iCal) - Standard RFC 5546 <https://datatracker.ietf.org/doc/html/rfc5546>
|
|
Works with iCloud, Google Calendar and Outlook.
|
|
Provides a subscribeable URL for iCal subscriptions """
|
|
|
|
logger.log(f'Receiving iCal request from {self.request.remote_ip}')
|
|
|
|
# Limit dates
|
|
past_date = (dt_date.today() + timedelta(weeks=-52)).toordinal()
|
|
future_date = (dt_date.today() + timedelta(weeks=52)).toordinal()
|
|
utc = tz.gettz('GMT', zoneinfo_priority=True)
|
|
|
|
# Get all the shows that are not paused and are currently on air
|
|
my_db = db.DBConnection()
|
|
show_list = my_db.select(
|
|
'SELECT show_name, indexer AS tv_id, indexer_id AS prod_id, network, airs, runtime'
|
|
' FROM tv_shows'
|
|
' WHERE (status = \'Continuing\' OR status = \'Returning Series\' ) AND paused != \'1\'')
|
|
|
|
nl = '\\n\\n'
|
|
crlf = '\r\n'
|
|
|
|
# Create iCal header
|
|
appname = 'SickGear'
|
|
ical = 'BEGIN:VCALENDAR%sVERSION:2.0%sX-WR-CALNAME:%s%sX-WR-CALDESC:%s%sPRODID://%s Upcoming Episodes//%s' \
|
|
% (crlf, crlf, appname, crlf, appname, crlf, appname, crlf)
|
|
|
|
for show in show_list:
|
|
# Get all episodes of this show airing between today and next month
|
|
|
|
episode_list = my_db.select(
|
|
'SELECT name, season, episode, description, airdate'
|
|
' FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' AND airdate >= ? AND airdate < ? ',
|
|
[show['tv_id'], show['prod_id']]
|
|
+ [past_date, future_date])
|
|
|
|
for episode in episode_list:
|
|
air_date_time = network_timezones.parse_date_time(episode['airdate'], show['airs'],
|
|
show['network']).astimezone(utc)
|
|
air_date_time_end = air_date_time + timedelta(minutes=helpers.try_int(show['runtime'], 60))
|
|
|
|
# Create event for episode
|
|
desc = '' if not episode['description'] else f'{nl}{episode["description"].splitlines()[0]}'
|
|
ical += (f'BEGIN:VEVENT{crlf}'
|
|
f'DTSTART:{air_date_time.strftime("%Y%m%d")}T{air_date_time.strftime("%H%M%S")}Z{crlf}'
|
|
f'DTEND:{air_date_time_end.strftime("%Y%m%d")}T{air_date_time_end.strftime("%H%M%S")}Z{crlf}'
|
|
f'SUMMARY:{show["show_name"]} - {episode["season"]}x{episode["episode"]}'
|
|
f' - {episode["name"]}{crlf}'
|
|
f'UID:{appname}-{dt_date.today().isoformat()}-{show["show_name"].replace(" ", "-")}'
|
|
f'-E{episode["episode"]}S{episode["season"]}{crlf}'
|
|
f'DESCRIPTION:{(show["airs"] or "(Unknown airs)")} on {(show["network"] or "Unknown network")}'
|
|
f'{desc}{crlf}'
|
|
f'END:VEVENT{crlf}')
|
|
|
|
# Ending the iCal
|
|
return ical + 'END:VCALENDAR'
|
|
|
|
|
|
class RepoHandler(BaseStaticFileHandler):
|
|
kodi_include = None
|
|
kodi_exclude = None
|
|
kodi_legacy = None
|
|
kodi_is_legacy = None
|
|
|
|
def parse_url_path(self, url_path):
|
|
logger.debug('Kodi req... get(path): %s' % url_path)
|
|
return super(RepoHandler, self).parse_url_path(url_path)
|
|
|
|
def set_extra_headers(self, *args, **kwargs):
|
|
super(RepoHandler, self).set_extra_headers(*args, **kwargs)
|
|
self.set_header('Cache-Control', 'no-store, no-cache, must-revalidate, max-age=0')
|
|
|
|
def initialize(self, *args, **kwargs):
|
|
self.kodi_legacy = '-helix-leia'
|
|
self.kodi_exclude = '' if kwargs.get('legacy') else self.kodi_legacy
|
|
self.kodi_include = '' if not kwargs.pop('legacy', None) else self.kodi_legacy
|
|
self.kodi_is_legacy = bool(self.kodi_include)
|
|
|
|
super(RepoHandler, self).initialize(*args, **kwargs)
|
|
|
|
logger.debug('Kodi req... initialize(path): %s' % kwargs['path'])
|
|
cache_client = os.path.join(sickgear.CACHE_DIR, 'clients')
|
|
cache_client_kodi = os.path.join(cache_client, 'kodi')
|
|
cache_client_kodi_watchedstate = os.path.join(cache_client_kodi, 'service.sickgear.watchedstate.updater')
|
|
|
|
cache_resources = os.path.join(cache_client_kodi_watchedstate, 'resources')
|
|
cache_lang = os.path.join(cache_resources, 'language')
|
|
cache_other_lang = os.path.join(cache_lang, ('English', 'resource.language.en_gb')[self.kodi_is_legacy])
|
|
os.path.exists(cache_other_lang) and remove_file(cache_other_lang, tree=True)
|
|
|
|
cache_lang_sub = os.path.join(cache_lang, ('resource.language.en_gb', 'English')[self.kodi_is_legacy])
|
|
for folder in (cache_client,
|
|
cache_client_kodi,
|
|
os.path.join(cache_client_kodi, 'repository.sickgear'),
|
|
cache_client_kodi_watchedstate,
|
|
os.path.join(cache_resources),
|
|
cache_lang, cache_lang_sub,
|
|
):
|
|
if not os.path.exists(folder):
|
|
os.mkdir(folder)
|
|
|
|
with io.open(os.path.join(cache_client_kodi, 'index.html'), 'w') as fh:
|
|
fh.write(self.render_kodi_index())
|
|
with io.open(os.path.join(cache_client_kodi, 'repository.sickgear', 'index.html'), 'w') as fh:
|
|
fh.write(self.render_kodi_repository_sickgear_index())
|
|
with io.open(os.path.join(cache_client_kodi_watchedstate, 'index.html'), 'w') as fh:
|
|
fh.write(self.render_kodi_service_sickgear_watchedstate_updater_index())
|
|
with io.open(os.path.join(cache_resources, 'index.html'), 'w') as fh:
|
|
fh.write(self.render_kodi_service_sickgear_watchedstate_updater_resources_index())
|
|
with io.open(os.path.join(cache_lang, 'index.html'), 'w') as fh:
|
|
fh.write(self.render_kodi_service_sickgear_watchedstate_updater_resources_language_index())
|
|
with io.open(os.path.join(cache_lang_sub, 'index.html'), 'w') as fh:
|
|
fh.write(self.render_kodi_service_sickgear_watchedstate_updater_resources_language_english_index())
|
|
|
|
'''
|
|
|
|
if add-on rendered md5 changes, update its zip and then flag to update repo addon
|
|
if repo rendered md5 changes or flag is true, update the repo addon, where repo version *must* be increased
|
|
|
|
'''
|
|
repo_md5_file = os.path.join(cache_client_kodi, 'addons.xml.md5')
|
|
saved_md5 = None
|
|
try:
|
|
with io.open(repo_md5_file, 'r', encoding='utf8') as fh:
|
|
saved_md5 = fh.readline()
|
|
except (BaseException, Exception):
|
|
pass
|
|
rendered_md5 = self.render_kodi_repo_addons_xml_md5()
|
|
if saved_md5 != rendered_md5:
|
|
with io.open(os.path.join(cache_client_kodi, 'repository.sickgear', 'addon.xml'), 'w') as fh:
|
|
fh.write(self.render_kodi_repo_addon_xml())
|
|
with io.open(os.path.join(cache_client_kodi_watchedstate, 'addon.xml'), 'w') as fh:
|
|
fh.write(self.get_watchedstate_updater_addon_xml())
|
|
with io.open(os.path.join(cache_client_kodi, 'addons.xml'), 'w') as fh:
|
|
fh.write(self.render_kodi_repo_addons_xml())
|
|
with io.open(os.path.join(cache_client_kodi, 'addons.xml.md5'), 'w') as fh:
|
|
fh.write(rendered_md5)
|
|
|
|
def save_zip(name, version, zip_path, zip_method):
|
|
zip_name = '%s-%s.zip' % (name, version)
|
|
zip_file = os.path.join(zip_path, zip_name)
|
|
for direntry in helpers.scantree(zip_path, ['resources'], [r'\.(?:md5|zip)$'], filter_kind=False):
|
|
remove_file_perm(direntry.path)
|
|
zip_data = zip_method()
|
|
with io.open(zip_file, 'wb') as zh:
|
|
zh.write(zip_data)
|
|
|
|
# Force a UNIX line ending, like the md5sum utility.
|
|
with io.open(os.path.join(zip_path, '%s.md5' % zip_name), 'w', newline='\n') as zh:
|
|
zh.write(f'{self.md5ify(zip_data)} *{zip_name}\n')
|
|
|
|
aid, ver = self.repo_sickgear_details()
|
|
save_zip(aid, ver, os.path.join(cache_client_kodi, 'repository.sickgear'),
|
|
self.kodi_repository_sickgear_zip)
|
|
|
|
aid, ver = self.addon_watchedstate_details()
|
|
save_zip(aid, ver, cache_client_kodi_watchedstate,
|
|
self.kodi_service_sickgear_watchedstate_updater_zip)
|
|
|
|
wsu_path = 'service.sickgear.watchedstate.updater'
|
|
for (src, dst) in (
|
|
(('repository.sickgear', 'icon.png'),
|
|
(cache_client_kodi, 'repository.sickgear', 'icon.png')),
|
|
((wsu_path, 'icon.png'),
|
|
(cache_client_kodi_watchedstate, 'icon.png')),
|
|
((wsu_path, 'resources', 'settings%s.xml' % self.kodi_include),
|
|
(cache_resources, 'settings%s.xml' % self.kodi_include.replace(self.kodi_legacy, ''))),
|
|
((wsu_path, 'icon.png'),
|
|
(cache_resources, 'icon.png')),
|
|
(((wsu_path, 'resources', 'language', 'resource.language.en_gb', 'strings.po'),
|
|
(cache_lang_sub, 'strings.po')),
|
|
((wsu_path, 'resources', 'language', 'English', 'strings.xml'),
|
|
(cache_lang_sub, 'strings.xml')
|
|
))[self.kodi_is_legacy],
|
|
):
|
|
helpers.copy_file(
|
|
os.path.join(*(sickgear.PROG_DIR, 'sickgear', 'clients', 'kodi') + src), os.path.join(*dst))
|
|
|
|
def get_content_type(self):
|
|
if '.md5' == self.absolute_path[-4:] or '.po' == self.absolute_path[-3:]:
|
|
return 'text/plain'
|
|
return super(RepoHandler, self).get_content_type()
|
|
|
|
def index(self, filelist):
|
|
t = PageTemplate(web_handler=self, file='repo_index.tmpl')
|
|
t.basepath = self.request.path
|
|
t.kodi_is_legacy = self.kodi_is_legacy
|
|
t.filelist = filelist
|
|
t.repo = '%s-%s.zip' % self.repo_sickgear_details()
|
|
t.addon = '%s-%s.zip' % self.addon_watchedstate_details()
|
|
|
|
try:
|
|
with open(os.path.join(sickgear.PROG_DIR, 'CHANGES.md')) as fh:
|
|
t.version = re.findall(r'###[^0-9x]+([0-9]+\.[0-9]+\.[0-9x]+)', fh.readline())[0]
|
|
except (BaseException, Exception):
|
|
t.version = ''
|
|
|
|
return t.respond()
|
|
|
|
def render_kodi_index(self):
|
|
return self.index(['repository.sickgear/',
|
|
'service.sickgear.watchedstate.updater/',
|
|
'addons.xml',
|
|
'addons.xml.md5',
|
|
])
|
|
|
|
def render_kodi_repository_sickgear_index(self):
|
|
aid, version = self.repo_sickgear_details()
|
|
return self.index(['addon.xml',
|
|
'icon.png',
|
|
'%s-%s.zip' % (aid, version),
|
|
'%s-%s.zip.md5' % (aid, version),
|
|
])
|
|
|
|
def render_kodi_service_sickgear_watchedstate_updater_index(self):
|
|
aid, version = self.addon_watchedstate_details()
|
|
return self.index(['resources/',
|
|
'addon.xml',
|
|
'icon.png',
|
|
'%s-%s.zip' % (aid, version),
|
|
'%s-%s.zip.md5' % (aid, version),
|
|
])
|
|
|
|
def render_kodi_service_sickgear_watchedstate_updater_resources_index(self):
|
|
return self.index(['language/',
|
|
'settings.xml',
|
|
'icon.png',
|
|
])
|
|
|
|
def render_kodi_service_sickgear_watchedstate_updater_resources_language_index(self):
|
|
return self.index([('resource.language.en_gb/', 'English/')[self.kodi_is_legacy]])
|
|
|
|
def render_kodi_service_sickgear_watchedstate_updater_resources_language_english_index(self):
|
|
return self.index([('strings.po', 'strings.xml')[self.kodi_is_legacy]])
|
|
|
|
def repo_sickgear_details(self):
|
|
return re.findall(r'(?si)addon\sid="(repository\.[^"]+)[^>]+version="([^"]+)',
|
|
self.render_kodi_repo_addon_xml())[0]
|
|
|
|
def addon_watchedstate_details(self):
|
|
return re.findall(r'(?si)addon\sid="([^"]+)[^>]+version="([^"]+)',
|
|
self.get_watchedstate_updater_addon_xml())[0]
|
|
|
|
def get_watchedstate_updater_addon_xml(self):
|
|
mem_key = 'kodi_xml'
|
|
if SGDatetime.timestamp_near() < sickgear.MEMCACHE.get(mem_key, {}).get('last_update', 0):
|
|
return sickgear.MEMCACHE.get(mem_key).get('data')
|
|
|
|
filename = 'addon%s.xml' % self.kodi_include
|
|
with io.open(os.path.join(sickgear.PROG_DIR, 'sickgear', 'clients', 'kodi',
|
|
'service.sickgear.watchedstate.updater', filename), 'r', encoding='utf8') as fh:
|
|
xml = fh.read().strip() % dict(ADDON_VERSION=self.get_addon_version(self.kodi_include))
|
|
|
|
sickgear.MEMCACHE[mem_key] = dict(last_update=30 + SGDatetime.timestamp_near(), data=xml)
|
|
return xml
|
|
|
|
@staticmethod
|
|
def get_addon_version(kodi_include):
|
|
"""
|
|
:param kodi_include: kodi variant to use
|
|
:type kodi_include: AnyStr
|
|
:return: Version of addon
|
|
:rtype: AnyStr
|
|
|
|
Must use an arg here instead of `self` due to static call use case from external class
|
|
"""
|
|
mem_key = 'kodi_ver'
|
|
if SGDatetime.timestamp_near() < sickgear.MEMCACHE.get(mem_key, {}).get('last_update', 0):
|
|
return sickgear.MEMCACHE.get(mem_key).get('data')
|
|
|
|
filename = 'service%s.py' % kodi_include
|
|
with io.open(os.path.join(sickgear.PROG_DIR, 'sickgear', 'clients', 'kodi',
|
|
'service.sickgear.watchedstate.updater', filename), 'r', encoding='utf8') as fh:
|
|
version = re.findall(r'ADDON_VERSION\s*?=\s*?\'([^\']+)', fh.read())[0]
|
|
|
|
sickgear.MEMCACHE[mem_key] = dict(last_update=30 + SGDatetime.timestamp_near(), data=version)
|
|
return version
|
|
|
|
def render_kodi_repo_addon_xml(self):
|
|
t = PageTemplate(web_handler=self, file='repo_kodi_addon.tmpl')
|
|
|
|
t.endpoint = 'kodi' + ('', '-legacy')[self.kodi_is_legacy]
|
|
|
|
return re.sub(r'#\s.*\n', '', t.respond())
|
|
|
|
def render_kodi_repo_addons_xml(self):
|
|
t = PageTemplate(web_handler=self, file='repo_kodi_addons.tmpl')
|
|
# noinspection PyTypeChecker
|
|
t.watchedstate_updater_addon_xml = re.sub(
|
|
r'(?m)^(\s*<)', r'\t\1',
|
|
'\n'.join(self.get_watchedstate_updater_addon_xml().split('\n')[1:])) # skip xml header
|
|
|
|
t.repo_xml = re.sub(
|
|
r'(?m)^(\s*<)', r'\t\1',
|
|
'\n'.join(self.render_kodi_repo_addon_xml().split('\n')[1:]))
|
|
|
|
return t.respond()
|
|
|
|
def render_kodi_repo_addons_xml_md5(self):
|
|
return self.md5ify('\n'.join(self.render_kodi_repo_addons_xml().split('\n')[1:]))
|
|
|
|
@staticmethod
|
|
def md5ify(string):
|
|
if not isinstance(string, binary_type):
|
|
string = string.encode('utf-8')
|
|
return f'{hashlib.new("md5", string).hexdigest()}'
|
|
|
|
def kodi_repository_sickgear_zip(self):
|
|
bfr = io.BytesIO()
|
|
|
|
try:
|
|
with zipfile.ZipFile(bfr, 'w') as zh:
|
|
zh.writestr('repository.sickgear/addon.xml', self.render_kodi_repo_addon_xml(), zipfile.ZIP_DEFLATED)
|
|
|
|
with io.open(os.path.join(sickgear.PROG_DIR, 'sickgear', 'clients', 'kodi',
|
|
'repository.sickgear', 'icon.png'), 'rb') as fh:
|
|
infile = fh.read()
|
|
zh.writestr('repository.sickgear/icon.png', infile, zipfile.ZIP_DEFLATED)
|
|
except OSError as e:
|
|
logger.warning('Unable to zip: %r / %s' % (e, ex(e)))
|
|
|
|
zip_data = bfr.getvalue()
|
|
bfr.close()
|
|
return zip_data
|
|
|
|
def kodi_service_sickgear_watchedstate_updater_zip(self):
|
|
bfr = io.BytesIO()
|
|
|
|
basepath = os.path.join(sickgear.PROG_DIR, 'sickgear', 'clients', 'kodi')
|
|
|
|
zip_path = os.path.join(basepath, 'service.sickgear.watchedstate.updater')
|
|
devenv_src = os.path.join(sickgear.PROG_DIR, 'tests', '_devenv.py')
|
|
devenv_dst = os.path.join(zip_path, '_devenv.py')
|
|
if sickgear.ENV.get('DEVENV') and os.path.exists(devenv_src):
|
|
helpers.copy_file(devenv_src, devenv_dst)
|
|
else:
|
|
helpers.remove_file_perm(devenv_dst)
|
|
|
|
for direntry in helpers.scantree(zip_path,
|
|
exclude=[r'\.xcf$',
|
|
'addon%s.xml$' % self.kodi_exclude,
|
|
'settings%s.xml$' % self.kodi_exclude,
|
|
'service%s.py' % self.kodi_exclude,
|
|
('^strings.xml$', r'\.po$')[self.kodi_is_legacy]],
|
|
filter_kind=False):
|
|
try:
|
|
infile = None
|
|
filename = 'addon%s.xml' % self.kodi_include
|
|
if 'service.sickgear.watchedstate.updater' in direntry.path and direntry.path.endswith(filename):
|
|
infile = self.get_watchedstate_updater_addon_xml()
|
|
if not infile:
|
|
with io.open(direntry.path, 'rb') as fh:
|
|
infile = fh.read()
|
|
|
|
with zipfile.ZipFile(bfr, 'a') as zh:
|
|
zh.writestr(os.path.relpath(direntry.path.replace(self.kodi_legacy, ''), basepath),
|
|
infile, zipfile.ZIP_DEFLATED)
|
|
except OSError as e:
|
|
logger.warning('Unable to zip %s: %r / %s' % (direntry.path, e, ex(e)))
|
|
|
|
zip_data = bfr.getvalue()
|
|
bfr.close()
|
|
return zip_data
|
|
|
|
|
|
class NoXSRFHandler(RouteHandler):
|
|
|
|
def __init__(self, *arg, **kwargs):
|
|
self.kodi_include = '' if not kwargs.pop('legacy', None) else '-helix-leia'
|
|
super(NoXSRFHandler, self).__init__(*arg, **kwargs)
|
|
self.lock = threading.Lock()
|
|
|
|
def check_xsrf_cookie(self):
|
|
pass
|
|
|
|
# noinspection PyUnusedLocal
|
|
@gen.coroutine
|
|
def post(self, route, *args, **kwargs):
|
|
|
|
yield self.route_method(route, limit_route=False, xsrf_filter=False)
|
|
|
|
def update_watched_state_kodi(self, payload=None, as_json=True, **kwargs):
|
|
data = {}
|
|
try:
|
|
data = json_loads(payload)
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
mapped = 0
|
|
mapping = None
|
|
maps = [x.split('=') for x in sickgear.KODI_PARENT_MAPS.split(',') if any(x)]
|
|
for k, d in iteritems(data):
|
|
try:
|
|
d['label'] = '%s%s{Kodi}' % (d['label'], bool(d['label']) and ' ' or '')
|
|
except (BaseException, Exception):
|
|
return
|
|
try:
|
|
d['played'] = 100 * int(d['played'])
|
|
except (BaseException, Exception):
|
|
d['played'] = 0
|
|
|
|
for m in maps:
|
|
result, change = helpers.path_mapper(m[0], m[1], d['path_file'])
|
|
if change:
|
|
if not mapping:
|
|
mapping = (d['path_file'], result)
|
|
mapped += 1
|
|
d['path_file'] = result
|
|
break
|
|
|
|
if mapping:
|
|
logger.log('Folder mappings used, the first of %s is [%s] in Kodi is [%s] in SickGear' %
|
|
(mapped, mapping[0], mapping[1]))
|
|
|
|
req_version = tuple([int(x) for x in kwargs.get('version', '0.0.0').split('.')])
|
|
this_version = RepoHandler.get_addon_version(self.kodi_include)
|
|
if not kwargs or (req_version < tuple([int(x) for x in this_version.split('.')])):
|
|
logger.log('Kodi Add-on update available. To upgrade to version %s; '
|
|
'select "Check for updates" on menu of "SickGear Add-on repository"' % this_version)
|
|
|
|
return MainHandler.update_watched_state(data, as_json)
|
|
|
|
|
|
class IsAliveHandler(BaseHandler):
|
|
|
|
# noinspection PyUnusedLocal
|
|
@gen.coroutine
|
|
def get(self, *args, **kwargs):
|
|
kwargs = self.request.arguments
|
|
if 'callback' in kwargs and '_' in kwargs:
|
|
callback, _ = kwargs['callback'][0], kwargs['_']
|
|
else:
|
|
self.write('Error: Unsupported Request. Send jsonp request with callback variable in the query string.')
|
|
return
|
|
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
self.set_header('Content-Type', 'text/javascript')
|
|
self.set_header('Access-Control-Allow-Origin', '*')
|
|
self.set_header('Access-Control-Allow-Headers', 'x-requested-with')
|
|
|
|
if sickgear.started:
|
|
results = decode_str(callback) + '(' + json_dumps(
|
|
{'msg': str(sickgear.PID)}) + ');'
|
|
else:
|
|
results = decode_str(callback) + '(' + json_dumps({'msg': 'nope'}) + ');'
|
|
|
|
self.write(results)
|
|
|
|
|
|
class WrongHostWebHandler(BaseHandler):
|
|
def __init__(self, *arg, **kwargs):
|
|
super(BaseHandler, self).__init__(*arg, **kwargs)
|
|
self.lock = threading.Lock()
|
|
|
|
@gen.coroutine
|
|
def prepare(self):
|
|
self.send_error(404)
|
|
|
|
|
|
class LoadingWebHandler(BaseHandler):
|
|
|
|
def __init__(self, *arg, **kwargs):
|
|
super(BaseHandler, self).__init__(*arg, **kwargs)
|
|
self.lock = threading.Lock()
|
|
|
|
def loading_page(self):
|
|
t = PageTemplate(web_handler=self, file='loading.tmpl')
|
|
t.message = classes.loading_msg.message
|
|
return t.respond()
|
|
|
|
@staticmethod
|
|
def get_message():
|
|
return json_dumps({'message': classes.loading_msg.message})
|
|
|
|
# noinspection PyUnusedLocal
|
|
@authenticated
|
|
@gen.coroutine
|
|
def get(self, route, *args, **kwargs):
|
|
yield self.route_method(route, use_404=True, limit_route=(
|
|
lambda _route: not re.search('get[_-]message', _route) and 'loading-page' or _route))
|
|
|
|
post = get
|
|
|
|
|
|
class LogfileHandler(BaseHandler):
|
|
|
|
def __init__(self, application, request, **kwargs):
|
|
super(LogfileHandler, self).__init__(application, request, **kwargs)
|
|
self.lock = threading.Lock()
|
|
|
|
# noinspection PyUnusedLocal
|
|
@authenticated
|
|
@gen.coroutine
|
|
def get(self, *args, **kwargs):
|
|
logfile_name = logger.current_log_file()
|
|
|
|
try:
|
|
self.set_header('Content-Type', 'text/html; charset=utf-8')
|
|
self.set_header('Content-Description', 'Logfile Download')
|
|
self.set_header('Content-Disposition', 'attachment; filename=sickgear.log')
|
|
# self.set_header('Content-Length', os.path.getsize(logfile_name))
|
|
auths = sickgear.GenericProvider.dedupe_auths(True)
|
|
rxc_auths = re.compile('(?i)%s' % '|'.join([(re.escape(_a)) for _a in auths]))
|
|
replacements = dict([(_a, starify(_a)) for _a in auths])
|
|
data_to_write = ''
|
|
with io.open(logfile_name, 'rt', encoding='utf8') as logfile:
|
|
while 1:
|
|
# read 1M bytes of line + up to next line
|
|
data_lines = logfile.readlines(1000000)
|
|
if not data_lines:
|
|
return
|
|
line_count = len(data_lines)
|
|
for l_n, cur_line in enumerate(data_lines, 1):
|
|
# noinspection HttpUrlsUsage
|
|
if 'https://' in cur_line or 'http://' in cur_line:
|
|
for cur_change in rxc_auths.finditer(cur_line):
|
|
cur_line = '%s%s%s' % (cur_line[:cur_change.start()],
|
|
replacements[cur_line[cur_change.start():cur_change.end()]],
|
|
cur_line[cur_change.end():])
|
|
data_to_write += cur_line
|
|
if 10000 < len(data_to_write) or l_n == line_count:
|
|
try:
|
|
self.write(data_to_write)
|
|
data_to_write = ''
|
|
yield self.flush()
|
|
except iostream.StreamClosedError:
|
|
return
|
|
finally:
|
|
# pause the coroutine so other handlers can run
|
|
yield gen.sleep(0.000000001)
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
|
|
class WebHandler(BaseHandler):
|
|
|
|
def __init__(self, *arg, **kwargs):
|
|
super(BaseHandler, self).__init__(*arg, **kwargs)
|
|
self.lock = threading.Lock()
|
|
|
|
@authenticated
|
|
@gen.coroutine
|
|
def get(self, route, *args, **kwargs):
|
|
yield self.route_method(route, use_404=True)
|
|
|
|
def send_message(self, message):
|
|
with self.lock:
|
|
self.write(message)
|
|
self.flush()
|
|
|
|
post = get
|
|
|
|
|
|
class MainHandler(WebHandler):
|
|
|
|
def index(self):
|
|
self.redirect('/home/')
|
|
|
|
@staticmethod
|
|
def http_error_401_handler():
|
|
""" Custom handler for 401 error """
|
|
return r'''<!DOCTYPE html>
|
|
<html>
|
|
<head>
|
|
<title>%s</title>
|
|
</head>
|
|
<body>
|
|
<br/>
|
|
<font color="#0000FF">Error %s: You need to provide a valid username and password.</font>
|
|
</body>
|
|
</html>
|
|
''' % ('Access denied', 401)
|
|
|
|
def write_error(self, status_code, **kwargs):
|
|
if 401 == status_code:
|
|
self.finish(self.http_error_401_handler())
|
|
elif 404 == status_code:
|
|
self.redirect(sickgear.WEB_ROOT + '/home/')
|
|
elif self.settings.get('debug') and 'exc_info' in kwargs:
|
|
exc_info = kwargs['exc_info']
|
|
trace_info = ''.join(['%s<br/>' % line for line in traceback.format_exception(*exc_info)])
|
|
request_info = ''.join(['<strong>%s</strong>: %s<br/>' % (k, self.request.__dict__[k]) for k in
|
|
iterkeys(self.request.__dict__)])
|
|
error = exc_info[1]
|
|
|
|
self.set_header('Content-Type', 'text/html')
|
|
self.finish('''<html>
|
|
<title>%s</title>
|
|
<body>
|
|
<h2>Error</h2>
|
|
<p>%s</p>
|
|
<h2>Traceback</h2>
|
|
<p>%s</p>
|
|
<h2>Request Info</h2>
|
|
<p>%s</p>
|
|
</body>
|
|
</html>''' % (error, error,
|
|
trace_info, request_info))
|
|
|
|
def robots_txt(self):
|
|
""" Keep web crawlers out """
|
|
self.set_header('Content-Type', 'text/plain')
|
|
return 'User-agent: *\nDisallow: /'
|
|
|
|
def set_layout_view_shows(self, layout):
|
|
|
|
if layout not in ('poster', 'small', 'banner', 'simple'):
|
|
layout = 'poster'
|
|
|
|
sickgear.HOME_LAYOUT = layout
|
|
|
|
self.redirect('/view-shows/')
|
|
|
|
@staticmethod
|
|
def set_display_show_glide(slidetime=None, tvid_prodid=None, start_at=None):
|
|
|
|
if tvid_prodid and start_at:
|
|
sickgear.DISPLAY_SHOW_GLIDE.setdefault(tvid_prodid, {}).update({'start_at': start_at})
|
|
|
|
if slidetime:
|
|
sickgear.DISPLAY_SHOW_GLIDE_SLIDETIME = sg_helpers.try_int(slidetime, 3000)
|
|
sickgear.save_config()
|
|
|
|
@staticmethod
|
|
def set_poster_sortby(sort):
|
|
|
|
if sort not in ('name', 'date', 'network', 'progress', 'quality'):
|
|
sort = 'name'
|
|
|
|
sickgear.POSTER_SORTBY = sort
|
|
sickgear.save_config()
|
|
|
|
@staticmethod
|
|
def set_poster_sortdir(direction):
|
|
|
|
sickgear.POSTER_SORTDIR = int(direction)
|
|
sickgear.save_config()
|
|
|
|
def view_shows(self):
|
|
return Home(self.application, self.request).view_shows()
|
|
|
|
def set_layout_daily_schedule(self, layout):
|
|
if layout not in ('poster', 'banner', 'list', 'daybyday'):
|
|
layout = 'banner'
|
|
|
|
if 'daybyday' == layout:
|
|
sickgear.EPISODE_VIEW_SORT = 'time'
|
|
|
|
sickgear.EPISODE_VIEW_LAYOUT = layout
|
|
|
|
sickgear.save_config()
|
|
|
|
self.redirect('/daily-schedule/')
|
|
|
|
def set_display_paused_daily_schedule(self, state=True):
|
|
|
|
sickgear.EPISODE_VIEW_DISPLAY_PAUSED = sg_helpers.try_int(state, 1)
|
|
|
|
sickgear.save_config()
|
|
|
|
self.redirect('/daily-schedule/')
|
|
|
|
def set_cards_daily_schedule(self, redir=0):
|
|
|
|
sickgear.EPISODE_VIEW_POSTERS = not sickgear.EPISODE_VIEW_POSTERS
|
|
|
|
sickgear.save_config()
|
|
|
|
if int(redir):
|
|
self.redirect('/daily-schedule/')
|
|
|
|
def set_sort_daily_schedule(self, sort, redir=1):
|
|
if sort not in ('time', 'network', 'show'):
|
|
sort = 'time'
|
|
|
|
sickgear.EPISODE_VIEW_SORT = sort
|
|
|
|
sickgear.save_config()
|
|
|
|
if int(redir):
|
|
self.redirect('/daily-schedule/')
|
|
|
|
@staticmethod
|
|
def get_daily_schedule():
|
|
# type: (...) -> Tuple[List[Dict], Dict, Dict, dt_date, integer_types, integer_types]
|
|
""" display the episodes """
|
|
today_dt = dt_date.today()
|
|
today = today_dt.toordinal()
|
|
yesterday_dt = today_dt - timedelta(days=1)
|
|
yesterday = yesterday_dt.toordinal()
|
|
tomorrow = (dt_date.today() + timedelta(days=1)).toordinal()
|
|
next_week_dt = (dt_date.today() + timedelta(days=7))
|
|
next_week = (next_week_dt + timedelta(days=1)).toordinal()
|
|
recently = (yesterday_dt - timedelta(days=sickgear.EPISODE_VIEW_MISSED_RANGE)).toordinal()
|
|
|
|
done_show_list = []
|
|
qualities = Quality.SNATCHED + Quality.DOWNLOADED + Quality.ARCHIVED + [IGNORED, SKIPPED]
|
|
|
|
my_db = db.DBConnection()
|
|
sql_result = my_db.select(
|
|
'SELECT *, tv_episodes.network as episode_network, tv_shows.status AS show_status,'
|
|
' tv_shows.network as show_network, tv_shows.timezone as show_timezone, tv_shows.airtime as show_airtime,'
|
|
' tv_episodes.timezone as ep_timezone, tv_episodes.airtime as ep_airtime'
|
|
' FROM tv_episodes, tv_shows'
|
|
' WHERE tv_shows.indexer = tv_episodes.indexer AND tv_shows.indexer_id = tv_episodes.showid'
|
|
' AND season != 0 AND airdate >= ? AND airdate <= ?'
|
|
' AND tv_episodes.status NOT IN (%s)' % ','.join(['?'] * len(qualities)),
|
|
[yesterday, next_week] + qualities)
|
|
|
|
for cur_result in sql_result:
|
|
done_show_list.append('%s-%s' % (cur_result['indexer'], cur_result['showid']))
|
|
|
|
# noinspection SqlRedundantOrderingDirection
|
|
sql_result += my_db.select(
|
|
'SELECT *, outer_eps.network as episode_network, tv_shows.status AS show_status,'
|
|
' tv_shows.network as show_network, tv_shows.timezone as show_timezone, tv_shows.airtime as show_airtime,'
|
|
' outer_eps.timezone as ep_timezone, outer_eps.airtime as ep_airtime'
|
|
' FROM tv_episodes outer_eps, tv_shows'
|
|
' WHERE season != 0'
|
|
' AND tv_shows.indexer || \'-\' || showid NOT IN (%s)' % ','.join(done_show_list)
|
|
+ ' AND tv_shows.indexer = outer_eps.indexer AND tv_shows.indexer_id = outer_eps.showid'
|
|
' AND airdate = (SELECT airdate FROM tv_episodes inner_eps'
|
|
' WHERE inner_eps.season != 0'
|
|
' AND inner_eps.indexer = outer_eps.indexer AND inner_eps.showid = outer_eps.showid'
|
|
' AND inner_eps.airdate >= ?'
|
|
' ORDER BY inner_eps.airdate ASC LIMIT 1) AND outer_eps.status NOT IN (%s)'
|
|
% ','.join(['?'] * len(Quality.SNATCHED + Quality.DOWNLOADED)),
|
|
[next_week] + Quality.SNATCHED + Quality.DOWNLOADED)
|
|
|
|
sql_result += my_db.select(
|
|
'SELECT *, tv_episodes.network as episode_network, tv_shows.status AS show_status,'
|
|
' tv_shows.network as show_network, tv_shows.timezone as show_timezone, tv_shows.airtime as show_airtime,'
|
|
' tv_episodes.timezone as ep_timezone, tv_episodes.airtime as ep_airtime'
|
|
' FROM tv_episodes, tv_shows'
|
|
' WHERE season != 0'
|
|
' AND tv_shows.indexer = tv_episodes.indexer AND tv_shows.indexer_id = tv_episodes.showid'
|
|
' AND airdate <= ? AND airdate >= ? AND tv_episodes.status = ? AND tv_episodes.status NOT IN (%s)'
|
|
% ','.join(['?'] * len(qualities)),
|
|
[tomorrow, recently, WANTED] + qualities)
|
|
sql_result = list(set(sql_result))
|
|
|
|
# make a dict out of the sql results
|
|
sql_result = [dict(row) for row in sql_result
|
|
if Quality.split_composite_status(helpers.try_int(row['status']))[0] not in
|
|
SNATCHED_ANY + [DOWNLOADED, ARCHIVED, IGNORED, SKIPPED]]
|
|
|
|
# multi dimension sort
|
|
sorts = {
|
|
'network': lambda a: (a['data_network'], a['localtime'], a['data_show_name'], a['season'], a['episode']),
|
|
'show': lambda a: (a['data_show_name'], a['localtime'], a['season'], a['episode']),
|
|
'time': lambda a: (a['localtime'], a['data_show_name'], a['season'], a['episode'])
|
|
}
|
|
|
|
def value_maybe_article(value=None):
|
|
if None is value:
|
|
return ''
|
|
return (remove_article(value.lower()), value.lower())[sickgear.SORT_ARTICLE]
|
|
|
|
# add localtime to the dict
|
|
cache_obj = image_cache.ImageCache()
|
|
fanarts = {}
|
|
cur_prodid = None
|
|
for index, item in enumerate(sql_result):
|
|
tvid_prodid_obj = TVidProdid({item['indexer']: item['showid']})
|
|
tvid_prodid = str(tvid_prodid_obj)
|
|
sql_result[index]['tv_id'] = item['indexer']
|
|
sql_result[index]['prod_id'] = item['showid']
|
|
sql_result[index]['tvid_prodid'] = tvid_prodid
|
|
if cur_prodid != tvid_prodid:
|
|
cur_prodid = tvid_prodid
|
|
|
|
sql_result[index]['network'] = (item['show_network'], item['episode_network'])[
|
|
isinstance(item['episode_network'], string_types) and 0 != len(item['episode_network'].strip())]
|
|
|
|
val = network_timezones.get_episode_time(
|
|
item['airdate'], item['airs'], item['show_network'], item['show_airtime'], item['show_timezone'],
|
|
item['timestamp'], item['episode_network'], item['ep_airtime'], item['ep_timezone'])
|
|
|
|
# noinspection PyCallByClass,PyTypeChecker
|
|
sql_result[index]['parsed_datetime'] = val
|
|
sql_result[index]['localtime'] = SGDatetime.convert_to_setting(val)
|
|
sql_result[index]['data_show_name'] = value_maybe_article(item['show_name'])
|
|
sql_result[index]['data_network'] = value_maybe_article(item['network'])
|
|
if not sql_result[index]['runtime']:
|
|
sql_result[index]['runtime'] = 5
|
|
|
|
imdb_id = None
|
|
if item['imdb_id']:
|
|
try:
|
|
imdb_id = helpers.try_int(re.search(r'(\d+)', item['imdb_id']).group(1))
|
|
except (BaseException, Exception):
|
|
pass
|
|
if imdb_id:
|
|
sql_result[index]['imdb_url'] = sickgear.indexers.indexer_config.tvinfo_config[
|
|
sickgear.indexers.indexer_config.TVINFO_IMDB][
|
|
'show_url'] % imdb_id
|
|
else:
|
|
sql_result[index]['imdb_url'] = ''
|
|
|
|
if tvid_prodid in fanarts:
|
|
continue
|
|
|
|
for img in glob.glob(cache_obj.fanart_path(*tvid_prodid_obj.tuple).replace('fanart.jpg', '*')) or []:
|
|
match = re.search(r'(\d+(?:\.\w*)?\.\w{5,8})\.fanart\.', img, re.I)
|
|
if not match:
|
|
continue
|
|
fanart = [(match.group(1), sickgear.FANART_RATINGS.get(tvid_prodid, {}).get(match.group(1), ''))]
|
|
if tvid_prodid not in fanarts:
|
|
fanarts[tvid_prodid] = fanart
|
|
else:
|
|
fanarts[tvid_prodid] += fanart
|
|
|
|
for tvid_prodid in fanarts:
|
|
fanart_rating = [(n, v) for n, v in fanarts[tvid_prodid] if 20 == v]
|
|
if fanart_rating:
|
|
fanarts[tvid_prodid] = fanart_rating
|
|
else:
|
|
rnd = [(n, v) for (n, v) in fanarts[tvid_prodid] if 30 != v]
|
|
grouped = [(n, v) for (n, v) in rnd if 10 == v]
|
|
if grouped:
|
|
fanarts[tvid_prodid] = [grouped[random.randint(0, len(grouped) - 1)]]
|
|
elif rnd:
|
|
fanarts[tvid_prodid] = [rnd[random.randint(0, len(rnd) - 1)]]
|
|
|
|
return sql_result, fanarts, sorts, next_week_dt, today, next_week
|
|
|
|
def daily_schedule(self, layout='None'):
|
|
""" display the episodes """
|
|
t = PageTemplate(web_handler=self, file='episodeView.tmpl')
|
|
sql_result, t.fanart, sorts, next_week_dt, today, next_week = self.get_daily_schedule()
|
|
# Allow local overriding of layout parameter
|
|
if layout and layout in ('banner', 'daybyday', 'list', 'poster'):
|
|
t.layout = layout
|
|
else:
|
|
t.layout = sickgear.EPISODE_VIEW_LAYOUT
|
|
|
|
t.has_art = bool(len(t.fanart))
|
|
t.css = ' '.join([t.layout] +
|
|
([], [('landscape', 'portrait')[sickgear.EPISODE_VIEW_POSTERS]])['daybyday' == t.layout] +
|
|
([], ['back-art'])[sickgear.EPISODE_VIEW_BACKGROUND and t.has_art] +
|
|
([], ['translucent'])[sickgear.EPISODE_VIEW_BACKGROUND_TRANSLUCENT] +
|
|
[{0: 'reg', 1: 'pro', 2: 'pro ii'}.get(sickgear.EPISODE_VIEW_VIEWMODE)])
|
|
t.fanart_panel = sickgear.FANART_PANEL
|
|
|
|
sql_result.sort(key=sorts[sickgear.EPISODE_VIEW_SORT])
|
|
|
|
t.next_week = datetime.combine(next_week_dt, dt_time(tzinfo=network_timezones.SG_TIMEZONE))
|
|
t.today = datetime.now(network_timezones.SG_TIMEZONE)
|
|
t.sql_results = sql_result
|
|
|
|
return t.respond()
|
|
|
|
@staticmethod
|
|
def live_panel(**kwargs):
|
|
|
|
if 'allseasons' in kwargs:
|
|
sickgear.DISPLAY_SHOW_MINIMUM = bool(config.minimax(kwargs['allseasons'], 0, 0, 1))
|
|
elif 'rate' in kwargs:
|
|
which = kwargs['which'].replace('fanart_', '')
|
|
rating = int(kwargs['rate'])
|
|
if rating:
|
|
sickgear.FANART_RATINGS.setdefault(kwargs['tvid_prodid'], {}).update({which: rating})
|
|
elif sickgear.FANART_RATINGS.get(kwargs['tvid_prodid'], {}).get(which):
|
|
del sickgear.FANART_RATINGS[kwargs['tvid_prodid']][which]
|
|
if not sickgear.FANART_RATINGS[kwargs['tvid_prodid']]:
|
|
del sickgear.FANART_RATINGS[kwargs['tvid_prodid']]
|
|
else:
|
|
translucent = bool(config.minimax(kwargs.get('translucent'), 0, 0, 1))
|
|
backart = bool(config.minimax(kwargs.get('backart'), 0, 0, 1))
|
|
viewmode = config.minimax(kwargs.get('viewmode'), 0, 0, 2)
|
|
|
|
if 'ds' == kwargs.get('pg'):
|
|
if 'viewart' in kwargs:
|
|
sickgear.DISPLAY_SHOW_VIEWART = config.minimax(kwargs['viewart'], 0, 0, 2)
|
|
elif 'translucent' in kwargs:
|
|
sickgear.DISPLAY_SHOW_BACKGROUND_TRANSLUCENT = translucent
|
|
elif 'backart' in kwargs:
|
|
sickgear.DISPLAY_SHOW_BACKGROUND = backart
|
|
elif 'viewmode' in kwargs:
|
|
sickgear.DISPLAY_SHOW_VIEWMODE = viewmode
|
|
elif 'ev' == kwargs.get('pg'):
|
|
if 'translucent' in kwargs:
|
|
sickgear.EPISODE_VIEW_BACKGROUND_TRANSLUCENT = translucent
|
|
elif 'backart' in kwargs:
|
|
sickgear.EPISODE_VIEW_BACKGROUND = backart
|
|
sickgear.FANART_PANEL = 'highlight-off' == sickgear.FANART_PANEL and 'highlight-off' or \
|
|
'highlight2' == sickgear.FANART_PANEL and 'highlight1' or \
|
|
'highlight1' == sickgear.FANART_PANEL and 'highlight' or 'highlight-off'
|
|
elif 'viewmode' in kwargs:
|
|
sickgear.EPISODE_VIEW_VIEWMODE = viewmode
|
|
|
|
sickgear.save_config()
|
|
|
|
@staticmethod
|
|
def get_footer_time(change_layout=True, json_dump=True):
|
|
|
|
now = datetime.now()
|
|
events = [
|
|
('recent', sickgear.search_recent_scheduler.time_left),
|
|
('backlog', sickgear.search_backlog_scheduler.next_backlog_timeleft),
|
|
]
|
|
|
|
if sickgear.DOWNLOAD_PROPERS:
|
|
events += [('propers', sickgear.properFinder.next_proper_timeleft)]
|
|
|
|
if change_layout not in (False, 0, '0', '', None):
|
|
sickgear.FOOTER_TIME_LAYOUT += 1
|
|
if 2 == sickgear.FOOTER_TIME_LAYOUT: # 2 layouts = time + delta
|
|
sickgear.FOOTER_TIME_LAYOUT = 0
|
|
sickgear.save_config()
|
|
|
|
next_event = []
|
|
for k, v in events:
|
|
try:
|
|
t = v()
|
|
except AttributeError:
|
|
t = None
|
|
if 0 == sickgear.FOOTER_TIME_LAYOUT:
|
|
next_event += [{k + '_time': t and SGDatetime.sbftime(now + t, markup=True) or 'soon'}]
|
|
else:
|
|
next_event += [{k + '_timeleft': t and str(t).split('.')[0] or 'soon'}]
|
|
|
|
if json_dump not in (False, 0, '0', '', None):
|
|
next_event = json_dumps(next_event)
|
|
|
|
return next_event
|
|
|
|
@staticmethod
|
|
def update_watched_state(payload=None, as_json=True):
|
|
"""
|
|
Update db with details of media file that is watched or unwatched
|
|
:param payload: Payload is a dict of dicts
|
|
:type payload: JSON or Dict
|
|
Each dict key in payload is an arbitrary value used to return its associated success or fail response.
|
|
Each dict value in payload comprises a dict of key value pairs where,
|
|
key: path_file: Path and filename of media, required for media to be found.
|
|
type: path_file: String
|
|
key: played: Optional default=100. Percentage times media has played. If 0, show is set as unwatched.
|
|
type: played: String
|
|
key: label: Optional default=''. Profile name or label in use while playing media.
|
|
type: label: String
|
|
key: date_watched: Optional default=current time. Datetime stamp that episode changed state.
|
|
type: date_watched: Timestamp
|
|
|
|
Example:
|
|
dict(
|
|
key01=dict(path_file='\\media\\', played=100, label='Bob', date_watched=1509850398.0),
|
|
key02=dict(path_file='\\media\\file-played1.mkv', played=100, label='Sue', date_watched=1509850398.0),
|
|
key03=dict(path_file='\\media\\file-played2.mkv', played=0, label='Rita', date_watched=1509850398.0)
|
|
)
|
|
JSON:
|
|
'{"key01": {"path_file": "\\media\\file_played1.mkv", "played": 100,
|
|
"label": "Bob", "date_watched": 1509850398.0}}'
|
|
|
|
:param as_json: True returns result as JSON otherwise Dict
|
|
:type as_json: Boolean
|
|
:return: if OK, the value of each dict is '' else fail reason string else None if payload is invalid.
|
|
:rtype: JSON if as_json is True otherwise None but with payload dict modified
|
|
Example:
|
|
Dict: {'key123': {''}} : on success
|
|
As JSON: '{"key123": {""}}' : on success
|
|
Dict: {'key123': {'error reason'}}
|
|
As JSON: '{"key123": {"error reason"}}'
|
|
Dict: {'error': {'error reason'}} : 'error' used as default key when bad key, value, or json
|
|
JSON: '{"error": {"error reason"}}' : 'error' used as default key when bad key, value, or json
|
|
|
|
Example case code using API endpoint, copy/paste, edit to suit, save, then run with: python sg_watched.py
|
|
```
|
|
import json
|
|
import urllib2
|
|
|
|
# SickGear APIkey
|
|
sg_apikey = '0123456789abcdef'
|
|
# SickGear server detail
|
|
sg_host = 'http://localhost:8081'
|
|
|
|
url = '%s/api/%s/?cmd=sg.updatewatchedstate' % (sg_host, sg_apikey)
|
|
payload = json_dumps(dict(
|
|
key01=dict(path_file='\\media\\path\\', played=100, label='Bob', date_watched=1509850398.0),
|
|
key02=dict(path_file='\\media\\path\\file-played1.mkv', played=100, label='Sue', date_watched=1509850398.0),
|
|
key03=dict(path_file='\\media\\path\\file-played2.mkv', played=0, label='Rita', date_watched=1509850398.0)
|
|
))
|
|
# payload is POST'ed to SG
|
|
rq = urllib2.Request(url, data=payload)
|
|
r = urllib2.urlopen(rq)
|
|
print json_load(r)
|
|
r.close()
|
|
```
|
|
"""
|
|
try:
|
|
if isinstance(payload, string_types):
|
|
data = json_loads(payload)
|
|
else:
|
|
data = payload
|
|
except ValueError:
|
|
payload = {}
|
|
data = payload
|
|
except TypeError:
|
|
data = payload
|
|
|
|
sql_result = []
|
|
if data:
|
|
my_db = db.DBConnection(row_type='dict')
|
|
|
|
media_paths = list(map(lambda arg: os.path.basename(arg[1]['path_file']), iteritems(data)))
|
|
|
|
def chunks(lines, n):
|
|
for c in range(0, len(lines), n):
|
|
yield lines[c:c + n]
|
|
|
|
# noinspection PyTypeChecker
|
|
for x in chunks(media_paths, 100):
|
|
# noinspection PyTypeChecker
|
|
sql_result += my_db.select(
|
|
'SELECT episode_id, status, location, file_size FROM tv_episodes WHERE file_size > 0 AND (%s)' %
|
|
' OR '.join(['location LIKE "%%%s"' % i for i in x]))
|
|
|
|
if sql_result:
|
|
cl = []
|
|
|
|
ep_results = {}
|
|
map_consume(lambda r: ep_results.update({'%s' % os.path.basename(r['location']).lower(): dict(
|
|
episode_id=r['episode_id'], status=r['status'], location=r['location'],
|
|
file_size=r['file_size'])}), sql_result)
|
|
|
|
for (k, v) in iteritems(data):
|
|
|
|
bname = (os.path.basename(v.get('path_file')) or '').lower()
|
|
if not bname:
|
|
msg = 'Missing media file name provided'
|
|
data[k] = msg
|
|
logger.warning('Update watched state skipped an item: %s' % msg)
|
|
continue
|
|
|
|
if bname in ep_results:
|
|
date_watched = now = SGDatetime.timestamp_near()
|
|
if 1500000000 < date_watched:
|
|
date_watched = helpers.try_int(float(v.get('date_watched')))
|
|
|
|
ep_data = ep_results[bname]
|
|
# using label and location with upsert to list multi-client items at same location
|
|
# can omit label to have the latest scanned client upsert an existing client row based on location
|
|
cl.extend(db.mass_upsert_sql(
|
|
'tv_episodes_watched',
|
|
dict(tvep_id=ep_data['episode_id'], clientep_id=v.get('media_id', '') or '',
|
|
played=v.get('played', 1),
|
|
date_watched=date_watched, date_added=now,
|
|
status=ep_data['status'], file_size=ep_data['file_size']),
|
|
dict(location=ep_data['location'], label=v.get('label', '')), sanitise=False))
|
|
|
|
data[k] = ''
|
|
|
|
if cl:
|
|
# noinspection PyUnboundLocalVariable
|
|
my_db.mass_action(cl)
|
|
|
|
if as_json:
|
|
if not data:
|
|
data = dict(error='Request made to SickGear with invalid payload')
|
|
logger.warning('Update watched state failed: %s' % data['error'])
|
|
|
|
return json_dumps(data)
|
|
|
|
def toggle_specials_view_show(self, tvid_prodid):
|
|
sickgear.DISPLAY_SHOW_SPECIALS = not sickgear.DISPLAY_SHOW_SPECIALS
|
|
|
|
self.redirect('/home/view-show?tvid_prodid=%s' % tvid_prodid)
|
|
|
|
def set_layout_history(self, layout):
|
|
|
|
if layout not in ('compact', 'detailed', 'compact_watched', 'detailed_watched',
|
|
'compact_stats', 'graph_stats', 'connect_failures'):
|
|
if 'provider_failures' == layout: # layout renamed
|
|
layout = 'connect_failures'
|
|
else:
|
|
layout = 'detailed'
|
|
|
|
sickgear.HISTORY_LAYOUT = layout
|
|
|
|
self.redirect('/history/')
|
|
|
|
def _generic_message(self, subject, message):
|
|
t = PageTemplate(web_handler=self, file='genericMessage.tmpl')
|
|
t.submenu = Home(self.application, self.request).home_menu()
|
|
t.subject = subject
|
|
t.message = message
|
|
return t.respond()
|
|
|
|
|
|
class Home(MainHandler):
|
|
|
|
def home_menu(self):
|
|
return [
|
|
{'title': 'Process Media', 'path': 'home/process-media/'},
|
|
{'title': 'Update Emby', 'path': 'home/update-mb/', 'requires': self.have_emby},
|
|
{'title': 'Update Kodi', 'path': 'home/update-kodi/', 'requires': self.have_kodi},
|
|
{'title': 'Update XBMC', 'path': 'home/update-xbmc/', 'requires': self.have_xbmc},
|
|
{'title': 'Update Plex', 'path': 'home/update-plex/', 'requires': self.have_plex}
|
|
]
|
|
|
|
@staticmethod
|
|
def have_emby():
|
|
return sickgear.USE_EMBY
|
|
|
|
@staticmethod
|
|
def have_kodi():
|
|
return sickgear.USE_KODI
|
|
|
|
@staticmethod
|
|
def have_xbmc():
|
|
return sickgear.USE_XBMC and sickgear.XBMC_UPDATE_LIBRARY
|
|
|
|
@staticmethod
|
|
def have_plex():
|
|
return sickgear.USE_PLEX and sickgear.PLEX_UPDATE_LIBRARY
|
|
|
|
@staticmethod
|
|
def _get_episode(tvid_prodid, season=None, episode=None, absolute=None):
|
|
"""
|
|
|
|
:param tvid_prodid:
|
|
:type tvid_prodid:
|
|
:param season:
|
|
:type season:
|
|
:param episode:
|
|
:type episode:
|
|
:param absolute:
|
|
:type absolute:
|
|
:return:
|
|
:rtype: sickgear.tv.TVEpisode
|
|
"""
|
|
if None is tvid_prodid:
|
|
return 'Invalid show parameters'
|
|
|
|
show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
if None is show_obj:
|
|
return 'Invalid show paramaters'
|
|
|
|
if absolute:
|
|
ep_obj = show_obj.get_episode(absolute_number=int(absolute))
|
|
elif None is not season and None is not episode:
|
|
ep_obj = show_obj.get_episode(int(season), int(episode))
|
|
else:
|
|
return 'Invalid paramaters'
|
|
|
|
if None is ep_obj:
|
|
return "Episode couldn't be retrieved"
|
|
|
|
return ep_obj
|
|
|
|
def index(self):
|
|
if 'episodes' == sickgear.DEFAULT_HOME:
|
|
self.redirect('/daily-schedule/')
|
|
elif 'history' == sickgear.DEFAULT_HOME:
|
|
self.redirect('/history/')
|
|
else:
|
|
self.redirect('/view-shows/')
|
|
|
|
def view_shows(self):
|
|
t = PageTemplate(web_handler=self, file='home.tmpl')
|
|
t.showlists = []
|
|
index = 0
|
|
if 'custom' == sickgear.SHOWLIST_TAGVIEW:
|
|
for name in sickgear.SHOW_TAGS:
|
|
results = list(filter(lambda so: so.tag == name, sickgear.showList))
|
|
if results:
|
|
t.showlists.append(['container%s' % index, name, results])
|
|
index += 1
|
|
elif 'anime' == sickgear.SHOWLIST_TAGVIEW:
|
|
show_results = list(filter(lambda so: not so.anime, sickgear.showList))
|
|
anime_results = list(filter(lambda so: so.anime, sickgear.showList))
|
|
if show_results:
|
|
t.showlists.append(['container%s' % index, 'Show List', show_results])
|
|
index += 1
|
|
if anime_results:
|
|
t.showlists.append(['container%s' % index, 'Anime List', anime_results])
|
|
|
|
if 0 == len(t.showlists):
|
|
t.showlists.append(['container0', 'Show List', sickgear.showList])
|
|
else:
|
|
items = []
|
|
default = 0
|
|
for index, group in enumerate(t.showlists):
|
|
items += group[2]
|
|
default = (default, index)['Show List' == group[1]]
|
|
t.showlists[default][2] += [cur_so for cur_so in sickgear.showList if cur_so not in items]
|
|
|
|
if 'simple' != sickgear.HOME_LAYOUT:
|
|
t.network_images = {}
|
|
networks = {}
|
|
images_path = os.path.join(sickgear.PROG_DIR, 'gui', 'slick', 'images', 'network')
|
|
for cur_show_obj in sickgear.showList:
|
|
network_name = 'nonetwork' if None is cur_show_obj.network \
|
|
else cur_show_obj.network.replace('\u00C9', 'e').lower()
|
|
if network_name not in networks:
|
|
filename = f'{network_name}.png'
|
|
if not os.path.isfile(os.path.join(images_path, filename)):
|
|
filename = '%s.png' % re.sub(r'(?m)(.*)\s+\(\w{2}\)$', r'\1', network_name)
|
|
if not os.path.isfile(os.path.join(images_path, filename)):
|
|
filename = 'nonetwork.png'
|
|
networks.setdefault(network_name, filename)
|
|
t.network_images.setdefault(cur_show_obj.tvid_prodid, networks[network_name])
|
|
|
|
t.submenu = self.home_menu()
|
|
t.layout = sickgear.HOME_LAYOUT
|
|
|
|
# Get all show snatched / downloaded / next air date stats
|
|
my_db = db.DBConnection()
|
|
today = dt_date.today().toordinal()
|
|
status_quality = ','.join([str(x) for x in Quality.SNATCHED_ANY])
|
|
status_download = ','.join([str(x) for x in Quality.DOWNLOADED + Quality.ARCHIVED])
|
|
status_total = '%s, %s, %s' % (SKIPPED, WANTED, FAILED)
|
|
|
|
sql_result = my_db.select(
|
|
'SELECT indexer AS tvid, showid as prodid, '
|
|
+ '(SELECT COUNT(*) FROM tv_episodes'
|
|
' WHERE indexer = tv_eps.indexer AND showid = tv_eps.showid'
|
|
' AND season > 0 AND episode > 0 AND airdate > 1 AND status IN (%s)) AS ep_snatched,'
|
|
' (SELECT COUNT(*) FROM tv_episodes'
|
|
' WHERE indexer = tv_eps.indexer AND showid = tv_eps.showid'
|
|
' AND season > 0 AND episode > 0 AND airdate > 1 AND status IN (%s)) AS ep_downloaded,'
|
|
' (SELECT COUNT(*) FROM tv_episodes'
|
|
' WHERE indexer = tv_eps.indexer AND showid = tv_eps.showid'
|
|
' AND season > 0 AND episode > 0 AND airdate > 1'
|
|
' AND ('
|
|
'(airdate <= %s AND (status IN (%s)))'
|
|
' OR (status IN (%s)) OR (status IN (%s)))) AS ep_total,'
|
|
' (SELECT airdate FROM tv_episodes'
|
|
' WHERE indexer = tv_eps.indexer AND showid = tv_eps.showid'
|
|
' AND airdate >= %s AND (status = %s OR status = %s)'
|
|
' ORDER BY airdate ASC LIMIT 1) AS ep_airs_next'
|
|
' FROM tv_episodes tv_eps GROUP BY indexer, showid'
|
|
% (status_quality, status_download, today, status_total,
|
|
status_quality, status_download, today, UNAIRED, WANTED))
|
|
|
|
t.show_stat = {}
|
|
|
|
for cur_result in sql_result:
|
|
t.show_stat[TVidProdid({cur_result['tvid']: cur_result['prodid']})()] = cur_result
|
|
|
|
return t.respond()
|
|
|
|
def test_sabnzbd(self, host=None, username=None, password=None, apikey=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
host = config.clean_url(host)
|
|
connection, access_msg = sab.access_method(host)
|
|
if connection:
|
|
if None is not password and set('*') == set(password):
|
|
password = sickgear.SAB_PASSWORD
|
|
if None is not apikey and starify(apikey, True):
|
|
apikey = sickgear.SAB_APIKEY
|
|
|
|
authed, auth_msg = sab.test_authentication(host, username, password, apikey)
|
|
if authed:
|
|
return f'Success. Connected' \
|
|
f' {(f"using {access_msg}", "with no")["None" == auth_msg.lower()]} authentication'
|
|
return f'Authentication failed. {auth_msg}'
|
|
return 'Unable to connect to host'
|
|
|
|
def test_nzbget(self, host=None, use_https=None, username=None, password=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
host = config.clean_url(host)
|
|
if None is not password and set('*') == set(password):
|
|
password = sickgear.NZBGET_PASSWORD
|
|
|
|
authed, auth_msg, void = nzbget.test_nzbget(host, bool(config.checkbox_to_value(use_https)), username, password,
|
|
timeout=20)
|
|
return auth_msg
|
|
|
|
def test_torrent(self, torrent_method=None, host=None, username=None, password=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
host = config.clean_url(host)
|
|
if None is not password and set('*') == set(password):
|
|
password = sickgear.TORRENT_PASSWORD
|
|
|
|
client = clients.get_client_instance(torrent_method)
|
|
|
|
connection, acces_msg = client(host, username, password).test_authentication()
|
|
|
|
return acces_msg
|
|
|
|
def test_flaresolverr(self, host=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
hosts = config.clean_hosts(host, default_port=8191)
|
|
if not hosts:
|
|
return 'Fail: No valid host(s)'
|
|
|
|
try:
|
|
fs_ver = CloudflareScraper().test_flaresolverr(host)
|
|
result = 'Successful connection to FlareSolverr %s' % fs_ver
|
|
except(BaseException, Exception):
|
|
result = 'Failed host connection (is it running?)'
|
|
|
|
ui.notifications.message('Tested Flaresolverr:', unquote_plus(hosts))
|
|
return result
|
|
|
|
@staticmethod
|
|
def discover_emby():
|
|
return notifiers.NotifierFactory().get('EMBY').discover_server()
|
|
|
|
def test_emby(self, host=None, apikey=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
hosts = config.clean_hosts(host, default_port=8096)
|
|
if not hosts:
|
|
return 'Fail: No valid host(s)'
|
|
|
|
result = notifiers.NotifierFactory().get('EMBY').test_notify(hosts, apikey)
|
|
|
|
ui.notifications.message('Tested Emby:', unquote_plus(hosts.replace(',', ', ')))
|
|
return result
|
|
|
|
def test_kodi(self, host=None, username=None, password=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
hosts = config.clean_hosts(host, default_port=8080)
|
|
if not hosts:
|
|
return 'Fail: No valid host(s)'
|
|
|
|
if None is not password and set('*') == set(password):
|
|
password = sickgear.KODI_PASSWORD
|
|
|
|
result = notifiers.NotifierFactory().get('KODI').test_notify(hosts, username, password)
|
|
|
|
ui.notifications.message('Tested Kodi:', unquote_plus(hosts.replace(',', ', ')))
|
|
return result
|
|
|
|
def test_plex(self, host=None, username=None, password=None, server=False):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
hosts = config.clean_hosts(host, default_port=32400)
|
|
if not hosts:
|
|
return 'Fail: No valid host(s)'
|
|
|
|
if None is not password and set('*') == set(password):
|
|
password = sickgear.PLEX_PASSWORD
|
|
|
|
server = 'true' == server
|
|
n = notifiers.NotifierFactory().get('PLEX')
|
|
method = n.test_update_library if server else n.test_notify
|
|
result = method(hosts, username, password)
|
|
|
|
ui.notifications.message('Tested Plex %s(s): ' % ('client', 'Media Server host')[server],
|
|
unquote_plus(hosts.replace(',', ', ')))
|
|
return result
|
|
|
|
def test_nmj(self, host=None, database=None, mount=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
host = config.clean_host(host)
|
|
if not host:
|
|
return 'Fail: No valid host(s)'
|
|
|
|
return notifiers.NotifierFactory().get('NMJ').test_notify(unquote_plus(host), database, mount)
|
|
|
|
def settings_nmj(self, host=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
host = config.clean_host(host)
|
|
if not host:
|
|
return 'Fail: No valid host(s)'
|
|
|
|
return notifiers.NotifierFactory().get('NMJ').notify_settings(unquote_plus(host))
|
|
|
|
def test_nmj2(self, host=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
host = config.clean_host(host)
|
|
if not host:
|
|
return 'Fail: No valid host(s)'
|
|
|
|
return notifiers.NotifierFactory().get('NMJV2').test_notify(unquote_plus(host))
|
|
|
|
def settings_nmj2(self, host=None, dbloc=None, instance=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
host = config.clean_host(host)
|
|
return notifiers.NotifierFactory().get('NMJV2').notify_settings(unquote_plus(host), dbloc, instance)
|
|
|
|
def test_boxcar2(self, access_token=None, sound=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
if None is not access_token and starify(access_token, True):
|
|
access_token = sickgear.BOXCAR2_ACCESSTOKEN
|
|
|
|
return notifiers.NotifierFactory().get('BOXCAR2').test_notify(access_token, sound)
|
|
|
|
def test_pushbullet(self, access_token=None, device_iden=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
if None is not access_token and starify(access_token, True):
|
|
access_token = sickgear.PUSHBULLET_ACCESS_TOKEN
|
|
|
|
return notifiers.NotifierFactory().get('PUSHBULLET').test_notify(access_token, device_iden)
|
|
|
|
def get_pushbullet_devices(self, access_token=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
if None is not access_token and starify(access_token, True):
|
|
access_token = sickgear.PUSHBULLET_ACCESS_TOKEN
|
|
|
|
return notifiers.NotifierFactory().get('PUSHBULLET').get_devices(access_token)
|
|
|
|
def test_pushover(self, user_key=None, api_key=None, priority=None, device=None, sound=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
if None is not user_key and starify(user_key, True):
|
|
user_key = sickgear.PUSHOVER_USERKEY
|
|
|
|
if None is not api_key and starify(api_key, True):
|
|
api_key = sickgear.PUSHOVER_APIKEY
|
|
|
|
return notifiers.NotifierFactory().get('PUSHOVER').test_notify(user_key, api_key, priority, device, sound)
|
|
|
|
def get_pushover_devices(self, user_key=None, api_key=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
if None is not user_key and starify(user_key, True):
|
|
user_key = sickgear.PUSHOVER_USERKEY
|
|
|
|
if None is not api_key and starify(api_key, True):
|
|
api_key = sickgear.PUSHOVER_APIKEY
|
|
|
|
return notifiers.NotifierFactory().get('PUSHOVER').get_devices(user_key, api_key)
|
|
|
|
def test_growl(self, host=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
hosts = config.clean_hosts(host, default_port=23053)
|
|
if not hosts:
|
|
return 'Fail: No valid host(s)'
|
|
|
|
result = notifiers.NotifierFactory().get('GROWL').test_notify(None, hosts)
|
|
|
|
ui.notifications.message('Tested Growl:', unquote_plus(hosts.replace(',', ', ')))
|
|
return result
|
|
|
|
def test_prowl(self, prowl_api=None, prowl_priority=0):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
if None is not prowl_api and starify(prowl_api, True):
|
|
prowl_api = sickgear.PROWL_API
|
|
|
|
return notifiers.NotifierFactory().get('PROWL').test_notify(prowl_api, prowl_priority)
|
|
|
|
def test_libnotify(self):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
return notifiers.NotifierFactory().get('LIBNOTIFY').test_notify()
|
|
|
|
def trakt_authenticate(self, pin=None, account=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
if None is pin:
|
|
return json_dumps({'result': 'Fail', 'error_message': 'Trakt PIN required for authentication'})
|
|
|
|
if account and 'new' == account:
|
|
account = None
|
|
|
|
acc = None
|
|
if account:
|
|
acc = helpers.try_int(account, -1)
|
|
if 0 < acc and acc not in sickgear.TRAKT_ACCOUNTS:
|
|
return json_dumps({'result': 'Fail', 'error_message': 'Fail: cannot update non-existing account'})
|
|
|
|
json_fail_auth = json_dumps({'result': 'Fail', 'error_message': 'Trakt NOT authenticated'})
|
|
try:
|
|
resp = TraktAPI().trakt_token(pin, account=acc)
|
|
except TraktAuthException:
|
|
return json_fail_auth
|
|
if not account and isinstance(resp, bool) and not resp:
|
|
return json_fail_auth
|
|
|
|
if not sickgear.USE_TRAKT:
|
|
sickgear.USE_TRAKT = True
|
|
sickgear.save_config()
|
|
pick = resp if not account else acc
|
|
return json_dumps({'result': 'Success',
|
|
'account_id': sickgear.TRAKT_ACCOUNTS[pick].account_id,
|
|
'account_name': sickgear.TRAKT_ACCOUNTS[pick].name})
|
|
|
|
def trakt_delete(self, accountid=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
if accountid:
|
|
aid = helpers.try_int(accountid, None)
|
|
if None is not aid:
|
|
if aid in sickgear.TRAKT_ACCOUNTS:
|
|
account = {'result': 'Success',
|
|
'account_id': sickgear.TRAKT_ACCOUNTS[aid].account_id,
|
|
'account_name': sickgear.TRAKT_ACCOUNTS[aid].name}
|
|
if TraktAPI.delete_account(aid):
|
|
trakt_collection_remove_account(aid)
|
|
account['num_accounts'] = len(sickgear.TRAKT_ACCOUNTS)
|
|
return json_dumps(account)
|
|
|
|
return json_dumps({'result': 'Not found: Account to delete'})
|
|
return json_dumps({'result': 'Not found: Invalid account id'})
|
|
|
|
def load_show_notify_lists(self):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
my_db = db.DBConnection()
|
|
# noinspection SqlResolve
|
|
rows = my_db.select(
|
|
'SELECT indexer || ? || indexer_id AS tvid_prodid, notify_list'
|
|
' FROM tv_shows'
|
|
' WHERE notify_list NOTNULL'
|
|
' AND notify_list != ""',
|
|
[TVidProdid.glue])
|
|
notify_lists = {}
|
|
for r in filter(lambda x: x['notify_list'].strip(), rows):
|
|
# noinspection PyTypeChecker
|
|
notify_lists[r['tvid_prodid']] = r['notify_list']
|
|
|
|
sorted_show_lists = self.sorted_show_lists()
|
|
response = []
|
|
for current_group in sorted_show_lists:
|
|
data = []
|
|
for show_obj in current_group[1]:
|
|
data.append({
|
|
'id': show_obj.tvid_prodid,
|
|
'name': show_obj.name,
|
|
'list': '' if show_obj.tvid_prodid not in notify_lists else notify_lists[show_obj.tvid_prodid]})
|
|
if data:
|
|
response.append({current_group[0]: data})
|
|
|
|
return json_dumps(response)
|
|
|
|
def test_slack(self, channel=None, as_authed=False, bot_name=None, icon_url=None, access_token=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
return notifiers.NotifierFactory().get('SLACK').test_notify(
|
|
channel=channel, as_authed='true' == as_authed,
|
|
bot_name=bot_name, icon_url=icon_url, access_token=access_token)
|
|
|
|
def test_discord(self, as_authed=False, username=None, icon_url=None, as_tts=False, access_token=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
return notifiers.NotifierFactory().get('DISCORD').test_notify(
|
|
as_authed='true' == as_authed, username=username, icon_url=icon_url,
|
|
as_tts='true' == as_tts, access_token=access_token)
|
|
|
|
def test_gitter(self, room_name=None, access_token=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
return notifiers.NotifierFactory().get('GITTER').test_notify(
|
|
room_name=room_name, access_token=access_token)
|
|
|
|
def test_telegram(self, send_icon=False, access_token=None, chatid=None, quiet=False):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
r = notifiers.NotifierFactory().get('TELEGRAM').test_notify(
|
|
send_icon=bool(config.checkbox_to_value(send_icon)), access_token=access_token, chatid=chatid, quiet=quiet)
|
|
return json_dumps(r)
|
|
|
|
def test_email(self, host=None, port=None, smtp_from=None, use_tls=None, user=None, pwd=None, to=None):
|
|
self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
if None is not pwd and set('*') == set(pwd):
|
|
pwd = sickgear.EMAIL_PASSWORD
|
|
|
|
host = config.clean_host(host)
|
|
|
|
return notifiers.NotifierFactory().get('EMAIL').test_notify(host, port, smtp_from, use_tls, user, pwd, to)
|
|
|
|
@staticmethod
|
|
def save_show_email(show=None, emails=None):
|
|
# self.set_header('Cache-Control', 'max-age=0,no-cache,no-store')
|
|
|
|
my_db = db.DBConnection()
|
|
success = False
|
|
parse = show.split(TVidProdid.glue)
|
|
if 1 < len(parse) and \
|
|
my_db.action('UPDATE tv_shows SET notify_list = ?'
|
|
' WHERE indexer = ? AND indexer_id = ?',
|
|
[emails, parse[0], parse[1]]):
|
|
success = True
|
|
return json_dumps({'id': show, 'success': success})
|
|
|
|
def check_update(self):
|
|
# force a check to see if there is a new version
|
|
if sickgear.update_software_scheduler.action.check_for_new_version(force=True):
|
|
logger.log('Forced version check found results')
|
|
|
|
if sickgear.update_packages_scheduler.action.check_for_new_version(force=True):
|
|
logger.log('Forced package version check found results')
|
|
|
|
self.redirect('/home/')
|
|
|
|
def view_changes(self):
|
|
|
|
t = PageTemplate(web_handler=self, file='viewchanges.tmpl')
|
|
|
|
t.changelist = [{'type': 'rel', 'ver': '', 'date': 'Nothing to display at this time'}]
|
|
url = 'https://raw.githubusercontent.com/wiki/SickGear/SickGear/sickgear/CHANGES.md'
|
|
response = helpers.get_url(url)
|
|
if not response:
|
|
return t.respond()
|
|
|
|
data = response.replace('\xef\xbb\xbf', '').splitlines()
|
|
|
|
output, change, max_rel = [], {}, 5
|
|
for line in data:
|
|
if not line.strip():
|
|
continue
|
|
if line.startswith(' '):
|
|
change_parts = re.findall(r'^\W+(.*)$', line)
|
|
change['text'] += change_parts and (' %s' % change_parts[0].strip()) or ''
|
|
else:
|
|
if change:
|
|
output.append(change)
|
|
change = None
|
|
if line.startswith('* '):
|
|
change_parts = re.findall(r'^[*\W]+(Add|Change|Fix|Port|Remove|Update)\W(.*)', line)
|
|
change = change_parts and {'type': change_parts[0][0], 'text': change_parts[0][1].strip()} or {}
|
|
elif not max_rel:
|
|
break
|
|
elif line.startswith('### '):
|
|
rel_data = re.findall(r'(?im)^###\W*(\S+)\W\(([^)]+)\)', line)
|
|
rel_data and output.append({'type': 'rel', 'ver': rel_data[0][0], 'date': rel_data[0][1]})
|
|
max_rel -= 1
|
|
elif line.startswith('# '):
|
|
max_data = re.findall(r'^#\W*(\d+)\W*$', line)
|
|
max_rel = max_data and helpers.try_int(max_data[0], None) or 5
|
|
if change:
|
|
output.append(change)
|
|
|
|
t.changelist = output
|
|
return t.respond()
|
|
|
|
def shutdown(self, pid=None):
|
|
|
|
if str(pid) != str(sickgear.PID):
|
|
return self.redirect('/home/')
|
|
|
|
if self.maybe_ignore('Shutdown'):
|
|
return
|
|
|
|
t = PageTemplate(web_handler=self, file='restart.tmpl')
|
|
t.shutdown = True
|
|
|
|
sickgear.events.put(sickgear.events.SystemEvent.SHUTDOWN)
|
|
|
|
return t.respond()
|
|
|
|
def restart(self, pid=None, update_pkg=None):
|
|
|
|
if str(pid) != str(sickgear.PID):
|
|
return self.redirect('/home/')
|
|
|
|
if self.maybe_ignore('Restart'):
|
|
return
|
|
|
|
t = PageTemplate(web_handler=self, file='restart.tmpl')
|
|
t.shutdown = False
|
|
|
|
sickgear.restart(soft=False, update_pkg=bool(helpers.try_int(update_pkg)))
|
|
|
|
return t.respond()
|
|
|
|
def maybe_ignore(self, task):
|
|
response = Scheduler.blocking_jobs()
|
|
if response:
|
|
task and logger.log('%s aborted because %s' % (task, response.lower()), logger.DEBUG)
|
|
|
|
self.redirect(self.request.headers['Referer'])
|
|
if task:
|
|
ui.notifications.message(u'Fail %s because %s, please try later' % (task.lower(), response.lower()))
|
|
return True
|
|
return False
|
|
|
|
def update(self, pid=None):
|
|
|
|
if str(pid) != str(sickgear.PID):
|
|
return self.redirect('/home/')
|
|
|
|
if sickgear.update_software_scheduler.action.update():
|
|
return self.restart(pid)
|
|
|
|
return self._generic_message('Update Failed',
|
|
'Update wasn\'t successful, not restarting. Check your log for more information.')
|
|
|
|
def branch_checkout(self, branch):
|
|
sickgear.BRANCH = branch
|
|
ui.notifications.message('Checking out branch: ', branch)
|
|
return self.update(sickgear.PID)
|
|
|
|
def pull_request_checkout(self, branch):
|
|
pull_request = branch
|
|
branch = branch.split(':')[1]
|
|
fetched = sickgear.update_software_scheduler.action.fetch(pull_request)
|
|
if fetched:
|
|
sickgear.BRANCH = branch
|
|
ui.notifications.message('Checking out branch: ', branch)
|
|
return self.update(sickgear.PID)
|
|
else:
|
|
self.redirect('/home/')
|
|
|
|
# noinspection PyUnusedLocal
|
|
def season_render(self, tvid_prodid=None, season=None, **kwargs):
|
|
|
|
response = {'success': False}
|
|
# noinspection PyTypeChecker
|
|
show_obj = None
|
|
if tvid_prodid:
|
|
show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
if not show_obj:
|
|
return json_dumps(response)
|
|
|
|
re_season = re.compile(r'(?i)^showseason-(\d+)$')
|
|
season = None if not any(re_season.findall(season)) else \
|
|
helpers.try_int(re_season.findall(season)[0], None)
|
|
if None is season:
|
|
return json_dumps(response)
|
|
|
|
t = PageTemplate(web_handler=self, file='inc_displayShow.tmpl')
|
|
t.show_obj = show_obj
|
|
|
|
my_db = db.DBConnection()
|
|
sql_result = my_db.select('SELECT *'
|
|
' FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' AND season = ?'
|
|
' ORDER BY episode DESC',
|
|
[show_obj.tvid, show_obj.prodid,
|
|
season])
|
|
t.episodes = sql_result
|
|
|
|
ep_cats = {}
|
|
for cur_result in sql_result:
|
|
status_overview = show_obj.get_overview(int(cur_result['status']))
|
|
if status_overview:
|
|
ep_cats['%sx%s' % (season, cur_result['episode'])] = status_overview
|
|
t.ep_cats = ep_cats
|
|
|
|
args = (int(show_obj.tvid), int(show_obj.prodid))
|
|
t.scene_numbering = get_scene_numbering_for_show(*args)
|
|
t.xem_numbering = get_xem_numbering_for_show(*args)
|
|
t.scene_absolute_numbering = get_scene_absolute_numbering_for_show(*args)
|
|
t.xem_absolute_numbering = get_xem_absolute_numbering_for_show(*args)
|
|
|
|
return json_dumps({'success': t.respond()})
|
|
|
|
@staticmethod
|
|
def fix_show_obj_db_data(show_obj):
|
|
# adjust show_obj db data
|
|
if 'genres' not in show_obj.imdb_info or None is show_obj.imdb_info.get('genres'):
|
|
show_obj.imdb_info['genres'] = ''
|
|
if show_obj.genre and not show_obj.genre[1:-1]:
|
|
show_obj.genre = ''
|
|
if 'country_codes' not in show_obj.imdb_info or None is show_obj.imdb_info.get('country_codes'):
|
|
show_obj.imdb_info['country_codes'] = ''
|
|
return show_obj
|
|
|
|
def view_show(self, tvid_prodid=None):
|
|
|
|
if None is tvid_prodid:
|
|
return self._generic_message('Error', 'Invalid show ID')
|
|
|
|
show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
if None is show_obj:
|
|
return self._generic_message('Error', 'Show not in show list')
|
|
|
|
show_obj = self.fix_show_obj_db_data(show_obj)
|
|
|
|
t = PageTemplate(web_handler=self, file='displayShow.tmpl')
|
|
t.submenu = [{'title': 'Edit', 'path': 'home/edit-show?tvid_prodid=%s' % tvid_prodid}]
|
|
|
|
try:
|
|
t.showLoc = (show_obj.location, True)
|
|
except exceptions_helper.ShowDirNotFoundException:
|
|
# noinspection PyProtectedMember
|
|
t.showLoc = (show_obj._location, False)
|
|
|
|
show_message = []
|
|
|
|
if sickgear.show_queue_scheduler.action.is_being_added(show_obj):
|
|
show_message = ['Downloading this show, the information below is incomplete']
|
|
|
|
elif sickgear.show_queue_scheduler.action.is_being_updated(show_obj):
|
|
show_message = ['Updating information for this show']
|
|
|
|
elif sickgear.show_queue_scheduler.action.is_being_refreshed(show_obj):
|
|
show_message = ['Refreshing episodes from disk for this show']
|
|
|
|
elif sickgear.show_queue_scheduler.action.is_being_subtitled(show_obj):
|
|
show_message = ['Downloading subtitles for this show']
|
|
|
|
elif sickgear.show_queue_scheduler.action.is_in_refresh_queue(show_obj):
|
|
show_message = ['Refresh queued for this show']
|
|
|
|
elif sickgear.show_queue_scheduler.action.is_in_update_queue(show_obj):
|
|
show_message = ['Update queued for this show']
|
|
|
|
elif sickgear.show_queue_scheduler.action.is_in_subtitle_queue(show_obj):
|
|
show_message = ['Subtitle download queued for this show']
|
|
|
|
if sickgear.show_queue_scheduler.action.is_show_being_switched(show_obj):
|
|
show_message += ['Switching TV info source and awaiting update for this show']
|
|
|
|
elif sickgear.show_queue_scheduler.action.is_show_switch_queued(show_obj):
|
|
show_message += ['Queuing a switch of TV info source for this show']
|
|
|
|
if sickgear.people_queue_scheduler.action.show_in_queue(show_obj, check_inprogress=True):
|
|
show_message += ['Updating cast for this show']
|
|
elif sickgear.people_queue_scheduler.action.show_in_queue(show_obj):
|
|
show_message += ['Cast update queued for this show']
|
|
|
|
if 0 != show_obj.not_found_count:
|
|
last_found = ('', ' since %s' % SGDatetime.fromordinal(
|
|
show_obj.last_found_on_indexer).sbfdate())[1 < show_obj.last_found_on_indexer]
|
|
show_message += [
|
|
'The main ID of this show has been <span class="addQTip" title="many reasons exist, including: ' +
|
|
'<br>show flagged as a duplicate, removed completely... etc">abandoned</span>%s, ' % last_found +
|
|
'<a href="%s/home/edit-show?tvid_prodid=%s&tvsrc=0&srcid=%s#core-component-group3">replace it here</a>'
|
|
% (sickgear.WEB_ROOT, tvid_prodid, show_obj.prodid)]
|
|
|
|
show_message = '.<br>'.join(show_message)
|
|
|
|
t.force_update = 'home/update-show?tvid_prodid=%s&force=1&web=1' % tvid_prodid
|
|
if not sickgear.show_queue_scheduler.action.is_being_added(show_obj):
|
|
if not sickgear.show_queue_scheduler.action.is_being_updated(show_obj):
|
|
t.submenu.append(
|
|
{'title': 'Remove',
|
|
'path': 'home/delete-show?tvid_prodid=%s' % tvid_prodid, 'confirm': True})
|
|
t.submenu.append(
|
|
{'title': 'Re-scan files', 'path': 'home/refresh-show?tvid_prodid=%s' % tvid_prodid})
|
|
t.submenu.append(
|
|
{'title': 'Force Full Update', 'path': t.force_update})
|
|
t.submenu.append(
|
|
{'title': 'Cast Update', 'path': 'home/update-cast?tvid_prodid=%s' % tvid_prodid})
|
|
t.submenu.append(
|
|
{'title': 'Update show in Emby',
|
|
'path': 'home/update-mb%s' % (
|
|
TVINFO_TVDB == show_obj.tvid and ('?tvid_prodid=%s' % tvid_prodid) or '/'),
|
|
'requires': self.have_emby})
|
|
t.submenu.append(
|
|
{'title': 'Update show in Kodi', 'path': 'home/update-kodi?show_name=%s' % quote_plus(
|
|
show_obj.name.encode('utf-8')), 'requires': self.have_kodi})
|
|
t.submenu.append(
|
|
{'title': 'Update show in XBMC',
|
|
'path': 'home/update-xbmc?show_name=%s' % quote_plus(
|
|
show_obj.name.encode('utf-8')), 'requires': self.have_xbmc})
|
|
t.submenu.append(
|
|
{'title': 'Media Rename',
|
|
'path': 'home/rename-media?tvid_prodid=%s' % tvid_prodid})
|
|
if sickgear.USE_SUBTITLES and not sickgear.show_queue_scheduler.action.is_being_subtitled(
|
|
show_obj) and show_obj.subtitles:
|
|
t.submenu.append(
|
|
{'title': 'Download Subtitles',
|
|
'path': 'home/subtitle-show?tvid_prodid=%s' % tvid_prodid})
|
|
|
|
t.show_obj = show_obj
|
|
with BS4Parser('<html><body>%s</body></html>' % show_obj.overview, features=['html5lib', 'permissive']) as soup:
|
|
try:
|
|
soup.a.replace_with(soup.new_tag(''))
|
|
except (BaseException, Exception):
|
|
pass
|
|
overview = re.sub('(?i)full streaming', '', soup.get_text().strip())
|
|
t.show_obj.overview = overview
|
|
t.show_message = show_message
|
|
|
|
ep_counts = {}
|
|
ep_cats = {}
|
|
ep_counts[Overview.SKIPPED] = 0
|
|
ep_counts[Overview.WANTED] = 0
|
|
ep_counts[Overview.QUAL] = 0
|
|
ep_counts[Overview.GOOD] = 0
|
|
ep_counts[Overview.UNAIRED] = 0
|
|
ep_counts[Overview.SNATCHED] = 0
|
|
ep_counts['videos'] = {}
|
|
ep_counts['status'] = {}
|
|
ep_counts['archived'] = {}
|
|
ep_counts['totals'] = {}
|
|
ep_counts['eps_most'] = 0
|
|
ep_counts['eps_all'] = 0
|
|
t.latest_season = 0
|
|
t.has_special = False
|
|
|
|
my_db = db.DBConnection()
|
|
|
|
failed_check = my_db.select('SELECT status FROM tv_src_switch WHERE old_indexer = ? AND old_indexer_id = ?'
|
|
' AND status != ?', [show_obj.tvid, show_obj.prodid, TVSWITCH_NORMAL])
|
|
if failed_check:
|
|
t.show_message = '%s%s%s' % \
|
|
(t.show_message, ('<br>', '')[0 == len(t.show_message)],
|
|
'Failed to switch tv info source: %s' %
|
|
tvswitch_names.get(failed_check[0]['status'], 'Unknown reason'))
|
|
|
|
for row in my_db.select('SELECT season, count(*) AS cnt'
|
|
' FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' GROUP BY season',
|
|
[show_obj.tvid, show_obj.prodid]):
|
|
ep_counts['totals'][row['season']] = row['cnt']
|
|
|
|
if None is not ep_counts['totals'].get(0):
|
|
t.has_special = True
|
|
if not sickgear.DISPLAY_SHOW_SPECIALS:
|
|
del (ep_counts['totals'][0])
|
|
|
|
ep_counts['eps_all'] = sum(itervalues(ep_counts['totals']))
|
|
ep_counts['eps_most'] = max(list(ep_counts['totals'].values()) + [0])
|
|
all_seasons = sorted(iterkeys(ep_counts['totals']), reverse=True)
|
|
t.lowest_season, t.highest_season = all_seasons and (all_seasons[-1], all_seasons[0]) or (0, 0)
|
|
|
|
# 55 == seasons 1-10 and excludes the random season 0
|
|
force_display_show_minimum = 30 < ep_counts['eps_most'] or 55 < sum(ep_counts['totals'])
|
|
display_show_minimum = sickgear.DISPLAY_SHOW_MINIMUM or force_display_show_minimum
|
|
|
|
for row in my_db.select('SELECT max(season) AS latest'
|
|
' FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' AND 1000 < airdate AND ? < status',
|
|
[show_obj.tvid, show_obj.prodid,
|
|
UNAIRED]):
|
|
t.latest_season = row['latest'] or {0: 1, 1: 1, 2: -1}.get(sickgear.DISPLAY_SHOW_VIEWMODE)
|
|
|
|
t.season_min = ([], [1])[2 < t.latest_season] + [t.latest_season]
|
|
t.other_seasons = (list(set(all_seasons) - set(t.season_min)), [])[display_show_minimum]
|
|
t.seasons = []
|
|
for cur_season in all_seasons:
|
|
t.seasons += [(cur_season, [None] if cur_season not in (t.season_min + t.other_seasons) else my_db.select(
|
|
'SELECT *'
|
|
' FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' AND season = ?'
|
|
' ORDER BY episode DESC',
|
|
[show_obj.tvid, show_obj.prodid, cur_season]
|
|
), scene_exceptions.ReleaseMap().has_season_exceptions(show_obj.tvid, show_obj.prodid, cur_season))]
|
|
|
|
for row in my_db.select('SELECT season, episode, status'
|
|
' FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' AND season IN (%s)' % ','.join(['?'] * len(t.season_min + t.other_seasons)),
|
|
[show_obj.tvid, show_obj.prodid]
|
|
+ t.season_min + t.other_seasons):
|
|
status_overview = show_obj.get_overview(row['status'])
|
|
if status_overview:
|
|
ep_cats['%sx%s' % (row['season'], row['episode'])] = status_overview
|
|
t.ep_cats = ep_cats
|
|
|
|
for row in my_db.select('SELECT season, count(*) AS cnt, status'
|
|
' FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' GROUP BY season, status',
|
|
[show_obj.tvid, show_obj.prodid]):
|
|
status_overview = show_obj.get_overview(row['status'])
|
|
if status_overview:
|
|
ep_counts[status_overview] += row['cnt']
|
|
if ARCHIVED == Quality.split_composite_status(row['status'])[0]:
|
|
ep_counts['archived'].setdefault(row['season'], 0)
|
|
ep_counts['archived'][row['season']] = row['cnt'] + ep_counts['archived'].get(row['season'], 0)
|
|
else:
|
|
ep_counts['status'].setdefault(row['season'], {})
|
|
ep_counts['status'][row['season']][status_overview] = row['cnt'] + \
|
|
ep_counts['status'][row['season']].get(status_overview, 0)
|
|
|
|
for row in my_db.select('SELECT season, count(*) AS cnt FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' AND \'\' != location'
|
|
' GROUP BY season',
|
|
[show_obj.tvid, show_obj.prodid]):
|
|
ep_counts['videos'][row['season']] = row['cnt']
|
|
t.ep_counts = ep_counts
|
|
|
|
t.sortedShowLists = self.sorted_show_lists()
|
|
t.tvshow_id_csv = []
|
|
tvshow_names = []
|
|
cur_sel = None
|
|
for cur_tvshow_types in t.sortedShowLists:
|
|
for cur_show_obj in cur_tvshow_types[1]:
|
|
t.tvshow_id_csv.append(cur_show_obj.tvid_prodid)
|
|
tvshow_names.append(cur_show_obj.name)
|
|
if show_obj.tvid_prodid == cur_show_obj.tvid_prodid:
|
|
cur_sel = len(tvshow_names)
|
|
|
|
last_item = len(tvshow_names)
|
|
t.prev_title = ''
|
|
t.next_title = ''
|
|
if cur_sel:
|
|
t.prev_title = 'Prev show, %s' % tvshow_names[(cur_sel - 2, last_item - 1)[1 == cur_sel]]
|
|
t.next_title = 'Next show, %s' % tvshow_names[(cur_sel, 0)[last_item == cur_sel]]
|
|
|
|
t.anigroups = None
|
|
if show_obj.is_anime:
|
|
t.anigroups = show_obj.release_groups
|
|
|
|
t.fanart = []
|
|
cache_obj = image_cache.ImageCache()
|
|
for img in glob.glob(cache_obj.fanart_path(show_obj.tvid, show_obj.prodid).replace('fanart.jpg', '*')) or []:
|
|
match = re.search(r'(\d+(?:\.(\w*?(\d*)))?\.\w{5,8})\.fanart\.', img, re.I)
|
|
if match and match.group(1):
|
|
t.fanart += [(match.group(1),
|
|
sickgear.FANART_RATINGS.get(tvid_prodid, {}).get(match.group(1), ''))]
|
|
|
|
t.start_image = None
|
|
ratings = [v for n, v in t.fanart]
|
|
if 20 in ratings:
|
|
t.start_image = ratings.index(20)
|
|
else:
|
|
rnd = [(x, v) for x, (n, v) in enumerate(t.fanart) if 30 != v]
|
|
grouped = [n for (n, v) in rnd if 10 == v]
|
|
if grouped:
|
|
t.start_image = grouped[random.randint(0, len(grouped) - 1)]
|
|
elif rnd:
|
|
t.start_image = rnd[random.randint(0, len(rnd) - 1)][0]
|
|
t.has_art = bool(len(t.fanart))
|
|
t.css = ' '.join(([], ['back-art'])[sickgear.DISPLAY_SHOW_BACKGROUND and t.has_art] +
|
|
([], ['translucent'])[sickgear.DISPLAY_SHOW_BACKGROUND_TRANSLUCENT] +
|
|
{0: [], 1: ['poster-right'], 2: ['poster-off']}.get(sickgear.DISPLAY_SHOW_VIEWART) +
|
|
([], ['min'])[display_show_minimum] +
|
|
([], ['min-force'])[force_display_show_minimum] +
|
|
[{0: 'reg', 1: 'pro', 2: 'pro ii'}.get(sickgear.DISPLAY_SHOW_VIEWMODE)])
|
|
|
|
t.clean_show_name = quote_plus(sickgear.indexermapper.clean_show_name(show_obj.name))
|
|
|
|
t.min_initial = Quality.get_quality_ui(min(Quality.split_quality(show_obj.quality)[0]))
|
|
t.show_obj.exceptions = scene_exceptions.ReleaseMap().get_alt_names(show_obj.tvid, show_obj.prodid)
|
|
# noinspection PyUnresolvedReferences
|
|
t.all_scene_exceptions = show_obj.exceptions # normally Unresolved as not a class attribute, force set above
|
|
t.scene_numbering = get_scene_numbering_for_show(show_obj.tvid, show_obj.prodid)
|
|
t.scene_absolute_numbering = get_scene_absolute_numbering_for_show(show_obj.tvid, show_obj.prodid)
|
|
t.xem_numbering = get_xem_numbering_for_show(show_obj.tvid, show_obj.prodid)
|
|
t.xem_absolute_numbering = get_xem_absolute_numbering_for_show(show_obj.tvid, show_obj.prodid)
|
|
|
|
return t.respond()
|
|
|
|
@staticmethod
|
|
def make_showlist_unique_names():
|
|
def titler(x):
|
|
return (remove_article(x), x)[not x or sickgear.SORT_ARTICLE].lower()
|
|
|
|
sorted_show_list = sorted(sickgear.showList, key=lambda x: titler(x.name))
|
|
year_check = re.compile(r' \(\d{4}\)$')
|
|
dups = {}
|
|
|
|
for i, val in enumerate(sorted_show_list):
|
|
if val.name not in dups:
|
|
# Store index of first occurrence and occurrence value
|
|
dups[val.name] = i
|
|
val.unique_name = val.name
|
|
else:
|
|
# remove cached parsed result
|
|
sickgear.name_parser.parser.name_parser_cache.flush(val)
|
|
if not year_check.search(sorted_show_list[dups[val.name]].name):
|
|
# add year to first show
|
|
first_ep = sorted_show_list[dups[val.name]].first_aired_regular_episode
|
|
start_year = (first_ep and first_ep.airdate and first_ep.airdate.year) or \
|
|
sorted_show_list[dups[val.name]].startyear
|
|
if start_year:
|
|
sorted_show_list[dups[val.name]].unique_name = '%s (%s)' % (
|
|
sorted_show_list[dups[val.name]].name, start_year)
|
|
dups[sorted_show_list[dups[val.name]].unique_name] = i
|
|
if not year_check.search(sorted_show_list[i].name):
|
|
# add year to duplicate
|
|
first_ep = sorted_show_list[i].first_aired_regular_episode
|
|
start_year = (first_ep and first_ep.airdate and first_ep.airdate.year) or sorted_show_list[
|
|
i].startyear
|
|
if start_year:
|
|
sorted_show_list[i].unique_name = '%s (%s)' % (sorted_show_list[i].name, start_year)
|
|
dups[sorted_show_list[i].unique_name] = i
|
|
|
|
name_cache.build_name_cache()
|
|
|
|
@staticmethod
|
|
def sorted_show_lists():
|
|
def titler(x):
|
|
return (remove_article(x), x)[not x or sickgear.SORT_ARTICLE].lower()
|
|
|
|
if 'custom' == sickgear.SHOWLIST_TAGVIEW:
|
|
sorted_show_lists = []
|
|
for tag in sickgear.SHOW_TAGS:
|
|
results = list(filter(lambda _so: _so.tag == tag, sickgear.showList))
|
|
if results:
|
|
sorted_show_lists.append([tag, sorted(results, key=lambda x: titler(x.unique_name))])
|
|
# handle orphaned shows
|
|
if len(sickgear.showList) != sum([len(so[1]) for so in sorted_show_lists]):
|
|
used_ids = set()
|
|
for so in sorted_show_lists:
|
|
for y in so[1]:
|
|
used_ids |= {y.tvid_prodid}
|
|
|
|
showlist = dict()
|
|
all_ids = set([cur_so.tvid_prodid for cur_so in sickgear.showList])
|
|
for iid in list(all_ids - used_ids):
|
|
show_obj = None
|
|
try:
|
|
show_obj = helpers.find_show_by_id(iid)
|
|
except (BaseException, Exception):
|
|
pass
|
|
if show_obj:
|
|
if show_obj.tag in showlist:
|
|
showlist[show_obj.tag] += [show_obj]
|
|
else:
|
|
showlist[show_obj.tag] = [show_obj]
|
|
|
|
sorted_show_lists += [[key, shows] for key, shows in iteritems(showlist)]
|
|
|
|
elif 'anime' == sickgear.SHOWLIST_TAGVIEW:
|
|
shows = []
|
|
anime = []
|
|
for cur_show_obj in sickgear.showList:
|
|
if cur_show_obj.is_anime:
|
|
anime.append(cur_show_obj)
|
|
else:
|
|
shows.append(cur_show_obj)
|
|
sorted_show_lists = [['Shows', sorted(shows, key=lambda x: titler(x.unique_name))],
|
|
['Anime', sorted(anime, key=lambda x: titler(x.unique_name))]]
|
|
|
|
else:
|
|
sorted_show_lists = [
|
|
['Show List', sorted(sickgear.showList, key=lambda x: titler(x.unique_name))]]
|
|
|
|
return sorted_show_lists
|
|
|
|
@staticmethod
|
|
def plot_details(tvid_prodid, season, episode):
|
|
|
|
my_db = db.DBConnection()
|
|
sql_result = my_db.select(
|
|
'SELECT description'
|
|
' FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' AND season = ? AND episode = ?',
|
|
TVidProdid(tvid_prodid).list + [int(season), int(episode)])
|
|
return 'Episode not found.' if not sql_result else (sql_result[0]['description'] or '')[:250:]
|
|
|
|
@staticmethod
|
|
def media_stats(tvid_prodid=None):
|
|
|
|
if None is tvid_prodid:
|
|
shows = sickgear.showList
|
|
else:
|
|
shows = [helpers.find_show_by_id(tvid_prodid)]
|
|
|
|
response = {}
|
|
for cur_show_obj in shows:
|
|
if cur_show_obj and cur_show_obj.path:
|
|
loc_size = helpers.get_size(cur_show_obj.path)
|
|
num_files, smallest, largest, average_size = get_media_stats(cur_show_obj.path)
|
|
response[cur_show_obj.tvid_prodid] = {'message': 'No media files'} if not num_files else \
|
|
{
|
|
'nFiles': num_files,
|
|
'bSmallest': smallest, 'hSmallest': helpers.human(smallest),
|
|
'bLargest': largest, 'hLargest': helpers.human(largest),
|
|
'bAverageSize': average_size, 'hAverageSize': helpers.human(average_size)
|
|
}
|
|
|
|
response[cur_show_obj.tvid_prodid].update({
|
|
'path': cur_show_obj.path, 'bSize': loc_size, 'hSize': helpers.human(loc_size)})
|
|
|
|
return json_dumps(response)
|
|
|
|
@staticmethod
|
|
def scene_exceptions(tvid_prodid, wanted_season=None):
|
|
|
|
exceptions_list = scene_exceptions.ReleaseMap().get_show_exceptions(tvid_prodid)
|
|
wanted_season = helpers.try_int(wanted_season, None)
|
|
wanted_not_found = None is not wanted_season and wanted_season not in exceptions_list
|
|
if not exceptions_list or wanted_not_found:
|
|
return ('No scene exceptions', 'No season exceptions')[wanted_not_found]
|
|
|
|
out = []
|
|
for season, names in iter(sorted(iteritems(exceptions_list))):
|
|
if None is wanted_season or wanted_season == season:
|
|
out.append('S%s: %s' % (('%02d' % season, '*')[-1 == season], ',<br>\n'.join(names)))
|
|
return '\n<hr class="exception-divider">\n'.join(out)
|
|
|
|
@staticmethod
|
|
def switch_infosrc(prodid, tvid, m_prodid, m_tvid, set_pause=False, mark_wanted=False):
|
|
tvid = helpers.try_int(tvid)
|
|
prodid = helpers.try_int(prodid)
|
|
m_tvid = helpers.try_int(m_tvid)
|
|
m_prodid = helpers.try_int(m_prodid)
|
|
show_obj = helpers.find_show_by_id({tvid: prodid}, no_mapped_ids=True)
|
|
try:
|
|
sickgear.show_queue_scheduler.action.switch_show(show_obj=show_obj, new_tvid=m_tvid,
|
|
new_prodid=m_prodid, force_id=True,
|
|
set_pause=set_pause, mark_wanted=mark_wanted)
|
|
except (BaseException, Exception) as e:
|
|
logger.warning('Could not add show %s to switch queue: %s' % (show_obj.tvid_prodid, ex(e)))
|
|
|
|
ui.notifications.message('TV info source switch', 'Queued switch of tv info source')
|
|
return {'Success': 'Switched to new TV info source'}
|
|
|
|
def save_mapping(self, tvid_prodid, **kwargs):
|
|
|
|
m_tvid = helpers.try_int(kwargs.get('m_tvid'))
|
|
m_prodid = helpers.try_int(kwargs.get('m_prodid'))
|
|
show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
response = {}
|
|
if not show_obj:
|
|
return json_dumps(response)
|
|
new_ids = {}
|
|
save_map = []
|
|
with show_obj.lock:
|
|
for k, v in iteritems(kwargs):
|
|
t = re.search(r'mid-(\d+)', k)
|
|
if t:
|
|
i = helpers.try_int(v, None)
|
|
if None is not i:
|
|
new_ids.setdefault(helpers.try_int(t.group(1)),
|
|
{'id': 0,
|
|
'status': MapStatus.NONE,
|
|
'date': dt_date.fromordinal(1)
|
|
})['id'] = i
|
|
else:
|
|
t = re.search(r'lockid-(\d+)', k)
|
|
if t:
|
|
new_ids.setdefault(helpers.try_int(t.group(1)), {
|
|
'id': 0, 'status': MapStatus.NONE,
|
|
'date': dt_date.fromordinal(1)})['status'] = \
|
|
(MapStatus.NONE, MapStatus.NO_AUTOMATIC_CHANGE)['true' == v]
|
|
if new_ids:
|
|
for k, v in iteritems(new_ids):
|
|
if None is v.get('id') or None is v.get('status'):
|
|
continue
|
|
if (show_obj.ids.get(k, {'id': 0}).get('id') != v.get('id')
|
|
or (MapStatus.NO_AUTOMATIC_CHANGE == v.get('status')
|
|
and MapStatus.NO_AUTOMATIC_CHANGE != show_obj.ids.get(
|
|
k, {'status': MapStatus.NONE}).get('status'))
|
|
or (MapStatus.NO_AUTOMATIC_CHANGE != v.get('status')
|
|
and MapStatus.NO_AUTOMATIC_CHANGE == show_obj.ids.get(
|
|
k, {'status': MapStatus.NONE}).get('status'))):
|
|
show_obj.ids[k]['id'] = (0, v['id'])[v['id'] >= 0]
|
|
show_obj.ids[k]['status'] = (MapStatus.NOT_FOUND, v['status'])[v['id'] != 0]
|
|
save_map.append(k)
|
|
if len(save_map):
|
|
save_mapping(show_obj, save_map=save_map)
|
|
ui.notifications.message('Mappings saved')
|
|
elif show_obj.tvid == m_tvid:
|
|
ui.notifications.message('Mappings unchanged, not saving.')
|
|
|
|
main_ids = [show_obj.prodid, helpers.try_int(kwargs.get('tvid')), m_prodid, m_tvid]
|
|
if all([0 < x for x in main_ids]) and sickgear.TVInfoAPI(m_tvid).config.get('active') and \
|
|
not sickgear.TVInfoAPI(m_tvid).config.get('defunct') and \
|
|
not sickgear.TVInfoAPI(m_tvid).config.get('mapped_only') and \
|
|
(m_tvid != show_obj.tvid or m_prodid != show_obj.prodid):
|
|
try:
|
|
new_show_obj = helpers.find_show_by_id({m_tvid: m_prodid}, no_mapped_ids=False, check_multishow=True)
|
|
mtvid_prodid = TVidProdid({m_tvid: m_prodid})()
|
|
if not new_show_obj or (new_show_obj.tvid == show_obj.tvid and new_show_obj.prodid == show_obj.prodid):
|
|
main_ids += [bool(helpers.try_int(kwargs.get(x))) for x in ('paused', 'markwanted')]
|
|
response = dict(switch=self.switch_infosrc(*main_ids), mtvid_prodid=mtvid_prodid)
|
|
else:
|
|
msg = 'Main ID unchanged, because show from %s with ID: %s exists in DB.' % \
|
|
(sickgear.TVInfoAPI(m_tvid).name, mtvid_prodid)
|
|
logger.warning(msg)
|
|
ui.notifications.message(*[s.strip() for s in msg.split(',')])
|
|
except MultipleShowObjectsException:
|
|
msg = 'Main ID unchanged, because show from %s with ID: %s exists in DB.' % \
|
|
(sickgear.TVInfoAPI(m_tvid).name, m_prodid)
|
|
logger.warning(msg)
|
|
ui.notifications.message(*[s.strip() for s in msg.split(',')])
|
|
|
|
response.update({
|
|
'map': {k: {r: w for r, w in iteritems(v) if 'date' != r} for k, v in iteritems(show_obj.ids)}
|
|
})
|
|
return json_dumps(response)
|
|
|
|
@staticmethod
|
|
def force_mapping(tvid_prodid, **kwargs):
|
|
|
|
show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
if not show_obj:
|
|
return json_dumps({})
|
|
save_map = []
|
|
with show_obj.lock:
|
|
for k, v in iteritems(kwargs):
|
|
t = re.search(r'lockid-(\d+)', k)
|
|
if t:
|
|
new_status = (MapStatus.NONE, MapStatus.NO_AUTOMATIC_CHANGE)['true' == v]
|
|
old_status = show_obj.ids.get(helpers.try_int(t.group(1)), {'status': MapStatus.NONE})['status']
|
|
if ((MapStatus.NO_AUTOMATIC_CHANGE == new_status and
|
|
MapStatus.NO_AUTOMATIC_CHANGE != old_status) or
|
|
(MapStatus.NO_AUTOMATIC_CHANGE != new_status and
|
|
MapStatus.NO_AUTOMATIC_CHANGE == old_status)):
|
|
locked_val = helpers.try_int(t.group(1))
|
|
if 'mid-%s' % locked_val in kwargs:
|
|
mid_val = helpers.try_int(kwargs['mid-%s' % locked_val], None)
|
|
if None is not mid_val and 0 <= mid_val:
|
|
show_obj.ids.setdefault(locked_val, {
|
|
'id': 0, 'status': MapStatus.NONE,
|
|
'date': dt_date.fromordinal(1)})['id'] = mid_val
|
|
show_obj.ids.setdefault(locked_val, {
|
|
'id': 0, 'status': MapStatus.NONE,
|
|
'date': dt_date.fromordinal(1)})['status'] = new_status
|
|
save_map.append(locked_val)
|
|
if len(save_map):
|
|
save_mapping(show_obj, save_map=save_map)
|
|
map_indexers_to_show(show_obj, force=True)
|
|
ui.notifications.message('Mapping Reloaded')
|
|
return json_dumps({k: {r: w for r, w in iteritems(v) if 'date' != r} for k, v in iteritems(show_obj.ids)})
|
|
|
|
@staticmethod
|
|
def fanart_tmpl(t):
|
|
t.fanart = []
|
|
cache_obj = image_cache.ImageCache()
|
|
show_obj = getattr(t, 'show_obj', None) or getattr(t, 'show', None)
|
|
for img in glob.glob(cache_obj.fanart_path(
|
|
show_obj.tvid, show_obj.prodid).replace('fanart.jpg', '*')) or []:
|
|
match = re.search(r'(\d+(?:\.(\w*?(\d*)))?\.\w{5,8})\.fanart\.', img, re.I)
|
|
if match and match.group(1):
|
|
t.fanart += [(match.group(1),
|
|
sickgear.FANART_RATINGS.get(show_obj.tvid_prodid, {}).get(match.group(1), ''))]
|
|
|
|
t.start_image = None
|
|
ratings = [v for n, v in t.fanart]
|
|
if 20 in ratings:
|
|
t.start_image = ratings.index(20)
|
|
else:
|
|
rnd = [(x, v) for x, (n, v) in enumerate(t.fanart) if 30 != v]
|
|
grouped = [n for (n, v) in rnd if 10 == v]
|
|
if grouped:
|
|
t.start_image = grouped[random.randint(0, len(grouped) - 1)]
|
|
elif rnd:
|
|
t.start_image = rnd[random.randint(0, len(rnd) - 1)][0]
|
|
|
|
t.has_art = bool(len(t.fanart))
|
|
t.css = ' '.join(([], ['back-art'])[sickgear.DISPLAY_SHOW_BACKGROUND and t.has_art] +
|
|
([], ['translucent'])[sickgear.DISPLAY_SHOW_BACKGROUND_TRANSLUCENT] +
|
|
[{0: 'reg', 1: 'pro', 2: 'pro ii'}.get(sickgear.DISPLAY_SHOW_VIEWMODE)])
|
|
|
|
def edit_show(self, tvid_prodid=None, location=None,
|
|
any_qualities=None, best_qualities=None, exceptions_list=None,
|
|
flatten_folders=None, paused=None, direct_call=False, air_by_date=None, sports=None, dvdorder=None,
|
|
tvinfo_lang=None, subs=None, upgrade_once=None, rls_ignore_words=None,
|
|
rls_require_words=None, anime=None, allowlist=None, blocklist=None,
|
|
scene=None, prune=None, tag=None, quality_preset=None, reset_fanart=None,
|
|
rls_global_exclude_ignore=None, rls_global_exclude_require=None, **kwargs):
|
|
|
|
any_qualities = any_qualities if None is not any_qualities else []
|
|
best_qualities = best_qualities if None is not best_qualities else []
|
|
exceptions_list = exceptions_list if None is not exceptions_list else []
|
|
|
|
if None is tvid_prodid:
|
|
err_string = 'Invalid show ID: ' + str(tvid_prodid)
|
|
if direct_call:
|
|
return [err_string]
|
|
return self._generic_message('Error', err_string)
|
|
|
|
show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
if not show_obj:
|
|
err_string = 'Unable to find the specified show: %s' % tvid_prodid
|
|
if direct_call:
|
|
return [err_string]
|
|
return self._generic_message('Error', err_string)
|
|
|
|
show_obj.exceptions = scene_exceptions.ReleaseMap().get_show_exceptions(tvid_prodid)
|
|
|
|
if None is not quality_preset and int(quality_preset):
|
|
best_qualities = []
|
|
|
|
if not location and not any_qualities and not best_qualities and not flatten_folders:
|
|
t = PageTemplate(web_handler=self, file='editShow.tmpl')
|
|
t.submenu = self.home_menu()
|
|
|
|
t.expand_ids = all([kwargs.get('tvsrc'), helpers.try_int(kwargs.get('srcid'))])
|
|
t.tvsrc = int(kwargs.get('tvsrc', 0))
|
|
t.srcid = helpers.try_int(kwargs.get('srcid'))
|
|
|
|
my_db = db.DBConnection()
|
|
# noinspection SqlRedundantOrderingDirection
|
|
t.seasonResults = my_db.select(
|
|
'SELECT DISTINCT season'
|
|
' FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' ORDER BY season DESC',
|
|
[show_obj.tvid, show_obj.prodid])
|
|
|
|
if show_obj.is_anime:
|
|
if not show_obj.release_groups:
|
|
show_obj.release_groups = AniGroupList(show_obj.tvid, show_obj.prodid, show_obj.tvid_prodid)
|
|
t.allowlist = show_obj.release_groups.allowlist
|
|
t.blocklist = show_obj.release_groups.blocklist
|
|
|
|
t.groups = pull_anidb_groups(show_obj.name)
|
|
if None is t.groups:
|
|
t.groups = [dict(name='Did not initialise AniDB. Check debug log if reqd.', rating='', range='')]
|
|
elif False is t.groups:
|
|
t.groups = [dict(name='Fail: AniDB connect. Restart SG else check debug log', rating='', range='')]
|
|
elif isinstance(t.groups, list) and 0 == len(t.groups):
|
|
t.groups = [dict(name='No groups listed in API response', rating='', range='')]
|
|
|
|
with show_obj.lock:
|
|
t.show_obj = show_obj
|
|
t.show_has_scene_map = sickgear.scene_numbering.has_xem_scene_mapping(
|
|
show_obj.tvid, show_obj.prodid)
|
|
|
|
# noinspection PyTypeChecker
|
|
self.fanart_tmpl(t)
|
|
t.num_ratings = len(sickgear.FANART_RATINGS.get(tvid_prodid, {}))
|
|
|
|
t.unlock_main_id = 0 != show_obj.not_found_count
|
|
t.showname_enc = quote_plus(show_obj.name.encode('utf-8'))
|
|
|
|
show_message = ''
|
|
|
|
if 0 != show_obj.not_found_count:
|
|
# noinspection PyUnresolvedReferences
|
|
last_found = ('', ' since %s' % SGDatetime.fromordinal(
|
|
show_obj.last_found_on_indexer).sbfdate())[1 < show_obj.last_found_on_indexer]
|
|
show_message = (
|
|
'The main ID of this show has been <span class="addQTip" title="many reasons exist, including: '
|
|
+ '\nshow flagged as a duplicate, removed completely... etc">abandoned</span>%s' % last_found
|
|
+ '<br>search for a replacement in the "<b>Related show IDs</b>" section of the "<b>Other</b>" tab')
|
|
|
|
t.show_message = show_message
|
|
|
|
return t.respond()
|
|
|
|
flatten_folders = config.checkbox_to_value(flatten_folders)
|
|
dvdorder = config.checkbox_to_value(dvdorder)
|
|
upgrade_once = config.checkbox_to_value(upgrade_once)
|
|
paused = config.checkbox_to_value(paused)
|
|
air_by_date = config.checkbox_to_value(air_by_date)
|
|
scene = config.checkbox_to_value(scene)
|
|
sports = config.checkbox_to_value(sports)
|
|
anime = config.checkbox_to_value(anime)
|
|
subs = config.checkbox_to_value(subs)
|
|
|
|
if config.checkbox_to_value(reset_fanart) and sickgear.FANART_RATINGS.get(tvid_prodid):
|
|
del sickgear.FANART_RATINGS[tvid_prodid]
|
|
sickgear.save_config()
|
|
|
|
t = sickgear.TVInfoAPI(show_obj.tvid).setup()
|
|
if tvinfo_lang and (tvinfo_lang in t.config['valid_languages'] or
|
|
tvinfo_lang in (_l.get('sg_lang') for _l in t.get_languages() or [])):
|
|
infosrc_lang = tvinfo_lang
|
|
else:
|
|
infosrc_lang = show_obj.lang
|
|
|
|
# if we changed the language then kick off an update
|
|
if infosrc_lang == show_obj.lang:
|
|
do_update = False
|
|
else:
|
|
do_update = True
|
|
|
|
if scene == show_obj.scene and anime == show_obj.anime:
|
|
do_update_scene_numbering = False
|
|
else:
|
|
do_update_scene_numbering = True
|
|
|
|
if type(any_qualities) != list:
|
|
any_qualities = [any_qualities]
|
|
|
|
if type(best_qualities) != list:
|
|
best_qualities = [best_qualities]
|
|
|
|
if type(exceptions_list) != list:
|
|
exceptions_list = [exceptions_list]
|
|
|
|
# If direct call from mass_edit_update no scene exceptions handling or blockandallow list handling or tags
|
|
if direct_call:
|
|
do_update_exceptions = False
|
|
else:
|
|
do_update_exceptions = True # TODO: make this smarter and only update on changes
|
|
|
|
with show_obj.lock:
|
|
if anime:
|
|
if not show_obj.release_groups:
|
|
show_obj.release_groups = AniGroupList(
|
|
show_obj.tvid, show_obj.prodid, show_obj.tvid_prodid)
|
|
if allowlist:
|
|
shortallowlist = short_group_names(allowlist)
|
|
show_obj.release_groups.set_allow_keywords(shortallowlist)
|
|
else:
|
|
show_obj.release_groups.set_allow_keywords([])
|
|
|
|
if blocklist:
|
|
shortblocklist = short_group_names(blocklist)
|
|
show_obj.release_groups.set_block_keywords(shortblocklist)
|
|
else:
|
|
show_obj.release_groups.set_block_keywords([])
|
|
|
|
errors = []
|
|
with show_obj.lock:
|
|
show_obj.quality = Quality.combine_qualities(list(map(int, any_qualities)), list(map(int, best_qualities)))
|
|
show_obj.upgrade_once = upgrade_once
|
|
|
|
# reversed for now
|
|
if bool(show_obj.flatten_folders) != bool(flatten_folders):
|
|
show_obj.flatten_folders = flatten_folders
|
|
try:
|
|
sickgear.show_queue_scheduler.action.refresh_show(show_obj)
|
|
except exceptions_helper.CantRefreshException as e:
|
|
errors.append('Unable to refresh this show: ' + ex(e))
|
|
|
|
if bool(anime) != show_obj.is_anime:
|
|
sickgear.name_parser.parser.name_parser_cache.flush(show_obj)
|
|
|
|
show_obj.paused = paused
|
|
show_obj.scene = scene
|
|
show_obj.anime = anime
|
|
show_obj.sports = sports
|
|
show_obj.subtitles = subs
|
|
show_obj.air_by_date = air_by_date
|
|
show_obj.tag = tag
|
|
show_obj.prune = config.minimax(prune, 0, 0, 9999)
|
|
|
|
if not direct_call:
|
|
show_obj.lang = infosrc_lang
|
|
show_obj.dvdorder = dvdorder
|
|
new_ignore_words, new_i_regex = helpers.split_word_str(rls_ignore_words.strip())
|
|
new_ignore_words -= sickgear.IGNORE_WORDS
|
|
if 0 == len(new_ignore_words):
|
|
new_i_regex = False
|
|
show_obj.rls_ignore_words, show_obj.rls_ignore_words_regex = new_ignore_words, new_i_regex
|
|
new_require_words, new_r_regex = helpers.split_word_str(rls_require_words.strip())
|
|
new_require_words -= sickgear.REQUIRE_WORDS
|
|
if 0 == len(new_require_words):
|
|
new_r_regex = False
|
|
show_obj.rls_require_words, show_obj.rls_require_words_regex = new_require_words, new_r_regex
|
|
if isinstance(rls_global_exclude_ignore, list):
|
|
show_obj.rls_global_exclude_ignore = set(r for r in rls_global_exclude_ignore if '.*' != r)
|
|
elif isinstance(rls_global_exclude_ignore, string_types) and '.*' != rls_global_exclude_ignore:
|
|
show_obj.rls_global_exclude_ignore = {rls_global_exclude_ignore}
|
|
else:
|
|
show_obj.rls_global_exclude_ignore = set()
|
|
if isinstance(rls_global_exclude_require, list):
|
|
show_obj.rls_global_exclude_require = set(r for r in rls_global_exclude_require if '.*' != r)
|
|
elif isinstance(rls_global_exclude_require, string_types) and '.*' != rls_global_exclude_require:
|
|
show_obj.rls_global_exclude_require = {rls_global_exclude_require}
|
|
else:
|
|
show_obj.rls_global_exclude_require = set()
|
|
clean_ignore_require_words()
|
|
|
|
# if we change location clear the db of episodes, change it, write to db, and rescan
|
|
# noinspection PyProtectedMember
|
|
old_path = os.path.normpath(show_obj._location)
|
|
new_path = os.path.normpath(location)
|
|
if old_path != new_path:
|
|
logger.debug(f'{old_path} != {new_path}')
|
|
if not os.path.isdir(new_path) and not sickgear.CREATE_MISSING_SHOW_DIRS:
|
|
errors.append(f'New location <tt>{new_path}</tt> does not exist')
|
|
|
|
# don't bother if we're going to update anyway
|
|
elif not do_update:
|
|
# change it
|
|
try:
|
|
show_obj.location = new_path
|
|
try:
|
|
sickgear.show_queue_scheduler.action.refresh_show(show_obj)
|
|
except exceptions_helper.CantRefreshException as e:
|
|
errors.append('Unable to refresh this show:' + ex(e))
|
|
# grab updated info from TVDB
|
|
# show_obj.load_episodes_from_tvinfo()
|
|
# rescan the episodes in the new folder
|
|
except exceptions_helper.NoNFOException:
|
|
errors.append(f'The folder at <tt>{new_path}</tt> doesn"t contain a tvshow.nfo -'
|
|
f' copy your files to that folder before you change the directory in SickGear.')
|
|
|
|
# save it to the DB
|
|
show_obj.save_to_db()
|
|
|
|
# force the update
|
|
if do_update:
|
|
try:
|
|
sickgear.show_queue_scheduler.action.update_show(show_obj, True)
|
|
helpers.cpu_sleep()
|
|
except exceptions_helper.CantUpdateException:
|
|
errors.append('Unable to force an update on the show.')
|
|
|
|
if do_update_exceptions:
|
|
try:
|
|
scene_exceptions.ReleaseMap().update_exceptions(show_obj, exceptions_list)
|
|
helpers.cpu_sleep()
|
|
except exceptions_helper.CantUpdateException:
|
|
errors.append('Unable to force an update on scene exceptions of the show.')
|
|
|
|
if do_update_scene_numbering:
|
|
try:
|
|
sickgear.scene_numbering.xem_refresh(show_obj.tvid, show_obj.prodid)
|
|
helpers.cpu_sleep()
|
|
except exceptions_helper.CantUpdateException:
|
|
errors.append('Unable to force an update on scene numbering of the show.')
|
|
|
|
if direct_call:
|
|
return errors
|
|
|
|
if 0 < len(errors):
|
|
ui.notifications.error('%d error%s while saving changes:' % (len(errors), '' if 1 == len(errors) else 's'),
|
|
'<ul>' + '\n'.join(['<li>%s</li>' % error for error in errors]) + '</ul>')
|
|
|
|
self.redirect('/home/view-show?tvid_prodid=%s' % tvid_prodid)
|
|
|
|
def delete_show(self, tvid_prodid=None, full=0):
|
|
|
|
if None is tvid_prodid:
|
|
return self._generic_message('Error', 'Invalid show ID')
|
|
|
|
show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
|
|
if None is show_obj:
|
|
return self._generic_message('Error', 'Unable to find the specified show')
|
|
|
|
if sickgear.show_queue_scheduler.action.is_being_added(
|
|
show_obj) or sickgear.show_queue_scheduler.action.is_being_updated(show_obj):
|
|
return self._generic_message("Error", "Shows can't be deleted while they're being added or updated.")
|
|
|
|
# if sickgear.USE_TRAKT and sickgear.TRAKT_SYNC:
|
|
# # remove show from trakt.tv library
|
|
# sickgear.trakt_checker_scheduler.action.removeShowFromTraktLibrary(show_obj)
|
|
|
|
show_obj.delete_show(bool(full))
|
|
|
|
ui.notifications.message('%s with %s' % (('Deleting', 'Trashing')[sickgear.TRASH_REMOVE_SHOW],
|
|
('media left untouched', 'all related media')[bool(full)]),
|
|
'<b>%s</b>' % show_obj.unique_name)
|
|
self.redirect('/home/')
|
|
|
|
def update_cast(self, tvid_prodid=None):
|
|
if None is tvid_prodid:
|
|
return self._generic_message('Error', 'Invalid show ID')
|
|
|
|
show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
|
|
if None is show_obj:
|
|
return self._generic_message('Error', 'Unable to find the specified show')
|
|
|
|
# force the update from the DB
|
|
try:
|
|
sickgear.people_queue_scheduler.action.add_cast_update(show_obj=show_obj, show_info_cast=None)
|
|
except (BaseException, Exception) as e:
|
|
ui.notifications.error('Unable to refresh this show.', ex(e))
|
|
|
|
helpers.cpu_sleep()
|
|
|
|
self.redirect('/home/view-show?tvid_prodid=%s' % show_obj.tvid_prodid)
|
|
|
|
def refresh_show(self, tvid_prodid=None):
|
|
|
|
if None is tvid_prodid:
|
|
return self._generic_message('Error', 'Invalid show ID')
|
|
|
|
show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
|
|
if None is show_obj:
|
|
return self._generic_message('Error', 'Unable to find the specified show')
|
|
|
|
# force the update from the DB
|
|
try:
|
|
sickgear.show_queue_scheduler.action.refresh_show(show_obj)
|
|
except exceptions_helper.CantRefreshException as e:
|
|
ui.notifications.error('Unable to refresh this show.', ex(e))
|
|
|
|
helpers.cpu_sleep()
|
|
|
|
self.redirect('/home/view-show?tvid_prodid=%s' % show_obj.tvid_prodid)
|
|
|
|
def update_show(self, tvid_prodid=None, force=0, web=0):
|
|
|
|
if None is tvid_prodid:
|
|
return self._generic_message('Error', 'Invalid show ID')
|
|
|
|
show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
|
|
if None is show_obj:
|
|
return self._generic_message('Error', 'Unable to find the specified show')
|
|
|
|
# force the update
|
|
try:
|
|
sickgear.show_queue_scheduler.action.update_show(show_obj, bool(force), bool(web))
|
|
except exceptions_helper.CantUpdateException as e:
|
|
ui.notifications.error('Unable to update this show.',
|
|
ex(e))
|
|
|
|
helpers.cpu_sleep()
|
|
|
|
self.redirect('/home/view-show?tvid_prodid=%s' % show_obj.tvid_prodid)
|
|
|
|
# noinspection PyUnusedLocal
|
|
def subtitle_show(self, tvid_prodid=None, force=0):
|
|
|
|
if None is tvid_prodid:
|
|
return self._generic_message('Error', 'Invalid show ID')
|
|
|
|
show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
|
|
if None is show_obj:
|
|
return self._generic_message('Error', 'Unable to find the specified show')
|
|
|
|
# search and download subtitles
|
|
if sickgear.USE_SUBTITLES:
|
|
sickgear.show_queue_scheduler.action.download_subtitles(show_obj)
|
|
|
|
helpers.cpu_sleep()
|
|
|
|
self.redirect('/home/view-show?tvid_prodid=%s' % show_obj.tvid_prodid)
|
|
|
|
# noinspection PyUnusedLocal
|
|
def update_mb(self, tvid_prodid=None, **kwargs):
|
|
|
|
if notifiers.NotifierFactory().get('EMBY').update_library(
|
|
helpers.find_show_by_id(tvid_prodid), force=True):
|
|
ui.notifications.message('Library update command sent to Emby host(s): ' + sickgear.EMBY_HOST)
|
|
else:
|
|
ui.notifications.error('Unable to contact one or more Emby host(s): ' + sickgear.EMBY_HOST)
|
|
self.redirect('/home/')
|
|
|
|
def update_kodi(self, show_name=None):
|
|
|
|
# only send update to first host in the list -- workaround for kodi sql backend users
|
|
if sickgear.KODI_UPDATE_ONLYFIRST:
|
|
# only send update to first host in the list -- workaround for kodi sql backend users
|
|
host = sickgear.KODI_HOST.split(',')[0].strip()
|
|
else:
|
|
host = sickgear.KODI_HOST
|
|
|
|
if notifiers.NotifierFactory().get('KODI').update_library(show_name=show_name):
|
|
ui.notifications.message('Library update command sent to Kodi host(s): ' + host)
|
|
else:
|
|
ui.notifications.error('Unable to contact one or more Kodi host(s): ' + host)
|
|
self.redirect('/home/')
|
|
|
|
def update_plex(self):
|
|
result = notifiers.NotifierFactory().get('PLEX').update_library()
|
|
if 'Fail' not in result:
|
|
ui.notifications.message(
|
|
'Library update command sent to',
|
|
'Plex Media Server host(s): ' + sickgear.PLEX_SERVER_HOST.replace(',', ', '))
|
|
else:
|
|
ui.notifications.error('Unable to contact', 'Plex Media Server host(s): ' + result)
|
|
self.redirect('/home/')
|
|
|
|
def set_show_status(self, tvid_prodid=None, eps=None, status=None, direct=False):
|
|
|
|
if None is tvid_prodid or None is eps or None is status:
|
|
err_msg = 'You must specify a show and at least one episode'
|
|
if direct:
|
|
ui.notifications.error('Error', err_msg)
|
|
return json_dumps({'result': 'error'})
|
|
return self._generic_message('Error', err_msg)
|
|
|
|
use_default = False
|
|
if isinstance(status, string_types) and '-' in status:
|
|
use_default = True
|
|
status = status.replace('-', '')
|
|
status = int(status)
|
|
|
|
if status not in statusStrings:
|
|
err_msg = 'Invalid status'
|
|
if direct:
|
|
ui.notifications.error('Error', err_msg)
|
|
return json_dumps({'result': 'error'})
|
|
return self._generic_message('Error', err_msg)
|
|
|
|
show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
|
|
if None is show_obj:
|
|
err_msg = 'Error', 'Show not in show list'
|
|
if direct:
|
|
ui.notifications.error('Error', err_msg)
|
|
return json_dumps({'result': 'error'})
|
|
return self._generic_message('Error', err_msg)
|
|
|
|
min_initial = min(Quality.split_quality(show_obj.quality)[0])
|
|
segments = {}
|
|
if None is not eps:
|
|
|
|
sql_l = []
|
|
# sort episode numbers
|
|
eps_list = eps.split('|')
|
|
eps_list.sort()
|
|
for cur_ep in eps_list:
|
|
|
|
logger.debug(f'Attempting to set status on episode {cur_ep} to {status}')
|
|
|
|
ep_obj = show_obj.get_episode(*tuple([int(x) for x in cur_ep.split('x')]))
|
|
|
|
if None is ep_obj:
|
|
return self._generic_message('Error', 'Episode couldn\'t be retrieved')
|
|
|
|
if status in [WANTED, FAILED]:
|
|
# figure out what episodes are wanted so we can backlog them
|
|
if ep_obj.season in segments:
|
|
segments[ep_obj.season].append(ep_obj)
|
|
else:
|
|
segments[ep_obj.season] = [ep_obj]
|
|
|
|
with ep_obj.lock:
|
|
required = Quality.SNATCHED_ANY + Quality.DOWNLOADED
|
|
err_msg = ''
|
|
# don't let them mess up UNAIRED episodes
|
|
if UNAIRED == ep_obj.status:
|
|
err_msg = 'because it is unaired'
|
|
|
|
elif FAILED == status and ep_obj.status not in required:
|
|
err_msg = 'to failed because it\'s not snatched/downloaded'
|
|
|
|
elif status in Quality.DOWNLOADED \
|
|
and ep_obj.status not in required + Quality.ARCHIVED + [IGNORED, SKIPPED] \
|
|
and not os.path.isfile(ep_obj.location):
|
|
err_msg = 'to downloaded because it\'s not snatched/downloaded/archived'
|
|
|
|
if err_msg:
|
|
logger.error('Refusing to change status of %s %s' % (cur_ep, err_msg))
|
|
continue
|
|
|
|
if ARCHIVED == status:
|
|
if ep_obj.status in Quality.DOWNLOADED or direct:
|
|
ep_obj.status = Quality.composite_status(
|
|
ARCHIVED, (Quality.split_composite_status(ep_obj.status)[1], min_initial)[use_default])
|
|
elif DOWNLOADED == status:
|
|
if ep_obj.status in Quality.ARCHIVED:
|
|
ep_obj.status = Quality.composite_status(
|
|
DOWNLOADED, Quality.split_composite_status(ep_obj.status)[1])
|
|
else:
|
|
ep_obj.status = status
|
|
|
|
# mass add to database
|
|
result = ep_obj.get_sql()
|
|
if None is not result:
|
|
sql_l.append(result)
|
|
|
|
if 0 < len(sql_l):
|
|
my_db = db.DBConnection()
|
|
my_db.mass_action(sql_l)
|
|
|
|
if WANTED == status:
|
|
season_list = ''
|
|
season_wanted = []
|
|
if sickgear.search_backlog.BacklogSearcher.providers_active(scheduled=False):
|
|
for season, segment in iteritems(segments): # type: int, List[sickgear.tv.TVEpisode]
|
|
if not show_obj.paused:
|
|
cur_backlog_queue_item = search_queue.BacklogQueueItem(show_obj, segment)
|
|
sickgear.search_queue_scheduler.action.add_item(cur_backlog_queue_item)
|
|
|
|
if season not in season_wanted:
|
|
season_wanted += [season]
|
|
season_list += f'<li>Season {season}</li>'
|
|
logger.log(('Not adding wanted eps to backlog search for %s season %s because show is paused',
|
|
'Starting backlog search for %s season %s because eps were set to wanted')[
|
|
not show_obj.paused] % (show_obj.unique_name, season))
|
|
|
|
(title, msg) = (('Not starting backlog', 'Paused show prevented backlog search'),
|
|
('Backlog started', 'Backlog search started'))[not show_obj.paused]
|
|
|
|
if segments:
|
|
ui.notifications.message(title,
|
|
f'{msg} for the following seasons of <b>{show_obj.unique_name}</b>:<br>'
|
|
f'<ul>{season_list}</ul>')
|
|
else:
|
|
ui.notifications.message('Not starting backlog', 'No provider has active searching enabled')
|
|
|
|
elif FAILED == status:
|
|
msg = f'Retrying search automatically for the following season of <b>{show_obj.unique_name}</b>:<br><ul>'
|
|
|
|
for season, segment in iteritems(segments): # type: int, List[sickgear.tv.TVEpisode]
|
|
cur_failed_queue_item = search_queue.FailedQueueItem(show_obj, segment)
|
|
sickgear.search_queue_scheduler.action.add_item(cur_failed_queue_item)
|
|
|
|
msg += '<li>Season %s</li>' % season
|
|
logger.log(f'Retrying search for {show_obj.unique_name} season {season}'
|
|
f' because some eps were set to failed')
|
|
|
|
msg += '</ul>'
|
|
|
|
if segments:
|
|
ui.notifications.message('Retry search started', msg)
|
|
|
|
if direct:
|
|
return json_dumps({'result': 'success'})
|
|
self.redirect('/home/view-show?tvid_prodid=%s' % tvid_prodid)
|
|
|
|
def rename_media(self, tvid_prodid=None):
|
|
|
|
if None is tvid_prodid:
|
|
return self._generic_message('Error', 'You must specify a show')
|
|
|
|
show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
|
|
if None is show_obj:
|
|
return self._generic_message('Error', 'Show not in show list')
|
|
|
|
try:
|
|
_ = show_obj.location
|
|
except exceptions_helper.ShowDirNotFoundException:
|
|
return self._generic_message('Error', "Can't rename episodes when the show dir is missing.")
|
|
|
|
ep_obj_rename_list = []
|
|
|
|
ep_obj_list = show_obj.get_all_episodes(has_location=True)
|
|
|
|
for cur_ep_obj in ep_obj_list:
|
|
# Only want to rename if we have a location
|
|
if cur_ep_obj.location:
|
|
if cur_ep_obj.related_ep_obj:
|
|
# do we have one of multi-episodes in the rename list already
|
|
for _cur_ep_obj in cur_ep_obj.related_ep_obj + [cur_ep_obj]:
|
|
if _cur_ep_obj in ep_obj_rename_list:
|
|
break
|
|
ep_status, ep_qual = Quality.split_composite_status(_cur_ep_obj.status)
|
|
if not ep_qual:
|
|
continue
|
|
ep_obj_rename_list.append(cur_ep_obj)
|
|
else:
|
|
ep_status, ep_qual = Quality.split_composite_status(cur_ep_obj.status)
|
|
if not ep_qual:
|
|
continue
|
|
ep_obj_rename_list.append(cur_ep_obj)
|
|
|
|
if ep_obj_rename_list:
|
|
# present season DESC episode DESC on screen
|
|
ep_obj_rename_list.reverse()
|
|
|
|
t = PageTemplate(web_handler=self, file='testRename.tmpl')
|
|
t.submenu = [{'title': 'Edit', 'path': 'home/edit-show?tvid_prodid=%s' % show_obj.tvid_prodid}]
|
|
t.ep_obj_list = ep_obj_rename_list
|
|
t.show_obj = show_obj
|
|
|
|
# noinspection PyTypeChecker
|
|
self.fanart_tmpl(t)
|
|
|
|
return t.respond()
|
|
|
|
def do_rename(self, tvid_prodid=None, eps=None):
|
|
|
|
if None is tvid_prodid or None is eps:
|
|
err_msg = 'You must specify a show and at least one episode'
|
|
return self._generic_message('Error', err_msg)
|
|
|
|
show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
|
|
if None is show_obj:
|
|
err_msg = 'Error', 'Show not in show list'
|
|
return self._generic_message('Error', err_msg)
|
|
|
|
try:
|
|
_ = show_obj.location
|
|
except exceptions_helper.ShowDirNotFoundException:
|
|
return self._generic_message('Error', "Can't rename episodes when the show dir is missing.")
|
|
|
|
if None is eps:
|
|
return self.redirect('/home/view-show?tvid_prodid=%s' % tvid_prodid)
|
|
|
|
my_db = db.DBConnection()
|
|
tvid_prodid_obj = TVidProdid(tvid_prodid)
|
|
for cur_ep in eps.split('|'):
|
|
|
|
ep_info = cur_ep.split('x')
|
|
|
|
# noinspection SqlConstantCondition
|
|
sql_result = my_db.select(
|
|
'SELECT * FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' AND season = ? AND episode = ? AND 5=5',
|
|
tvid_prodid_obj.list
|
|
+ [ep_info[0], ep_info[1]])
|
|
if not sql_result:
|
|
logger.warning(f'Unable to find an episode for {cur_ep}, skipping')
|
|
continue
|
|
related_ep_result = my_db.select('SELECT * FROM tv_episodes WHERE location = ? AND episode != ?',
|
|
[sql_result[0]['location'], ep_info[1]])
|
|
|
|
root_ep_obj = show_obj.get_episode(int(ep_info[0]), int(ep_info[1]))
|
|
root_ep_obj.related_ep_obj = []
|
|
|
|
for cur_ep_result in related_ep_result:
|
|
ep_obj = show_obj.get_episode(int(cur_ep_result['season']), int(cur_ep_result['episode']))
|
|
if ep_obj not in root_ep_obj.related_ep_obj:
|
|
root_ep_obj.related_ep_obj.append(ep_obj)
|
|
|
|
root_ep_obj.rename()
|
|
|
|
self.redirect('/home/view-show?tvid_prodid=%s' % tvid_prodid)
|
|
|
|
def search_episode(self, tvid_prodid=None, season=None, episode=None, retry=False, **kwargs):
|
|
|
|
result = dict(result='failure')
|
|
|
|
# retrieve the episode object and fail if we can't get one
|
|
ep_obj = self._get_episode(tvid_prodid, season, episode)
|
|
if not isinstance(ep_obj, str):
|
|
if UNKNOWN == Quality.split_composite_status(ep_obj.status)[0]:
|
|
ep_obj.status = SKIPPED
|
|
|
|
# make a queue item for the TVEpisode and put it on the queue
|
|
ep_queue_item = (search_queue.ManualSearchQueueItem(ep_obj.show_obj, ep_obj),
|
|
search_queue.FailedQueueItem(ep_obj.show_obj, [ep_obj]))[retry]
|
|
|
|
sickgear.search_queue_scheduler.action.add_item(ep_queue_item)
|
|
|
|
if None is ep_queue_item.success: # invocation
|
|
result.update(dict(result=('success', 'queuing')[not ep_queue_item.started]))
|
|
# elif ep_queue_item.success:
|
|
# return self.search_q_status(
|
|
# '%s%s%s' % (ep_obj.show_obj.tvid, TVidProdid.glue, ep_obj.show_obj.prodid)) # page refresh
|
|
|
|
return json_dumps(result)
|
|
|
|
def episode_retry(self, tvid_prodid, season, episode):
|
|
|
|
return self.search_episode(tvid_prodid, season, episode, True)
|
|
|
|
# Return progress for queued, active and finished episodes
|
|
def search_q_status(self, tvid_prodid=None, **kwargs):
|
|
|
|
ep_data_list = []
|
|
seen_eps = set([])
|
|
|
|
# Queued searches
|
|
queued = sickgear.search_queue_scheduler.action.get_queued_manual(tvid_prodid)
|
|
|
|
# Active search
|
|
active = sickgear.search_queue_scheduler.action.get_current_manual_item(tvid_prodid)
|
|
|
|
# Finished searches
|
|
sickgear.search_queue.remove_old_fifo(sickgear.search_queue.MANUAL_SEARCH_HISTORY)
|
|
results = sickgear.search_queue.MANUAL_SEARCH_HISTORY
|
|
|
|
for item in filter(lambda q: hasattr(q, 'segment_ns'), queued):
|
|
for ep_ns in item.segment_ns:
|
|
ep_data, uniq_sxe = self.prepare_episode(ep_ns, 'queued')
|
|
ep_data_list.append(ep_data)
|
|
seen_eps.add(uniq_sxe)
|
|
|
|
if active and hasattr(active, 'segment_ns'):
|
|
episode_params = dict(([('searchstate', 'finished'), ('statusoverview', True)],
|
|
[('searchstate', 'searching'), ('statusoverview', False)])[None is active.success],
|
|
retrystate=True)
|
|
for ep_ns in active.segment_ns:
|
|
ep_data, uniq_sxe = self.prepare_episode(ep_ns, **episode_params)
|
|
ep_data_list.append(ep_data)
|
|
seen_eps.add(uniq_sxe)
|
|
|
|
episode_params = dict(searchstate='finished', retrystate=True, statusoverview=True)
|
|
for item in filter(lambda r: hasattr(r, 'segment_ns') and (
|
|
not tvid_prodid or tvid_prodid == str(r.show_ns.tvid_prodid)), results):
|
|
for ep_ns in filter(
|
|
lambda e: (e.show_ns.tvid, e.show_ns.prodid, e.season, e.episode) not in seen_eps, item.segment_ns):
|
|
ep_obj = getattr(ep_ns, 'ep_obj', None)
|
|
if not ep_obj:
|
|
continue
|
|
# try:
|
|
# show_obj = helpers.find_show_by_id(dict({ep_ns.show_ns.tvid: ep_ns.show_ns.prodid}))
|
|
# ep_obj = show_obj.get_episode(season=ep_ns.season, episode=ep_ns.episode)
|
|
# except (BaseException, Exception):
|
|
# continue
|
|
ep_data, uniq_sxe = self.prepare_episode(ep_obj, **episode_params)
|
|
ep_data_list.append(ep_data)
|
|
seen_eps.add(uniq_sxe)
|
|
|
|
for snatched in filter(lambda s: ((s.tvid, s.prodid, s.season, s.episode) not in seen_eps),
|
|
item.snatched_eps):
|
|
ep_obj = getattr(snatched, 'ep_obj', None)
|
|
if not ep_obj:
|
|
continue
|
|
# try:
|
|
# show_obj = helpers.find_show_by_id(snatched[0])
|
|
# ep_obj = show_obj.get_episode(season=snatched[1], episode=snatched[2])
|
|
# except (BaseException, Exception):
|
|
# continue
|
|
ep_data, uniq_sxe = self.prepare_episode(ep_obj, **episode_params)
|
|
ep_data_list.append(ep_data)
|
|
seen_eps.add(uniq_sxe)
|
|
|
|
if not ep_data_list:
|
|
return '{"episodes":[]}'
|
|
return json_dumps(dict(episodes=ep_data_list))
|
|
|
|
@staticmethod
|
|
def prepare_episode(ep_type, searchstate, retrystate=False, statusoverview=False):
|
|
"""
|
|
Prepare episode data and its unique id
|
|
:param ep_type: Episode structure containing the show that it relates to
|
|
:type ep_type: sickgear.tv.TVEpisode object or Episode Base Namespace
|
|
:param searchstate: Progress of search
|
|
:type searchstate: string
|
|
:param retrystate: True to add retrystate to data
|
|
:type retrystate: bool
|
|
:param statusoverview: True to add statusoverview to data
|
|
:type statusoverview: bool
|
|
:return: Episode data and its unique episode id
|
|
:rtype: tuple containing a dict and a tuple
|
|
"""
|
|
# Find the quality class for the episode
|
|
quality_class = Quality.qualityStrings[Quality.UNKNOWN]
|
|
ep_status, ep_quality = Quality.split_composite_status(ep_type.status)
|
|
for x in (SD, HD720p, HD1080p, UHD2160p):
|
|
if ep_quality in Quality.split_quality(x)[0]:
|
|
quality_class = qualityPresetStrings[x]
|
|
break
|
|
|
|
# show_item: ep_type.show_ns or ep_type.show_obj
|
|
show_item = getattr(ep_type, 'show_%s' % ('ns', 'obj')[isinstance(ep_type, sickgear.tv.TVEpisode)])
|
|
|
|
ep_data = dict(showindexer=show_item.tvid, showindexid=show_item.prodid,
|
|
season=ep_type.season, episode=ep_type.episode, quality=quality_class,
|
|
searchstate=searchstate, status=statusStrings[ep_type.status])
|
|
if retrystate:
|
|
retry_statuses = SNATCHED_ANY + [DOWNLOADED, ARCHIVED]
|
|
ep_data.update(dict(retrystate=sickgear.USE_FAILED_DOWNLOADS and ep_status in retry_statuses))
|
|
if statusoverview:
|
|
ep_data.update(dict(statusoverview=Overview.overviewStrings[
|
|
helpers.get_overview(ep_type.status, show_item.quality, show_item.upgrade_once)]))
|
|
|
|
return ep_data, (show_item.tvid, show_item.prodid, ep_type.season, ep_type.episode)
|
|
|
|
def search_episode_subtitles(self, tvid_prodid=None, season=None, episode=None):
|
|
|
|
if not sickgear.USE_SUBTITLES:
|
|
return json_dumps({'result': 'failure'})
|
|
|
|
# retrieve the episode object and fail if we can't get one
|
|
ep_obj = self._get_episode(tvid_prodid, season, episode)
|
|
if isinstance(ep_obj, str):
|
|
return json_dumps({'result': 'failure'})
|
|
|
|
# try to download subtitles for that episode
|
|
try:
|
|
previous_subtitles = set([subliminal.language.Language(x) for x in ep_obj.subtitles])
|
|
ep_obj.subtitles = set([x.language for x in next(itervalues(ep_obj.download_subtitles()))])
|
|
except (BaseException, Exception):
|
|
return json_dumps({'result': 'failure'})
|
|
|
|
# return the correct json value
|
|
if previous_subtitles != ep_obj.subtitles:
|
|
status = 'New subtitles downloaded: %s' % ' '.join([
|
|
"<img src='" + sickgear.WEB_ROOT + "/images/flags/" + x.alpha2 +
|
|
".png' alt='" + x.name + "'/>" for x in
|
|
sorted(list(ep_obj.subtitles.difference(previous_subtitles)))])
|
|
else:
|
|
status = 'No subtitles downloaded'
|
|
ui.notifications.message('Subtitles Search', status)
|
|
return json_dumps({'result': status,
|
|
'subtitles': ','.join(sorted([x.alpha2 for x in
|
|
ep_obj.subtitles.union(previous_subtitles)]))})
|
|
|
|
@staticmethod
|
|
def set_scene_numbering(tvid_prodid=None, for_season=None, for_episode=None, for_absolute=None,
|
|
scene_season=None, scene_episode=None, scene_absolute=None):
|
|
# TODO: ui does not currently send for_absolute
|
|
tvid, prodid = TVidProdid(tvid_prodid).list
|
|
result = set_scene_numbering_helper(tvid, prodid, for_season, for_episode, for_absolute, scene_season,
|
|
scene_episode, scene_absolute)
|
|
|
|
return json_dumps(result)
|
|
|
|
@staticmethod
|
|
def fetch_releasegroups(show_name):
|
|
|
|
result = pull_anidb_groups(show_name)
|
|
if None is result:
|
|
result = dict(result='fail', resp='init')
|
|
elif False is result:
|
|
result = dict(result='fail', resp='connect')
|
|
elif isinstance(result, list) and 0 == len(result):
|
|
result = dict(result='success',
|
|
groups=[dict(name='No groups fetched in API response', rating='', range='')])
|
|
else:
|
|
result = dict(result='success', groups=result)
|
|
return json_dumps(result)
|
|
|
|
@staticmethod
|
|
def csv_items(text):
|
|
# type: (AnyStr) -> AnyStr
|
|
"""Return a text list of items separated by comma instead of '/' """
|
|
|
|
return (isinstance(text, string_types) and re.sub(r'\b\s?/\s?\b', ', ', text)) or text
|
|
|
|
def role(self, rid, tvid_prodid, **kwargs):
|
|
_ = kwargs.get('oid') # suppress pyc non used var highlight, oid (original id) is a visual ui key
|
|
t = PageTemplate(web_handler=self, file='cast_role.tmpl')
|
|
|
|
character_id = usable_id(rid)
|
|
if not character_id:
|
|
return self._generic_message('Error', 'Invalid character ID')
|
|
|
|
try:
|
|
t.show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
except (BaseException, Exception):
|
|
return
|
|
|
|
# after this point, only use character.id
|
|
character = TVCharacter(sid=character_id, show_obj=t.show_obj)
|
|
if not character:
|
|
return self._generic_message('Error', 'character ID not found')
|
|
|
|
t.character = character
|
|
t.roles = []
|
|
|
|
# this basic relate logic uses the link from a character to a person to find the same character name being known
|
|
# across multiple shows. The character name is more likely to be static compared to an actor name.
|
|
# A hardcoded exclusion may be needed if used actor plays a same named character in an _unrelated_ universe,
|
|
# but the chance of this is negligible, therefore, it's worth it to give this a try than not to :)
|
|
known = {}
|
|
main_role_known = False
|
|
rc_clean = re.compile(r'[^a-z0-9]')
|
|
char_name = rc_clean.sub('', (character.name or 'unknown name').lower())
|
|
for cur_person in (character.person or []):
|
|
person, roles, msg = self.cast(cur_person.id)
|
|
if not msg:
|
|
for cur_role in roles:
|
|
if known.get(person.id, {}).get(cur_role['character_id']) == cur_role['show_obj'].tvid_prodid:
|
|
continue
|
|
# 1 person plays 1-n roles across 1-n shows
|
|
known.setdefault(person.id, {}).setdefault(
|
|
cur_role['character_id'], cur_role['show_obj'].tvid_prodid)
|
|
|
|
# mark names that are a subset as the same character
|
|
# case1 Detective Lt. Louie Provenza became Louie Provenza
|
|
# case2 Commander Russell Taylor became Commander Taylor
|
|
# case3 Micheal Tao became Mike Tao (surname is static except for women post marriage)
|
|
lower_name = cur_role['character_name'].lower().strip() or 'unknown name'
|
|
name_parts = lower_name.split()
|
|
|
|
is_main = character.id == cur_role['character_id'] \
|
|
and character.show_obj.tvid_prodid == cur_role['show_obj'].tvid_prodid
|
|
|
|
# exclusion exceptions, ignoring main role
|
|
if not is_main and any([
|
|
# 1 person playing multiple same role surname but each are distinct characters
|
|
rc_clean.sub('', lower_name).endswith('griffin')
|
|
]):
|
|
continue
|
|
|
|
if any([
|
|
char_name in rc_clean.sub('', lower_name),
|
|
char_name in ('', '%s%s' % (name_parts[0], name_parts[-1]))[2 < len(name_parts)],
|
|
# case3 surname only, provided more than one name part exist
|
|
(False, name_parts[-1] in char_name)[1 < len(name_parts)],
|
|
# inclusion exceptions
|
|
re.search('(?i)^(?:Host|Presenter)[0-9]*$', char_name),
|
|
re.search('(?i)^(?:Host|Prese)', lower_name) and re.search('(?i)JeremyClarkson', char_name),
|
|
re.search('(?i)(?:AnnaBaker|BelleStone)', char_name),
|
|
re.search('(?i)(?:JimmyMcgill|SaulGoodman)', char_name)
|
|
]):
|
|
|
|
t.roles.append({
|
|
'character_name': cur_role['character_name'],
|
|
'character_id': cur_role['character_id'],
|
|
'character_rid': cur_role['character_rid'],
|
|
'show_obj': cur_role['show_obj'],
|
|
'person_name': person.name,
|
|
'person_id': person.id
|
|
})
|
|
|
|
# ensure main role is first
|
|
if not main_role_known and is_main:
|
|
main_role_known = True
|
|
t.roles.insert(0, t.roles.pop(-1))
|
|
|
|
return t.respond()
|
|
|
|
def cast(self, rid):
|
|
person = roles = None
|
|
msg = None
|
|
|
|
person_id = usable_id(rid)
|
|
if person_id:
|
|
person = TVPerson(sid=person_id)
|
|
if person:
|
|
my_db = db.DBConnection()
|
|
sql_result = my_db.select(
|
|
"""
|
|
SELECT DISTINCT characters.id AS id, name, indexer, indexer_id,
|
|
cpy.start_year AS start_year, cpy.end_year AS end_year,
|
|
c.indexer AS c_tvid, c.indexer_id AS c_prodid,
|
|
(SELECT group_concat(character_ids.src || ':' || character_ids.src_id, ';;;')
|
|
FROM character_ids WHERE character_ids.character_id = characters.id) as c_ids
|
|
FROM characters
|
|
LEFT JOIN castlist c ON characters.id = c.character_id
|
|
LEFT JOIN character_person_map cpm ON characters.id = cpm.character_id
|
|
LEFT JOIN character_person_years cpy ON characters.id = cpy.character_id
|
|
AND cpy.person_id = ?
|
|
WHERE cpm.person_id = ?
|
|
""", [person.id, person.id])
|
|
|
|
pref = [TVINFO_IMDB, TVINFO_TVMAZE, TVINFO_TMDB, TVINFO_TRAKT]
|
|
roles = []
|
|
for cur_char in sql_result or []:
|
|
ref_id = None
|
|
pri = 9999
|
|
for cur_ref_id in (cur_char['c_ids'] and cur_char['c_ids'].split(';;;')) or []:
|
|
k, v = [helpers.try_int(_v, None) for _v in cur_ref_id.split(':')]
|
|
if None is not k and None is not v:
|
|
if k in pref:
|
|
test_pri = pref.index(k)
|
|
if test_pri < pri:
|
|
pri = test_pri
|
|
ref_id = cur_ref_id
|
|
|
|
roles.append({
|
|
'character_name': self.csv_items(cur_char['name']) or 'unknown name',
|
|
'character_id': cur_char['id'],
|
|
'character_rid': ref_id,
|
|
'show_obj': helpers.find_show_by_id({cur_char['c_tvid']: cur_char['c_prodid']}),
|
|
'start_year': cur_char['start_year'], 'end_year': cur_char['end_year']
|
|
})
|
|
else:
|
|
msg = 'Person ID not found'
|
|
else:
|
|
msg = 'Invalid person ID'
|
|
|
|
return person, roles, msg
|
|
|
|
def person(self, rid, **kwargs):
|
|
_ = kwargs.get('oid') # suppress pyc non used var highlight, oid (original id) is a visual ui key
|
|
t = PageTemplate(web_handler=self, file='cast_person.tmpl')
|
|
person, roles, msg = self.cast(rid)
|
|
if msg:
|
|
return self._generic_message('Error', msg)
|
|
|
|
t.person = person
|
|
t.roles = roles
|
|
|
|
return t.respond()
|
|
|
|
@staticmethod
|
|
def _convert_person_data(person_dict):
|
|
event = {}
|
|
|
|
for cur_date_kind in ('birthdate', 'deathdate'):
|
|
if person_dict[cur_date_kind]:
|
|
try:
|
|
doe = dt_date.fromordinal(person_dict[cur_date_kind])
|
|
event[cur_date_kind] = doe
|
|
person_dict[cur_date_kind] = doe.strftime('%Y-%m-%d')
|
|
person_dict['%s_user' % cur_date_kind] = SGDatetime.sbfdate(doe)
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
person_dict['age'] = sg_helpers.calc_age(event['birthdate'], event['deathdate'])
|
|
|
|
def _select_person_by_date(self, date_str, date_kind):
|
|
# type: (AnyStr, AnyStr) -> List[Dict]
|
|
|
|
if date_kind not in ('birthdate', 'deathdate'):
|
|
return []
|
|
|
|
try:
|
|
dt = dateutil.parser.parse(date_str).date()
|
|
except (BaseException, Exception):
|
|
raise Exception('invalid date')
|
|
|
|
possible_dates = []
|
|
for cur_year in moves.xrange((1850, 1920)['deathdate' == date_kind], dt_date.today().year + 1):
|
|
try:
|
|
possible_dates.append(dt_date(year=cur_year, month=dt.month, day=dt.day).toordinal())
|
|
if 2 == dt.month and 28 == dt.day:
|
|
try:
|
|
dt_date(year=dt.year, month=dt.month, day=29)
|
|
except (BaseException, Exception):
|
|
possible_dates.append(dt_date(year=cur_year, month=dt.month, day=29).toordinal())
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
my_db = db.DBConnection(row_type='dict')
|
|
sql_result = my_db.select(
|
|
"""
|
|
SELECT * FROM persons
|
|
WHERE %s IN (%s)
|
|
""" % (date_kind, ','.join(['?'] * len(possible_dates))), possible_dates)
|
|
for cur_person in sql_result:
|
|
self._convert_person_data(cur_person)
|
|
|
|
return sql_result
|
|
|
|
def get_persons(self, names=None, **kwargs):
|
|
# type: (AnyStr, dict) -> AnyStr
|
|
"""
|
|
:param names:
|
|
:param kwargs: optional `birthday`, optional `deathday`
|
|
:return:
|
|
"""
|
|
results = {}
|
|
for cur_date_kind in ('birthdate', 'deathdate'):
|
|
date_arg = kwargs.get(cur_date_kind)
|
|
if date_arg:
|
|
try:
|
|
results[cur_date_kind] = self._select_person_by_date(date_arg, cur_date_kind)
|
|
except (BaseException, Exception) as e:
|
|
return json_dumps({'result': 'error', 'error': ex(e)})
|
|
|
|
names = names and names.split('|')
|
|
if names:
|
|
my_db = db.DBConnection(row_type='dict')
|
|
sql_result = my_db.select(
|
|
"""
|
|
SELECT * FROM persons
|
|
WHERE name IN (%s)
|
|
""" % ','.join(['?'] * len(names)), names)
|
|
for cur_person in sql_result:
|
|
self._convert_person_data(cur_person)
|
|
results['names'] = sql_result
|
|
|
|
return json_dumps({'result': 'success', 'person_list': results})
|
|
|
|
def get_switch_changed(self):
|
|
t = PageTemplate(web_handler=self, file='switch_show_result.tmpl')
|
|
t.show_list = {}
|
|
my_db = db.DBConnection()
|
|
sql_result = my_db.select(
|
|
"""
|
|
SELECT DISTINCT new_indexer, new_indexer_id, COUNT(reason) AS count, reason
|
|
FROM switch_ep_result
|
|
GROUP BY new_indexer, new_indexer_id, reason
|
|
""")
|
|
for cur_show in sql_result:
|
|
try:
|
|
show_obj = helpers.find_show_by_id({cur_show['new_indexer']: cur_show['new_indexer_id']})
|
|
except (BaseException, Exception):
|
|
# todo: what to do with unknown entries
|
|
continue
|
|
if not show_obj:
|
|
continue
|
|
t.show_list.setdefault(show_obj, {}).update(
|
|
{('changed', 'deleted')[TVSWITCH_EP_DELETED == cur_show['reason']]: cur_show['count']})
|
|
|
|
return t.respond()
|
|
|
|
def get_switch_changed_episodes(self, tvid_prodid):
|
|
t = PageTemplate(web_handler=self, file='switch_episode_result.tmpl')
|
|
try:
|
|
show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
except (BaseException, Exception):
|
|
# todo: what to do with unknown entries
|
|
show_obj = None
|
|
if not show_obj:
|
|
return self.page_not_found()
|
|
|
|
t.show_obj = show_obj
|
|
t.ep_list = []
|
|
my_db = db.DBConnection()
|
|
sql_result = my_db.select(
|
|
"""
|
|
SELECT * FROM switch_ep_result
|
|
WHERE new_indexer = ? AND new_indexer_id = ?
|
|
ORDER BY season, episode
|
|
""", TVidProdid(tvid_prodid).list)
|
|
for cur_episode in sql_result:
|
|
try:
|
|
ep_obj = show_obj.get_episode(season=cur_episode['season'], episode=cur_episode['episode'],
|
|
existing_only=True)
|
|
except (BaseException, Exception):
|
|
ep_obj = None
|
|
t.ep_list.append({'season': cur_episode['season'], 'episode': cur_episode['episode'],
|
|
'reason': tvswitch_ep_names.get(cur_episode['reason'], 'unknown'), 'ep_obj': ep_obj})
|
|
|
|
return t.respond()
|
|
|
|
|
|
class HomeProcessMedia(Home):
|
|
|
|
def get(self, route, *args, **kwargs):
|
|
route = route.strip('/')
|
|
if 'files' == route:
|
|
route = 'process_files'
|
|
return super(HomeProcessMedia, self).get(route, *args, **kwargs)
|
|
|
|
def index(self):
|
|
|
|
t = PageTemplate(web_handler=self, file='home_postprocess.tmpl')
|
|
t.submenu = [x for x in self.home_menu() if 'process-media' not in x['path']]
|
|
return t.respond()
|
|
|
|
def process_files(self, dir_name=None, nzb_name=None, quiet=None, process_method=None, force=None,
|
|
force_replace=None, failed='0', process_type='auto', stream='0', dupekey=None, is_basedir='1',
|
|
client=None, **kwargs):
|
|
|
|
if 'test' in kwargs and kwargs['test'] in ['True', True, 1, '1']:
|
|
return 'Connection success!'
|
|
|
|
if not dir_name and ('0' == failed or not nzb_name):
|
|
self.redirect('/home/process-media/')
|
|
else:
|
|
show_id_regex = re.compile(r'^SickGear-([A-Za-z]*)(\d+)-')
|
|
tvid = 0
|
|
show_obj = None
|
|
nzbget_call = isinstance(client, string_types) and 'nzbget' == client
|
|
nzbget_dupekey = nzbget_call and isinstance(dupekey, string_types) and \
|
|
None is not show_id_regex.search(dupekey)
|
|
if nzbget_dupekey:
|
|
m = show_id_regex.match(dupekey)
|
|
istr = m.group(1)
|
|
for i in sickgear.TVInfoAPI().sources:
|
|
if istr == sickgear.TVInfoAPI(i).config.get('dupekey'):
|
|
tvid = i
|
|
break
|
|
show_obj = helpers.find_show_by_id({tvid: int(m.group(2))}, no_mapped_ids=True)
|
|
|
|
skip_failure_processing = nzbget_call and not nzbget_dupekey
|
|
|
|
if nzbget_call and sickgear.NZBGET_SCRIPT_VERSION != kwargs.get('pp_version', '0'):
|
|
logger.error(f'Calling SickGear-NG.py script {kwargs.get("pp_version", "0")} is not current version'
|
|
f' {sickgear.NZBGET_SCRIPT_VERSION}, please update.')
|
|
|
|
if sickgear.NZBGET_SKIP_PM and nzbget_call and nzbget_dupekey and nzb_name and show_obj:
|
|
processTV.process_minimal(nzb_name, show_obj,
|
|
failed in (1, '1', True, 'True', 'true'),
|
|
webhandler=None if '0' == stream else self.send_message)
|
|
else:
|
|
|
|
cleanup = kwargs.get('cleanup') in ('on', '1')
|
|
if isinstance(dir_name, string_types):
|
|
dir_name = decode_str(dir_name)
|
|
if 'auto' != process_type:
|
|
sickgear.PROCESS_LAST_DIR = dir_name
|
|
sickgear.PROCESS_LAST_METHOD = process_method
|
|
if 'move' == process_method:
|
|
sickgear.PROCESS_LAST_CLEANUP = cleanup
|
|
sickgear.save_config()
|
|
|
|
if nzbget_call and isinstance(sickgear.NZBGET_MAP, string_types) and sickgear.NZBGET_MAP:
|
|
m = sickgear.NZBGET_MAP.split('=')
|
|
dir_name, not_used = helpers.path_mapper(m[0], m[1], dir_name)
|
|
|
|
result = processTV.process_dir(dir_name if dir_name else None,
|
|
None if not nzb_name else decode_str(nzb_name),
|
|
process_method=process_method, pp_type=process_type,
|
|
cleanup=cleanup,
|
|
force=force in ('on', '1'),
|
|
force_replace=force_replace in ('on', '1'),
|
|
failed='0' != failed,
|
|
webhandler=None if '0' == stream else self.send_message,
|
|
show_obj=show_obj, is_basedir=is_basedir in ('on', '1'),
|
|
skip_failure_processing=skip_failure_processing, client=client)
|
|
|
|
if '0' == stream:
|
|
regexp = re.compile(r'(?i)<br[\s/]+>', flags=re.UNICODE)
|
|
result = regexp.sub('\n', result)
|
|
if None is not quiet and 1 == int(quiet):
|
|
regexp = re.compile('(?i)<a[^>]+>([^<]+)</a>', flags=re.UNICODE)
|
|
return regexp.sub(r'\1', result)
|
|
return self._generic_message('Postprocessing results', f'<pre>{result}</pre>')
|
|
|
|
# noinspection PyPep8Naming
|
|
@staticmethod
|
|
def processEpisode(**kwargs):
|
|
""" legacy function name, stubbed and will be removed
|
|
"""
|
|
logger.error('This endpoint is no longer to be used,'
|
|
' nzbToMedia users please follow: https://github.com/SickGear/SickGear/wiki/FAQ-nzbToMedia')
|
|
sickgear.MEMCACHE['DEPRECATE_PP_LEGACY'] = True
|
|
|
|
class AddShows(Home):
|
|
|
|
def get(self, route, *args, **kwargs):
|
|
route = route.strip('/')
|
|
if 'import' == route:
|
|
route = 'import_shows'
|
|
elif 'find' == route:
|
|
route = 'new_show'
|
|
return super(AddShows, self).get(route, *args, **kwargs)
|
|
|
|
def index(self):
|
|
t = PageTemplate(web_handler=self, file='home_addShows.tmpl')
|
|
t.submenu = self.home_menu()
|
|
return t.respond()
|
|
|
|
@staticmethod
|
|
def get_infosrc_languages():
|
|
result = sickgear.TVInfoAPI().config['valid_languages'].copy()
|
|
|
|
# sort list alphabetically with sickgear.ADD_SHOWS_METALANG as the first item
|
|
if sickgear.ADD_SHOWS_METALANG in result:
|
|
del result[result.index(sickgear.ADD_SHOWS_METALANG)]
|
|
result.sort()
|
|
result.insert(0, sickgear.ADD_SHOWS_METALANG)
|
|
|
|
for src in sickgear.TVInfoAPI().search_sources:
|
|
tvinfo_config = sickgear.TVInfoAPI(src).api_params.copy()
|
|
t = sickgear.TVInfoAPI(src).setup(**tvinfo_config)
|
|
try:
|
|
all_langs = t.get_languages()
|
|
except (BaseException, Exception):
|
|
continue
|
|
if all_langs:
|
|
result.extend([lang['sg_lang'] for lang in all_langs if lang['sg_lang'] not in result])
|
|
|
|
try:
|
|
# noinspection PyPep8Naming
|
|
from langcodes import Language as lang_obj, LanguageTagError, standardize_tag
|
|
except ImportError:
|
|
lang_obj = None
|
|
result_ext = []
|
|
if None is not lang_obj:
|
|
prio_abbr = ''
|
|
prio_lang = []
|
|
try:
|
|
lang = lang_obj.get(sickgear.ADD_SHOWS_METALANG)
|
|
prio_abbr = lang.to_alpha3()
|
|
prio_lang = [dict(orig_abbr=sickgear.ADD_SHOWS_METALANG, std_abbr=sickgear.ADD_SHOWS_METALANG,
|
|
abbr=prio_abbr, en=lang.display_name(), native=lang.autonym())]
|
|
except (BaseException, Exception) as _:
|
|
pass
|
|
dedupe = []
|
|
for cur_lang in result:
|
|
try:
|
|
lang = lang_obj.get(cur_lang)
|
|
abbr = lang.to_alpha3()
|
|
except (BaseException, Exception) as _:
|
|
continue
|
|
|
|
try:
|
|
std_abbr = standardize_tag(cur_lang, macro=True)
|
|
except (BaseException, Exception) as _:
|
|
std_abbr = None
|
|
|
|
if abbr not in dedupe and abbr != prio_abbr:
|
|
dedupe += [abbr]
|
|
result_ext += [dict(orig_abbr=cur_lang, std_abbr=std_abbr, abbr=abbr, en=lang.display_name(), native=lang.autonym())]
|
|
|
|
result_ext = prio_lang + sorted(result_ext, key=lambda x: x['en'])
|
|
|
|
return json_dumps({'results': [] if result_ext else result, 'results_ext': result_ext})
|
|
|
|
@staticmethod
|
|
def generate_show_dir_name(show_name):
|
|
return helpers.generate_show_dir_name(None, show_name)
|
|
|
|
@staticmethod
|
|
def _generate_search_text_list(search_term):
|
|
# type: (AnyStr) -> Set[AnyStr]
|
|
used_search_term = re.sub(r'\(?(19|20)\d{2}\)?', '', search_term).strip()
|
|
# fix for users that don't know the correct title
|
|
used_search_term = re.sub(r'(?i)(grown|mixed)(ish)', r'\1-\2', used_search_term)
|
|
|
|
b_term = decode_str(used_search_term).strip()
|
|
terms = []
|
|
try:
|
|
for cur_term in [unidecode(b_term), b_term]:
|
|
if cur_term not in terms:
|
|
terms += [cur_term]
|
|
except (BaseException, Exception):
|
|
text = used_search_term.strip()
|
|
terms = text
|
|
|
|
return set(s for s in set([used_search_term] + terms) if s)
|
|
|
|
# noinspection PyPep8Naming
|
|
def search_tvinfo_for_showname(self, search_term, lang='en', search_tvid=None):
|
|
if not lang or 'null' == lang:
|
|
lang = sickgear.ADD_SHOWS_METALANG or 'en'
|
|
if lang != sickgear.ADD_SHOWS_METALANG:
|
|
sickgear.ADD_SHOWS_METALANG = lang
|
|
sickgear.save_config()
|
|
|
|
search_tvid = sg_helpers.try_int(search_tvid, None)
|
|
search_term = search_term and search_term.strip()
|
|
ids_to_search, id_srcs, searchable = {}, [], \
|
|
(list(iterkeys(sickgear.TVInfoAPI().search_sources)), [search_tvid])[
|
|
search_tvid in sickgear.TVInfoAPI().search_sources]
|
|
id_check = re.finditer(r'((\w+):\W*([t0-9]+))', search_term)
|
|
if id_check:
|
|
for cur_match in id_check:
|
|
total, slug, id_str = cur_match.groups()
|
|
for cur_tvid in sickgear.TVInfoAPI().all_sources:
|
|
if sickgear.TVInfoAPI(cur_tvid).config.get('slug') \
|
|
and (slug.lower() == sickgear.TVInfoAPI(cur_tvid).config['slug']
|
|
or cur_tvid == sg_helpers.try_int(slug, None)):
|
|
try:
|
|
ids_to_search[cur_tvid] = int(id_str.strip().replace('tt', ''))
|
|
except (BaseException, Exception):
|
|
pass
|
|
try:
|
|
search_term = re.sub(r' *%s *' % re.escape(total), ' ', search_term).strip()
|
|
if cur_tvid in searchable:
|
|
id_srcs.append(cur_tvid)
|
|
break
|
|
except (BaseException, Exception):
|
|
continue
|
|
|
|
id_check = re.finditer(
|
|
r'(?P<imdb_full>[^ ]+imdb\.com/title/(?P<imdb>tt\d+)[^ ]*)|'
|
|
r'(?P<imdb_id_full>[^ ]*(?P<imdb_id>' + helpers.RE_IMDB_ID + '))[^ ]*|'
|
|
r'(?P<tmdb_full>[^ ]+themoviedb\.org/tv/(?P<tmdb>\d+)[^ ]*)|'
|
|
r'(?P<trakt_full>[^ ]+trakt\.tv/shows/(?P<trakt>[^ /]+)[^ ]*)|'
|
|
r'(?P<tvdb_full>[^ ]+thetvdb\.com/series/(?P<tvdb>[^ /]+)[^ ]*)|'
|
|
r'(?P<tvdb_id_full>[^ ]+thetvdb\.com/\D+(?P<tvdb_id>[^ /]+)[^ ]*)|'
|
|
r'(?P<tvmaze_full>[^ ]+tvmaze\.com/shows/(?P<tvmaze>\d+)/?[^ ]*)', search_term)
|
|
if id_check:
|
|
for cur_match in id_check:
|
|
for cur_tvid, cur_slug in [
|
|
(TVINFO_IMDB, 'imdb'), (TVINFO_IMDB, 'imdb_id'), (TVINFO_TMDB, 'tmdb'),
|
|
(TVINFO_TRAKT_SLUG, 'trakt'), (TVINFO_TVDB_SLUG, 'tvdb'), (TVINFO_TVDB, 'tvdb_id'),
|
|
(TVINFO_TVMAZE, 'tvmaze')]:
|
|
if cur_match.group(cur_slug):
|
|
try:
|
|
slug_match = cur_match.group(cur_slug).strip()
|
|
if TVINFO_IMDB == cur_tvid:
|
|
slug_match = slug_match.replace('tt', '')
|
|
if cur_tvid not in (TVINFO_TVDB_SLUG, TVINFO_TRAKT_SLUG):
|
|
slug_match = sg_helpers.try_int(slug_match, slug_match)
|
|
ids_to_search[cur_tvid] = slug_match
|
|
search_term = re.sub(r' *%s *' % re.escape(cur_match.group('%s_full' % cur_slug)), ' ',
|
|
search_term).strip()
|
|
if TVINFO_TVDB_SLUG == cur_tvid:
|
|
cur_tvid = TVINFO_TVDB
|
|
elif TVINFO_TRAKT_SLUG == cur_tvid:
|
|
cur_tvid = TVINFO_TRAKT
|
|
if cur_tvid in searchable:
|
|
id_srcs.append(cur_tvid)
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
# term is used for relevancy
|
|
term = decode_str(search_term).strip()
|
|
used_search_term = self._generate_search_text_list(search_term)
|
|
text_search_used = bool(used_search_term)
|
|
|
|
exclude_results = []
|
|
if TVINFO_TVMAZE in ids_to_search:
|
|
id_srcs = [TVINFO_TVMAZE] + [i for i in id_srcs if TVINFO_TVMAZE != i]
|
|
if TVINFO_TVMAZE not in searchable:
|
|
exclude_results.append(TVINFO_TVMAZE)
|
|
|
|
results = {}
|
|
final_results = []
|
|
sources_to_search = id_srcs + [s for s in [TVINFO_TRAKT] + searchable if s not in id_srcs]
|
|
ids_search_used = ids_to_search.copy()
|
|
|
|
for cur_tvid in sources_to_search:
|
|
tvinfo_config = sickgear.TVInfoAPI(cur_tvid).api_params.copy()
|
|
tvinfo_config['language'] = lang
|
|
tvinfo_config['custom_ui'] = classes.AllShowInfosNoFilterListUI
|
|
t = sickgear.TVInfoAPI(cur_tvid).setup(**tvinfo_config)
|
|
results.setdefault(cur_tvid, {})
|
|
try:
|
|
for cur_result in t.search_show(list(used_search_term),
|
|
ids=ids_search_used,
|
|
lang=lang): # type: TVInfoShow
|
|
if TVINFO_TRAKT == cur_tvid and not cur_result['ids'].tvdb:
|
|
continue
|
|
tv_src_id = int(cur_result['id'])
|
|
if cur_tvid in exclude_results:
|
|
ids_search_used.update({k: v for k, v in iteritems(cur_result.get('ids', {}))
|
|
if v and k not in iterkeys(ids_to_search)})
|
|
else:
|
|
if type(cur_result) == dict:
|
|
results[cur_tvid][tv_src_id] = cur_result.copy()
|
|
else:
|
|
results[cur_tvid][tv_src_id] = cur_result.to_dict()
|
|
results[cur_tvid][tv_src_id]['direct_id'] = \
|
|
(cur_tvid in ids_to_search and ids_to_search.get(cur_tvid)
|
|
and tv_src_id == ids_to_search.get(cur_tvid)) or \
|
|
(TVINFO_TVDB == cur_tvid and cur_result.get('slug') and
|
|
ids_to_search.get(TVINFO_TVDB_SLUG) == cur_result.get('slug')) or False
|
|
if results[cur_tvid][tv_src_id]['direct_id'] or \
|
|
any(ids_to_search[si] == results[cur_tvid][tv_src_id].get('ids', {})[si]
|
|
for si in ids_to_search):
|
|
ids_search_used.update({k: v for k, v in iteritems(
|
|
results[cur_tvid][tv_src_id].get('ids', {}))
|
|
if v and k not in iterkeys(ids_to_search)})
|
|
results[cur_tvid][tv_src_id]['rename_suggest'] = '' \
|
|
if not results[cur_tvid][tv_src_id]['firstaired'] \
|
|
else dateutil.parser.parse(results[cur_tvid][tv_src_id]['firstaired']).year
|
|
if not text_search_used and cur_tvid in ids_to_search and tv_src_id == ids_to_search.get(cur_tvid):
|
|
used_search_term.update(self._generate_search_text_list(cur_result['seriesname']))
|
|
if not term:
|
|
term = decode_str(cur_result['seriesname']).strip()
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
if TVINFO_TVDB not in searchable:
|
|
try:
|
|
results.pop(TVINFO_TRAKT)
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
id_names = {tvid: (name, '%s via %s' % (sickgear.TVInfoAPI(TVINFO_TVDB).name, name))[TVINFO_TRAKT == tvid]
|
|
for tvid, name in iteritems(sickgear.TVInfoAPI().all_sources)}
|
|
|
|
if TVINFO_TRAKT in results and TVINFO_TVDB in results:
|
|
tvdb_ids = list(results[TVINFO_TVDB])
|
|
results[TVINFO_TRAKT] = {k: v for k, v in iteritems(results[TVINFO_TRAKT]) if v['ids'].tvdb not in tvdb_ids}
|
|
|
|
def in_db(tvid, prod_id):
|
|
show_obj = helpers.find_show_by_id({(tvid, TVINFO_TVDB)[TVINFO_TRAKT == tvid]: prod_id},
|
|
no_mapped_ids=False, no_exceptions=True)
|
|
return any([show_obj]) and '/home/view-show?tvid_prodid=%s' % show_obj.tvid_prodid
|
|
|
|
def _parse_date(dt_str):
|
|
try:
|
|
return dateutil.parser.parse(dt_str)
|
|
except (BaseException, Exception):
|
|
return ''
|
|
|
|
# noinspection PyUnboundLocalVariable
|
|
map_consume(final_results.extend,
|
|
[[[id_names[tvid], in_db(*((tvid, int(show['id'])),
|
|
(TVINFO_TVDB, show['ids'][TVINFO_TVDB]))[TVINFO_TRAKT == tvid]),
|
|
tvid, (tvid, TVINFO_TVDB)[TVINFO_TRAKT == tvid],
|
|
sickgear.TVInfoAPI((tvid, TVINFO_TVDB)[TVINFO_TRAKT == tvid]).config['slug'],
|
|
(sickgear.TVInfoAPI((tvid, TVINFO_TVDB)[TVINFO_TRAKT == tvid]).config['show_url'] %
|
|
show['ids'][(tvid, TVINFO_TVDB)[TVINFO_TRAKT == tvid]])
|
|
+ ('', '&lid=%s' % sickgear.TVInfoAPI().config.get('langabbv_to_id', {}).get(lang, lang))[
|
|
TVINFO_TVDB == tvid],
|
|
(int(show['id']), show['ids'][TVINFO_TVDB])[TVINFO_TRAKT == tvid],
|
|
show['seriesname'], helpers.xhtml_escape(show['seriesname']), show['firstaired'],
|
|
(isinstance(show['firstaired'], string_types) and show['firstaired']
|
|
and SGDatetime.sbfdate(_parse_date(show['firstaired'])) or ''),
|
|
show.get('network', '') or '', # 11
|
|
(show.get('genres', '') or show.get('genre', '') or '').replace('|', ', '), # 12
|
|
show.get('language', ''), show.get('language_country_code') or '', # 13 - 14
|
|
re.sub(r'([,.!][^,.!]*?)$', '...',
|
|
re.sub(r'([.!?])(?=\w)', r'\1 ',
|
|
helpers.xhtml_escape((show.get('overview', '') or '')[:250:].strip()))), # 15
|
|
self._make_cache_image_url(tvid, show, default_transparent_img=False), # 16
|
|
100 - ((show['direct_id'] and 100)
|
|
or self.get_uw_ratio(term, show['seriesname'], show.get('aliases') or [],
|
|
show.get('language_country_code') or '')),
|
|
None, None, None, None, None, None, None, None, None, # 18 - 26
|
|
show['direct_id'], show.get('rename_suggest')
|
|
] for show in itervalues(shows)] for tvid, shows in iteritems(results)])
|
|
|
|
def final_order(sortby_index, data, final_sort):
|
|
idx_is_indb = 1
|
|
for (_n, x) in enumerate(data):
|
|
x[sortby_index] = _n + (1000, 0)[x[idx_is_indb] and 'notop' not in sickgear.RESULTS_SORTBY]
|
|
return data if not final_sort else sorted(data, reverse=False, key=lambda _x: _x[sortby_index])
|
|
|
|
def sort_newest(data_result, is_last_sort, combine):
|
|
return sort_date(data_result, is_last_sort, 19, as_combined=combine)
|
|
|
|
def sort_oldest(data_result, is_last_sort, combine):
|
|
return sort_date(data_result, is_last_sort, 21, False, combine)
|
|
|
|
def sort_date(data_result, is_last_sort, idx_sort, reverse=True, as_combined=False):
|
|
idx_aired = 9
|
|
date_sorted = sorted(data_result, reverse=reverse, key=lambda x: (dateutil.parser.parse(
|
|
re.match(r'^(?:19|20)\d\d$', str(x[idx_aired])) and ('%s-12-31' % str(x[idx_aired]))
|
|
or (x[idx_aired] and str(x[idx_aired])) or '1900')))
|
|
combined = final_order(idx_sort + 1, date_sorted, is_last_sort)
|
|
|
|
idx_src = 2
|
|
grouped = final_order(idx_sort, sorted(date_sorted, key=lambda x: x[idx_src]), is_last_sort)
|
|
|
|
return (grouped, combined)[as_combined]
|
|
|
|
def sort_az(data_result, is_last_sort, combine):
|
|
return sort_zaaz(data_result, is_last_sort, 23, as_combined=combine)
|
|
|
|
def sort_za(data_result, is_last_sort, combine):
|
|
return sort_zaaz(data_result, is_last_sort, 25, True, combine)
|
|
|
|
def sort_zaaz(data_result, is_last_sort, idx_sort, reverse=False, as_combined=False):
|
|
idx_title = 7
|
|
zaaz_sorted = sorted(data_result, reverse=reverse, key=lambda x: (
|
|
(remove_article(x[idx_title].lower()), x[idx_title].lower())[sickgear.SORT_ARTICLE]))
|
|
combined = final_order(idx_sort + 1, zaaz_sorted, is_last_sort)
|
|
|
|
idx_src = 2
|
|
grouped = final_order(idx_sort, sorted(zaaz_sorted, key=lambda x: x[idx_src]), is_last_sort)
|
|
|
|
return (grouped, combined)[as_combined]
|
|
|
|
def sort_rel(data_result, is_last_sort, as_combined):
|
|
idx_rel_sort, idx_rel, idx_direct_id = 17, 17, 27
|
|
idx_title = 7
|
|
idx_src = 2
|
|
rel_sorted = sorted(data_result, key=lambda x: (not x[idx_direct_id], x[idx_rel], x[idx_title], x[idx_src]))
|
|
combined = final_order(idx_rel_sort + 1, rel_sorted, is_last_sort)
|
|
|
|
grouped = final_order(idx_rel_sort, sorted(rel_sorted, key=lambda x: (x[idx_src])), is_last_sort)
|
|
|
|
return (grouped, combined)[as_combined]
|
|
|
|
sort_methods = [sort_oldest, sort_newest, sort_za, sort_az, sort_rel]
|
|
if re.match('az|za|ne|ol', sickgear.RESULTS_SORTBY[:2]):
|
|
if 'az' == sickgear.RESULTS_SORTBY[:2]:
|
|
new_default = sort_az
|
|
elif 'za' == sickgear.RESULTS_SORTBY[:2]:
|
|
new_default = sort_za
|
|
elif 'newest' == sickgear.RESULTS_SORTBY[:6]:
|
|
new_default = sort_newest
|
|
else: # 'oldest' == sickgear.RESULTS_SORTBY[:6]:
|
|
new_default = sort_oldest
|
|
|
|
sort_methods.remove(new_default)
|
|
sort_methods += [new_default]
|
|
|
|
idx_last_sort = len(sort_methods) - 1
|
|
sort_nogroup = 'nogroup' == sickgear.RESULTS_SORTBY[-7:]
|
|
for n, cur_method in enumerate(sort_methods):
|
|
final_results = cur_method(final_results, n == idx_last_sort, sort_nogroup)
|
|
|
|
return json_dumps({'results': final_results})
|
|
|
|
@staticmethod
|
|
def _make_cache_image_url(iid, show_info, default_transparent_img=True, use_source_id=False):
|
|
img_url = ''
|
|
trans_param = ('1', '0')[not default_transparent_img]
|
|
if TVINFO_TRAKT == iid:
|
|
img_url = 'imagecache?path=browse/thumb/trakt&filename=%s&trans=%s&tmdbid=%s&tvdbid=%s' % \
|
|
('%s.jpg' % show_info['ids'].trakt, trans_param, show_info['ids'].tmdb, show_info['ids'].tvdb)
|
|
elif use_source_id and TVINFO_TVMAZE == iid:
|
|
img_url = 'imagecache?path=browse/thumb/tvmaze&filename=%s&trans=%s&tvmazeid=%s' % \
|
|
('%s.jpg' % show_info['ids'].tvmaze, trans_param, show_info['ids'].tvmaze)
|
|
elif iid in (TVINFO_TVDB, TVINFO_TVMAZE, TVINFO_TMDB) and show_info.get('poster'):
|
|
img_url = 'imagecache?path=browse/thumb/%s&filename=%s&trans=%s&source=%s' % \
|
|
(tv_src_names[iid], '%s.jpg' % show_info['id'], trans_param, show_info['poster'])
|
|
sickgear.CACHE_IMAGE_URL_LIST.add_url(show_info['poster'])
|
|
return img_url
|
|
|
|
@classmethod
|
|
def get_uw_ratio(cls, search_term, showname, aliases, lang=None):
|
|
search_term = decode_str(search_term, errors='replace')
|
|
showname = decode_str(showname, errors='replace')
|
|
try:
|
|
s = fuzz.UWRatio(search_term, showname)
|
|
# check aliases and give them a little lower score
|
|
lower_alias = 0
|
|
for cur_alias in aliases or []:
|
|
ns = fuzz.UWRatio(search_term, cur_alias)
|
|
if (ns - 1) > s:
|
|
s = ns
|
|
lower_alias = 1
|
|
except (BaseException, Exception) as e:
|
|
if getattr(cls, 'levenshtein_error', None) != dt_date.today():
|
|
cls.levenshtein_error = dt_date.today()
|
|
logger.error('Error generating relevance rating: %s' % ex(e))
|
|
logger.debug('Traceback: %s' % traceback.format_exc())
|
|
return 0
|
|
|
|
# if lang param is supplied, add scale in order to reorder elements 1) en:lang 2) other:lang 3) alias
|
|
# this spacer behaviour may improve the original logic, but currently isn't due to lang used as off switch
|
|
# scale = 3 will enable spacing for all use cases
|
|
scale = (1, 3)[None is not lang]
|
|
|
|
score_scale = (s * scale)
|
|
if score_scale:
|
|
score_scale -= lower_alias
|
|
|
|
# if lang param is supplied, and does not specify English, then lower final score
|
|
score_scale -= (1, 0)[None is lang or lang in ('gb',) or not score_scale]
|
|
|
|
return score_scale
|
|
|
|
def mass_add_table(self, root_dir=None, hash_dir=None, **kwargs):
|
|
|
|
root_dir = root_dir or kwargs.get('root_dir[]')
|
|
if not root_dir:
|
|
return 'No folders selected.'
|
|
|
|
t = PageTemplate(web_handler=self, file='home_massAddTable.tmpl')
|
|
t.submenu = self.home_menu()
|
|
t.kwargs = {'hash_dir': hash_dir}
|
|
t.dir_list = []
|
|
|
|
root_d = sickgear.ROOT_DIRS.split('|')
|
|
if hash_dir:
|
|
root_dirs = root_d[1:]
|
|
else:
|
|
default_i = 0 if not sickgear.ROOT_DIRS else int(root_d[0])
|
|
|
|
root_dirs = [unquote_plus(x) for x in ([root_dir], root_dir)[type(root_dir) == list]]
|
|
if len(root_dirs) > default_i:
|
|
tmp = root_dirs[default_i]
|
|
root_dirs.remove(tmp)
|
|
root_dirs.insert(0, tmp)
|
|
|
|
dir_data = {}
|
|
display_one_dir = None
|
|
|
|
for cur_root_dir in root_dirs:
|
|
try:
|
|
for cur_dir in scantree(cur_root_dir, filter_kind=True, recurse=False):
|
|
|
|
normpath = os.path.normpath(cur_dir.path)
|
|
highlight = hash_dir == re.sub('[^a-z]', '', sg_helpers.md5_for_text(normpath))
|
|
if hash_dir:
|
|
display_one_dir = highlight
|
|
if not hash_dir or display_one_dir:
|
|
dir_data.setdefault(cur_root_dir, {
|
|
'highlight': [], 'rename_suggest': [], 'normpath': [], 'name': [], 'sql': []})
|
|
|
|
dir_data[cur_root_dir]['highlight'].append(highlight)
|
|
dir_data[cur_root_dir]['normpath'].append(normpath)
|
|
suggest = None
|
|
if display_one_dir:
|
|
rename_suggest = ' '
|
|
if kwargs.get('rename_suggest'):
|
|
rename_suggest = ' %s ' % kwargs.get('rename_suggest')
|
|
suggestions = ([], [rename_suggest.rstrip()])[bool(rename_suggest.strip())] + \
|
|
['%s(%s)' % (rename_suggest, x) for x in range(10) if 1 < x]
|
|
for cur_suggestion in suggestions:
|
|
if not os.path.exists('%s%s' % (normpath, cur_suggestion)):
|
|
suggest = cur_suggestion
|
|
break
|
|
dir_data[cur_root_dir]['rename_suggest'].append(suggest)
|
|
dir_data[cur_root_dir]['name'].append(cur_dir.name)
|
|
dir_data[cur_root_dir]['sql'].append([
|
|
"""
|
|
SELECT indexer FROM tv_shows WHERE location = ? LIMIT 1
|
|
""", [normpath]])
|
|
if display_one_dir:
|
|
break
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
if display_one_dir:
|
|
break
|
|
|
|
my_db = db.DBConnection()
|
|
for _, cur_data in iteritems(dir_data):
|
|
cur_data['exists'] = my_db.mass_action(cur_data['sql'])
|
|
|
|
for cur_enum, cur_normpath in enumerate(cur_data['normpath']):
|
|
if display_one_dir and not cur_data['highlight'][cur_enum]:
|
|
continue
|
|
|
|
dir_item = dict(normpath=cur_normpath, rootpath='%s%s' % (os.path.dirname(cur_normpath), os.sep),
|
|
name=cur_data['name'][cur_enum], added_already=any(cur_data['exists'][cur_enum]),
|
|
highlight=cur_data['highlight'][cur_enum])
|
|
|
|
if display_one_dir and cur_data['rename_suggest'][cur_enum]:
|
|
dir_item['rename_suggest'] = cur_data['rename_suggest'][cur_enum]
|
|
|
|
tvid = prodid = show_name = None
|
|
for cur_provider in itervalues(sickgear.metadata_provider_dict):
|
|
if prodid and show_name:
|
|
break
|
|
|
|
(tvid, prodid, show_name) = cur_provider.retrieve_show_metadata(cur_normpath)
|
|
|
|
# default to TVDB if TV info src was not detected
|
|
if show_name and (not tvid or not prodid):
|
|
(sn, idx, pid) = helpers.search_infosrc_for_show_id(show_name, tvid, prodid)
|
|
|
|
# set TV info vars from found info
|
|
if idx and pid:
|
|
(tvid, prodid, show_name) = (idx, pid, sn)
|
|
|
|
# in case we don't have both requirements, set both to None
|
|
if not tvid or not prodid:
|
|
tvid = prodid = None
|
|
|
|
dir_item['existing_info'] = (tvid, prodid, show_name)
|
|
|
|
if helpers.find_show_by_id({tvid: prodid}):
|
|
dir_item['added_already'] = True
|
|
|
|
t.dir_list.append(dir_item)
|
|
|
|
return t.respond()
|
|
|
|
def new_show(self, show_to_add=None, other_shows=None, use_show_name=False, **kwargs):
|
|
"""
|
|
Display the new show page which collects a tvdb id, folder, and extra options and
|
|
posts them to add_new_show
|
|
"""
|
|
self.set_header('Cache-Control', 'no-cache, no-store, must-revalidate')
|
|
self.set_header('Pragma', 'no-cache')
|
|
self.set_header('Expires', '0')
|
|
|
|
t = PageTemplate(web_handler=self, file='home_newShow.tmpl')
|
|
t.submenu = self.home_menu()
|
|
t.enable_anime_options = True
|
|
t.enable_default_wanted = True
|
|
t.kwargs = kwargs
|
|
|
|
tvid, show_dir, prodid, show_name = self.split_extra_show(show_to_add)
|
|
|
|
# use the given show_dir for the TV info search if available
|
|
if use_show_name:
|
|
t.default_show_name = show_name
|
|
elif not show_dir:
|
|
t.default_show_name = ''
|
|
elif not show_name:
|
|
t.default_show_name = os.path.basename(os.path.normpath(show_dir)).replace('.', ' ')
|
|
else:
|
|
t.default_show_name = show_name
|
|
|
|
# carry a list of other dirs if given
|
|
if not other_shows:
|
|
other_shows = []
|
|
elif type(other_shows) != list:
|
|
other_shows = [other_shows]
|
|
|
|
# tell the template whether we're providing it show name & TV info src
|
|
t.use_provided_info = bool(prodid and tvid and show_name)
|
|
if t.use_provided_info:
|
|
t.provided_prodid = int(prodid or 0)
|
|
t.provided_show_name = show_name
|
|
|
|
t.provided_show_dir = show_dir
|
|
t.other_shows = other_shows
|
|
t.infosrc = sickgear.TVInfoAPI().search_sources
|
|
search_tvid = None
|
|
if use_show_name and 1 == show_name.count(':'): # if colon is found once
|
|
search_tvid = list(filter(lambda x: bool(x),
|
|
[('%s:' % sickgear.TVInfoAPI(_tvid).config['slug']) in show_name and _tvid
|
|
for _tvid, _ in iteritems(t.infosrc)]))
|
|
search_tvid = 1 == len(search_tvid) and search_tvid[0]
|
|
t.provided_tvid = search_tvid or int(tvid or sickgear.TVINFO_DEFAULT)
|
|
t.infosrc_icons = [sickgear.TVInfoAPI(cur_tvid).config.get('icon') for cur_tvid in t.infosrc]
|
|
t.meta_lang = sickgear.ADD_SHOWS_METALANG
|
|
t.allowlist = []
|
|
t.blocklist = []
|
|
t.groups = []
|
|
|
|
t.show_scene_maps = list(itervalues(scene_exceptions.MEMCACHE['release_map_xem']))
|
|
|
|
has_shows = len(sickgear.showList)
|
|
t.try_id = [] # [dict try_tip: try_term]
|
|
t.try_id_name = [] # [dict try_tip: try_term]
|
|
t.try_url = [] # [dict try_tip: try_term]
|
|
url_num = 0
|
|
for cur_idx, (cur_tvid, cur_try, cur_id_def, cur_url_def) in enumerate([
|
|
(TVINFO_IMDB, '%s:tt%s', '0944947', 'https://www.imdb.com/title/tt0944947'),
|
|
(TVINFO_TMDB, '%s:%s', '1399', 'https://www.themoviedb.org/tv/1399'),
|
|
(TVINFO_TRAKT, None, None, 'https://trakt.tv/shows/game-of-thrones'),
|
|
(TVINFO_TVDB, None, None, 'https://thetvdb.com/series/game-of-thrones'),
|
|
(TVINFO_TVDB, '%s:%s', '121361', 'https://thetvdb.com/?tab=series&id=121361&lid=7'),
|
|
(TVINFO_TVMAZE, '%s:%s', '82', None)]
|
|
):
|
|
slug = sickgear.TVInfoAPI(cur_tvid).config['slug']
|
|
try_id = has_shows and cur_try and sickgear.showList[-1].ids[cur_tvid].get('id')
|
|
if not cur_idx:
|
|
t.try_name = [{
|
|
'showname': 'Game of Thrones' if not try_id else sickgear.showList[-1].name.replace("'", "\\'")}]
|
|
|
|
if cur_try:
|
|
id_key = '%s:id%s' % (slug, ('', ' (GoT)')[not try_id])
|
|
id_val = cur_try % (slug, try_id or cur_id_def)
|
|
t.try_id += [{id_key: id_val}]
|
|
t.try_id_name += [{'%s show name' % id_key: '%s %s' % (id_val, t.try_name[0]['showname'])}]
|
|
|
|
if cur_url_def:
|
|
url_num += 1
|
|
t.try_url += [{
|
|
'url .. %s%s' % (url_num, ('', ' (GoT)')[not cur_try]):
|
|
cur_url_def if not try_id else sickgear.TVInfoAPI(cur_tvid).config['show_url'] % try_id}]
|
|
|
|
return t.respond()
|
|
|
|
def anime_default(self):
|
|
return self.randomhot_anidb()
|
|
|
|
def randomhot_anidb(self, **kwargs):
|
|
|
|
browse_type = 'AniDB'
|
|
filtered = []
|
|
|
|
# xref_src = 'https://raw.githubusercontent.com/ScudLee/anime-lists/master/anime-list.xml'
|
|
xref_src = 'https://raw.githubusercontent.com/Anime-Lists/anime-lists/master/anime-list.xml'
|
|
xml_data = helpers.get_url(xref_src)
|
|
xref_root = xml_data and helpers.parse_xml(xml_data)
|
|
if None is not xref_root and not len(xref_root):
|
|
xref_root = None
|
|
|
|
# noinspection HttpUrlsUsage
|
|
url = 'http://api.anidb.net:9001/httpapi?client=sickgear&clientver=1&protover=1&request=main'
|
|
response = helpers.get_url(url)
|
|
if response and None is not xref_root:
|
|
oldest, newest = None, None
|
|
try:
|
|
anime_root = helpers.parse_xml(response)
|
|
hot_anime, random_rec = [anime_root.find(node) for node in ['hotanime', 'randomrecommendation']]
|
|
random_rec = [item.find('./anime') for item in random_rec]
|
|
oldest_dt, newest_dt = 9999999, 0
|
|
for list_type, items in [('hot', list(hot_anime)), ('recommended', random_rec)]:
|
|
for anime in items:
|
|
ids = dict(anidb=config.to_int(anime.get('id')))
|
|
xref_node = xref_root.find('./anime[@anidbid="%s"]' % ids['anidb'])
|
|
if None is xref_node:
|
|
continue
|
|
# noinspection PyUnresolvedReferences
|
|
tvdbid = config.to_int(xref_node.get('tvdbid'))
|
|
if None is tvdbid:
|
|
continue
|
|
ids.update(dict(tvdb=tvdbid))
|
|
first_aired, title, image = [None is not y and y.text or y for y in [
|
|
anime.find(node) for node in ['startdate', 'title', 'picture']]]
|
|
|
|
ord_premiered, str_premiered, started_past, oldest_dt, newest_dt, oldest, newest, _, _, _, _ \
|
|
= self.sanitise_dates(first_aired, oldest_dt, newest_dt, oldest, newest)
|
|
|
|
# img_uri = 'http://img7.anidb.net/pics/anime/%s' % image
|
|
img_uri = 'https://cdn-eu.anidb.net/images/main/%s' % image
|
|
images = dict(poster=dict(thumb='imagecache?path=browse/thumb/anidb&source=%s' % img_uri))
|
|
sickgear.CACHE_IMAGE_URL_LIST.add_url(img_uri)
|
|
|
|
votes = rating = 0
|
|
counts = anime.find('./ratings/permanent')
|
|
if isinstance(counts, object):
|
|
# noinspection PyUnresolvedReferences
|
|
votes = counts.get('count')
|
|
# noinspection PyUnresolvedReferences
|
|
rated = float(counts.text)
|
|
rating = 100 < rated and rated / 10 or 10 > rated and 10 * rated or rated
|
|
|
|
# noinspection HttpUrlsUsage
|
|
filtered.append(dict(
|
|
type=list_type,
|
|
ord_premiered=ord_premiered,
|
|
str_premiered=str_premiered,
|
|
started_past=started_past, # air time not poss. 16.11.2015
|
|
genres='',
|
|
ids=ids,
|
|
images=images,
|
|
overview='',
|
|
rating=rating,
|
|
title=title.strip(),
|
|
url_src_db='http://anidb.net/perl-bin/animedb.pl?show=anime&aid=%s' % ids['anidb'],
|
|
url_tvdb=sickgear.TVInfoAPI(TVINFO_TVDB).config['show_url'] % ids['tvdb'],
|
|
votes=votes
|
|
))
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
kwargs.update(dict(oldest=oldest, newest=newest))
|
|
|
|
return self.browse_shows(browse_type, 'Random and Hot at AniDB', filtered, **kwargs)
|
|
|
|
def info_anidb(self, ids, show_name):
|
|
|
|
if not list(filter(lambda tvid_prodid: helpers.find_show_by_id(tvid_prodid), ids.split(' '))):
|
|
return self.new_show('|'.join(['', '', '', ' '.join([ids, show_name])]), use_show_name=True, is_anime=True)
|
|
|
|
@staticmethod
|
|
def watchlist_config(**kwargs):
|
|
|
|
if not isinstance(sickgear.IMDB_ACCOUNTS, type([])):
|
|
sickgear.IMDB_ACCOUNTS = list(sickgear.IMDB_ACCOUNTS)
|
|
accounts = dict(map_none(*[iter(sickgear.IMDB_ACCOUNTS)] * 2))
|
|
|
|
if 'enable' == kwargs.get('action'):
|
|
account_id = re.findall(r'\d{6,32}', kwargs.get('input', ''))
|
|
if not account_id:
|
|
return json_dumps({'result': 'Fail: Invalid IMDb ID'})
|
|
acc_id = account_id[0]
|
|
|
|
url = 'https://www.imdb.com/user/ur%s/watchlist' % acc_id + \
|
|
'?sort=date_added,desc&title_type=tvSeries,tvEpisode,tvMiniSeries&view=detail'
|
|
html = helpers.get_url(url, nocache=True)
|
|
if not html:
|
|
return json_dumps({'result': 'Fail: No list found with id: %s' % acc_id})
|
|
if 'id="unavailable"' in html or 'list is not public' in html or 'not enabled public view' in html:
|
|
return json_dumps({'result': 'Fail: List is not public with id: %s' % acc_id})
|
|
|
|
try:
|
|
list_name = re.findall(r'(?i)og:title[^>]+?content[^"]+?"([^"]+?)\s+Watchlist\s*"',
|
|
html)[0].replace('\'s', '')
|
|
accounts[acc_id] = list_name or 'noname'
|
|
except (BaseException, Exception):
|
|
return json_dumps({'result': 'Fail: No list found with id: %s' % acc_id})
|
|
|
|
else:
|
|
acc_id = kwargs.get('select', '')
|
|
if acc_id not in accounts:
|
|
return json_dumps({'result': 'Fail: Unknown IMDb ID'})
|
|
|
|
if 'disable' == kwargs.get('action'):
|
|
accounts[acc_id] = '(Off) %s' % accounts[acc_id].replace('(Off) ', '')
|
|
else:
|
|
del accounts[acc_id]
|
|
|
|
gears = [[k, v] for k, v in iteritems(accounts) if 'sickgear' in v.lower()]
|
|
if gears:
|
|
del accounts[gears[0][0]]
|
|
yours = [[k, v] for k, v in iteritems(accounts) if 'your' == v.replace('(Off) ', '').lower()]
|
|
if yours:
|
|
del accounts[yours[0][0]]
|
|
sickgear.IMDB_ACCOUNTS = [x for tup in sorted(list(iteritems(accounts)), key=lambda t: t[1]) for x in tup]
|
|
if gears:
|
|
sickgear.IMDB_ACCOUNTS.insert(0, gears[0][1])
|
|
sickgear.IMDB_ACCOUNTS.insert(0, gears[0][0])
|
|
if yours:
|
|
sickgear.IMDB_ACCOUNTS.insert(0, yours[0][1])
|
|
sickgear.IMDB_ACCOUNTS.insert(0, yours[0][0])
|
|
sickgear.save_config()
|
|
|
|
return json_dumps({'result': 'Success', 'accounts': sickgear.IMDB_ACCOUNTS})
|
|
|
|
@staticmethod
|
|
def parse_imdb_overview(tag):
|
|
paragraphs = tag.select('.dli-plot-container .ipc-html-content-inner-div')
|
|
filtered = []
|
|
for item in paragraphs:
|
|
if not (item.select('span.certificate') or item.select('span.genre') or
|
|
item.select('span.runtime') or item.select('span.ghost')):
|
|
filtered.append(item.get_text().strip())
|
|
split_lines = [element.split('\n') for element in filtered]
|
|
filtered = []
|
|
least_lines = 10
|
|
for item_lines in split_lines:
|
|
if len(item_lines) < least_lines:
|
|
least_lines = len(item_lines)
|
|
filtered = [item_lines]
|
|
elif len(item_lines) == least_lines:
|
|
filtered.append(item_lines)
|
|
overview = ''
|
|
for item_lines in filtered:
|
|
text = ' '.join([item_lines.strip() for item_lines in item_lines]).strip()
|
|
if len(text) and (not overview or (len(text) > len(overview))):
|
|
overview = text
|
|
return overview
|
|
|
|
def parse_imdb(self, data, filtered, kwargs):
|
|
|
|
oldest, newest, oldest_dt, newest_dt = None, None, 9999999, 0
|
|
show_list = (data or {}).get('list', {}).get('items', {})
|
|
idx_ids = dict(map(lambda so: (so.imdbid, (so.tvid, so.prodid)),
|
|
filter(lambda _so: getattr(_so, 'imdbid', None), sickgear.showList)))
|
|
|
|
# list_id = (data or {}).get('list', {}).get('id', {})
|
|
for row in show_list:
|
|
row = data.get('titles', {}).get(row.get('const'))
|
|
if not row:
|
|
continue
|
|
try:
|
|
ids = dict(imdb=row.get('id', ''))
|
|
year, ended = 2 * [None]
|
|
if 2 == len(row.get('primary').get('year')):
|
|
year, ended = row.get('primary').get('year')
|
|
ord_premiered = 0
|
|
started_past = False
|
|
if year:
|
|
ord_premiered, str_premiered, started_past, oldest_dt, newest_dt, oldest, newest, _, _, _, _ \
|
|
= self.sanitise_dates('01-01-%s' % year, oldest_dt, newest_dt, oldest, newest)
|
|
|
|
overview = row.get('plot')
|
|
rating = row.get('ratings', {}).get('rating', 0)
|
|
voting = row.get('ratings', {}).get('votes', 0)
|
|
images = {}
|
|
img_uri = '%s' % row.get('poster', {}).get('url', '')
|
|
if img_uri and 'tv_series.gif' not in img_uri and 'nopicture' not in img_uri:
|
|
scale = (lambda low1, high1: int((float(450) / high1) * low1))
|
|
dims = [row.get('poster', {}).get('width', 0), row.get('poster', {}).get('height', 0)]
|
|
s = [scale(x, int(max(dims))) for x in dims]
|
|
img_uri = re.sub(r'(?im)(.*V1_?)(\..*?)$', r'\1UX%s_CR0,0,%s,%s_AL_\2'
|
|
% (s[0], s[0], s[1]), img_uri)
|
|
images = dict(poster=dict(thumb='imagecache?path=browse/thumb/imdb&source=%s' % img_uri))
|
|
sickgear.CACHE_IMAGE_URL_LIST.add_url(img_uri)
|
|
|
|
filtered.append(dict(
|
|
ord_premiered=ord_premiered,
|
|
str_premiered=year or 'No year',
|
|
ended_str=ended or '',
|
|
started_past=started_past, # air time not poss. 16.11.2015
|
|
genres=', '.join(row.get('metadata', {}).get('genres', {})) or 'No genre yet',
|
|
ids=ids,
|
|
images='' if not img_uri else images,
|
|
overview=self.clean_overview(overview),
|
|
rating=int(helpers.try_float(rating) * 10),
|
|
title=row.get('primary').get('title'),
|
|
url_src_db='https://www.imdb.com/%s/' % row.get('primary').get('href').strip('/'),
|
|
votes=helpers.try_int(voting, 'TBA')))
|
|
|
|
tvid, prodid = idx_ids.get(ids['imdb'], (None, None))
|
|
if tvid and tvid in [_tvid for _tvid in sickgear.TVInfoAPI().search_sources]:
|
|
infosrc_slug, infosrc_url = (sickgear.TVInfoAPI(tvid).config[x] for x in ('slug', 'show_url'))
|
|
filtered[-1]['ids'][infosrc_slug] = prodid
|
|
filtered[-1]['url_' + infosrc_slug] = infosrc_url % prodid
|
|
except (AttributeError, TypeError, KeyError, IndexError):
|
|
pass
|
|
|
|
kwargs.update(dict(oldest=oldest, newest=newest))
|
|
|
|
return show_list and True or None
|
|
|
|
def parse_imdb_html(self, html, filtered, kwargs):
|
|
|
|
img_size = re.compile(r'(?im)(V1[^XY]+([XY]))(\d+)(\D+)(\d+)(\D+)(\d+)(\D+)(\d+)(\D+)(\d+)(.*?)$')
|
|
|
|
with BS4Parser(html, features=['html5lib', 'permissive']) as soup:
|
|
show_list = soup.select('.detailed-list-view ')
|
|
shows = [] if not show_list else show_list[0].select('li')
|
|
oldest, newest, oldest_dt, newest_dt = None, None, 9999999, 0
|
|
|
|
for row in shows:
|
|
try:
|
|
title = re.sub(r'\d+\.\s(.*)', r'\1', row.select('.ipc-title__text')[0].get_text(strip=True))
|
|
url_path = re.sub(r'(.*?)(\?ref_=.*)?', r'\1', row.select('.ipc-title-link-wrapper')[0]['href'])
|
|
ids = dict(imdb=helpers.parse_imdb_id(url_path))
|
|
year, ended = 2 * [None]
|
|
first_aired = row.select('.dli-title-metadata .dli-title-metadata-item')
|
|
if len(first_aired):
|
|
years = re.findall(r'.*?(\d{4})(?:.*?(\d{4}))?.*', first_aired[0].get_text(strip=True))
|
|
year, ended = years and years[0] or 2 * [None]
|
|
ord_premiered = 0
|
|
started_past = False
|
|
if year:
|
|
ord_premiered, str_premiered, started_past, oldest_dt, newest_dt, oldest, newest, _, _, _, _ \
|
|
= self.sanitise_dates('01-01-%s' % year, oldest_dt, newest_dt, oldest, newest)
|
|
|
|
images = {}
|
|
img = row.select('img.ipc-image')
|
|
overview = self.parse_imdb_overview(row)
|
|
rating = row.select_one('.ipc-rating-star').get_text()
|
|
rating = rating and rating.split()[0] or ''
|
|
try:
|
|
voting = row.select_one('.ipc-rating-star--voteCount').get_text(strip=True).strip('()').lower()
|
|
if voting.endswith('k'):
|
|
voting = helpers.try_float(voting[:-1]) * 1000
|
|
elif voting.endswith('m'):
|
|
voting = helpers.try_float(voting[:-1]) * 1000000
|
|
except (BaseException, Exception):
|
|
voting = ''
|
|
img_uri = None
|
|
if len(img):
|
|
img_uri = img[0].get('src')
|
|
match = img_size.search(img_uri)
|
|
if match and 'tv_series.gif' not in img_uri and 'nopicture' not in img_uri:
|
|
scale = (lambda low1, high1: int((float(450) / high1) * low1))
|
|
high = int(max([match.group(9), match.group(11)]))
|
|
scaled = [scale(x, high) for x in
|
|
[(int(match.group(n)), high)[high == int(match.group(n))] for n in
|
|
(3, 5, 7, 9, 11)]]
|
|
parts = [match.group(1), match.group(4), match.group(6), match.group(8), match.group(10),
|
|
match.group(12)]
|
|
img_uri = img_uri.replace(match.group(), ''.join(
|
|
[str(y) for x in map_none(parts, scaled) for y in x if None is not y]))
|
|
images = dict(poster=dict(thumb='imagecache?path=browse/thumb/imdb&source=%s' % img_uri))
|
|
sickgear.CACHE_IMAGE_URL_LIST.add_url(img_uri)
|
|
|
|
filtered.append(dict(
|
|
ord_premiered=ord_premiered,
|
|
str_premiered=year or 'No year',
|
|
ended_str=ended or '',
|
|
started_past=started_past, # air time not poss. 16.11.2015
|
|
genres='',
|
|
ids=ids,
|
|
images='' if not img_uri else images,
|
|
overview=self.clean_overview(overview),
|
|
rating=0 if not len(rating) else int(helpers.try_float(rating) * 10),
|
|
title=title,
|
|
url_src_db='https://www.imdb.com/%s/' % url_path.strip('/'),
|
|
votes=helpers.try_int(voting, 'TBA')))
|
|
|
|
show_obj = helpers.find_show_by_id({TVINFO_IMDB: int(ids['imdb'].replace('tt', ''))},
|
|
no_mapped_ids=False)
|
|
for tvid in filter(lambda _tvid: _tvid == show_obj.tvid, sickgear.TVInfoAPI().search_sources):
|
|
infosrc_slug, infosrc_url = (sickgear.TVInfoAPI(tvid).config[x] for x in
|
|
('slug', 'show_url'))
|
|
filtered[-1]['ids'][infosrc_slug] = show_obj.prodid
|
|
filtered[-1]['url_' + infosrc_slug] = infosrc_url % show_obj.prodid
|
|
except (AttributeError, TypeError, KeyError, IndexError):
|
|
continue
|
|
|
|
kwargs.update(dict(oldest=oldest, newest=newest))
|
|
|
|
return show_list and True or None
|
|
|
|
def watchlist_imdb(self, **kwargs):
|
|
|
|
if 'add' == kwargs.get('action'):
|
|
return self.redirect('/config/general/#core-component-group2')
|
|
|
|
if kwargs.get('action') in ('delete', 'enable', 'disable'):
|
|
return self.watchlist_config(**kwargs)
|
|
|
|
browse_type = 'IMDb'
|
|
|
|
filtered = []
|
|
footnote = None
|
|
start_year, end_year = (dt_date.today().year - 10, dt_date.today().year + 1)
|
|
periods = [(start_year, end_year)] + [(x - 10, x) for x in range(start_year, start_year - 40, -10)]
|
|
|
|
accounts = dict(map_none(*[iter(sickgear.IMDB_ACCOUNTS)] * 2))
|
|
acc_id, list_name = (sickgear.IMDB_DEFAULT_LIST_ID, sickgear.IMDB_DEFAULT_LIST_NAME) if \
|
|
0 == helpers.try_int(kwargs.get('account')) or \
|
|
kwargs.get('account') not in accounts or \
|
|
accounts.get(kwargs.get('account'), '').startswith('(Off) ') else \
|
|
(kwargs.get('account'), accounts.get(kwargs.get('account')))
|
|
|
|
list_name += ('\'s', '')['your' == list_name.replace('(Off) ', '').lower()]
|
|
|
|
mode = 'watchlist-%s' % acc_id
|
|
|
|
url = 'https://www.imdb.com/user/ur%s/watchlist' % acc_id
|
|
url_ui = '?mode=detail&page=1&sort=date_added,desc&' \
|
|
'title_type=tvSeries,tvEpisode,tvMiniSeries&ref_=wl_ref_typ'
|
|
|
|
html = helpers.get_url(url + url_ui, headers={'Accept-Language': 'en-US'})
|
|
if html:
|
|
show_list_found = None
|
|
try:
|
|
data = json_loads((re.findall(r'(?im)IMDb.*?Initial.*?\.push\((.*)\).*?$', html) or ['{}'])[0])
|
|
show_list_found = self.parse_imdb(data, filtered, kwargs)
|
|
except (BaseException, Exception):
|
|
pass
|
|
if not show_list_found:
|
|
show_list_found = self.parse_imdb_html(html, filtered, kwargs)
|
|
kwargs.update(dict(start_year=start_year))
|
|
|
|
if len(filtered):
|
|
footnote = ('Note; Some images on this page may be cropped at source: ' +
|
|
'<a target="_blank" href="%s">%s watchlist at IMDb</a>' % (
|
|
helpers.anon_url(url + url_ui), list_name))
|
|
elif None is not show_list_found or (None is show_list_found and list_name in html):
|
|
kwargs['show_header'] = True
|
|
kwargs['error_msg'] = 'No TV titles in the <a target="_blank" href="%s">%s watchlist at IMDb</a>' % (
|
|
helpers.anon_url(url + url_ui), list_name)
|
|
|
|
kwargs.update(dict(footnote=footnote, mode='watchlist-%s' % acc_id, periods=periods))
|
|
|
|
if mode:
|
|
sickgear.IMDB_MRU = mode
|
|
sickgear.save_config()
|
|
|
|
return self.browse_shows(browse_type, '%s IMDb Watchlist' % list_name, filtered, **kwargs)
|
|
|
|
def imdb_default(self, **kwargs):
|
|
if 'popular-' in sickgear.IMDB_MRU:
|
|
kwargs.update(dict(period=sickgear.IMDB_MRU.split('-')[1]))
|
|
return self.popular_imdb(**kwargs)
|
|
if 'watchlist-' in sickgear.IMDB_MRU:
|
|
kwargs.update(dict(account=sickgear.IMDB_MRU.split('-')[1]))
|
|
return self.watchlist_imdb(**kwargs)
|
|
method = getattr(self, sickgear.IMDB_MRU, None)
|
|
if not callable(method):
|
|
return self.popular_imdb(**kwargs)
|
|
return method(**kwargs)
|
|
|
|
def popular_imdb(self, **kwargs):
|
|
|
|
browse_type = 'IMDb'
|
|
|
|
filtered = []
|
|
footnote = None
|
|
start_year, end_year = (dt_date.today().year - 10, dt_date.today().year + 1)
|
|
periods = [(start_year, end_year)] + [(x - 10, x) for x in range(start_year, start_year - 40, -10)]
|
|
|
|
start_year_in, end_year_in = [helpers.try_int(x) for x in (('0,0', kwargs.get('period'))[
|
|
',' in kwargs.get('period', '')]).split(',')]
|
|
if 1900 < start_year_in < 2050 and 2050 > end_year_in > 1900:
|
|
start_year, end_year = (start_year_in, end_year_in)
|
|
|
|
mode = 'popular-%s,%s' % (start_year, end_year)
|
|
|
|
page = 'more' in kwargs and '51' or ''
|
|
if page:
|
|
mode += '-more'
|
|
url = 'https://www.imdb.com/search/title?at=0&sort=moviemeter&' \
|
|
'title_type=tvSeries,tvEpisode,tvMiniSeries&year=%s,%s&start=%s' % (start_year, end_year, page)
|
|
html = helpers.get_url(url, headers={'Accept-Language': 'en-US'})
|
|
if html:
|
|
show_list_found = None
|
|
try:
|
|
data = json_loads((re.findall(r'(?im)IMDb.*?Initial.*?\.push\((.*)\).*?$', html) or ['{}'])[0])
|
|
show_list_found = self.parse_imdb(data, filtered, kwargs)
|
|
except (BaseException, Exception):
|
|
pass
|
|
if not show_list_found:
|
|
self.parse_imdb_html(html, filtered, kwargs)
|
|
kwargs.update(dict(mode=mode, periods=periods))
|
|
|
|
if len(filtered):
|
|
footnote = ('Note; Some images on this page may be cropped at source: ' +
|
|
'<a target="_blank" href="%s">IMDb</a>' % helpers.anon_url(url))
|
|
|
|
kwargs.update(dict(footnote=footnote))
|
|
|
|
if mode:
|
|
sickgear.IMDB_MRU = mode
|
|
sickgear.save_config()
|
|
|
|
return self.browse_shows(browse_type, 'Most Popular IMDb TV', filtered, **kwargs)
|
|
|
|
def info_imdb(self, ids, show_name):
|
|
|
|
return self.new_show('|'.join(['', '', '', helpers.parse_imdb_id(ids) and ' '.join([ids, show_name])]),
|
|
use_show_name=True)
|
|
|
|
def mc_default(self):
|
|
method = getattr(self, sickgear.MC_MRU, None)
|
|
if not callable(method):
|
|
return self.mc_newseries()
|
|
return method()
|
|
|
|
def mc_newseries(self, **kwargs):
|
|
return self.browse_mc(
|
|
'/all/all/all-time/new/', 'New Series at Metacritic', mode='newseries', **kwargs)
|
|
|
|
def mc_explore(self, **kwargs):
|
|
return self.browse_mc(
|
|
'/', 'Explore at Metacritic', mode='explore', **kwargs)
|
|
|
|
def mc_popular(self, **kwargs):
|
|
return self.browse_mc(
|
|
'/all/all/all-time/popular/', 'Popular at Metacritic', mode='popular', **kwargs)
|
|
|
|
def mc_metascore(self, **kwargs):
|
|
return self.browse_mc(
|
|
'/all/all/all-time/metascore/', 'By metascore at Metacritic', mode='metascore', **kwargs)
|
|
|
|
def mc_userscore(self, **kwargs):
|
|
return self.browse_mc(
|
|
'/all/all/all-time/userscore/', 'By userscore at Metacritic', mode='userscore', **kwargs)
|
|
|
|
def browse_mc(self, url_path, browse_title, **kwargs):
|
|
|
|
browse_type = 'Metacritic'
|
|
|
|
footnote = None
|
|
|
|
page = 'more' in kwargs and '&page=2' or ''
|
|
if page:
|
|
kwargs['mode'] += '-more'
|
|
|
|
filtered = []
|
|
|
|
import browser_ua
|
|
this_year = dt_date.today().strftime('%Y')
|
|
url = f'https://www.metacritic.com/browse/tv{url_path}' \
|
|
f'?releaseYearMin={this_year}&releaseYearMax={this_year}{page}'
|
|
html = helpers.get_url(url, headers={'User-Agent': browser_ua.get_ua()})
|
|
if html:
|
|
items_data = []
|
|
try:
|
|
items_html = html[6 + html.index('items:[{awards'):]
|
|
items_bufr = re.split(r'\btype:"', items_html)
|
|
for cur_item in items_bufr[1:]: # iterates from the first true type:"show" record
|
|
if not cur_item.startswith('show'):
|
|
break
|
|
items_data.append(f'type:"{cur_item}')
|
|
del items_html
|
|
del items_bufr
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
try:
|
|
if re.findall('(c-navigationPagination_item--next)', html)[0]:
|
|
kwargs.update(dict(more=1))
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
with BS4Parser(html, parse_only=dict(div={'class': (lambda at: at and 'c-productListings' in at)})) as soup:
|
|
items = [] if not soup else soup.select('.c-finderProductCard_container')
|
|
oldest, newest, oldest_dt, newest_dt = None, None, 9999999, 0
|
|
rc_title = re.compile(r'(?i)(?::\s*season\s*\d+|\s*\((?:19|20)\d{2}\))?$')
|
|
rc_id = re.compile(r'(?i)[^A-Z0-9]')
|
|
rc_img = re.compile(r'(.*?)(/resize/[^?]+)?(/catalog/provider.*?\.(?:jpg|png)).*')
|
|
rc_season = re.compile(r'(\d+)(?:[.]\d*?)?$')
|
|
for cur_idx, cur_row in enumerate(items):
|
|
try:
|
|
title = rc_title.sub(
|
|
'', cur_row.find('div', class_='c-finderProductCard_title').get('data-title').strip())
|
|
|
|
# 2023-09-23 deprecated id at site, using title as id
|
|
# ids = dict(custom=cur_row.select('input[type="checkbox"]')[0].attrs['id'], name='mc')
|
|
ids = dict(custom=rc_id.sub('', title), name='mc')
|
|
|
|
url_path = cur_row['href'].strip()
|
|
if not url_path.startswith('/tv/'):
|
|
continue
|
|
|
|
images = None
|
|
img_src = (cur_row.find('img') or {}).get('src', '').strip()
|
|
if not img_src and items_data: # items_data is the sites' image method from 2024
|
|
buffer_idx = None
|
|
if title in items_data[cur_idx]:
|
|
buffer_idx = cur_idx
|
|
else:
|
|
for cur_data_idx, cur_item in enumerate(items_data):
|
|
if title in cur_item:
|
|
buffer_idx = cur_data_idx
|
|
break
|
|
if None is not buffer_idx:
|
|
try:
|
|
img_rel = re.findall(
|
|
r'bucketPath[^:]*?:[^"]*?"([^"]+?)"',
|
|
items_data[buffer_idx], re.I)[0].encode().decode('unicode-escape')
|
|
img_src = f'https://www.metacritic.com/a/img/catalog/{img_rel.strip("/")}'
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
if img_src:
|
|
img_uri = rc_img.sub(r'\1\3', img_src)
|
|
images = dict(poster=dict(thumb=f'imagecache?path=browse/thumb/metac&source={img_uri}'))
|
|
sickgear.CACHE_IMAGE_URL_LIST.add_url(img_uri)
|
|
|
|
ord_premiered = 0
|
|
str_premiered = ''
|
|
started_past = False
|
|
|
|
dated = None
|
|
rating = None
|
|
rating_user = None # 2023-09-23 deprecated at site
|
|
meta_tags = cur_row.find_all('div', class_='c-finderProductCard_meta')
|
|
for tag in meta_tags:
|
|
meta_tag = tag.find('span', class_='u-text-uppercase')
|
|
if not dated and meta_tag:
|
|
dated = meta_tag
|
|
try: # a bad date caused a sanitise exception here
|
|
ord_premiered, str_premiered, started_past, oldest_dt, newest_dt, oldest, newest, \
|
|
_, _, _, _ = self.sanitise_dates(dated.get_text().strip(), oldest_dt, newest_dt,
|
|
oldest, newest)
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
meta_tag = tag.find('div', class_='c-siteReviewScore')
|
|
if not rating and meta_tag:
|
|
rating = meta_tag
|
|
rating = rating.get_text().strip()
|
|
|
|
if dated and rating:
|
|
break
|
|
|
|
overview = cur_row.find('div', class_='c-finderProductCard_description')
|
|
if overview:
|
|
overview = overview.get_text()
|
|
|
|
try:
|
|
season = rc_season.findall(url_path)[0]
|
|
except(BaseException, Exception):
|
|
season = -1
|
|
|
|
filtered.append(dict(
|
|
ord_premiered=ord_premiered,
|
|
str_premiered=str_premiered,
|
|
started_past=started_past,
|
|
episode_season=int(season),
|
|
genres='',
|
|
ids=ids,
|
|
images=images or '',
|
|
overview=self.clean_overview(overview),
|
|
rating=0 if not rating else rating or 'TBD',
|
|
rating_user='tbd' if not rating_user else int(helpers.try_float(rating_user) * 10) or 'tbd',
|
|
title=title,
|
|
url_src_db=f'https://www.metacritic.com/{url_path.strip("/")}/',
|
|
votes=None))
|
|
|
|
except (AttributeError, IndexError, KeyError, TypeError):
|
|
continue
|
|
|
|
kwargs.update(dict(oldest=oldest, newest=newest))
|
|
|
|
kwargs.update(dict(footnote=footnote, use_votes=False))
|
|
|
|
mode = kwargs.get('mode', '')
|
|
if mode:
|
|
func = f'mc_{mode}'
|
|
if callable(getattr(self, func, None)):
|
|
sickgear.MC_MRU = func
|
|
sickgear.save_config()
|
|
return self.browse_shows(browse_type, browse_title, filtered, **kwargs)
|
|
|
|
# noinspection PyUnusedLocal
|
|
def info_metacritic(self, ids, show_name):
|
|
|
|
return self.new_show('|'.join(['', '', '', show_name]), use_show_name=True)
|
|
|
|
def ne_default(self):
|
|
method = getattr(self, sickgear.NE_MRU, None)
|
|
if not callable(method):
|
|
return self.ne_newpop()
|
|
return method()
|
|
|
|
def ne_newpop(self, **kwargs):
|
|
return self.browse_ne(
|
|
'hot', 'Popular recent premiered at Next Episode', mode='newpop', **kwargs)
|
|
|
|
def ne_newtop(self, **kwargs):
|
|
return self.browse_ne(
|
|
'hot', 'Top rated recent premiered at Next Episode', mode='newtop', **kwargs)
|
|
|
|
def ne_upcoming(self, **kwargs):
|
|
return self.browse_ne(
|
|
'upcoming', 'Upcoming Season 1 at Next Episode', mode='upcoming', **kwargs)
|
|
|
|
def ne_upcoming2(self, **kwargs):
|
|
return self.browse_ne(
|
|
'upcoming2', 'Upcoming Season 2 at Next Episode', mode='upcoming2', **kwargs)
|
|
|
|
def ne_trending(self, **kwargs):
|
|
return self.browse_ne(
|
|
'trends', 'Trending at Next Episode', mode='trending', **kwargs)
|
|
|
|
def browse_ne(self, url_path, browse_title, **kwargs):
|
|
|
|
browse_type = 'Nextepisode'
|
|
|
|
footnote = None
|
|
|
|
page = 1
|
|
if 'more' in kwargs:
|
|
page = 2
|
|
kwargs['mode'] += '-more'
|
|
|
|
filtered = []
|
|
|
|
import browser_ua
|
|
|
|
started_past = True
|
|
if 'upcoming' in url_path:
|
|
started_past = False
|
|
params = f'premiering_period={("1", "2")["2" in url_path]}&sort=2'
|
|
url_path = 'upcoming'
|
|
elif 'trends' in url_path:
|
|
params = 'trending_within=2'
|
|
else:
|
|
params = f'chart_type={("most_popular", "top_rated")["Top" in browse_title]}&premiered_within=2'
|
|
url = f'https://next-episode.net/{url_path}/?{params}&page={page}'
|
|
html = helpers.get_url(url, headers={'User-Agent': browser_ua.get_ua()})
|
|
if html:
|
|
try:
|
|
if re.findall(r'(?i)id="paginationDiv"', html)[0]:
|
|
kwargs.update(dict(more=1))
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
with BS4Parser(html) as soup:
|
|
shows = [] if not soup else soup.find_all(class_='list_item')
|
|
oldest, newest, oldest_dt, newest_dt = None, None, 9999999, 0
|
|
rc = [(k, re.compile(r'(?i).*?(\d+)\s*%s.*' % v)) for (k, v) in iteritems(
|
|
dict(months='months?', weeks='weeks?', days='days?', hours='hours?', minutes='min'))]
|
|
rc_show = re.compile(r'^namelink_(\d+)$')
|
|
rc_title_clean = re.compile(r'(?i)(?:\s*\((?:19|20)\d{2}\))?$')
|
|
for row in shows:
|
|
try:
|
|
info_tag = row.find('a', id=rc_show)
|
|
if not info_tag:
|
|
continue
|
|
|
|
ids = dict(custom=rc_show.findall(info_tag['id'])[0], name='ne')
|
|
url_path = info_tag['href'].strip()
|
|
|
|
images = {}
|
|
img_uri = None
|
|
img_tag = info_tag.find('img')
|
|
if img_tag and isinstance(img_tag.attrs, dict):
|
|
img_src = img_tag.attrs.get('src').strip()
|
|
img_uri = img_src.startswith('//') and ('https:' + img_src) or img_src
|
|
images = dict(poster=dict(thumb='imagecache?path=browse/thumb/ne&source=%s' % img_uri))
|
|
sickgear.CACHE_IMAGE_URL_LIST.add_url(img_uri)
|
|
|
|
title = info_tag.get_text('strip=True')
|
|
title = rc_title_clean.sub('', title.strip())
|
|
|
|
channel_tag = row.find('span', class_='channel_name')
|
|
|
|
network, date_info, dt = None, None, None
|
|
channel_tag_copy = copy.copy(channel_tag)
|
|
if channel_tag_copy:
|
|
network = channel_tag_copy.a.extract().get_text(strip=True)
|
|
date_info = re.sub(r'^\D+', '', channel_tag_copy.get_text(strip=True))
|
|
if date_info:
|
|
date_info = (date_info, '%s.01.01' % date_info)[4 == len(date_info)]
|
|
|
|
if not started_past and channel_tag:
|
|
tag = [t for t in channel_tag.next_siblings if hasattr(t, 'attrs')
|
|
and 'printed' in ' '.join(t.get('class', ''))]
|
|
if len(tag):
|
|
age_args = {}
|
|
future = re.sub(r'\D+(.*)', r'\1', tag[0].get_text(strip=True))
|
|
for (dim, rcx) in rc:
|
|
value = helpers.try_int(rcx.sub(r'\1', future), None)
|
|
if value:
|
|
age_args.update({dim: value})
|
|
|
|
if age_args:
|
|
dt = datetime.now(timezone.utc)
|
|
if 'months' in age_args and 'days' in age_args:
|
|
age_args['days'] -= 1
|
|
dt += relativedelta(day=1)
|
|
dt += relativedelta(**age_args)
|
|
date_info = dt
|
|
|
|
ord_premiered = 0
|
|
str_premiered = ''
|
|
if date_info:
|
|
ord_premiered, str_premiered, _, oldest_dt, newest_dt, oldest, newest, _, _, _, _ \
|
|
= self.sanitise_dates(date_info, oldest_dt, newest_dt, oldest, newest)
|
|
if started_past:
|
|
# started_past is false for relative future dates, and those can be output.
|
|
# however, it is set true for response data that doesn't contain an accurate date,
|
|
# therefore, a too broad fuzzy date is prevented from UI output.
|
|
str_premiered = ''
|
|
|
|
genres = row.find(class_='genre')
|
|
if genres:
|
|
genres = re.sub(r',(\S)', r', \1', genres.get_text(strip=True)).lower()
|
|
overview = row.find(class_='summary')
|
|
if overview:
|
|
overview = overview.get_text(strip=True)
|
|
|
|
rating = None
|
|
rating_tag = row.find(class_='rating')
|
|
if rating_tag:
|
|
label_tag = rating_tag.find('label')
|
|
if label_tag:
|
|
rating = re.sub(r'.*?width:\s*(\d+).*', r'\1', label_tag.get('style', ''))
|
|
|
|
filtered.append(dict(
|
|
ord_premiered=ord_premiered,
|
|
str_premiered=str_premiered,
|
|
started_past=started_past,
|
|
genres=('No genre yet' if not genres else genres),
|
|
ids=ids,
|
|
images='' if not img_uri else images,
|
|
network=network or None,
|
|
overview=self.clean_overview(overview),
|
|
rating=(rating, 'TBD')[None is rating],
|
|
title=title,
|
|
url_src_db='https://next-episode.net/%s/' % url_path.strip('/'),
|
|
votes=None))
|
|
|
|
except (AttributeError, IndexError, KeyError, TypeError):
|
|
continue
|
|
|
|
kwargs.update(dict(oldest=oldest, newest=newest))
|
|
|
|
kwargs.update(dict(footnote=footnote, use_votes=False))
|
|
|
|
mode = kwargs.get('mode', '')
|
|
if mode:
|
|
func = 'ne_%s' % mode
|
|
if callable(getattr(self, func, None)):
|
|
sickgear.NE_MRU = func
|
|
sickgear.save_config()
|
|
return self.browse_shows(browse_type, browse_title, filtered, **kwargs)
|
|
|
|
# noinspection PyUnusedLocal
|
|
def info_nextepisode(self, ids, show_name):
|
|
|
|
return self.new_show('|'.join(['', '', '', show_name]), use_show_name=True)
|
|
|
|
@staticmethod
|
|
def _make_char_person_list(cur_show_info):
|
|
# type: (TVInfoShow) -> List[Tuple[str, int, str, int]]
|
|
return [(ch.name.replace('"', "'"), r_t, RoleTypes.reverse[r_t], ch.episode_count)
|
|
for r_t in cur_show_info.cast or [] for ch in cur_show_info.cast[r_t] if ch.name]
|
|
|
|
@staticmethod
|
|
def allow_browse_mru(mode_or_mru):
|
|
# Fix an issue where a default view mixed with a deriviative view that requires a param will break the default
|
|
# Disallows default views from using derivative mru's
|
|
return 'person' not in mode_or_mru
|
|
|
|
def tmdb_default(self):
|
|
method = getattr(self, sickgear.TMDB_MRU, None)
|
|
if not callable(method) or not self.allow_browse_mru(sickgear.TMDB_MRU):
|
|
return self.tmdb_upcoming()
|
|
return method()
|
|
|
|
def tmdb_upcoming(self, **kwargs):
|
|
return self.browse_tmdb(
|
|
'Upcoming at TMDB', mode='upcoming', **kwargs)
|
|
|
|
def tmdb_popular(self, **kwargs):
|
|
return self.browse_tmdb(
|
|
'Popular at TMDB', mode='popular', **kwargs)
|
|
|
|
def tmdb_toprated(self, **kwargs):
|
|
return self.browse_tmdb(
|
|
'Top rated at TMDB', mode='toprated', **kwargs)
|
|
|
|
def tmdb_trending_today(self, **kwargs):
|
|
return self.browse_tmdb(
|
|
'Trending today at TMDB', mode='trending_today', **kwargs)
|
|
|
|
def tmdb_trending_week(self, **kwargs):
|
|
return self.browse_tmdb(
|
|
'Trending this week at TMDB', mode='trending_week', **kwargs)
|
|
|
|
def tmdb_person(self, person_tmdb_id=None, **kwargs):
|
|
return self.browse_tmdb(
|
|
'Person at TMDB', mode='person', p_id=person_tmdb_id, **kwargs)
|
|
|
|
def browse_tmdb(self, browse_title, **kwargs):
|
|
|
|
browse_type = 'TMDB'
|
|
mode = kwargs.get('mode', '')
|
|
|
|
footnote = None
|
|
filtered = []
|
|
p_ref = None
|
|
|
|
tvid = TVINFO_TMDB
|
|
tvinfo_config = sickgear.TVInfoAPI(tvid).api_params.copy()
|
|
t = sickgear.TVInfoAPI(tvid).setup(**tvinfo_config) # type: Union[TmdbIndexer, TVInfoBase]
|
|
if 'popular' == mode:
|
|
items = t.get_popular()
|
|
elif 'toprated' == mode:
|
|
items = t.get_top_rated()
|
|
elif 'trending_today' == mode:
|
|
items = t.get_trending()
|
|
elif 'trending_week' == mode:
|
|
items = t.get_trending(time_window='week')
|
|
elif 'person' == mode:
|
|
items = []
|
|
p_item = t.get_person(get_show_credits=True, **kwargs) # type: TVInfoPerson
|
|
if p_item:
|
|
p_ref = f'{TVINFO_TMDB}:{p_item.id}'
|
|
dup = {} # type: Dict[int, TVInfoShow]
|
|
for c in p_item.characters: # type: TVInfoCharacter
|
|
c.ti_show.cast[RoleTypes.ActorMain].append(c)
|
|
if c.ti_show.id not in dup:
|
|
dup[c.ti_show.id] = c.ti_show
|
|
items.append(c.ti_show)
|
|
else:
|
|
dup[c.ti_show.id].cast[RoleTypes.ActorMain].extend(c.ti_show.cast[RoleTypes.ActorMain])
|
|
del dup
|
|
else:
|
|
items = t.discover()
|
|
|
|
oldest, newest, oldest_dt, newest_dt, dedupe = None, None, 9999999, 0, []
|
|
use_networks = False
|
|
parseinfo = dateutil.parser.parserinfo(dayfirst=False, yearfirst=True)
|
|
base_url = sickgear.TVInfoAPI(TVINFO_TMDB).config['show_url']
|
|
for cur_show_info in items:
|
|
if cur_show_info.id in dedupe or not cur_show_info.seriesname:
|
|
continue
|
|
dedupe += [cur_show_info.id]
|
|
|
|
try:
|
|
airtime = cur_show_info.airs_time
|
|
if not airtime or (0, 0) == (airtime.hour, airtime.minute):
|
|
airtime = dateutil.parser.parse('23:59').time()
|
|
try:
|
|
dt = datetime.combine(dateutil.parser.parse(cur_show_info.firstaired, parseinfo).date(), airtime)
|
|
except (BaseException, Exception):
|
|
dt = None
|
|
ord_premiered, str_premiered, started_past, oldest_dt, newest_dt, oldest, newest, _, _, _, _ \
|
|
= self.sanitise_dates(dt, oldest_dt, newest_dt, oldest, newest)
|
|
|
|
image = self._make_cache_image_url(tvid, cur_show_info)
|
|
images = {} if not image else dict(poster=dict(thumb=image))
|
|
|
|
network_name = cur_show_info.network
|
|
cc = 'US'
|
|
if network_name:
|
|
use_networks = True
|
|
cc = cur_show_info.network_country_code or cc
|
|
|
|
language = ((cur_show_info.language and 'jap' in cur_show_info.language.lower())
|
|
and 'jp' or 'en')
|
|
filtered.append(dict(
|
|
ord_premiered=ord_premiered,
|
|
str_premiered=str_premiered,
|
|
started_past=started_past,
|
|
episode_overview=self.clean_overview(cur_show_info),
|
|
episode_season=cur_show_info.season,
|
|
genres=(', '.join(cur_show_info.genre_list)
|
|
or (cur_show_info.genre and (cur_show_info.genre.strip('|').replace('|', ', ')) or '')
|
|
).lower(),
|
|
ids=cur_show_info.ids.__dict__,
|
|
images=images,
|
|
overview=self.clean_overview(cur_show_info),
|
|
title=cur_show_info.seriesname,
|
|
language=language,
|
|
language_img=sickgear.MEMCACHE_FLAG_IMAGES.get(language, False),
|
|
country=cc,
|
|
country_img=sickgear.MEMCACHE_FLAG_IMAGES.get(cc.lower(), False),
|
|
network=network_name,
|
|
url_src_db=base_url % cur_show_info.id,
|
|
rating=0 < (cur_show_info.vote_average or 0) and
|
|
('%.2f' % (cur_show_info.vote_average * 10)).replace('.00', '') or 0,
|
|
votes=('%.2f' % cur_show_info.popularity) or 0,
|
|
))
|
|
if p_ref:
|
|
filtered[-1].update(dict(
|
|
p_name=p_item.name,
|
|
p_ref=p_ref,
|
|
p_chars=self._make_char_person_list(cur_show_info)
|
|
))
|
|
except (BaseException, Exception):
|
|
pass
|
|
kwargs.update(dict(oldest=oldest, newest=newest, use_filter=True, term_vote='Score'))
|
|
|
|
kwargs.update(dict(footnote=footnote, use_networks=use_networks))
|
|
|
|
if mode and self.allow_browse_mru(mode):
|
|
func = 'tmdb_%s' % mode
|
|
if callable(getattr(self, func, None)):
|
|
sickgear.TMDB_MRU = func
|
|
sickgear.save_config()
|
|
return self.browse_shows(browse_type, browse_title, filtered, **kwargs)
|
|
|
|
# noinspection PyUnusedLocal
|
|
def info_tmdb(self, ids, show_name):
|
|
|
|
if not list(filter(lambda tvid_prodid: helpers.find_show_by_id(tvid_prodid), ids.split(' '))):
|
|
return self.new_show('|'.join(['', '', '', ' '.join([ids, show_name])]), use_show_name=True)
|
|
|
|
def trakt_default(self):
|
|
method = getattr(self, sickgear.TRAKT_MRU, None)
|
|
if not callable(method) or not self.allow_browse_mru(sickgear.TMDB_MRU):
|
|
return self.trakt_trending()
|
|
return method()
|
|
|
|
def trakt_anticipated(self):
|
|
|
|
return self.browse_trakt(
|
|
'get_anticipated',
|
|
'Anticipated at Trakt',
|
|
mode='anticipated',
|
|
footnote='Note; Expect default placeholder images in this list'
|
|
)
|
|
|
|
def trakt_newseasons(self):
|
|
|
|
return self.browse_trakt(
|
|
'get_new_seasons',
|
|
'Returning at Trakt',
|
|
mode='returning',
|
|
footnote='Note; Expect default placeholder images in this list')
|
|
|
|
def trakt_newshows(self):
|
|
|
|
return self.browse_trakt(
|
|
'get_new_shows',
|
|
'Brand new at Trakt',
|
|
mode='newshows',
|
|
footnote='Note; Expect default placeholder images in this list')
|
|
|
|
def trakt_popular(self):
|
|
|
|
return self.browse_trakt(
|
|
'get_popular',
|
|
'Popular at Trakt',
|
|
mode='popular')
|
|
|
|
def trakt_trending(self):
|
|
|
|
return self.browse_trakt(
|
|
'get_trending',
|
|
'Trending at Trakt',
|
|
mode='trending',
|
|
footnote='Tip: For more Trakt, use "Show" near the top of this view')
|
|
|
|
def trakt_watched(self, **kwargs):
|
|
|
|
return self.trakt_action('watch', **kwargs)
|
|
|
|
def trakt_played(self, **kwargs):
|
|
|
|
return self.trakt_action('play', **kwargs)
|
|
|
|
def trakt_collected(self, **kwargs):
|
|
|
|
return self.trakt_action('collect', **kwargs)
|
|
|
|
def trakt_action(self, action, **kwargs):
|
|
|
|
cycle, desc, ext = (('month', 'month', ''), ('year', '12 months', '-year'))['year' == kwargs.get('period', '')]
|
|
return self.browse_trakt(
|
|
f'get_most_{action}ed',
|
|
f'Most {action}ed at Trakt during the last {desc}',
|
|
mode=f'{action}ed{ext}', period=f'{cycle}ly')
|
|
|
|
def trakt_recommended(self, **kwargs):
|
|
|
|
if 'add' == kwargs.get('action'):
|
|
return self.redirect('/config/notifications/#tabs-3')
|
|
|
|
account = helpers.try_int(kwargs.get('account'))
|
|
try:
|
|
name = sickgear.TRAKT_ACCOUNTS[account].name
|
|
except KeyError:
|
|
return self.trakt_default()
|
|
return self.browse_trakt(
|
|
'get_recommended_for_account',
|
|
'Recommended for <b class="grey-text">%s</b> by Trakt' % name,
|
|
mode='recommended-%s' % account, account=account, ignore_collected=True, ignore_watchlisted=True)
|
|
|
|
def trakt_watchlist(self, **kwargs):
|
|
|
|
if 'add' == kwargs.get('action'):
|
|
return self.redirect('/config/notifications/#tabs-3')
|
|
|
|
account = helpers.try_int(kwargs.get('account'))
|
|
try:
|
|
name = sickgear.TRAKT_ACCOUNTS[account].name
|
|
except KeyError:
|
|
return self.trakt_default()
|
|
return self.browse_trakt(
|
|
'get_watchlisted_for_account',
|
|
'WatchList for <b class="grey-text">%s</b> by Trakt' % name,
|
|
mode='watchlist-%s' % account, account=account, ignore_collected=True)
|
|
|
|
def get_trakt_data(self, api_method, **kwargs):
|
|
|
|
mode = kwargs.get('mode', '')
|
|
items, filtered = ([], [])
|
|
error_msg = None
|
|
p_item = None
|
|
p_ref = None
|
|
tvid = TVINFO_TRAKT
|
|
tvinfo_config = sickgear.TVInfoAPI(tvid).api_params.copy()
|
|
t = sickgear.TVInfoAPI(tvid).setup(**tvinfo_config) # type: Union[TraktIndexer, TVInfoBase]
|
|
try:
|
|
trakt_func = getattr(t, api_method, None) # type: callable
|
|
if not callable(trakt_func):
|
|
raise TraktException(f'missing api_trakt lib func: ({api_method})')
|
|
|
|
if 'get_anticipated' == api_method:
|
|
items = t.get_anticipated()
|
|
elif 'get_new_seasons' == api_method:
|
|
items = t.get_new_seasons()
|
|
elif 'get_new_shows' == api_method:
|
|
items = t.get_new_shows()
|
|
elif 'get_popular' == api_method:
|
|
items = t.get_popular()
|
|
elif 'get_trending' == api_method:
|
|
items = t.get_trending()
|
|
elif 'get_most_watched' == api_method:
|
|
items = t.get_most_watched(**kwargs)
|
|
elif 'get_most_played' == api_method:
|
|
items = t.get_most_played(**kwargs)
|
|
elif 'get_most_collected' == api_method:
|
|
items = t.get_most_collected(**kwargs)
|
|
elif 'get_recommended_for_account' == api_method:
|
|
items = t.get_recommended_for_account(**kwargs)
|
|
elif 'get_watchlisted_for_account' == api_method:
|
|
items = t.get_watchlisted_for_account(**kwargs)
|
|
if not items:
|
|
error_msg = 'No items in watchlist. Use the "Add to watchlist" button at the Trakt website'
|
|
raise ValueError(error_msg)
|
|
elif 'person' == mode:
|
|
items = []
|
|
p_item = t.get_person(get_show_credits=True, **kwargs) # type: TVInfoPerson
|
|
if p_item:
|
|
p_ref = f'{TVINFO_TRAKT}:{p_item.id}'
|
|
dup = {} # type: Dict[int, TVInfoShow]
|
|
for c in p_item.characters: # type: TVInfoCharacter
|
|
if c.ti_show.id not in dup:
|
|
dup[c.ti_show.id] = c.ti_show
|
|
items.append(c.ti_show)
|
|
del dup
|
|
else:
|
|
items = t.get_trending()
|
|
except TraktAuthException as e:
|
|
logger.warning(f'Pin authorisation needed to connect to Trakt service: {ex(e)}')
|
|
error_msg = 'Unauthorized: Get another pin in the Notifications Trakt settings'
|
|
except TraktException as e:
|
|
logger.warning(f'Could not connect to Trakt service: {ex(e)}')
|
|
except exceptions_helper.ConnectionSkipException as e:
|
|
logger.log('Skipping Trakt because of previous failure: %s' % ex(e))
|
|
except ValueError as e:
|
|
raise e
|
|
except (IndexError, KeyError):
|
|
pass
|
|
|
|
oldest, newest, oldest_dt, newest_dt, dedupe = None, None, 9999999, 0, []
|
|
use_networks = False
|
|
rx_ignore = re.compile(r'''
|
|
((bbc|channel\s*?5.*?|itv)\s*?(drama|documentaries))|bbc\s*?(comedy|music)|music\s*?specials|tedtalks
|
|
''', re.I | re.X)
|
|
for cur_show_info in items:
|
|
if cur_show_info.id in dedupe or not cur_show_info.seriesname:
|
|
continue
|
|
dedupe += [cur_show_info.id]
|
|
network_name = cur_show_info.network
|
|
if network_name:
|
|
use_networks = True
|
|
language = (cur_show_info.language or '').lower()
|
|
language_en = 'en' == language
|
|
country = (cur_show_info.network_country or '').lower()
|
|
country_ok = country in ('uk', 'gb', 'ie', 'ca', 'us', 'au', 'nz', 'za')
|
|
try:
|
|
season = next(iter(cur_show_info))
|
|
if 1 == season and 'returning' == mode:
|
|
# new shows and new seasons have season 1 shows, filter S1 from new seasons list
|
|
continue
|
|
episode_info = cur_show_info[season][next(iter(cur_show_info[season]))]
|
|
except(BaseException, Exception):
|
|
episode_info = TVInfoEpisode()
|
|
|
|
if 'get_person' != api_method and \
|
|
(rx_ignore.search(cur_show_info.seriesname.strip()) or
|
|
not (language_en or country_ok) or
|
|
not (cur_show_info.overview or episode_info.overview)):
|
|
continue
|
|
try:
|
|
ord_premiered, str_premiered, started_past, oldest_dt, newest_dt, oldest, newest, \
|
|
ok_returning, ord_returning, str_returning, return_past \
|
|
= self.sanitise_dates(cur_show_info.firstaired, oldest_dt, newest_dt, oldest, newest, episode_info)
|
|
if 'returning' == mode and not ok_returning:
|
|
continue
|
|
|
|
image = self._make_cache_image_url(tvid, cur_show_info)
|
|
images = {} if not image else dict(poster=dict(thumb=image))
|
|
|
|
filtered.append(dict(
|
|
ord_premiered=ord_premiered,
|
|
str_premiered=str_premiered,
|
|
ord_returning=ord_returning,
|
|
str_returning=str_returning,
|
|
started_past=started_past, # air time not yet available 16.11.2015
|
|
return_past=return_past,
|
|
episode_number=episode_info.episodenumber,
|
|
episode_overview=self.clean_overview(episode_info),
|
|
episode_season=getattr(episode_info.season, 'number', 1),
|
|
genres=(', '.join(['%s' % v for v in cur_show_info.genre_list])),
|
|
ids=cur_show_info.ids.__dict__,
|
|
images=images,
|
|
network=network_name,
|
|
overview=self.clean_overview(cur_show_info),
|
|
rating=0 < (cur_show_info.rating or 0) and
|
|
('%.2f' % (cur_show_info.rating * 10)).replace('.00', '') or 0,
|
|
title=(cur_show_info.seriesname or '').strip(),
|
|
language=language,
|
|
language_img=not language_en and sickgear.MEMCACHE_FLAG_IMAGES.get(language, False),
|
|
country=country,
|
|
country_img=sickgear.MEMCACHE_FLAG_IMAGES.get(country, False),
|
|
url_src_db='https://trakt.tv/shows/%s' % cur_show_info.slug,
|
|
url_tvdb=(
|
|
'' if not (isinstance(cur_show_info.ids.tvdb, integer_types) and 0 < cur_show_info.ids.tvdb)
|
|
else sickgear.TVInfoAPI(TVINFO_TVDB).config['show_url'] % cur_show_info.ids.tvdb),
|
|
votes=cur_show_info.vote_count or '0'
|
|
))
|
|
if p_ref:
|
|
filtered[-1].update(dict(
|
|
p_name=p_item.name,
|
|
p_ref=p_ref,
|
|
p_chars=self._make_char_person_list(cur_show_info)
|
|
))
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
if 'web_ui' in kwargs:
|
|
return filtered, oldest, newest, use_networks, error_msg
|
|
|
|
return filtered, oldest, newest
|
|
|
|
def trakt_person(self, person_trakt_id=None):
|
|
|
|
return self.browse_trakt(
|
|
'get_person',
|
|
'Person at Trakt',
|
|
mode='person',
|
|
footnote='Note; Expect default placeholder images in this list',
|
|
p_id=person_trakt_id
|
|
)
|
|
|
|
def browse_trakt(self, api_method, browse_title, **kwargs):
|
|
|
|
browse_type = 'Trakt'
|
|
mode = kwargs.get('mode', '')
|
|
filtered = []
|
|
|
|
if not sickgear.USE_TRAKT \
|
|
and ('recommended' in mode or 'watchlist' in mode):
|
|
error_msg = 'To browse personal recommendations, enable Trakt.tv in Config/Notifications/Social'
|
|
return self.browse_shows(browse_type, browse_title, filtered, error_msg=error_msg, show_header=1, **kwargs)
|
|
|
|
try:
|
|
filtered, oldest, newest, use_networks, error_msg = self.get_trakt_data(api_method, web_ui=True, **kwargs)
|
|
except (BaseException, Exception):
|
|
error_msg = 'No items in watchlist. Use the "Add to watchlist" button at the Trakt website'
|
|
return self.browse_shows(browse_type, browse_title, filtered, error_msg=error_msg, show_header=1, **kwargs)
|
|
|
|
kwargs.update(dict(oldest=oldest, newest=newest, error_msg=error_msg, use_networks=use_networks))
|
|
|
|
if not any(m in mode for m in ('recommended', 'watchlist', 'person')):
|
|
mode = mode.split('-')
|
|
if mode and self.allow_browse_mru(mode):
|
|
func = 'trakt_%s' % mode[0]
|
|
if callable(getattr(self, func, None)):
|
|
param = '' if 1 == len(mode) or mode[1] not in ['year', 'month', 'week', 'all'] else \
|
|
'?period=' + mode[1]
|
|
sickgear.TRAKT_MRU = '%s%s' % (func, param)
|
|
sickgear.save_config()
|
|
return self.browse_shows(browse_type, browse_title, filtered, **kwargs)
|
|
|
|
def info_trakt(self, ids, show_name):
|
|
|
|
if not list(filter(lambda tvid_prodid: helpers.find_show_by_id(tvid_prodid), ids.split(' '))):
|
|
return self.new_show('|'.join(['', '', '', ' '.join([ids, show_name])]), use_show_name=True)
|
|
|
|
def tvc_default(self):
|
|
method = getattr(self, sickgear.TVC_MRU, None)
|
|
if not callable(method):
|
|
return self.tvc_newshows()
|
|
return method()
|
|
|
|
def tvc_newshows(self, **kwargs):
|
|
return self.browse_tvc(
|
|
'TV-shows-starting-', 'New at TV Calendar', mode='newshows', **kwargs)
|
|
|
|
def tvc_returning(self, **kwargs):
|
|
return self.browse_tvc(
|
|
'TV-shows-starting-', 'Returning at TV Calendar', mode='returning', **kwargs)
|
|
|
|
def tvc_latest(self, **kwargs):
|
|
return self.browse_tvc(
|
|
'recent-additions', 'Latest new at TV Calendar', mode='latest', **kwargs)
|
|
|
|
def browse_tvc(self, url_path, browse_title, **kwargs):
|
|
|
|
browse_type = 'TVCalendar'
|
|
mode = kwargs.get('mode', '')
|
|
|
|
footnote = None
|
|
filtered = []
|
|
today = datetime.today()
|
|
months = ['', 'January', 'February', 'March', 'April', 'May', 'June', 'July',
|
|
'August', 'September', 'October', 'November', 'December']
|
|
this_month = '%s-%s' % (months[today.month], today.strftime('%Y'))
|
|
section = ''
|
|
if mode in ('newshows', 'returning'):
|
|
section = (kwargs.get('page') or this_month)
|
|
url_path += section
|
|
|
|
import browser_ua
|
|
url = 'https://www.pogdesign.co.uk/cat/%s' % url_path
|
|
html = helpers.get_url(url, headers={'User-Agent': browser_ua.get_ua()})
|
|
use_votes = False
|
|
if html:
|
|
prev_month = dt_prev_month = None
|
|
if mode in ('newshows', 'returning'):
|
|
try:
|
|
prev_month = re.findall('(?i)href="/cat/tv-shows-starting-([^"]+)', html)[0]
|
|
dt_prev_month = dateutil.parser.parse('1-%s' % prev_month)
|
|
except (BaseException, Exception):
|
|
prev_month = None
|
|
get_prev_month = (lambda _dt: _dt.replace(day=1) - timedelta(days=1))
|
|
get_next_month = (lambda _dt: _dt.replace(day=28) + timedelta(days=5))
|
|
get_month_year = (lambda _dt: '%s-%s' % (months[_dt.month], _dt.strftime('%Y')))
|
|
if prev_month:
|
|
dt_next_month = get_next_month(dt_prev_month)
|
|
while True:
|
|
next_month = get_month_year(dt_next_month)
|
|
if next_month not in (this_month, kwargs.get('page')):
|
|
break
|
|
dt_next_month = get_next_month(dt_next_month)
|
|
while True:
|
|
if prev_month not in (this_month, kwargs.get('page')):
|
|
break
|
|
dt_prev_month = get_prev_month(dt_prev_month)
|
|
prev_month = get_month_year(dt_prev_month)
|
|
|
|
else:
|
|
dt_next_month = get_next_month(today)
|
|
next_month = get_month_year(dt_next_month)
|
|
dt_prev_month = get_prev_month(today)
|
|
prev_month = get_month_year(dt_prev_month)
|
|
|
|
suppress_item = not kwargs.get('page') or kwargs.get('page') == this_month
|
|
get_date_text = (lambda m_y: m_y.replace('-', ' '))
|
|
next_month_text = get_date_text(next_month)
|
|
prev_month_text = get_date_text(prev_month)
|
|
page_month_text = suppress_item and 'void' or get_date_text(kwargs.get('page'))
|
|
kwargs.update(dict(pages=[
|
|
('tvc_newshows?page=%s' % this_month, 'New this month'),
|
|
('tvc_newshows?page=%s' % next_month, '...in %s' % next_month_text)] +
|
|
([('tvc_newshows?page=%s' % kwargs.get('page'), '...in %s' % page_month_text)], [])[suppress_item] +
|
|
[('tvc_newshows?page=%s' % prev_month, '...in %s' % prev_month_text)] +
|
|
[('tvc_returning?page=%s' % this_month, 'Returning this month'),
|
|
('tvc_returning?page=%s' % next_month, '...in %s' % next_month_text)] +
|
|
([('tvc_returning?page=%s' % kwargs.get('page'), '...in %s' % page_month_text)], [])[suppress_item] +
|
|
[('tvc_returning?page=%s' % prev_month, '...in %s' % prev_month_text)]
|
|
))
|
|
|
|
with BS4Parser(html, parse_only=dict(div={'class': (lambda at: at and 'pgwidth' in at)})) as tbl:
|
|
shows = []
|
|
if mode in ('latest', 'newshows', 'returning'):
|
|
tags = tbl.select('h2[class*="midtitle"], div[class*="contbox"]')
|
|
collect = False
|
|
for cur_tag in tags:
|
|
if re.match(r'(?i)h\d+', cur_tag.name) and 'midtitle' in cur_tag.attrs.get('class', []):
|
|
text = cur_tag.get_text(strip=True)
|
|
if mode in ('latest', 'newshows'):
|
|
if not collect and ('Latest' in text or 'New' in text):
|
|
collect = True
|
|
continue
|
|
break
|
|
elif 'New' in text:
|
|
continue
|
|
elif 'Return' in text:
|
|
collect = True
|
|
continue
|
|
break
|
|
if collect:
|
|
shows += [cur_tag]
|
|
|
|
if not len(shows):
|
|
kwargs['error_msg'] = 'No TV titles found in <a target="_blank" href="%s">%s</a>%s,' \
|
|
' try another selection' % (
|
|
helpers.anon_url(url), browse_title, (' for %s' % section.replace('-', ' '), '')[not section])
|
|
|
|
# build batches to correct htmlentity typos in overview
|
|
from html5lib.constants import entities
|
|
batches = []
|
|
for cur_n, cur_name in enumerate(entities):
|
|
if 0 == cur_n % 150:
|
|
if cur_n:
|
|
batches += [batch]
|
|
batch = []
|
|
batch += [cur_name]
|
|
else:
|
|
batches += [batch]
|
|
|
|
oldest, newest, oldest_dt, newest_dt = None, None, 9999999, 0
|
|
for row in shows:
|
|
try:
|
|
ids = dict(custom=row.select('input[type="checkbox"]')[0].attrs['value'], name='tvc')
|
|
info = row.find('a', href=re.compile('^/cat'))
|
|
url_path = info['href'].strip()
|
|
|
|
title = info.find('h2').get_text(strip=True)
|
|
img_uri = info.get('data-original', '').strip()
|
|
if not img_uri:
|
|
img_uri = re.findall(r'(?i).*?image:\s*url\(([^)]+)', info.attrs['style'])[0].strip()
|
|
images = dict(poster=dict(thumb='imagecache?path=browse/thumb/tvc&source=%s' % img_uri))
|
|
sickgear.CACHE_IMAGE_URL_LIST.add_url(img_uri)
|
|
title = re.sub(r'(?i)(?::\s*season\s*\d+|\s*\((?:19|20)\d{2}\))?$', '', title.strip())
|
|
|
|
ord_premiered = 0
|
|
str_premiered = ''
|
|
ord_returning = 0
|
|
str_returning = ''
|
|
date = genre = network = ''
|
|
date_tag = row.find('span', class_='startz')
|
|
if date_tag:
|
|
date_network = re.split(r'(?i)\son\s', ''.join(
|
|
[t.name and t.get_text() or str(t) for t in date_tag][0:2]))
|
|
date = re.sub('(?i)^(starts|returns)', '', date_network[0]).strip()
|
|
network = ('', date_network[1].strip())[2 == len(date_network)]
|
|
else:
|
|
date_tag = row.find('span', class_='selby')
|
|
if date_tag:
|
|
date = date_tag.get_text(strip=True)
|
|
network_genre = info.find('span')
|
|
if network_genre:
|
|
network_genre = network_genre.get_text(strip=True).split('//')
|
|
network = network_genre[0]
|
|
genre = ('', network_genre[1].strip())[2 == len(network_genre)]
|
|
|
|
started_past = return_past = False
|
|
if date:
|
|
ord_premiered, str_premiered, started_past, oldest_dt, newest_dt, oldest, newest, \
|
|
_, _, _, _ \
|
|
= self.sanitise_dates(date, oldest_dt, newest_dt, oldest, newest)
|
|
if mode in ('returning',):
|
|
ord_returning, str_returning = ord_premiered, str_premiered
|
|
ord_premiered, str_premiered = 0, ''
|
|
|
|
overview = row.find('span', class_='shwtxt')
|
|
if overview:
|
|
overview = re.sub(r'(?sim)(.*?)(?:[.\s]*\*\*NOTE.*?)?(\.{1,3})$', r'\1\2',
|
|
overview.get_text(strip=True))
|
|
for cur_entities in batches:
|
|
overview = re.sub(r'and(%s)' % '|'.join(cur_entities), r'&\1', overview)
|
|
|
|
votes = None
|
|
votes_tag = row.find('span', class_='selby')
|
|
if votes_tag:
|
|
votes_tag = votes_tag.find('strong')
|
|
if votes_tag:
|
|
votes = re.sub(r'(?i)\s*users', '', votes_tag.get_text()).strip()
|
|
use_votes = True
|
|
|
|
filtered.append(dict(
|
|
ord_premiered=ord_premiered,
|
|
str_premiered=str_premiered,
|
|
ord_returning=ord_returning,
|
|
str_returning=str_returning,
|
|
episode_season='',
|
|
started_past=started_past,
|
|
return_past=return_past,
|
|
genres=genre,
|
|
network=network or None,
|
|
ids=ids,
|
|
images='' if not img_uri else images,
|
|
overview=self.clean_overview(overview),
|
|
rating=None,
|
|
title=title,
|
|
url_src_db='https://www.pogdesign.co.uk/%s' % url_path.strip('/'),
|
|
votes=votes or 'n/a'))
|
|
|
|
except (AttributeError, IndexError, KeyError, TypeError):
|
|
continue
|
|
|
|
kwargs.update(dict(oldest=oldest, newest=newest))
|
|
|
|
kwargs.update(dict(footnote=footnote, use_ratings=False, use_votes=use_votes, show_header=True))
|
|
|
|
if mode:
|
|
func = 'tvc_%s' % mode
|
|
if callable(getattr(self, func, None)):
|
|
sickgear.TVC_MRU = func
|
|
sickgear.save_config()
|
|
|
|
return self.browse_shows(browse_type, browse_title, filtered, **kwargs)
|
|
|
|
# noinspection PyUnusedLocal
|
|
def info_tvcalendar(self, ids, show_name):
|
|
|
|
return self.new_show('|'.join(['', '', '', show_name]), use_show_name=True)
|
|
|
|
def tvm_default(self):
|
|
method = getattr(self, sickgear.TVM_MRU, None)
|
|
if not callable(method) or not self.allow_browse_mru(sickgear.TMDB_MRU):
|
|
return self.tvm_premieres()
|
|
return method()
|
|
|
|
def tvm_premieres(self, **kwargs):
|
|
return self.browse_tvm(
|
|
'New at TVmaze', mode='premieres', **kwargs)
|
|
|
|
def tvm_returning(self, **kwargs):
|
|
return self.browse_tvm(
|
|
'Returning at TVmaze', mode='returning', **kwargs)
|
|
|
|
def tvm_person(self, person_tvm_id=None, **kwargs):
|
|
return self.browse_tvm(
|
|
'Person at TVmaze', mode='person', p_id=person_tvm_id, **kwargs)
|
|
|
|
@staticmethod
|
|
def clean_overview(info=None):
|
|
# type (AnyStr, TVInfoShow) -> AnyStr
|
|
text = info if isinstance(info, str) else info.overview
|
|
if text:
|
|
result = helpers.xhtml_escape(re.sub(r'[\r\n]+', ' ', text[:250:])).strip('*').strip()
|
|
result = re.sub(r'([!?.])(?=\w)', r'\1 ', result)
|
|
result = re.sub(r'([,.!][^,.!]*?)$', '...', result)
|
|
return result.replace('.....', '...')
|
|
return 'No overview yet'
|
|
|
|
def tvm_get_showinfo(self, tvid_prodid=None, oldest_dt=9999999, newest_dt=0):
|
|
result = {}
|
|
if 'tvmaze' in tvid_prodid:
|
|
tvid = TVINFO_TVMAZE
|
|
tvinfo_config = sickgear.TVInfoAPI(tvid).api_params.copy()
|
|
t = sickgear.TVInfoAPI(tvid).setup(**tvinfo_config) # type: Union[TvmazeIndexer, TVInfoBase]
|
|
show_info = t.get_show(int(tvid_prodid.replace('tvmaze:','')), load_episodes=False)
|
|
|
|
oldest_dt, newest_dt = int(oldest_dt), int(newest_dt)
|
|
ord_premiered, str_premiered, started_past, old_dt, new_dt, oldest, newest, \
|
|
ok_returning, ord_returning, str_returning, return_past \
|
|
= self.sanitise_dates(show_info.firstaired, oldest_dt, newest_dt, None, None)
|
|
result = dict(
|
|
ord_premiered=ord_premiered,
|
|
str_premiered=str_premiered,
|
|
#ord_returning=ord_returning,
|
|
#str_returning=str_returning,
|
|
started_past=started_past,
|
|
#return_past=return_past,
|
|
genres=((show_info.genre or '')
|
|
or ', '.join(show_info.genre_list)
|
|
or ', '.join(show_info.show_type) or '').strip('|').replace('|', ', ').lower(),
|
|
overview=self.clean_overview(show_info),
|
|
network=show_info.network or ', '.join(show_info.networks) or '',
|
|
)
|
|
if old_dt < oldest_dt:
|
|
result['oldest_dt'] = old_dt
|
|
result['oldest'] = oldest
|
|
elif new_dt > newest_dt:
|
|
result['newest_dt'] = old_dt
|
|
result['newest'] = newest,
|
|
|
|
return json_dumps(result)
|
|
|
|
def browse_tvm(self, browse_title, **kwargs):
|
|
|
|
browse_type = 'TVmaze'
|
|
mode = kwargs.get('mode', '')
|
|
|
|
footnote = None
|
|
filtered = []
|
|
p_ref = None
|
|
|
|
tvid = TVINFO_TVMAZE
|
|
tvinfo_config = sickgear.TVInfoAPI(tvid).api_params.copy()
|
|
t = sickgear.TVInfoAPI(tvid).setup(**tvinfo_config) # type: Union[TvmazeIndexer, TVInfoBase]
|
|
if 'premieres' == mode:
|
|
items = t.get_premieres()
|
|
elif 'person' == mode:
|
|
items = []
|
|
p_item = t.get_person(get_show_credits=True, **kwargs) # type: TVInfoPerson
|
|
if p_item:
|
|
p_ref = f'{TVINFO_TVMAZE}:{p_item.id}'
|
|
dup = {} # type: Dict[int, TVInfoShow]
|
|
for c in p_item.characters: # type: TVInfoCharacter
|
|
c.ti_show.cast[(RoleTypes.ActorGuest, RoleTypes.ActorMain)[True is c.regular]].append(c)
|
|
if c.ti_show.id not in dup:
|
|
dup[c.ti_show.id] = c.ti_show
|
|
items.append(c.ti_show)
|
|
else:
|
|
dup[c.ti_show.id].cast[RoleTypes.ActorMain].extend(c.ti_show.cast[RoleTypes.ActorMain])
|
|
dup[c.ti_show.id].cast[RoleTypes.ActorGuest].extend(c.ti_show.cast[RoleTypes.ActorGuest])
|
|
del dup
|
|
else:
|
|
items = t.get_returning()
|
|
|
|
# handle switching between returning and premieres
|
|
sickgear.BROWSELIST_MRU.setdefault(browse_type, dict())
|
|
if mode in ('premieres', 'returning'):
|
|
showfilter = ('by_returning', 'by_premiered')['premieres' == mode]
|
|
saved_showsort = sickgear.BROWSELIST_MRU[browse_type].get('tvm_%s' % mode) or '*,asc'
|
|
showsort = saved_showsort + (f',{showfilter}', '')[3 == len(saved_showsort.split(','))]
|
|
sickgear.BROWSELIST_MRU[browse_type].update(dict(showfilter=showfilter, showsort=showsort))
|
|
|
|
oldest, newest, oldest_dt, newest_dt, dedupe = None, None, 9999999, 0, []
|
|
use_networks = False
|
|
base_url = sickgear.TVInfoAPI(tvid).config['show_url']
|
|
for cur_show_info in items:
|
|
if cur_show_info.id in dedupe or not cur_show_info.seriesname:
|
|
continue
|
|
dedupe += [cur_show_info.id]
|
|
|
|
try:
|
|
season = next(iter(cur_show_info))
|
|
episode_info = cur_show_info[season][next(iter(cur_show_info[season]))] # type: Optional[TVInfoEpisode]
|
|
except(BaseException, Exception):
|
|
episode_info = TVInfoEpisode()
|
|
|
|
try:
|
|
ord_premiered, str_premiered, started_past, oldest_dt, newest_dt, oldest, newest, \
|
|
ok_returning, ord_returning, str_returning, return_past \
|
|
= self.sanitise_dates(cur_show_info.firstaired, oldest_dt, newest_dt, oldest, newest, episode_info)
|
|
if 'returning' == mode and not ok_returning:
|
|
continue
|
|
|
|
image = self._make_cache_image_url(tvid, cur_show_info, use_source_id='person' == mode)
|
|
images = {} if not image else dict(poster=dict(thumb=image))
|
|
|
|
network_name = cur_show_info.network
|
|
cc = ''
|
|
if network_name:
|
|
use_networks = True
|
|
cc = cur_show_info.network_country_code or cc # ensure string type not None
|
|
country_ok = cc.lower() in ('uk', 'gb', 'ie', 'ca', 'us', 'au', 'nz', 'za')
|
|
country = (country_ok and cc) or (cur_show_info.network_country or '').lower()
|
|
|
|
language = (('jap' in (cur_show_info.language or '').lower()) and 'jp' or 'en')
|
|
|
|
overview = self.clean_overview(cur_show_info)
|
|
overview_ajax = ("No overview yet" == overview
|
|
and p_ref and not bool(cur_show_info.cast[RoleTypes.ActorMain]))
|
|
|
|
filtered.append(dict(
|
|
ord_premiered=ord_premiered,
|
|
str_premiered=str_premiered,
|
|
ord_returning=ord_returning,
|
|
str_returning=str_returning,
|
|
started_past=started_past,
|
|
return_past=return_past,
|
|
episode_number=episode_info.episodenumber or '',
|
|
episode_overview=self.clean_overview(episode_info),
|
|
episode_season=getattr(episode_info.season, 'number', episode_info.seasonnumber),
|
|
genres=((cur_show_info.genre or '')
|
|
or ', '.join(cur_show_info.genre_list)
|
|
or ', '.join(cur_show_info.show_type) or '').strip('|').replace('|', ', ').lower(),
|
|
ids=cur_show_info.ids.__dict__,
|
|
images=images,
|
|
overview_ajax=(0, 1)[overview_ajax],
|
|
overview=overview,
|
|
rating=cur_show_info.rating or cur_show_info.popularity or 0,
|
|
title=cur_show_info.seriesname,
|
|
language=language,
|
|
language_img=sickgear.MEMCACHE_FLAG_IMAGES.get(language, False),
|
|
country=country,
|
|
country_img=sickgear.MEMCACHE_FLAG_IMAGES.get(country.lower(), False),
|
|
network=network_name,
|
|
url_src_db=base_url % cur_show_info.id,
|
|
))
|
|
if p_ref:
|
|
filtered[-1].update(dict(
|
|
p_name=p_item.name or None,
|
|
p_ref=p_ref,
|
|
p_chars=self._make_char_person_list(cur_show_info)
|
|
))
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
kwargs.update(dict(oldest=oldest, newest=newest, oldest_dt=oldest_dt, newest_dt=newest_dt))
|
|
|
|
params = dict(footnote=footnote, use_votes=False, use_networks=use_networks)
|
|
if p_ref:
|
|
params.update(dict(use_ratings=False))
|
|
kwargs.update(params)
|
|
|
|
if mode and self.allow_browse_mru(mode):
|
|
func = 'tvm_%s' % mode
|
|
if callable(getattr(self, func, None)):
|
|
sickgear.TVM_MRU = func
|
|
sickgear.save_config()
|
|
return self.browse_shows(browse_type, browse_title, filtered, **kwargs)
|
|
|
|
# noinspection PyUnusedLocal
|
|
def info_tvmaze(self, ids, show_name):
|
|
|
|
if not list(filter(lambda tvid_prodid: helpers.find_show_by_id(tvid_prodid), ids.split(' '))):
|
|
return self.new_show('|'.join(['', '', '', ' '.join([ids, show_name])]), use_show_name=True)
|
|
|
|
@staticmethod
|
|
def sanitise_dates(date, oldest_dt, newest_dt, oldest, newest, episode_info=None, combine_ep_airtime=False):
|
|
# in case of person search (tvmaze) guest starring entires have only show name/id, no dates
|
|
if None is date:
|
|
return 9, '', True, oldest_dt, newest_dt, oldest, newest, True, 9, 'TBC', False
|
|
parseinfo = dateutil.parser.parserinfo(dayfirst=False, yearfirst=True)
|
|
dt = date if isinstance(date, datetime) else dateutil.parser.parse(date)
|
|
if episode_info:
|
|
airtime = episode_info.airtime \
|
|
or episode_info.timestamp and SGDatetime.from_timestamp(episode_info.timestamp).time()
|
|
if not airtime or (0, 0) == (airtime.hour, airtime.minute):
|
|
airtime = dateutil.parser.parse('23:59').time()
|
|
if combine_ep_airtime:
|
|
dt = datetime.combine(dateutil.parser.parse(date, parseinfo).date(), airtime)
|
|
|
|
ord_premiered = dt.toordinal()
|
|
ord_now = datetime.now().toordinal()
|
|
started_past = ord_premiered < ord_now
|
|
str_premiered = SGDatetime.sbfdate(dt) # an invalid dt becomes '' (e.g. 0202-12-28)
|
|
if str_premiered:
|
|
# to prevent UI issues, this logic only runs from a valid dt
|
|
if ord_premiered < oldest_dt:
|
|
oldest_dt = ord_premiered
|
|
oldest = str_premiered
|
|
if ord_premiered > newest_dt:
|
|
newest_dt = ord_premiered
|
|
newest = str_premiered
|
|
|
|
ok_returning = True
|
|
ord_returning = 9
|
|
str_returning = 'TBC'
|
|
return_past = False
|
|
if episode_info:
|
|
# noinspection PyUnboundLocalVariable
|
|
dt_returning = datetime.combine(dateutil.parser.parse(episode_info.firstaired, parseinfo).date(), airtime)
|
|
|
|
ord_returning = dt_returning.toordinal()
|
|
return_past = ord_returning < ord_now
|
|
|
|
str_returning = SGDatetime.sbfdate(dt_returning)
|
|
if dt.year == dt_returning.year and 1900 < dt.year and 1900 < dt_returning.year:
|
|
# drop first aired year == returning year as most likely to be a new, not a returning show
|
|
ok_returning = False
|
|
|
|
return ord_premiered, str_premiered, started_past, oldest_dt, newest_dt, oldest, newest, \
|
|
ok_returning, ord_returning, str_returning, return_past
|
|
|
|
@staticmethod
|
|
def browse_mru(browse_type, **kwargs):
|
|
save_config = False
|
|
if browse_type in ('AniDB', 'IMDb', 'Metacritic', 'Trakt', 'TVCalendar',
|
|
'TMDB', 'TVmaze', 'Nextepisode'):
|
|
save_config = True
|
|
if browse_type in ('TVmaze',) and kwargs.get('showfilter') and kwargs.get('showsort'):
|
|
sickgear.BROWSELIST_MRU.setdefault(browse_type, dict()) \
|
|
.update({kwargs.get('showfilter'): kwargs.get('showsort')})
|
|
else:
|
|
sickgear.BROWSELIST_MRU[browse_type] = dict(
|
|
showfilter=kwargs.get('showfilter', ''), showsort=kwargs.get('showsort', ''))
|
|
if save_config:
|
|
sickgear.save_config()
|
|
return json_dumps({'success': save_config})
|
|
|
|
@staticmethod
|
|
def show_toggle_hide(ids):
|
|
save_config = False
|
|
for sid in ids.split(' '):
|
|
save_config = True
|
|
if sid in sickgear.BROWSELIST_HIDDEN:
|
|
sickgear.BROWSELIST_HIDDEN.remove(sid)
|
|
else:
|
|
sickgear.BROWSELIST_HIDDEN += [sid]
|
|
if save_config:
|
|
sickgear.save_config()
|
|
return json_dumps({'success': save_config})
|
|
|
|
def browse_shows(self, browse_type, browse_title, shows, **kwargs):
|
|
"""
|
|
Display the new show page which collects a tvdb id, folder, and extra options and
|
|
posts them to add_new_show
|
|
"""
|
|
t = PageTemplate(web_handler=self, file='home_browseShows.tmpl')
|
|
t.submenu = self.home_menu()
|
|
t.browse_type = browse_type
|
|
t.browse_title = browse_title if ('person' != kwargs.get('mode') or not shows) \
|
|
else f'{shows[0].get("p_name", "")} (Person) on {browse_type}'
|
|
t.p_ref = (0 < len(shows) and shows[0].get('p_ref')) or None
|
|
t.saved_showfilter = sickgear.BROWSELIST_MRU.get(browse_type, {}).get('showfilter', '')
|
|
t.saved_showsort = sickgear.BROWSELIST_MRU.get(browse_type, {}).get('showsort', '*,asc,by_order')
|
|
showsort = t.saved_showsort.split(',')
|
|
t.saved_showsort_sortby = 3 == len(showsort) and showsort[2] or 'by_order'
|
|
t.reset_showsort_sortby = ('votes' in t.saved_showsort_sortby and not kwargs.get('use_votes', True)
|
|
or 'rating' in t.saved_showsort_sortby and not kwargs.get('use_ratings', True))
|
|
t.is_showsort_desc = ('desc' == (2 <= len(showsort) and showsort[1] or 'asc')) and not t.reset_showsort_sortby
|
|
t.saved_showsort_view = 1 <= len(showsort) and showsort[0] or '*'
|
|
t.all_shows = []
|
|
t.kwargs = kwargs
|
|
if None is t.kwargs.get('footnote') and kwargs.get('mode', 'nomode') in ('upcoming',):
|
|
t.kwargs['footnote'] = 'Note; Expect default placeholder images in this list'
|
|
known = []
|
|
|
|
t.num_inlibrary = 0
|
|
t.num_hidden = 0
|
|
n_p = NameParser(indexer_lookup=False)
|
|
rc_base = re.compile(r"(?i)^(?:dc|marvel)(?:['s]+\s)?")
|
|
rc_nopost = re.compile(r'(?i)(?:\s*\([^)]+\))?$')
|
|
rc_nopre = re.compile(r'(?i)(?:^\([^)]+\)\s*)?')
|
|
for order, item in enumerate(shows):
|
|
item['order'] = order
|
|
tvid_prodid_list = []
|
|
|
|
# first, process known ids
|
|
for tvid, infosrc_slug in filter(
|
|
lambda tvid_slug: item['ids'].get(tvid_slug[1])
|
|
and not sickgear.TVInfoAPI(tvid_slug[0]).config.get('defunct'),
|
|
map(lambda _tvid: (_tvid, sickgear.TVInfoAPI(_tvid).config['slug']),
|
|
iterkeys(sickgear.TVInfoAPI().all_sources))):
|
|
try:
|
|
src_id = item['ids'][infosrc_slug]
|
|
tvid_prodid_list += ['%s:%s' % (infosrc_slug, src_id)]
|
|
imdb = helpers.parse_imdb_id(src_id)
|
|
if imdb:
|
|
src_id = sg_helpers.try_int(imdb.replace('tt', ''))
|
|
show_obj = helpers.find_show_by_id({tvid: src_id}, no_mapped_ids=False, no_exceptions=True)
|
|
except (BaseException, Exception):
|
|
continue
|
|
if not item.get('indb') and show_obj:
|
|
item['indb'] = sickgear.TVInfoAPI(tvid).config.get('name')
|
|
t.num_inlibrary += 1
|
|
|
|
# then, process custom ids
|
|
if 'custom' in item['ids']:
|
|
base_title = rc_base.sub('', item['title'])
|
|
nopost_title = rc_nopost.sub('', item['title'])
|
|
nopre_title = rc_nopre.sub('', item['title'])
|
|
nopost_base_title = rc_nopost.sub('', base_title)
|
|
nopre_base_title = rc_nopre.sub('', base_title)
|
|
nopost_nopre_base_title = rc_nopost.sub('', nopre_base_title)
|
|
titles = [item['title']]
|
|
titles += ([], [base_title])[base_title not in titles]
|
|
titles += ([], [nopost_title])[nopost_title not in titles]
|
|
titles += ([], [nopre_title])[nopre_title not in titles]
|
|
titles += ([], [nopost_base_title])[nopost_base_title not in titles]
|
|
titles += ([], [nopre_base_title])[nopre_base_title not in titles]
|
|
titles += ([], [nopost_nopre_base_title])[nopost_nopre_base_title not in titles]
|
|
if 'ord_premiered' in item and 1 == item.get('season', -1):
|
|
titles += ['%s.%s' % (_t, dt_date.fromordinal(item['ord_premiered']).year) for _t in titles]
|
|
|
|
tvid_prodid_list += ['%s:%s' % (item['ids']['name'], item['ids']['custom'])]
|
|
for cur_title in titles:
|
|
try:
|
|
_ = n_p.parse('%s.s01e01.mp4' % cur_title)
|
|
item['indb'] = item['ids']['name']
|
|
t.num_inlibrary += 1
|
|
break
|
|
except (InvalidNameException, InvalidShowException):
|
|
pass
|
|
|
|
item['show_id'] = '%s' % ' '.join(tvid_prodid_list)
|
|
|
|
if not item['show_id']:
|
|
if 'tt' in item['ids'].get('imdb', ''):
|
|
item['show_id'] = item['ids']['imdb']
|
|
|
|
if item['ids'].get('custom'):
|
|
item['show_id'] = item['ids']['custom']
|
|
|
|
if item['show_id'] not in known:
|
|
known.append(item['show_id'])
|
|
t.all_shows.append(item)
|
|
|
|
if any(filter(lambda tp: tp in sickgear.BROWSELIST_HIDDEN, tvid_prodid_list)):
|
|
item['hide'] = True
|
|
t.num_hidden += 1
|
|
|
|
def _title(text):
|
|
return ((remove_article(text), text)[sickgear.SORT_ARTICLE]).lower()
|
|
if 'order' not in t.saved_showsort_sortby or t.is_showsort_desc:
|
|
for sort_when, sort_type in (
|
|
('order', lambda _x: _x['order']),
|
|
('name', lambda _x: _title(_x['title'])),
|
|
('premiered', lambda _x: (_x['ord_premiered'], _title(_x['title']))),
|
|
('returning', lambda _x: (_x['ord_returning'], _title(_x['title']))),
|
|
('votes', lambda _x: (helpers.try_int(_x['votes']), _title(_x['title']))),
|
|
('rating', lambda _x: (helpers.try_float(_x['rating']), _title(_x['title']))),
|
|
('rating_votes', lambda _x: (helpers.try_float(_x['rating']), helpers.try_int(_x['votes']),
|
|
_title(_x['title'])))):
|
|
if sort_when in t.saved_showsort_sortby:
|
|
t.all_shows.sort(key=sort_type, reverse=t.is_showsort_desc)
|
|
break
|
|
|
|
return t.respond()
|
|
|
|
def import_shows(self, **kwargs):
|
|
"""
|
|
Prints out the page to add existing shows from a root dir
|
|
"""
|
|
t = PageTemplate(web_handler=self, file='home_addExistingShow.tmpl')
|
|
t.submenu = self.home_menu()
|
|
t.enable_anime_options = False
|
|
t.kwargs = kwargs
|
|
t.multi_parents = helpers.maybe_plural(sickgear.ROOT_DIRS.split('|')[1:]) and 's are' or ' is'
|
|
|
|
return t.respond()
|
|
|
|
def add_new_show(self, root_dir=None, full_show_path=None, which_series=None, provided_tvid=None, tvinfo_lang='en',
|
|
other_shows=None, skip_show=None,
|
|
quality_preset=None, any_qualities=None, best_qualities=None, upgrade_once=None,
|
|
wanted_begin=None, wanted_latest=None, tag=None,
|
|
pause=None, prune=None, default_status=None, scene=None, subs=None, flatten_folders=None,
|
|
anime=None, allowlist=None, blocklist=None,
|
|
return_to=None, cancel_form=None, rename_suggest=None, **kwargs):
|
|
"""
|
|
Receive tvdb id, dir, and other options and create a show from them. If extra show dirs are
|
|
provided then it forwards back to new_show, if not it goes to /home.
|
|
"""
|
|
if None is not return_to:
|
|
tvid, void, prodid, show_name = self.split_extra_show(which_series)
|
|
if bool(helpers.try_int(cancel_form)):
|
|
tvid = tvid or provided_tvid or '0'
|
|
prodid = re.findall(r'tvid_prodid=[^%s]+%s(\d+)' % tuple(2 * [TVidProdid.glue]), return_to)[0]
|
|
return self.redirect(return_to % (tvid, prodid))
|
|
|
|
# grab our list of other dirs if given
|
|
if not other_shows:
|
|
other_shows = []
|
|
elif type(other_shows) != list:
|
|
other_shows = [other_shows]
|
|
|
|
def finish_add_show():
|
|
# if there are no extra shows then go home
|
|
if not other_shows:
|
|
return self.redirect('/home/')
|
|
|
|
# peel off the next one
|
|
next_show_dir = other_shows[0]
|
|
rest_of_show_dirs = other_shows[1:]
|
|
|
|
# go to add the next show
|
|
return self.new_show(next_show_dir, rest_of_show_dirs)
|
|
|
|
# if we're skipping then behave accordingly
|
|
if skip_show:
|
|
return finish_add_show()
|
|
|
|
# sanity check on our inputs
|
|
if (not root_dir and not full_show_path) or not which_series:
|
|
return 'Missing params, no production id or folder:' + repr(which_series) + ' and ' + repr(
|
|
root_dir) + '/' + repr(full_show_path)
|
|
|
|
# figure out what show we're adding and where
|
|
series_pieces = which_series.split('|')
|
|
if (which_series and root_dir) or (which_series and full_show_path and 1 < len(series_pieces)):
|
|
if 4 > len(series_pieces):
|
|
logger.error(f'Unable to add show due to show selection. Not enough arguments: {repr(series_pieces)}')
|
|
ui.notifications.error('Unknown error. Unable to add show due to problem with show selection.')
|
|
return self.redirect('/add-shows/import/')
|
|
|
|
tvid = int(series_pieces[0])
|
|
prodid = int(series_pieces[2])
|
|
show_name = kwargs.get('folder') or series_pieces[3]
|
|
else:
|
|
# if no TV info source was provided use the default one set in General settings
|
|
if not provided_tvid:
|
|
provided_tvid = sickgear.TVINFO_DEFAULT
|
|
|
|
tvid = int(provided_tvid)
|
|
prodid = int(which_series)
|
|
show_name = os.path.basename(os.path.normpath(full_show_path))
|
|
|
|
# use the whole path if it's given, or else append the show name to the root dir to get the full show path
|
|
if full_show_path:
|
|
show_dir = os.path.normpath(full_show_path)
|
|
new_show = False
|
|
else:
|
|
show_dir = helpers.generate_show_dir_name(root_dir, show_name)
|
|
new_show = True
|
|
|
|
# if the dir exists, do 'add existing show'
|
|
if os.path.isdir(show_dir) and not full_show_path:
|
|
ui.notifications.error('Unable to add show', f'Found existing folder: {show_dir}')
|
|
return self.redirect(
|
|
'/add-shows/import?tvid_prodid=%s%s%s&hash_dir=%s%s' %
|
|
(tvid, TVidProdid.glue, prodid, re.sub('[^a-z]', '', sg_helpers.md5_for_text(show_dir)),
|
|
rename_suggest and ('&rename_suggest=%s' % rename_suggest) or ''))
|
|
|
|
# don't create show dir if config says not to
|
|
if sickgear.ADD_SHOWS_WO_DIR:
|
|
logger.log('Skipping initial creation due to config.ini setting (add_shows_wo_dir)')
|
|
else:
|
|
if not helpers.make_dir(show_dir):
|
|
logger.error(f"Unable to add show because can't create folder: {show_dir}")
|
|
ui.notifications.error('Unable to add show', f"Can't create folder: {show_dir}")
|
|
return self.redirect('/home/')
|
|
|
|
helpers.chmod_as_parent(show_dir)
|
|
|
|
# prepare the inputs for passing along
|
|
if not any_qualities:
|
|
any_qualities = []
|
|
if not best_qualities or int(quality_preset):
|
|
best_qualities = []
|
|
if type(any_qualities) != list:
|
|
any_qualities = [any_qualities]
|
|
if type(best_qualities) != list:
|
|
best_qualities = [best_qualities]
|
|
new_quality = Quality.combine_qualities(list(map(int, any_qualities)), list(map(int, best_qualities)))
|
|
upgrade_once = config.checkbox_to_value(upgrade_once)
|
|
|
|
wanted_begin = config.minimax(wanted_begin, 0, -1, 10)
|
|
wanted_latest = config.minimax(wanted_latest, 0, -1, 10)
|
|
prune = config.minimax(prune, 0, 0, 9999)
|
|
|
|
pause = config.checkbox_to_value(pause)
|
|
scene = config.checkbox_to_value(scene)
|
|
subs = config.checkbox_to_value(subs)
|
|
flatten_folders = config.checkbox_to_value(flatten_folders)
|
|
|
|
anime = config.checkbox_to_value(anime)
|
|
if allowlist:
|
|
allowlist = short_group_names(allowlist)
|
|
if blocklist:
|
|
blocklist = short_group_names(blocklist)
|
|
|
|
# add the show
|
|
sickgear.show_queue_scheduler.action.add_show(
|
|
tvid, prodid, show_dir,
|
|
quality=new_quality, upgrade_once=upgrade_once,
|
|
wanted_begin=wanted_begin, wanted_latest=wanted_latest, tag=tag,
|
|
paused=pause, prune=prune, default_status=int(default_status), scene=scene, subtitles=subs,
|
|
flatten_folders=flatten_folders, anime=anime, allowlist=allowlist, blocklist=blocklist,
|
|
show_name=show_name, new_show=new_show, lang=tvinfo_lang
|
|
)
|
|
# ui.notifications.message('Show added', 'Adding the specified show into ' + show_dir)
|
|
|
|
return finish_add_show()
|
|
|
|
@staticmethod
|
|
def split_extra_show(extra_show):
|
|
if not extra_show:
|
|
return 4 * [None]
|
|
extra_show = decode_str(extra_show, errors='replace')
|
|
split_vals = extra_show.split('|')
|
|
tvid = helpers.try_int(split_vals[0], sickgear.TVINFO_DEFAULT)
|
|
show_dir = split_vals[1]
|
|
if 4 > len(split_vals):
|
|
return tvid, show_dir, None, None
|
|
prodid = split_vals[2]
|
|
show_name = '|'.join(split_vals[3:])
|
|
|
|
return tvid, show_dir, prodid, show_name
|
|
|
|
def add_existing_shows(self, shows_to_add=None, prompt_for_settings=None, **kwargs):
|
|
"""
|
|
Receives a dir list and add them. Adds the ones with given TVDB IDs first, then forwards
|
|
along to the new_show page.
|
|
"""
|
|
if kwargs.get('tvid_prodid'):
|
|
try:
|
|
search = '%s:%s' % [(sickgear.TVInfoAPI(c_tvid).config['slug'], c_prodid)
|
|
for c_tvid, c_prodid in [tuple(kwargs.get('tvid_prodid').split(':'))]][0]
|
|
except (BaseException, Exception):
|
|
search = kwargs.get('tvid_prodid', '')
|
|
return self.redirect(
|
|
'/add-shows/find/?show_to_add=%s&use_show_name=True%s' %
|
|
('|'.join(['', '', '', search]),
|
|
'|folder=' in shows_to_add and ('&folder=%s' % shows_to_add.split('|folder=')[-1]) or ''))
|
|
|
|
# grab a list of other shows to add, if provided
|
|
if not shows_to_add:
|
|
shows_to_add = []
|
|
elif type(shows_to_add) != list:
|
|
shows_to_add = [shows_to_add]
|
|
|
|
prompt_for_settings = config.checkbox_to_value(prompt_for_settings)
|
|
|
|
prodid_given = []
|
|
prompt_list = []
|
|
dirs_only = []
|
|
# separate all the ones with production ids
|
|
for cur_dir in shows_to_add:
|
|
if '|' in cur_dir:
|
|
split_vals = cur_dir.split('|')
|
|
if 3 > len(split_vals):
|
|
dirs_only.append(cur_dir)
|
|
if '|' not in cur_dir:
|
|
dirs_only.append(cur_dir)
|
|
else:
|
|
tvid, show_dir, prodid, show_name = self.split_extra_show(cur_dir)
|
|
|
|
if not show_dir or not prodid or not show_name:
|
|
continue
|
|
|
|
src_tvid, src_prodid = [helpers.try_int(x, None) for x in prodid.split(':')]
|
|
if tvid != src_tvid:
|
|
prompt_list.append(cur_dir.replace(prodid, ''))
|
|
continue
|
|
|
|
prodid_given.append((tvid, show_dir, src_prodid, show_name))
|
|
|
|
# if they don't want me to prompt for settings then I can just add all the nfo shows now
|
|
num_added = 0
|
|
for cur_show in prodid_given:
|
|
tvid, show_dir, prodid, show_name = cur_show
|
|
|
|
if None is not tvid and None is not prodid:
|
|
# add the show
|
|
sickgear.show_queue_scheduler.action.add_show(
|
|
tvid, prodid, show_dir,
|
|
quality=sickgear.QUALITY_DEFAULT,
|
|
paused=sickgear.PAUSE_DEFAULT, default_status=sickgear.STATUS_DEFAULT,
|
|
scene=sickgear.SCENE_DEFAULT, subtitles=sickgear.SUBTITLES_DEFAULT,
|
|
flatten_folders=sickgear.FLATTEN_FOLDERS_DEFAULT, anime=sickgear.ANIME_DEFAULT,
|
|
show_name=show_name
|
|
)
|
|
num_added += 1
|
|
|
|
if num_added:
|
|
ui.notifications.message('Shows Added',
|
|
'Automatically added ' + str(num_added) + ' from their existing metadata files')
|
|
|
|
if prompt_list:
|
|
shows_to_add = prompt_list
|
|
prompt_for_settings = True
|
|
# if they want me to prompt for settings then I will just carry on to the new_show page
|
|
if prompt_for_settings and shows_to_add:
|
|
return self.new_show(shows_to_add[0], shows_to_add[1:])
|
|
|
|
# if we're done then go home
|
|
if not dirs_only:
|
|
return self.redirect('/home/')
|
|
|
|
# for the remaining shows we need to prompt for each one, so forward this on to the new_show page
|
|
return self.new_show(dirs_only[0], dirs_only[1:])
|
|
|
|
|
|
class Manage(MainHandler):
|
|
|
|
@staticmethod
|
|
def manage_menu(exclude='n/a'):
|
|
menu = [
|
|
{'title': 'Backlog Overview', 'path': 'manage/backlog-overview/'},
|
|
{'title': 'Search Tasks', 'path': 'manage/search-tasks/'},
|
|
{'title': 'Show Tasks', 'path': 'manage/show-tasks/'},
|
|
{'title': 'Episode Overview', 'path': 'manage/episode-overview/'}, ]
|
|
|
|
if sickgear.USE_SUBTITLES:
|
|
menu.append({'title': 'Subtitles Missed', 'path': 'manage/subtitle-missed/'})
|
|
|
|
if sickgear.USE_FAILED_DOWNLOADS:
|
|
menu.append({'title': 'Failed Downloads', 'path': 'manage/failed-downloads/'})
|
|
|
|
return [x for x in menu if exclude not in x['title']]
|
|
|
|
def index(self):
|
|
t = PageTemplate(web_handler=self, file='manage.tmpl')
|
|
t.submenu = self.manage_menu('Bulk')
|
|
|
|
t.has_any_sports = False
|
|
t.has_any_anime = False
|
|
t.has_any_flat_folders = False
|
|
t.shows = []
|
|
t.shows_no_loc = []
|
|
for cur_show_obj in sorted(sickgear.showList, key=lambda _x: _x.name.lower()):
|
|
t.has_any_sports |= bool(cur_show_obj.sports)
|
|
t.has_any_anime |= bool(cur_show_obj.anime)
|
|
t.has_any_flat_folders |= bool(cur_show_obj.flatten_folders)
|
|
if not cur_show_obj.path:
|
|
t.shows_no_loc += [cur_show_obj]
|
|
else:
|
|
t.shows += [cur_show_obj]
|
|
|
|
return t.respond()
|
|
|
|
def get_status_episodes(self, tvid_prodid, which_status):
|
|
|
|
which_status = helpers.try_int(which_status)
|
|
status_list = ((([which_status],
|
|
Quality.SNATCHED_ANY)[SNATCHED == which_status],
|
|
Quality.DOWNLOADED)[DOWNLOADED == which_status],
|
|
Quality.ARCHIVED)[ARCHIVED == which_status]
|
|
|
|
my_db = db.DBConnection()
|
|
tvid_prodid_list = TVidProdid(tvid_prodid).list
|
|
# noinspection SqlResolve
|
|
sql_result = my_db.select(
|
|
'SELECT season, episode, name, airdate, status, location'
|
|
' FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ? AND season != 0 AND status IN (' + ','.join(
|
|
['?'] * len(status_list)) + ')', tvid_prodid_list + status_list)
|
|
|
|
result = {}
|
|
for cur_result in sql_result:
|
|
if not sickgear.SEARCH_UNAIRED and 1000 > cur_result['airdate']:
|
|
continue
|
|
cur_season = int(cur_result['season'])
|
|
cur_episode = int(cur_result['episode'])
|
|
|
|
if cur_season not in result:
|
|
result[cur_season] = {}
|
|
|
|
cur_quality = Quality.split_composite_status(int(cur_result['status']))[1]
|
|
result[cur_season][cur_episode] = {'name': cur_result['name'],
|
|
'airdateNever': 1000 > int(cur_result['airdate']),
|
|
'qualityCss': Quality.get_quality_css(cur_quality),
|
|
'qualityStr': Quality.qualityStrings[cur_quality],
|
|
'sxe': '%d x %02d' % (cur_season, cur_episode)}
|
|
|
|
if which_status in [SNATCHED, SKIPPED, IGNORED, WANTED]:
|
|
|
|
# noinspection SqlResolve
|
|
sql = 'SELECT action, date' \
|
|
' FROM history' \
|
|
' WHERE indexer = ? AND showid = ?' \
|
|
' AND season = ? AND episode = ? AND action in (%s)' \
|
|
' ORDER BY date DESC' % ','.join([str(q) for q in Quality.DOWNLOADED + Quality.SNATCHED_ANY])
|
|
event_sql_result = my_db.select(sql, tvid_prodid_list + [cur_season, cur_episode])
|
|
d_status, d_qual, s_status, s_quality, age = 5 * (None,)
|
|
if event_sql_result:
|
|
for cur_result_event in event_sql_result:
|
|
if None is d_status and cur_result_event['action'] in Quality.DOWNLOADED:
|
|
d_status, d_qual = Quality.split_composite_status(cur_result_event['action'])
|
|
if None is s_status and cur_result_event['action'] in Quality.SNATCHED_ANY:
|
|
s_status, s_quality = Quality.split_composite_status(cur_result_event['action'])
|
|
aged = ((datetime.now() - datetime.strptime(str(cur_result_event['date']),
|
|
sickgear.history.dateFormat)).total_seconds())
|
|
h = 60 * 60
|
|
d = 24 * h
|
|
days = aged // d
|
|
age = ([], ['%id' % days])[bool(days)]
|
|
hours, mins = 0, 0
|
|
if 7 > days:
|
|
hours = aged % d // h
|
|
mins = aged % d % h // 60
|
|
age = ', '.join(age + ([], ['%ih' % hours])[bool(hours)]
|
|
+ ([], ['%im' % mins])[not bool(days)])
|
|
|
|
if None is not d_status and None is not s_status:
|
|
break
|
|
|
|
undo_from_history, change_to, status = self.recommend_status(
|
|
cur_result['status'], cur_result['location'], d_qual, cur_quality)
|
|
if status:
|
|
result[cur_season][cur_episode]['recommend'] = [('. '.join(
|
|
(['snatched %s ago' % age], [])[None is age]
|
|
+ ([], ['file %sfound' % ('not ', '')[bool(cur_result['location'])]])[
|
|
None is d_status or not undo_from_history]
|
|
+ ['%s to <b>%s</b> ?' % (('undo from history',
|
|
'change')[None is d_status or not undo_from_history], change_to)])),
|
|
status]
|
|
|
|
return json_dumps(result)
|
|
|
|
@staticmethod
|
|
def recommend_status(cur_status, location=None, d_qual=None, cur_quality=None):
|
|
|
|
undo_from_history = False
|
|
change_to = ''
|
|
status = None
|
|
if Quality.NONE == cur_quality:
|
|
return undo_from_history, change_to, status
|
|
|
|
cur_status = Quality.split_composite_status(int(cur_status))[0]
|
|
if any([location]):
|
|
undo_from_history = True
|
|
change_to = statusStrings[DOWNLOADED]
|
|
status = [Quality.composite_status(DOWNLOADED, d_qual or cur_quality)]
|
|
elif cur_status in Quality.SNATCHED_ANY + [IGNORED, SKIPPED, WANTED]:
|
|
if None is d_qual:
|
|
if cur_status not in [IGNORED, SKIPPED]:
|
|
change_to = statusStrings[SKIPPED]
|
|
status = [SKIPPED]
|
|
else:
|
|
# downloaded and removed
|
|
if cur_status in Quality.SNATCHED_ANY + [WANTED] \
|
|
or sickgear.SKIP_REMOVED_FILES in [ARCHIVED, IGNORED, SKIPPED]:
|
|
undo_from_history = True
|
|
change_to = '%s %s' % (statusStrings[ARCHIVED], Quality.qualityStrings[d_qual])
|
|
status = [Quality.composite_status(ARCHIVED, d_qual)]
|
|
elif sickgear.SKIP_REMOVED_FILES in [IGNORED, SKIPPED] \
|
|
and cur_status not in [IGNORED, SKIPPED]:
|
|
change_to = statusStrings[statusStrings[sickgear.SKIP_REMOVED_FILES]]
|
|
status = [sickgear.SKIP_REMOVED_FILES]
|
|
|
|
return undo_from_history, change_to, status
|
|
|
|
def episode_overview(self, which_status=None):
|
|
|
|
which_status = helpers.try_int(which_status)
|
|
if which_status:
|
|
status_list = ((([which_status],
|
|
Quality.SNATCHED_ANY)[SNATCHED == which_status],
|
|
Quality.DOWNLOADED)[DOWNLOADED == which_status],
|
|
Quality.ARCHIVED)[ARCHIVED == which_status]
|
|
else:
|
|
status_list = []
|
|
|
|
t = PageTemplate(web_handler=self, file='manage_episodeStatuses.tmpl')
|
|
t.submenu = self.manage_menu('Episode')
|
|
t.which_status = which_status
|
|
|
|
my_db = db.DBConnection()
|
|
sql_result = my_db.select(
|
|
'SELECT COUNT(*) AS snatched FROM [tv_episodes] WHERE season > 0 AND episode > 0 AND airdate > 1 AND ' +
|
|
'status IN (%s)' % ','.join([str(quality) for quality in Quality.SNATCHED_ANY]))
|
|
t.default_manage = sql_result and sql_result[0]['snatched'] and SNATCHED or WANTED
|
|
|
|
# if we have no status then this is as far as we need to go
|
|
if not status_list:
|
|
return t.respond()
|
|
|
|
# noinspection SqlResolve
|
|
status_results = my_db.select(
|
|
'SELECT show_name, tv_shows.indexer AS tvid, tv_shows.indexer_id AS prod_id, airdate'
|
|
' FROM tv_episodes, tv_shows'
|
|
' WHERE tv_episodes.status IN (' + ','.join(['?'] * len(status_list)) +
|
|
') AND season != 0'
|
|
' AND tv_episodes.indexer = tv_shows.indexer AND tv_episodes.showid = tv_shows.indexer_id'
|
|
' ORDER BY show_name COLLATE NOCASE',
|
|
status_list)
|
|
|
|
ep_counts = {}
|
|
ep_count = 0
|
|
never_counts = {}
|
|
show_names = {}
|
|
sorted_show_ids = []
|
|
for cur_status_result in status_results:
|
|
if not sickgear.SEARCH_UNAIRED and 1000 > cur_status_result['airdate']:
|
|
continue
|
|
tvid_prodid = TVidProdid({cur_status_result['tvid']: cur_status_result['prod_id']})()
|
|
if tvid_prodid not in ep_counts:
|
|
ep_counts[tvid_prodid] = 1
|
|
else:
|
|
ep_counts[tvid_prodid] += 1
|
|
ep_count += 1
|
|
if tvid_prodid not in never_counts:
|
|
never_counts[tvid_prodid] = 0
|
|
if 1000 > int(cur_status_result['airdate']):
|
|
never_counts[tvid_prodid] += 1
|
|
|
|
show_names[tvid_prodid] = cur_status_result['show_name']
|
|
if tvid_prodid not in sorted_show_ids:
|
|
sorted_show_ids.append(tvid_prodid)
|
|
|
|
t.show_names = show_names
|
|
t.ep_counts = ep_counts
|
|
t.ep_count = ep_count
|
|
t.never_counts = never_counts
|
|
t.sorted_show_ids = sorted_show_ids
|
|
return t.respond()
|
|
|
|
def change_episode_statuses(self, old_status, new_status, wanted_status=sickgear.common.UNKNOWN, **kwargs):
|
|
status = int(old_status)
|
|
status_list = ((([status],
|
|
Quality.SNATCHED_ANY)[SNATCHED == status],
|
|
Quality.DOWNLOADED)[DOWNLOADED == status],
|
|
Quality.ARCHIVED)[ARCHIVED == status]
|
|
|
|
changes, new_status = self.status_changes(new_status, wanted_status, **kwargs)
|
|
|
|
my_db = None if not any(changes) else db.DBConnection()
|
|
for tvid_prodid, c_what_to in iteritems(changes):
|
|
tvid_prodid_list = TVidProdid(tvid_prodid).list
|
|
for what, to in iteritems(c_what_to):
|
|
if 'all' == what:
|
|
sql_result = my_db.select(
|
|
'SELECT season, episode'
|
|
' FROM tv_episodes'
|
|
' WHERE status IN (%s)' % ','.join(['?'] * len(status_list)) +
|
|
' AND season != 0'
|
|
' AND indexer = ? AND showid = ?',
|
|
status_list + tvid_prodid_list)
|
|
what = (sql_result and '|'.join(map(lambda r: '%sx%s' % (r['season'], r['episode']), sql_result))
|
|
or None)
|
|
to = new_status
|
|
|
|
Home(self.application, self.request).set_show_status(tvid_prodid, what, to, direct=True)
|
|
|
|
self.redirect('/manage/episode-overview/')
|
|
|
|
@staticmethod
|
|
def status_changes(new_status, wanted_status=sickgear.common.UNKNOWN, **kwargs):
|
|
|
|
# make a list of all shows and their associated args
|
|
to_change = {}
|
|
for arg in kwargs:
|
|
# only work with checked checkboxes
|
|
if kwargs[arg] == 'on':
|
|
|
|
tvid_prodid, _, what = arg.partition('-')
|
|
what, _, to = what.partition('-')
|
|
to = (to, new_status)[not to]
|
|
if 'recommended' != to:
|
|
to_change.setdefault(tvid_prodid, dict())
|
|
to_change[tvid_prodid].setdefault(to, [])
|
|
to_change[tvid_prodid][to] += [what]
|
|
|
|
wanted_status = int(wanted_status)
|
|
if wanted_status in (FAILED, WANTED):
|
|
new_status = wanted_status
|
|
|
|
changes = {}
|
|
for tvid_prodid, to_what in iteritems(to_change):
|
|
changes.setdefault(tvid_prodid, dict())
|
|
all_to = None
|
|
for to, what in iteritems(to_what):
|
|
if 'all' in what:
|
|
all_to = to
|
|
continue
|
|
changes[tvid_prodid].update({'|'.join(sorted(what)): (new_status, to)['recommended' == new_status]})
|
|
if None is not all_to and not any(changes[tvid_prodid]):
|
|
if 'recommended' == new_status:
|
|
del (changes[tvid_prodid])
|
|
else:
|
|
changes[tvid_prodid] = {'all': all_to}
|
|
|
|
return changes, new_status
|
|
|
|
@staticmethod
|
|
def show_subtitle_missed(tvid_prodid, which_subs):
|
|
|
|
my_db = db.DBConnection()
|
|
# noinspection SqlResolve
|
|
sql_result = my_db.select(
|
|
'SELECT season, episode, name, subtitles'
|
|
' FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' AND season != 0 AND status LIKE "%4"',
|
|
TVidProdid(tvid_prodid).list)
|
|
|
|
result = {}
|
|
for cur_result in sql_result:
|
|
if 'all' == which_subs:
|
|
if len(set(cur_result['subtitles'].split(',')).intersection(set(subtitles.wanted_languages()))) >= len(
|
|
subtitles.wanted_languages()):
|
|
continue
|
|
elif which_subs in cur_result['subtitles'].split(','):
|
|
continue
|
|
|
|
cur_season = '{0:02d}'.format(cur_result['season'])
|
|
cur_episode = '{0:02d}'.format(cur_result['episode'])
|
|
|
|
if cur_season not in result:
|
|
result[cur_season] = {}
|
|
|
|
if cur_episode not in result[cur_season]:
|
|
result[cur_season][cur_episode] = {}
|
|
|
|
result[cur_season][cur_episode]['name'] = cur_result['name']
|
|
|
|
result[cur_season][cur_episode]['subtitles'] = ','.join([
|
|
subliminal.language.Language(subtitle, strict=False).alpha2
|
|
for subtitle in cur_result['subtitles'].split(',')]) if '' != cur_result['subtitles'] else ''
|
|
|
|
return json_dumps(result)
|
|
|
|
def subtitle_missed(self, which_subs=None):
|
|
|
|
t = PageTemplate(web_handler=self, file='manage_subtitleMissed.tmpl')
|
|
t.submenu = self.manage_menu('Subtitle')
|
|
t.which_subs = which_subs
|
|
|
|
if not which_subs:
|
|
return t.respond()
|
|
|
|
my_db = db.DBConnection()
|
|
# noinspection SqlResolve
|
|
sql_result = my_db.select(
|
|
'SELECT tv_episodes.subtitles as subtitles, show_name,'
|
|
' tv_shows.indexer AS tv_id, tv_shows.indexer_id AS prod_id'
|
|
' FROM tv_episodes, tv_shows'
|
|
' WHERE tv_shows.subtitles = 1'
|
|
' AND tv_episodes.status LIKE "%4" AND tv_episodes.season != 0'
|
|
' AND tv_shows.indexer = tv_episodes.indexer AND tv_episodes.showid = tv_shows.indexer_id'
|
|
' ORDER BY show_name')
|
|
|
|
ep_counts = {}
|
|
show_names = {}
|
|
sorted_show_ids = []
|
|
for cur_result in sql_result:
|
|
if 'all' == which_subs:
|
|
if len(set(cur_result['subtitles'].split(',')).intersection(
|
|
set(subtitles.wanted_languages()))) >= len(subtitles.wanted_languages()):
|
|
continue
|
|
elif which_subs in cur_result['subtitles'].split(','):
|
|
continue
|
|
|
|
tvid_prodid = TVidProdid({cur_result['tv_id']: cur_result['prod_id']})()
|
|
if tvid_prodid not in ep_counts:
|
|
ep_counts[tvid_prodid] = 1
|
|
else:
|
|
ep_counts[tvid_prodid] += 1
|
|
|
|
show_names[tvid_prodid] = cur_result['show_name']
|
|
if tvid_prodid not in sorted_show_ids:
|
|
sorted_show_ids.append(tvid_prodid)
|
|
|
|
t.show_names = show_names
|
|
t.ep_counts = ep_counts
|
|
t.sorted_show_ids = sorted_show_ids
|
|
return t.respond()
|
|
|
|
def download_subtitle_missed(self, **kwargs):
|
|
|
|
if sickgear.USE_SUBTITLES:
|
|
to_download = {}
|
|
|
|
# make a list of all shows and their associated args
|
|
for arg in kwargs:
|
|
tvid_prodid, what = arg.split('-')
|
|
|
|
# we don't care about unchecked checkboxes
|
|
if kwargs[arg] != 'on':
|
|
continue
|
|
|
|
if tvid_prodid not in to_download:
|
|
to_download[tvid_prodid] = []
|
|
|
|
to_download[tvid_prodid].append(what)
|
|
|
|
for cur_tvid_prodid in to_download:
|
|
# get a list of all the eps we want to download subtitles if 'all' is selected
|
|
if 'all' in to_download[cur_tvid_prodid]:
|
|
my_db = db.DBConnection()
|
|
sql_result = my_db.select(
|
|
'SELECT season, episode'
|
|
' FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' AND season != 0 AND status LIKE \'%4\'',
|
|
TVidProdid(cur_tvid_prodid).list)
|
|
to_download[cur_tvid_prodid] = list(map(lambda x: '%sx%s' % (x['season'], x['episode']),
|
|
sql_result))
|
|
|
|
for epResult in to_download[cur_tvid_prodid]:
|
|
season, episode = epResult.split('x')
|
|
|
|
show_obj = helpers.find_show_by_id(cur_tvid_prodid)
|
|
_ = show_obj.get_episode(int(season), int(episode)).download_subtitles()
|
|
|
|
self.redirect('/manage/subtitle-missed/')
|
|
|
|
def backlog_show(self, tvid_prodid):
|
|
|
|
show_obj = helpers.find_show_by_id(tvid_prodid)
|
|
|
|
if show_obj:
|
|
sickgear.search_backlog_scheduler.action.search_backlog([show_obj])
|
|
|
|
self.redirect('/manage/backlog-overview/')
|
|
|
|
def backlog_overview(self):
|
|
|
|
t = PageTemplate(web_handler=self, file='manage_backlogOverview.tmpl')
|
|
t.submenu = self.manage_menu('Backlog')
|
|
|
|
show_counts = {}
|
|
show_cats = {}
|
|
t.ep_sql_results = {}
|
|
|
|
my_db = db.DBConnection(row_type='dict')
|
|
sql_cmds = []
|
|
show_objects = []
|
|
for cur_show_obj in sickgear.showList:
|
|
sql_cmds.append([
|
|
'SELECT season, episode, status, airdate, name'
|
|
' FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' ORDER BY season DESC, episode DESC',
|
|
[cur_show_obj.tvid, cur_show_obj.prodid]])
|
|
show_objects.append(cur_show_obj)
|
|
|
|
sql_results = my_db.mass_action(sql_cmds)
|
|
|
|
for i, sql_result in enumerate(sql_results):
|
|
ep_cats = {}
|
|
ep_counts = {
|
|
Overview.UNAIRED: 0, Overview.GOOD: 0, Overview.SKIPPED: 0,
|
|
Overview.WANTED: 0, Overview.QUAL: 0, Overview.SNATCHED: 0}
|
|
|
|
for cur_result in sql_result:
|
|
if not sickgear.SEARCH_UNAIRED and 1 == cur_result['airdate']:
|
|
continue
|
|
ep_cat = show_objects[i].get_overview(int(cur_result['status']), split_snatch=True)
|
|
if ep_cat in (Overview.WANTED, Overview.QUAL, Overview.SNATCHED_QUAL):
|
|
cur_result['backlog'] = True
|
|
if Overview.SNATCHED_QUAL == ep_cat:
|
|
ep_cat = Overview.SNATCHED
|
|
else:
|
|
cur_result['backlog'] = False
|
|
if ep_cat:
|
|
ep_cats['%sx%s' % (cur_result['season'], cur_result['episode'])] = ep_cat
|
|
ep_counts[ep_cat] += 1
|
|
|
|
tvid_prodid = show_objects[i].tvid_prodid
|
|
show_counts[tvid_prodid] = ep_counts
|
|
show_cats[tvid_prodid] = ep_cats
|
|
t.ep_sql_results[tvid_prodid] = sql_result
|
|
|
|
t.show_counts = show_counts
|
|
t.show_cats = show_cats
|
|
t.backlog_active_providers = sickgear.search_backlog.BacklogSearcher.providers_active(scheduled=False)
|
|
|
|
return t.respond()
|
|
|
|
def mass_edit(self, to_edit=None):
|
|
|
|
t = PageTemplate(web_handler=self, file='manage_massEdit.tmpl')
|
|
t.submenu = self.manage_menu()
|
|
|
|
if not to_edit:
|
|
return self.redirect('/manage/')
|
|
|
|
show_ids = to_edit.split('|')
|
|
show_list = []
|
|
for cur_tvid_prodid in show_ids:
|
|
show_obj = helpers.find_show_by_id(cur_tvid_prodid)
|
|
if show_obj:
|
|
show_list.append(show_obj)
|
|
|
|
upgrade_once_all_same = True
|
|
last_upgrade_once = None
|
|
|
|
flatten_folders_all_same = True
|
|
last_flatten_folders = None
|
|
|
|
paused_all_same = True
|
|
last_paused = None
|
|
|
|
prune_all_same = True
|
|
last_prune = None
|
|
|
|
tag_all_same = True
|
|
last_tag = None
|
|
|
|
anime_all_same = True
|
|
last_anime = None
|
|
|
|
sports_all_same = True
|
|
last_sports = None
|
|
|
|
quality_all_same = True
|
|
last_quality = None
|
|
|
|
subtitles_all_same = True
|
|
last_subtitles = None
|
|
|
|
scene_all_same = True
|
|
last_scene = None
|
|
|
|
air_by_date_all_same = True
|
|
last_air_by_date = None
|
|
|
|
tvid_all_same = True
|
|
last_tvid = None
|
|
|
|
root_dir_list = []
|
|
|
|
for cur_show_obj in show_list:
|
|
|
|
# noinspection PyProtectedMember
|
|
cur_root_dir = os.path.dirname(cur_show_obj._location)
|
|
if cur_root_dir not in root_dir_list:
|
|
root_dir_list.append(cur_root_dir)
|
|
|
|
if upgrade_once_all_same:
|
|
# if we had a value already and this value is different, then they're not all the same
|
|
if last_upgrade_once not in (None, cur_show_obj.upgrade_once):
|
|
upgrade_once_all_same = False
|
|
else:
|
|
last_upgrade_once = cur_show_obj.upgrade_once
|
|
|
|
# if we know they're not all the same then no point even bothering
|
|
if paused_all_same:
|
|
# if we had a value already and this value is different, then they're not all the same
|
|
if last_paused not in (None, cur_show_obj.paused):
|
|
paused_all_same = False
|
|
else:
|
|
last_paused = cur_show_obj.paused
|
|
|
|
if prune_all_same:
|
|
# if we had a value already and this value is different, then they're not all the same
|
|
if last_prune not in (None, cur_show_obj.prune):
|
|
prune_all_same = False
|
|
else:
|
|
last_prune = cur_show_obj.prune
|
|
|
|
if tag_all_same:
|
|
# if we had a value already and this value is different, then they're not all the same
|
|
if last_tag not in (None, cur_show_obj.tag):
|
|
tag_all_same = False
|
|
else:
|
|
last_tag = cur_show_obj.tag
|
|
|
|
if anime_all_same:
|
|
# if we had a value already and this value is different, then they're not all the same
|
|
if last_anime not in (None, cur_show_obj.is_anime):
|
|
anime_all_same = False
|
|
else:
|
|
last_anime = cur_show_obj.anime
|
|
|
|
if flatten_folders_all_same:
|
|
if last_flatten_folders not in (None, cur_show_obj.flatten_folders):
|
|
flatten_folders_all_same = False
|
|
else:
|
|
last_flatten_folders = cur_show_obj.flatten_folders
|
|
|
|
if quality_all_same:
|
|
if last_quality not in (None, cur_show_obj.quality):
|
|
quality_all_same = False
|
|
else:
|
|
last_quality = cur_show_obj.quality
|
|
|
|
if subtitles_all_same:
|
|
if last_subtitles not in (None, cur_show_obj.subtitles):
|
|
subtitles_all_same = False
|
|
else:
|
|
last_subtitles = cur_show_obj.subtitles
|
|
|
|
if scene_all_same:
|
|
if last_scene not in (None, cur_show_obj.scene):
|
|
scene_all_same = False
|
|
else:
|
|
last_scene = cur_show_obj.scene
|
|
|
|
if sports_all_same:
|
|
if last_sports not in (None, cur_show_obj.sports):
|
|
sports_all_same = False
|
|
else:
|
|
last_sports = cur_show_obj.sports
|
|
|
|
if air_by_date_all_same:
|
|
if last_air_by_date not in (None, cur_show_obj.air_by_date):
|
|
air_by_date_all_same = False
|
|
else:
|
|
last_air_by_date = cur_show_obj.air_by_date
|
|
|
|
if tvid_all_same:
|
|
if last_tvid not in (None, cur_show_obj.tvid):
|
|
tvid_all_same = False
|
|
else:
|
|
last_tvid = cur_show_obj.tvid
|
|
|
|
t.showList = to_edit
|
|
t.upgrade_once_value = last_upgrade_once if upgrade_once_all_same else None
|
|
t.paused_value = last_paused if paused_all_same else None
|
|
t.prune_value = last_prune if prune_all_same else None
|
|
t.tag_value = last_tag if tag_all_same else None
|
|
t.anime_value = last_anime if anime_all_same else None
|
|
t.flatten_folders_value = last_flatten_folders if flatten_folders_all_same else None
|
|
t.quality_value = last_quality if quality_all_same else None
|
|
t.subtitles_value = last_subtitles if subtitles_all_same else None
|
|
t.scene_value = last_scene if scene_all_same else None
|
|
t.sports_value = last_sports if sports_all_same else None
|
|
t.air_by_date_value = last_air_by_date if air_by_date_all_same else None
|
|
t.tvid_value = last_tvid if tvid_all_same else None
|
|
t.root_dir_list = root_dir_list
|
|
|
|
return t.respond()
|
|
|
|
def mass_edit_submit(self, to_edit=None, upgrade_once=None, paused=None, anime=None, sports=None, scene=None,
|
|
flatten_folders=None, quality_preset=False, subs=None, air_by_date=None, any_qualities=None,
|
|
best_qualities=None, prune=None, tag=None, tvid=None, **kwargs):
|
|
|
|
any_qualities = any_qualities if None is not any_qualities else []
|
|
best_qualities = best_qualities if None is not best_qualities else []
|
|
|
|
dir_map = {}
|
|
for cur_arg in kwargs:
|
|
if not cur_arg.startswith('orig_root_dir_'):
|
|
continue
|
|
which_index = cur_arg.replace('orig_root_dir_', '')
|
|
end_dir = kwargs['new_root_dir_' + which_index]
|
|
dir_map[kwargs[cur_arg]] = end_dir
|
|
|
|
switch_tvid = []
|
|
tvid = sg_helpers.try_int(tvid, tvid)
|
|
|
|
show_ids = to_edit.split('|')
|
|
errors = []
|
|
for cur_tvid_prodid in show_ids:
|
|
cur_errors = []
|
|
show_obj = helpers.find_show_by_id(cur_tvid_prodid)
|
|
if not show_obj:
|
|
continue
|
|
|
|
# noinspection PyProtectedMember
|
|
cur_root_dir = os.path.dirname(show_obj._location)
|
|
# noinspection PyProtectedMember
|
|
cur_show_dir = os.path.basename(show_obj._location)
|
|
if cur_root_dir in dir_map and cur_root_dir != dir_map[cur_root_dir]:
|
|
new_show_dir = os.path.join(dir_map[cur_root_dir], cur_show_dir)
|
|
if 'nt' != os.name and ':\\' in cur_show_dir:
|
|
# noinspection PyProtectedMember
|
|
cur_show_dir = show_obj._location.split('\\')[-1]
|
|
try:
|
|
base_dir = dir_map[cur_root_dir].rsplit(cur_show_dir)[0].rstrip('/')
|
|
except IndexError:
|
|
base_dir = dir_map[cur_root_dir]
|
|
new_show_dir = os.path.join(base_dir, cur_show_dir)
|
|
# noinspection PyProtectedMember
|
|
logger.log(f'For show {show_obj.unique_name} changing dir from {show_obj._location} to {new_show_dir}')
|
|
else:
|
|
# noinspection PyProtectedMember
|
|
new_show_dir = show_obj._location
|
|
|
|
if 'keep' == upgrade_once:
|
|
new_upgrade_once = show_obj.upgrade_once
|
|
else:
|
|
new_upgrade_once = True if 'enable' == upgrade_once else False
|
|
new_upgrade_once = 'on' if new_upgrade_once else 'off'
|
|
|
|
if 'keep' == paused:
|
|
new_paused = show_obj.paused
|
|
else:
|
|
new_paused = True if 'enable' == paused else False
|
|
new_paused = 'on' if new_paused else 'off'
|
|
|
|
new_prune = (config.minimax(prune, 0, 0, 9999), show_obj.prune)[prune in (None, '', 'keep')]
|
|
|
|
if 'keep' == tag:
|
|
new_tag = show_obj.tag
|
|
else:
|
|
new_tag = tag
|
|
|
|
if 'keep' != tvid and tvid != show_obj.tvid:
|
|
switch_tvid += ['%s-%s' % (cur_tvid_prodid, tvid)]
|
|
|
|
if 'keep' == anime:
|
|
new_anime = show_obj.anime
|
|
else:
|
|
new_anime = True if 'enable' == anime else False
|
|
new_anime = 'on' if new_anime else 'off'
|
|
|
|
if 'keep' == sports:
|
|
new_sports = show_obj.sports
|
|
else:
|
|
new_sports = True if 'enable' == sports else False
|
|
new_sports = 'on' if new_sports else 'off'
|
|
|
|
if 'keep' == scene:
|
|
new_scene = show_obj.is_scene
|
|
else:
|
|
new_scene = True if 'enable' == scene else False
|
|
new_scene = 'on' if new_scene else 'off'
|
|
|
|
if 'keep' == air_by_date:
|
|
new_air_by_date = show_obj.air_by_date
|
|
else:
|
|
new_air_by_date = True if 'enable' == air_by_date else False
|
|
new_air_by_date = 'on' if new_air_by_date else 'off'
|
|
|
|
if 'keep' == flatten_folders:
|
|
new_flatten_folders = show_obj.flatten_folders
|
|
else:
|
|
new_flatten_folders = True if 'enable' == flatten_folders else False
|
|
new_flatten_folders = 'on' if new_flatten_folders else 'off'
|
|
|
|
if 'keep' == subs:
|
|
new_subtitles = show_obj.subtitles
|
|
else:
|
|
new_subtitles = True if 'enable' == subs else False
|
|
|
|
new_subtitles = 'on' if new_subtitles else 'off'
|
|
|
|
if 'keep' == quality_preset:
|
|
any_qualities, best_qualities = Quality.split_quality(show_obj.quality)
|
|
elif int(quality_preset):
|
|
best_qualities = []
|
|
|
|
exceptions_list = []
|
|
|
|
cur_errors += Home(self.application, self.request).edit_show(
|
|
tvid_prodid=cur_tvid_prodid, location=new_show_dir,
|
|
any_qualities=any_qualities, best_qualities=best_qualities, exceptions_list=exceptions_list,
|
|
upgrade_once=new_upgrade_once, flatten_folders=new_flatten_folders, paused=new_paused,
|
|
sports=new_sports, subs=new_subtitles, anime=new_anime, scene=new_scene, air_by_date=new_air_by_date,
|
|
prune=new_prune, tag=new_tag, direct_call=True)
|
|
|
|
if cur_errors:
|
|
logger.error(f'Errors: {cur_errors}')
|
|
errors.append('<b>%s:</b>\n<ul>' % show_obj.unique_name + ' '.join(
|
|
['<li>%s</li>' % error for error in cur_errors]) + '</ul>')
|
|
|
|
if 0 < len(errors):
|
|
ui.notifications.error('%d error%s while saving changes:' % (len(errors), '' if 1 == len(errors) else 's'),
|
|
' '.join(errors))
|
|
|
|
if switch_tvid:
|
|
self.mass_switch(to_switch='|'.join(switch_tvid))
|
|
return self.redirect('/manage/show-tasks/')
|
|
|
|
self.redirect('/manage/')
|
|
|
|
def bulk_change(self, to_update='', to_refresh='', to_rename='',
|
|
to_subtitle='', to_delete='', to_remove='', **kwargs):
|
|
|
|
to_change = dict({_tvid_prodid: helpers.find_show_by_id(_tvid_prodid)
|
|
for _tvid_prodid in
|
|
next(iter([_x.split('|') for _x in (to_update, to_refresh, to_rename, to_subtitle,
|
|
to_delete, to_remove) if _x]), '')})
|
|
|
|
update, refresh, rename, subtitle, errors = [], [], [], [], []
|
|
for cur_tvid_prodid, cur_show_obj in iteritems(to_change):
|
|
|
|
if cur_tvid_prodid in to_delete:
|
|
cur_show_obj.delete_show(True)
|
|
|
|
elif cur_tvid_prodid in to_remove:
|
|
cur_show_obj.delete_show()
|
|
|
|
else:
|
|
if cur_tvid_prodid in to_update:
|
|
try:
|
|
sickgear.show_queue_scheduler.action.update_show(cur_show_obj, True, True)
|
|
update.append(cur_show_obj.name)
|
|
except exceptions_helper.CantUpdateException as e:
|
|
errors.append('Unable to update show %s: %s' % (cur_show_obj.unique_name, ex(e)))
|
|
|
|
elif cur_tvid_prodid in to_refresh:
|
|
try:
|
|
sickgear.show_queue_scheduler.action.refresh_show(cur_show_obj)
|
|
refresh.append(cur_show_obj.name)
|
|
except exceptions_helper.CantRefreshException as e:
|
|
errors.append('Unable to refresh show %s: %s' % (cur_show_obj.unique_name, ex(e)))
|
|
|
|
if cur_tvid_prodid in to_rename:
|
|
sickgear.show_queue_scheduler.action.rename_show_episodes(cur_show_obj)
|
|
rename.append(cur_show_obj.name)
|
|
|
|
if sickgear.USE_SUBTITLES and cur_tvid_prodid in to_subtitle:
|
|
sickgear.show_queue_scheduler.action.download_subtitles(cur_show_obj)
|
|
subtitle.append(cur_show_obj.name)
|
|
|
|
if len(errors):
|
|
ui.notifications.error('Errors encountered', '<br>\n'.join(errors))
|
|
|
|
if len(update + refresh + rename + subtitle):
|
|
ui.notifications.message(
|
|
'Queued the following actions:',
|
|
''.join(['%s:<br>* %s<br>' % (_to_do, '<br>'.join(_shows))
|
|
for (_to_do, _shows) in (('Updates', update), ('Refreshes', refresh),
|
|
('Renames', rename), ('Subtitles', subtitle)) if len(_shows)]))
|
|
self.redirect('/manage/')
|
|
|
|
def failed_downloads(self, limit=100, to_remove=None):
|
|
|
|
my_db = db.DBConnection('failed.db')
|
|
|
|
sql = 'SELECT * FROM failed ORDER BY ROWID DESC'
|
|
limit = helpers.try_int(limit, 100)
|
|
if not limit:
|
|
sql_result = my_db.select(sql)
|
|
else:
|
|
sql_result = my_db.select(sql + ' LIMIT ?', [limit + 1])
|
|
|
|
to_remove = to_remove.split('|') if None is not to_remove else []
|
|
|
|
for release in to_remove:
|
|
item = re.sub('_{3,}', '%', release)
|
|
my_db.action('DELETE FROM failed WHERE `release` like ?', [item])
|
|
|
|
if to_remove:
|
|
return self.redirect('/manage/failed-downloads/')
|
|
|
|
t = PageTemplate(web_handler=self, file='manage_failedDownloads.tmpl')
|
|
t.over_limit = limit and len(sql_result) > limit
|
|
t.failed_results = t.over_limit and sql_result[0:-1] or sql_result
|
|
t.limit = str(limit)
|
|
t.submenu = self.manage_menu('Failed')
|
|
|
|
return t.respond()
|
|
|
|
@staticmethod
|
|
def mass_switch(to_switch=None):
|
|
"""
|
|
switch multiple given shows to a new tvinfo source
|
|
shows are separated by |
|
|
value for show is: current_tvid:current_prodid-new_tvid:new_prodid-force_id
|
|
parts: new_prodid, force_id are optional
|
|
without new_prodid the mapped id will be used
|
|
with force set to '1' or 'true' the given new id will be used and NO verification for correct show will be done
|
|
|
|
to_switch examples:
|
|
to_switch=1:123-3|3:564-1|1:456-3:123|3:55-1:77-1|3:88-1-1
|
|
|
|
:param to_switch:
|
|
"""
|
|
if not to_switch:
|
|
return json_dumps({'error': 'No list given'})
|
|
|
|
shows = to_switch.split('|')
|
|
sl, tv_sources, errors = [], sickgear.TVInfoAPI().search_sources, []
|
|
for show in shows:
|
|
show_split = show.split('-')
|
|
if 2 == len(show_split):
|
|
old_show, new_show = show_split
|
|
force_id = False
|
|
else:
|
|
old_show, new_show, force_id = show_split
|
|
force_id = force_id in (1, True, '1', 'true', 'True')
|
|
old_show_id = old_show.split(':')
|
|
old_tvid, old_prodid = int(old_show_id[0]), int(old_show_id[1])
|
|
new_show_id = new_show.split(':')
|
|
new_tvid = int(new_show_id[0])
|
|
if new_tvid not in tv_sources:
|
|
logger.warning('Skipping %s because target is not a valid source' % show)
|
|
errors.append('Skipping %s because target is not a valid source' % show)
|
|
continue
|
|
try:
|
|
show_obj = helpers.find_show_by_id({old_tvid: old_prodid})
|
|
except (BaseException, Exception):
|
|
show_obj = None
|
|
if not show_obj:
|
|
logger.warning('Skipping %s because source is not a valid show' % show)
|
|
errors.append('Skipping %s because source is not a valid show' % show)
|
|
continue
|
|
if 2 == len(new_show_id):
|
|
new_prodid = int(new_show_id[1])
|
|
try:
|
|
new_show_obj = helpers.find_show_by_id({new_tvid: new_prodid})
|
|
except (BaseException, Exception):
|
|
new_show_obj = None
|
|
if new_show_obj:
|
|
logger.warning('Skipping %s because target show with that id already exists in db' % show)
|
|
errors.append('Skipping %s because target show with that id already exists in db' % show)
|
|
continue
|
|
else:
|
|
new_prodid = None
|
|
if show_obj.tvid == new_tvid and (not new_prodid or new_prodid == show_obj.prodid):
|
|
logger.warning('Skipping %s because target same as source' % show)
|
|
errors.append('Skipping %s because target same as source' % show)
|
|
continue
|
|
try:
|
|
sickgear.show_queue_scheduler.action.switch_show(show_obj=show_obj, new_tvid=new_tvid,
|
|
new_prodid=new_prodid, force_id=force_id)
|
|
except (BaseException, Exception) as e:
|
|
logger.warning('Could not add show %s to switch queue: %s' % (show_obj.tvid_prodid, ex(e)))
|
|
errors.append('Could not add show %s to switch queue: %s' % (show_obj.tvid_prodid, ex(e)))
|
|
|
|
return json_dumps(({'result': 'success'}, {'errors': ', '.join(errors)})[0 < len(errors)])
|
|
|
|
|
|
class ManageSearch(Manage):
|
|
|
|
def index(self):
|
|
t = PageTemplate(web_handler=self, file='manage_manageSearches.tmpl')
|
|
# t.backlog_pi = sickgear.search_backlog_scheduler.action.get_progress_indicator()
|
|
t.backlog_paused = sickgear.search_queue_scheduler.action.is_backlog_paused()
|
|
t.scheduled_backlog_active_providers = sickgear.search_backlog.BacklogSearcher.providers_active(scheduled=True)
|
|
t.backlog_running = sickgear.search_queue_scheduler.action.is_backlog_in_progress()
|
|
t.backlog_is_active = sickgear.search_backlog_scheduler.action.am_running()
|
|
t.standard_backlog_running = sickgear.search_queue_scheduler.action.is_standard_backlog_in_progress()
|
|
t.backlog_running_type = sickgear.search_queue_scheduler.action.type_of_backlog_in_progress()
|
|
t.recent_search_status = sickgear.search_queue_scheduler.action.is_recentsearch_in_progress()
|
|
t.find_propers_status = sickgear.search_queue_scheduler.action.is_propersearch_in_progress()
|
|
t.queue_length = sickgear.search_queue_scheduler.action.queue_length()
|
|
|
|
t.submenu = self.manage_menu('Search')
|
|
|
|
return t.respond()
|
|
|
|
@staticmethod
|
|
def remove_from_search_queue(to_remove=None):
|
|
if not to_remove:
|
|
return json_dumps({'error': 'nothing to do'})
|
|
to_remove = [int(r) for r in to_remove.split('|')]
|
|
sickgear.search_queue_scheduler.action.remove_from_queue(to_remove=to_remove)
|
|
return json_dumps({'result': 'success'})
|
|
|
|
@staticmethod
|
|
def clear_search_queue(search_type=None):
|
|
search_type = helpers.try_int(search_type, None)
|
|
if not search_type:
|
|
return json_dumps({'error': 'nothing to do'})
|
|
sickgear.search_queue_scheduler.action.clear_queue(action_types=search_type)
|
|
return json_dumps({'result': 'success'})
|
|
|
|
@staticmethod
|
|
def retry_provider(provider=None):
|
|
if not provider:
|
|
return
|
|
prov = [p for p in sickgear.provider_list + sickgear.newznab_providers if p.get_id() == provider]
|
|
if not prov:
|
|
return
|
|
prov[0].retry_next()
|
|
time.sleep(3)
|
|
return
|
|
|
|
def force_backlog(self):
|
|
# force it to run the next time it looks
|
|
if not sickgear.search_queue_scheduler.action.is_standard_backlog_in_progress():
|
|
sickgear.search_backlog_scheduler.force_search(force_type=FORCED_BACKLOG)
|
|
logger.log('Backlog search forced')
|
|
ui.notifications.message('Backlog search started')
|
|
|
|
time.sleep(5)
|
|
self.redirect('/manage/search-tasks/')
|
|
|
|
def force_search(self):
|
|
|
|
# force it to run the next time it looks
|
|
if not sickgear.search_queue_scheduler.action.is_recentsearch_in_progress():
|
|
result = sickgear.search_recent_scheduler.force_run()
|
|
if result:
|
|
logger.log('Recent search forced')
|
|
ui.notifications.message('Recent search started')
|
|
|
|
time.sleep(5)
|
|
self.redirect('/manage/search-tasks/')
|
|
|
|
def force_find_propers(self):
|
|
|
|
# force it to run the next time it looks
|
|
result = sickgear.search_propers_scheduler.force_run()
|
|
if result:
|
|
logger.log('Find propers search forced')
|
|
ui.notifications.message('Find propers search started')
|
|
|
|
time.sleep(5)
|
|
self.redirect('/manage/search-tasks/')
|
|
|
|
def pause_backlog(self, paused=None):
|
|
if '1' == paused:
|
|
sickgear.search_queue_scheduler.action.pause_backlog()
|
|
else:
|
|
sickgear.search_queue_scheduler.action.unpause_backlog()
|
|
|
|
time.sleep(5)
|
|
self.redirect('/manage/search-tasks/')
|
|
|
|
|
|
class ShowTasks(Manage):
|
|
|
|
def index(self):
|
|
t = PageTemplate(web_handler=self, file='manage_showProcesses.tmpl')
|
|
t.queue_length = sickgear.show_queue_scheduler.action.queue_length()
|
|
t.people_queue = sickgear.people_queue_scheduler.action.queue_data()
|
|
t.next_run = sickgear.update_show_scheduler.last_run.replace(
|
|
hour=sickgear.update_show_scheduler.start_time.hour)
|
|
t.show_update_running = sickgear.show_queue_scheduler.action.is_show_update_running() \
|
|
or sickgear.update_show_scheduler.is_running_job
|
|
|
|
my_db = db.DBConnection(row_type='dict')
|
|
sql_result = my_db.select('SELECT n.indexer || ? || n.indexer_id AS tvid_prodid,'
|
|
' n.indexer AS tvid, n.indexer_id AS prodid,'
|
|
' n.last_success, n.fail_count, s.show_name'
|
|
' FROM tv_shows_not_found AS n'
|
|
' INNER JOIN tv_shows AS s'
|
|
' ON (n.indexer == s.indexer AND n.indexer_id == s.indexer_id)',
|
|
[TVidProdid.glue])
|
|
for cur_result in sql_result:
|
|
date = helpers.try_int(cur_result['last_success'])
|
|
cur_result['last_success'] = ('never', SGDatetime.fromordinal(date).sbfdate())[1 < date]
|
|
cur_result['ignore_warning'] = 0 > cur_result['fail_count']
|
|
|
|
defunct_indexer = [i for i in sickgear.TVInfoAPI().all_sources if sickgear.TVInfoAPI(i).config.get('defunct')]
|
|
defunct_sql_result = None
|
|
if defunct_indexer:
|
|
defunct_sql_result = my_db.select('SELECT indexer || ? || indexer_id AS tvid_prodid, show_name'
|
|
' FROM tv_shows'
|
|
' WHERE indexer IN (%s)' % ','.join(['?'] * len(defunct_indexer)),
|
|
[TVidProdid.glue] + defunct_indexer)
|
|
t.defunct_indexer = defunct_sql_result
|
|
t.not_found_shows = sql_result
|
|
|
|
failed_result = my_db.select('SELECT * FROM tv_src_switch WHERE status != ?', [TVSWITCH_NORMAL])
|
|
t.failed_switch = []
|
|
for f in failed_result:
|
|
try:
|
|
show_obj = helpers.find_show_by_id({f['old_indexer']: f['old_indexer_id']})
|
|
except (BaseException, Exception):
|
|
show_obj = None
|
|
new_failed = {'tvid': f['old_indexer'], 'prodid': f['old_indexer_id'], 'new_tvid': f['new_indexer'],
|
|
'new_prodid': f['new_indexer_id'],
|
|
'status': tvswitch_names.get(f['status'], 'unknown %s' % f['status']), 'show_obj': show_obj,
|
|
'uid': f['uid']}
|
|
t.failed_switch.append(new_failed)
|
|
|
|
t.submenu = self.manage_menu('Show')
|
|
|
|
return t.respond()
|
|
|
|
@staticmethod
|
|
def remove_from_show_queue(to_remove=None, force=False):
|
|
if not to_remove:
|
|
return json_dumps({'error': 'nothing to do'})
|
|
force = force in (1, '1', 'true', 'True', True)
|
|
to_remove = [int(r) for r in to_remove.split('|')]
|
|
sickgear.show_queue_scheduler.action.remove_from_queue(to_remove=to_remove, force=force)
|
|
return json_dumps({'result': 'success'})
|
|
|
|
@staticmethod
|
|
def remove_from_people_queue(to_remove=None):
|
|
if not to_remove:
|
|
return json_dumps({'error': 'nothing to do'})
|
|
to_remove = [int(r) for r in to_remove.split('|')]
|
|
sickgear.people_queue_scheduler.action.remove_from_queue(to_remove=to_remove)
|
|
return json_dumps({'result': 'success'})
|
|
|
|
@staticmethod
|
|
def clear_show_queue(show_type=None):
|
|
show_type = helpers.try_int(show_type, None)
|
|
if not show_type:
|
|
return json_dumps({'error': 'nothing to do'})
|
|
if show_type in [sickgear.show_queue.ShowQueueActions.UPDATE,
|
|
sickgear.show_queue.ShowQueueActions.FORCEUPDATE,
|
|
sickgear.show_queue.ShowQueueActions.WEBFORCEUPDATE]:
|
|
show_type = [sickgear.show_queue.ShowQueueActions.UPDATE,
|
|
sickgear.show_queue.ShowQueueActions.FORCEUPDATE,
|
|
sickgear.show_queue.ShowQueueActions.WEBFORCEUPDATE]
|
|
sickgear.show_queue_scheduler.action.clear_queue(action_types=show_type)
|
|
return json_dumps({'result': 'success'})
|
|
|
|
@staticmethod
|
|
def clear_people_queue(people_type=None):
|
|
people_type = helpers.try_int(people_type, None)
|
|
if not people_type:
|
|
return json_dumps({'error': 'nothing to do'})
|
|
sickgear.people_queue_scheduler.action.clear_queue(action_types=people_type)
|
|
return json_dumps({'result': 'success'})
|
|
|
|
def force_show_update(self):
|
|
|
|
result = sickgear.update_show_scheduler.force_run()
|
|
if result:
|
|
logger.log('Show Update forced')
|
|
ui.notifications.message('Forced Show Update started')
|
|
|
|
time.sleep(5)
|
|
self.redirect('/manage/show-tasks/')
|
|
|
|
@staticmethod
|
|
def switch_ignore_warning(**kwargs):
|
|
|
|
for cur_tvid_prodid, state in iteritems(kwargs):
|
|
show_obj = helpers.find_show_by_id(cur_tvid_prodid)
|
|
if show_obj:
|
|
change = -1
|
|
if 'true' == state:
|
|
if 0 > show_obj.not_found_count:
|
|
change = 1
|
|
elif 0 < show_obj.not_found_count:
|
|
change = 1
|
|
show_obj.not_found_count *= change
|
|
|
|
return json_dumps({})
|
|
|
|
|
|
class History(MainHandler):
|
|
flagname_help_watched = 'ui_history_help_watched_supported_clients'
|
|
flagname_wdf = 'ui_history_watched_delete_files'
|
|
flagname_wdr = 'ui_history_watched_delete_records'
|
|
|
|
def toggle_help(self):
|
|
db.DBConnection().toggle_flag(self.flagname_help_watched)
|
|
|
|
@classmethod
|
|
def menu_tab(cls, limit):
|
|
|
|
result = []
|
|
my_db = db.DBConnection(row_type='dict') # type: db.DBConnection
|
|
history_detailed, history_compact = cls.query_history(my_db)
|
|
dedupe = set()
|
|
for item in history_compact:
|
|
if item.get('tvid_prodid') not in dedupe:
|
|
dedupe.add(item.get('tvid_prodid'))
|
|
item['show_name'] = abbr_showname(item['show_name'])
|
|
result += [item]
|
|
if limit == len(result):
|
|
break
|
|
return result
|
|
|
|
@classmethod
|
|
def query_history(cls, my_db, limit=100):
|
|
# type: (db.DBConnection, int) -> Tuple[List[dict], List[dict]]
|
|
"""Query db for historical data
|
|
:param my_db: connection should be instantiated with row_type='dict'
|
|
:param limit: number of db rows to fetch
|
|
:return: two data sets, detailed and compact
|
|
"""
|
|
|
|
sql = 'SELECT h.*, show_name, s.indexer || ? || s.indexer_id AS tvid_prodid' \
|
|
' FROM history h, tv_shows s' \
|
|
' WHERE h.indexer=s.indexer AND h.showid=s.indexer_id' \
|
|
' AND h.hide = 0' \
|
|
' ORDER BY date DESC' \
|
|
'%s' % (' LIMIT %s' % limit, '')['0' == limit]
|
|
sql_result = my_db.select(sql, [TVidProdid.glue])
|
|
|
|
compact = []
|
|
|
|
for cur_result in sql_result:
|
|
|
|
action = dict(time=cur_result['date'], action=cur_result['action'],
|
|
provider=cur_result['provider'], resource=cur_result['resource'])
|
|
|
|
if not any([(record['show_id'] == cur_result['showid']
|
|
and record['indexer'] == cur_result['indexer']
|
|
and record['season'] == cur_result['season']
|
|
and record['episode'] == cur_result['episode']
|
|
and record['quality'] == cur_result['quality']) for record in compact]):
|
|
show_obj = helpers.find_show_by_id({cur_result['indexer']: cur_result['showid']}, no_mapped_ids=False,
|
|
no_exceptions=True)
|
|
cur_res = dict(show_id=cur_result['showid'], indexer=cur_result['indexer'],
|
|
tvid_prodid=cur_result['tvid_prodid'],
|
|
show_name=(show_obj and show_obj.unique_name) or cur_result['show_name'],
|
|
season=cur_result['season'], episode=cur_result['episode'],
|
|
quality=cur_result['quality'], resource=cur_result['resource'], actions=[])
|
|
|
|
cur_res['actions'].append(action)
|
|
cur_res['actions'].sort(key=lambda _x: _x['time'])
|
|
|
|
compact.append(cur_res)
|
|
else:
|
|
index = [i for i, record in enumerate(compact)
|
|
if record['show_id'] == cur_result['showid']
|
|
and record['season'] == cur_result['season']
|
|
and record['episode'] == cur_result['episode']
|
|
and record['quality'] == cur_result['quality']][0]
|
|
|
|
cur_res = compact[index]
|
|
|
|
cur_res['actions'].append(action)
|
|
cur_res['actions'].sort(key=lambda _x: _x['time'], reverse=True)
|
|
|
|
return sql_result, compact
|
|
|
|
def index(self, limit=100, layout=None):
|
|
|
|
t = PageTemplate(web_handler=self, file='history.tmpl')
|
|
t.limit = limit
|
|
|
|
if 'provider_failures' == layout: # layout renamed
|
|
layout = 'connect_failures'
|
|
if layout in ('compact', 'detailed', 'compact_watched', 'detailed_watched', 'connect_failures'):
|
|
sickgear.HISTORY_LAYOUT = layout
|
|
|
|
my_db = db.DBConnection(row_type='dict')
|
|
|
|
result_sets = []
|
|
if sickgear.HISTORY_LAYOUT in ('compact', 'detailed'):
|
|
|
|
sql_result, compact = self.query_history(my_db, limit)
|
|
|
|
t.compact_results = compact
|
|
t.history_results = sql_result
|
|
t.submenu = [{'title': 'Clear History', 'path': 'history/clear-history'},
|
|
{'title': 'Trim History', 'path': 'history/trim-history'}]
|
|
|
|
result_sets = ['compact_results', 'history_results']
|
|
|
|
elif 'watched' in sickgear.HISTORY_LAYOUT:
|
|
|
|
t.hide_watched_help = my_db.has_flag(self.flagname_help_watched)
|
|
|
|
t.results = my_db.select(
|
|
'SELECT tvs.show_name, '
|
|
' tve.indexer AS tvid, tve.showid AS prodid,'
|
|
' tve.indexer || ? || tve.showid AS tvid_prodid,'
|
|
' tve.season, tve.episode, tve.status, tve.file_size,'
|
|
' tvew.rowid, tvew.tvep_id, tvew.label, tvew.played, tvew.date_watched,'
|
|
' tvew.status AS status_w, tvew.location, tvew.file_size AS file_size_w, tvew.hide'
|
|
' FROM [tv_shows] AS tvs'
|
|
' INNER JOIN [tv_episodes] AS tve ON (tvs.indexer = tve.indexer AND tvs.indexer_id = tve.showid)'
|
|
' INNER JOIN [tv_episodes_watched] AS tvew ON (tve.episode_id = tvew.tvep_id)'
|
|
' WHERE 0 = hide'
|
|
' ORDER BY tvew.date_watched DESC'
|
|
'%s' % (' LIMIT %s' % limit, '')['0' == limit],
|
|
[TVidProdid.glue])
|
|
|
|
mru_count = {}
|
|
t.mru_row_ids = []
|
|
for r in t.results:
|
|
r['deleted'] = False
|
|
no_file = not helpers.get_size(r['location'])
|
|
if no_file or not r['file_size']: # if not filesize, possible file recovered so restore known size
|
|
if no_file:
|
|
# file no longer available, can be due to upgrade, so use known details
|
|
r['deleted'] = True
|
|
r['status'] = r['status_w']
|
|
r['file_size'] = r['file_size_w']
|
|
|
|
r['status'], r['quality'] = Quality.split_composite_status(helpers.try_int(r['status']))
|
|
r['season'], r['episode'] = '%02i' % r['season'], '%02i' % r['episode']
|
|
if r['tvep_id'] not in mru_count:
|
|
# depends on SELECT ORDER BY date_watched DESC to determine mru_count
|
|
mru_count.update({r['tvep_id']: r['played']})
|
|
t.mru_row_ids += [r['rowid']]
|
|
r['mru_count'] = mru_count[r['tvep_id']]
|
|
|
|
result_sets = ['results']
|
|
|
|
# restore state of delete dialog
|
|
t.last_delete_files = my_db.has_flag(self.flagname_wdf)
|
|
t.last_delete_records = my_db.has_flag(self.flagname_wdr)
|
|
|
|
elif 'stats' in sickgear.HISTORY_LAYOUT:
|
|
|
|
prov_list = [p.name for p in (sickgear.provider_list
|
|
+ sickgear.newznab_providers
|
|
+ sickgear.torrent_rss_providers)]
|
|
# noinspection SqlResolve
|
|
sql = 'SELECT COUNT(1) AS count,' \
|
|
' MIN(DISTINCT date) AS earliest,' \
|
|
' MAX(DISTINCT date) AS latest,' \
|
|
' provider ' \
|
|
'FROM ' \
|
|
'(SELECT * FROM history h, tv_shows s' \
|
|
' WHERE h.showid=s.indexer_id' \
|
|
' AND h.provider in ("%s")' % '","'.join(prov_list) + \
|
|
' AND h.action in ("%s")' % '","'.join([str(x) for x in Quality.SNATCHED_ANY]) + \
|
|
' AND h.hide = 0' \
|
|
' ORDER BY date DESC%s)' % (' LIMIT %s' % limit, '')['0' == limit] + \
|
|
' GROUP BY provider' \
|
|
' ORDER BY count DESC'
|
|
t.stat_results = my_db.select(sql)
|
|
|
|
t.earliest = 0
|
|
t.latest = 0
|
|
for r in t.stat_results:
|
|
if r['latest'] > t.latest or not t.latest:
|
|
t.latest = r['latest']
|
|
if r['earliest'] < t.earliest or not t.earliest:
|
|
t.earliest = r['earliest']
|
|
|
|
elif 'failures' in sickgear.HISTORY_LAYOUT:
|
|
|
|
t.provider_fail_stats = list(filter(lambda stat: len(stat['fails']), [
|
|
dict(name=p.name, id=p.get_id(), active=p.is_active(), prov_img=p.image_name(),
|
|
prov_id=p.get_id(), # 2020.03.17 legacy var, remove at future date
|
|
fails=p.fails.fails_sorted, next_try=p.get_next_try_time,
|
|
has_limit=getattr(p, 'has_limit', False), tmr_limit_time=p.tmr_limit_time)
|
|
for p in sickgear.provider_list + sickgear.newznab_providers]))
|
|
|
|
t.provider_fail_cnt = len([p for p in t.provider_fail_stats if len(p['fails'])])
|
|
t.provider_fails = t.provider_fail_cnt # 2020.03.17 legacy var, remove at future date
|
|
|
|
t.provider_fail_stats = sorted([item for item in t.provider_fail_stats],
|
|
key=lambda y: y.get('fails')[0].get('timestamp'),
|
|
reverse=True)
|
|
t.provider_fail_stats = sorted([item for item in t.provider_fail_stats],
|
|
key=lambda y: y.get('next_try') or timedelta(weeks=65535),
|
|
reverse=False)
|
|
|
|
def img(_item, as_class=False):
|
|
# type: (AnyStr, bool) -> Optional[AnyStr]
|
|
"""
|
|
Return an image src, image class, or None based on a recognised identifier
|
|
:param _item: to search for a known domain identifier
|
|
:param as_class: whether a search should return an image (by default) or class
|
|
:return: image src, image class, or None if unknown identifier
|
|
"""
|
|
for identifier, result in (
|
|
(('fanart', 'fanart.png'), ('imdb', 'imdb16.png'), ('metac', 'metac16.png'),
|
|
('next-episode', 'nextepisode16.png'),
|
|
('predb', 'predb16.png'), ('srrdb', 'srrdb16.png'),
|
|
('thexem', 'xem.png'), ('tmdb', 'tmdb16.png'), ('trakt', 'trakt16.png'),
|
|
('tvdb', 'thetvdb16.png'), ('tvmaze', 'tvmaze16.png')),
|
|
(('anidb', 'img-anime-16 square-16'), ('github', 'icon16-github'),
|
|
('emby', 'sgicon-emby'), ('plex', 'sgicon-plex'))
|
|
)[as_class]:
|
|
if identifier in _item:
|
|
return result
|
|
|
|
with sg_helpers.DOMAIN_FAILURES.lock:
|
|
t.domain_fail_stats = list(filter(lambda stat: len(stat['fails']), [
|
|
dict(name=k, id=sickgear.GenericProvider.make_id(k), img=img(k), cls=img(k, True),
|
|
fails=v.fails_sorted, next_try=v.get_next_try_time,
|
|
has_limit=getattr(v, 'has_limit', False), tmr_limit_time=v.tmr_limit_time)
|
|
for k, v in iteritems(sg_helpers.DOMAIN_FAILURES.domain_list)]))
|
|
|
|
t.domain_fail_cnt = len([d for d in t.domain_fail_stats if len(d['fails'])])
|
|
|
|
t.domain_fail_stats = sorted([item for item in t.domain_fail_stats],
|
|
key=lambda y: y.get('fails')[0].get('timestamp'),
|
|
reverse=True)
|
|
t.domain_fail_stats = sorted([item for item in t.domain_fail_stats],
|
|
key=lambda y: y.get('next_try') or timedelta(weeks=65535),
|
|
reverse=False)
|
|
|
|
article_match = r'^((?:A(?!\s+to)n?)|The)\s+(.*)$'
|
|
for rs in [getattr(t, name, []) for name in result_sets]:
|
|
for r in rs:
|
|
r['name1'] = ''
|
|
r['name2'] = r['data_name'] = r['show_name']
|
|
if not sickgear.SORT_ARTICLE:
|
|
try:
|
|
r['name1'], r['name2'] = re.findall(article_match, r['show_name'])[0]
|
|
r['data_name'] = r['name2']
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
return t.respond()
|
|
|
|
@staticmethod
|
|
def check_site(site_name=''):
|
|
|
|
site_url = dict(
|
|
tvdb='api.thetvdb.com', thexem='thexem.info', github='github.com'
|
|
).get(site_name.replace('check_', ''))
|
|
|
|
result = {}
|
|
|
|
if site_url:
|
|
import requests
|
|
down_url = 'www.isitdownrightnow.com'
|
|
proto = 'https'
|
|
try:
|
|
requests.head('%s://%s' % (proto, down_url), timeout=5)
|
|
except (BaseException, Exception):
|
|
proto = 'http'
|
|
try:
|
|
requests.head('%s://%s' % (proto, down_url), timeout=5)
|
|
except (BaseException, Exception):
|
|
return json_dumps(result)
|
|
|
|
resp = helpers.get_url('%s://%s/check.php?domain=%s' % (proto, down_url, site_url))
|
|
if resp:
|
|
check = resp.lower()
|
|
day = re.findall(r'(\d+)\s*day', check)
|
|
hr = re.findall(r'(\d+)\s*hour', check)
|
|
mn = re.findall(r'(\d+)\s*min', check)
|
|
if any([day, hr, mn]):
|
|
period = ', '.join(
|
|
(day and ['%sd' % day[0]] or day)
|
|
+ (hr and ['%sh' % hr[0]] or hr)
|
|
+ (mn and ['%sm' % mn[0]] or mn))
|
|
else:
|
|
try:
|
|
period = re.findall('[^>]>([^<]+)ago', check)[0].strip()
|
|
except (BaseException, Exception):
|
|
try:
|
|
period = re.findall('[^>]>([^<]+week)', check)[0]
|
|
except (BaseException, Exception):
|
|
period = 'quite some time'
|
|
|
|
result = {('last_down', 'down_for')['up' not in check and 'down for' in check]: period}
|
|
|
|
return json_dumps(result)
|
|
|
|
def clear_history(self):
|
|
|
|
my_db = db.DBConnection()
|
|
# noinspection SqlConstantCondition
|
|
my_db.action('UPDATE history SET hide = ? WHERE hide = 0', [1])
|
|
|
|
ui.notifications.message('History cleared')
|
|
self.redirect('/history/')
|
|
|
|
def trim_history(self):
|
|
|
|
my_db = db.DBConnection()
|
|
my_db.action('UPDATE history SET hide = ? WHERE date < ' + str(
|
|
(datetime.now() - timedelta(days=30)).strftime(history.dateFormat)), [1])
|
|
|
|
ui.notifications.message('Removed history entries greater than 30 days old')
|
|
self.redirect('/history/')
|
|
|
|
@staticmethod
|
|
def retry_domain(domain=None):
|
|
|
|
if domain in sg_helpers.DOMAIN_FAILURES.domain_list:
|
|
sg_helpers.DOMAIN_FAILURES.domain_list[domain].retry_next()
|
|
time.sleep(3)
|
|
|
|
@staticmethod
|
|
def update_watched_state_emby():
|
|
|
|
import sickgear.notifiers.emby as emby
|
|
|
|
client = emby.EmbyNotifier()
|
|
hosts, keys, message = client.check_config(sickgear.EMBY_HOST, sickgear.EMBY_APIKEY)
|
|
|
|
if sickgear.USE_EMBY and hosts:
|
|
logger.debug('Updating Emby watched episode states')
|
|
|
|
rd = sickgear.ROOT_DIRS.split('|')[1:] \
|
|
+ [x.split('=')[0] for x in sickgear.EMBY_PARENT_MAPS.split(',') if any(x)]
|
|
rootpaths = sorted(['%s%s' % (os.path.splitdrive(x)[1], os.path.sep) for x in rd], key=len, reverse=True)
|
|
rootdirs = sorted([x for x in rd], key=len, reverse=True)
|
|
headers = {'Content-type': 'application/json'}
|
|
states = {}
|
|
idx = 0
|
|
mapped = 0
|
|
mapping = None
|
|
maps = [x.split('=') for x in sickgear.EMBY_PARENT_MAPS.split(',') if any(x)]
|
|
args = dict(params=dict(format='json'), timeout=10, parse_json=True, failure_monitor=False)
|
|
for i, cur_host in enumerate(hosts):
|
|
# noinspection HttpUrlsUsage
|
|
base_url = 'http://%s/emby' % cur_host
|
|
headers.update({'X-MediaBrowser-Token': keys[i]})
|
|
|
|
users = helpers.get_url(base_url + '/Users', headers=headers, **args)
|
|
|
|
for user_id in users and [u.get('Id') for u in users if u.get('Id')] or []:
|
|
user_url = '%s/Users/%s' % (base_url, user_id)
|
|
user = helpers.get_url(user_url, headers=headers, **args)
|
|
|
|
folder_ids = user.get('Policy', {}).get('EnabledFolders') or []
|
|
if not folder_ids and user.get('Policy', {}).get('EnableAllFolders'):
|
|
folders = helpers.get_url('%s/Library/MediaFolders' % base_url, headers=headers, **args)
|
|
folder_ids = [_f.get('Id') for _f in folders.get('Items', {}) if _f.get('IsFolder')
|
|
and 'tvshows' == _f.get('CollectionType', '') and _f.get('Id')]
|
|
|
|
for folder_id in folder_ids:
|
|
folder = helpers.get_url('%s/Items/%s' % (user_url, folder_id), headers=headers,
|
|
mute_http_error=True, **args)
|
|
|
|
if not folder or 'tvshows' != folder.get('CollectionType', ''):
|
|
continue
|
|
|
|
items = helpers.get_url('%s/Items' % user_url, failure_monitor=False, headers=headers,
|
|
params=dict(SortBy='DatePlayed,SeriesSortName,SortName',
|
|
SortOrder='Descending',
|
|
IncludeItemTypes='Episode',
|
|
Recursive='true',
|
|
Fields='Path,UserData',
|
|
IsMissing='false',
|
|
IsVirtualUnaired='false',
|
|
StartIndex='0', Limit='100',
|
|
ParentId=folder_id,
|
|
Filters='IsPlayed',
|
|
format='json'), timeout=10, parse_json=True) or {}
|
|
for d in filter(lambda item: 'Episode' == item.get('Type', ''), items.get('Items')):
|
|
try:
|
|
root_dir_found = False
|
|
path_file = d.get('Path')
|
|
if not path_file:
|
|
continue
|
|
for index, p in enumerate(rootpaths):
|
|
if p in path_file:
|
|
path_file = os.path.join(
|
|
rootdirs[index], re.sub('.*?%s' % re.escape(p), '', path_file))
|
|
root_dir_found = True
|
|
break
|
|
if not root_dir_found:
|
|
continue
|
|
states[idx] = dict(
|
|
path_file=path_file,
|
|
media_id=d.get('Id', ''),
|
|
played=(d.get('UserData', {}).get('PlayedPercentage') or
|
|
(d.get('UserData', {}).get('Played') and
|
|
d.get('UserData', {}).get('PlayCount') * 100) or 0),
|
|
label='%s%s{Emby}' % (user.get('Name', ''), bool(user.get('Name')) and ' ' or ''),
|
|
date_watched=SGDatetime.timestamp_far(
|
|
dateutil.parser.parse(d.get('UserData', {}).get('LastPlayedDate'))))
|
|
|
|
for m in maps:
|
|
result, change = helpers.path_mapper(m[0], m[1], states[idx]['path_file'])
|
|
if change:
|
|
if not mapping:
|
|
mapping = (states[idx]['path_file'], result)
|
|
mapped += 1
|
|
states[idx]['path_file'] = result
|
|
break
|
|
|
|
idx += 1
|
|
except (BaseException, Exception):
|
|
continue
|
|
if mapping:
|
|
logger.debug(f'Folder mappings used, the first of {mapped} is [{mapping[0]}] in Emby is'
|
|
f' [{mapping[1]}] in SickGear')
|
|
|
|
if states:
|
|
# Prune user removed items that are no longer being returned by API
|
|
media_paths = list(map(lambda arg: os.path.basename(arg[1]['path_file']), iteritems(states)))
|
|
sql = 'FROM tv_episodes_watched WHERE hide=1 AND label LIKE "%%{Emby}"'
|
|
my_db = db.DBConnection(row_type='dict')
|
|
files = my_db.select('SELECT location %s' % sql)
|
|
for i in filter(lambda f: os.path.basename(f['location']) not in media_paths, files):
|
|
loc = i.get('location')
|
|
if loc:
|
|
my_db.select('DELETE %s AND location="%s"' % (sql, loc))
|
|
|
|
MainHandler.update_watched_state(states, False)
|
|
|
|
logger.log('Finished updating Emby watched episode states')
|
|
|
|
@staticmethod
|
|
def update_watched_state_plex():
|
|
|
|
hosts = [x.strip().lower() for x in sickgear.PLEX_SERVER_HOST.split(',')]
|
|
if sickgear.USE_PLEX and hosts:
|
|
logger.debug('Updating Plex watched episode states')
|
|
|
|
from lib.plex import Plex
|
|
|
|
plex = Plex(dict(username=sickgear.PLEX_USERNAME, password=sickgear.PLEX_PASSWORD,
|
|
section_filter_path=sickgear.ROOT_DIRS.split('|')[1:] +
|
|
[x.split('=')[0] for x in sickgear.PLEX_PARENT_MAPS.split(',') if any(x)]))
|
|
|
|
states = {}
|
|
idx = 0
|
|
played = 0
|
|
mapped = 0
|
|
mapping = None
|
|
maps = [x.split('=') for x in sickgear.PLEX_PARENT_MAPS.split(',') if any(x)]
|
|
for cur_host in hosts:
|
|
# noinspection HttpUrlsUsage
|
|
parts = re.search(r'(.*):(\d+)$', urlparse('http://' + re.sub(r'^\w+://', '', cur_host)).netloc)
|
|
if not parts:
|
|
logger.warning('Skipping host not in min. host:port format : %s' % cur_host)
|
|
elif parts.group(1):
|
|
plex.plex_host = parts.group(1)
|
|
if None is not parts.group(2):
|
|
plex.plex_port = parts.group(2)
|
|
|
|
plex.fetch_show_states()
|
|
|
|
for k, v in iteritems(plex.show_states):
|
|
if 0 < v.get('played') or 0:
|
|
played += 1
|
|
states[idx] = v
|
|
states[idx]['label'] = '%s%s{Plex}' % (v['label'], bool(v['label']) and ' ' or '')
|
|
|
|
for m in maps:
|
|
result, change = helpers.path_mapper(m[0], m[1], states[idx]['path_file'])
|
|
if change:
|
|
if not mapping:
|
|
mapping = (states[idx]['path_file'], result)
|
|
mapped += 1
|
|
states[idx]['path_file'] = result
|
|
break
|
|
|
|
idx += 1
|
|
|
|
logger.debug('Fetched %s of %s played for host : %s' % (len(plex.show_states), played, cur_host))
|
|
if mapping:
|
|
logger.debug(f'Folder mappings used, the first of {mapped} is [{mapping[0]}] in Plex is'
|
|
f' [{mapping[1]}] in SickGear')
|
|
|
|
if states:
|
|
# Prune user removed items that are no longer being returned by API
|
|
media_paths = list(map(lambda arg: os.path.basename(arg[1]['path_file']), iteritems(states)))
|
|
sql = 'FROM tv_episodes_watched WHERE hide=1 AND label LIKE "%%{Plex}"'
|
|
my_db = db.DBConnection(row_type='dict')
|
|
files = my_db.select('SELECT location %s' % sql)
|
|
for i in filter(lambda f: os.path.basename(f['location']) not in media_paths, files):
|
|
loc = i.get('location')
|
|
if loc:
|
|
my_db.select('DELETE %s AND location="%s"' % (sql, loc))
|
|
|
|
MainHandler.update_watched_state(states, False)
|
|
|
|
logger.log('Finished updating Plex watched episode states')
|
|
|
|
def watched(self, tvew_id=None, files=None, records=None):
|
|
|
|
my_db = db.DBConnection(row_type='dict')
|
|
|
|
# remember state of dialog
|
|
my_db.set_flag(self.flagname_wdf, files)
|
|
my_db.set_flag(self.flagname_wdr, records)
|
|
|
|
ids = tvew_id.split('|')
|
|
if not (ids and any([files, records])):
|
|
return
|
|
|
|
row_show_ids = {}
|
|
for show_detail in ids:
|
|
rowid, tvid, prodid = show_detail.split('-')
|
|
row_show_ids.update({int(rowid): {int(tvid): int(prodid)}})
|
|
|
|
sql_result = my_db.select(
|
|
'SELECT rowid, tvep_id, label, location'
|
|
' FROM [tv_episodes_watched] WHERE `rowid` in (%s)' % ','.join([str(k) for k in row_show_ids])
|
|
)
|
|
|
|
h_records = []
|
|
removed = []
|
|
deleted = {}
|
|
attempted = []
|
|
refresh = []
|
|
for cur_result in sql_result:
|
|
if files and cur_result['location'] not in attempted and 0 < helpers.get_size(cur_result['location']) \
|
|
and os.path.isfile(cur_result['location']):
|
|
# locations repeat with watch events but attempt to delete once
|
|
attempted += [cur_result['location']]
|
|
|
|
result = helpers.remove_file(cur_result['location'])
|
|
if result:
|
|
logger.log(f'{result} file {cur_result["location"]}')
|
|
|
|
deleted.update({cur_result['tvep_id']: row_show_ids[cur_result['rowid']]})
|
|
if row_show_ids[cur_result['rowid']] not in refresh:
|
|
# schedule a show for one refresh after deleting an arbitrary number of locations
|
|
refresh += [row_show_ids[cur_result['rowid']]]
|
|
|
|
if records:
|
|
if not cur_result['label'].endswith('{Emby}') and not cur_result['label'].endswith('{Plex}'):
|
|
r_del = my_db.action('DELETE FROM [tv_episodes_watched] WHERE `rowid` == ?',
|
|
[cur_result['rowid']])
|
|
if 1 == r_del.rowcount:
|
|
h_records += ['%s-%s-%s' % (cur_result['rowid'], k, v)
|
|
for k, v in iteritems(row_show_ids[cur_result['rowid']])]
|
|
else:
|
|
r_del = my_db.action('UPDATE [tv_episodes_watched] SET hide=1 WHERE `rowid` == ?',
|
|
[cur_result['rowid']])
|
|
if 1 == r_del.rowcount:
|
|
removed += ['%s-%s-%s' % (cur_result['rowid'], k, v)
|
|
for k, v in iteritems(row_show_ids[cur_result['rowid']])]
|
|
|
|
updating = False
|
|
for epid, tvid_prodid_dict in iteritems(deleted):
|
|
sql_result = my_db.select('SELECT season, episode FROM [tv_episodes] WHERE `episode_id` = %s' % epid)
|
|
for cur_result in sql_result:
|
|
show_obj = helpers.find_show_by_id(tvid_prodid_dict)
|
|
ep_obj = show_obj.get_episode(cur_result['season'], cur_result['episode'])
|
|
for n in filter(lambda x: x.name.lower() in ('emby', 'kodi', 'plex'),
|
|
notifiers.NotifierFactory().get_enabled()):
|
|
if 'PLEX' == n.name:
|
|
if updating:
|
|
continue
|
|
updating = True
|
|
n.update_library(show_obj=show_obj, show_name=show_obj.name, ep_obj=ep_obj)
|
|
|
|
for tvid_prodid_dict in refresh:
|
|
try:
|
|
sickgear.show_queue_scheduler.action.refresh_show(
|
|
helpers.find_show_by_id(tvid_prodid_dict))
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
if not any([removed, h_records, len(deleted)]):
|
|
msg = 'No items removed and no files deleted'
|
|
else:
|
|
msg = []
|
|
if deleted:
|
|
msg += ['%s %s media file%s' % (
|
|
('Permanently deleted', 'Trashed')[sickgear.TRASH_REMOVE_SHOW],
|
|
len(deleted), helpers.maybe_plural(deleted))]
|
|
elif removed:
|
|
msg += ['Removed %s watched history item%s' % (len(removed), helpers.maybe_plural(removed))]
|
|
else:
|
|
msg += ['Deleted %s watched history item%s' % (len(h_records), helpers.maybe_plural(h_records))]
|
|
msg = '<br>'.join(msg)
|
|
|
|
ui.notifications.message('History : Watch', msg)
|
|
|
|
return json_dumps(dict(success=h_records))
|
|
|
|
|
|
class Config(MainHandler):
|
|
|
|
@staticmethod
|
|
def config_menu(exclude='n/a'):
|
|
menu = [
|
|
{'title': 'General', 'path': 'config/general/'},
|
|
{'title': 'Media Providers', 'path': 'config/providers/'},
|
|
{'title': 'Search', 'path': 'config/search/'},
|
|
{'title': 'Subtitles', 'path': 'config/subtitles/'},
|
|
{'title': 'Media Process', 'path': 'config/media-process/'},
|
|
{'title': 'Notifications', 'path': 'config/notifications/'},
|
|
{'title': 'Anime', 'path': 'config/anime/'},
|
|
]
|
|
return [x for x in menu if exclude not in x['title']]
|
|
|
|
def index(self):
|
|
t = PageTemplate(web_handler=self, file='config.tmpl')
|
|
t.submenu = self.config_menu()
|
|
|
|
try:
|
|
with open(os.path.join(sickgear.PROG_DIR, 'CHANGES.md')) as fh:
|
|
t.version = re.findall(r'###[^0-9]+([0-9]+\.[0-9]+\.[0-9x]+)', fh.readline())[0]
|
|
except (BaseException, Exception):
|
|
t.version = ''
|
|
|
|
current_file = zoneinfo.ZONEFILENAME
|
|
t.tz_fallback = False
|
|
t.tz_version = None
|
|
try:
|
|
if None is not current_file:
|
|
current_file = os.path.basename(current_file)
|
|
zonefile = real_path(os.path.join(sickgear.ZONEINFO_DIR, current_file))
|
|
if not os.path.isfile(zonefile):
|
|
t.tz_fallback = True
|
|
zonefile = os.path.join(os.path.dirname(zoneinfo.__file__), current_file)
|
|
if os.path.isfile(zonefile):
|
|
t.tz_version = zoneinfo.ZoneInfoFile(zoneinfo.getzoneinfofile_stream()).metadata['tzversion']
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
t.backup_db_path = sickgear.BACKUP_DB_MAX_COUNT and \
|
|
(sickgear.BACKUP_DB_PATH or os.path.join(sickgear.DATA_DIR, 'backup')) or 'Disabled'
|
|
|
|
return t.respond()
|
|
|
|
|
|
class ConfigGeneral(Config):
|
|
|
|
def index(self):
|
|
|
|
t = PageTemplate(web_handler=self, file='config_general.tmpl')
|
|
t.submenu = self.config_menu('General')
|
|
t.show_tags = ', '.join(sickgear.SHOW_TAGS)
|
|
t.infosrc = dict([(i, sickgear.TVInfoAPI().sources[i]) for i in sickgear.TVInfoAPI().sources
|
|
if sickgear.TVInfoAPI(i).config['active']])
|
|
t.request_host = helpers.xhtml_escape(self.request.host_name, False)
|
|
api_keys = '|||'.join([':::'.join(a) for a in sickgear.API_KEYS])
|
|
t.api_keys = api_keys and sickgear.API_KEYS or []
|
|
t.pip_user_arg = ('--user ', '')[is_virtualenv()]
|
|
if 'git' == sickgear.update_software_scheduler.action.install_type:
|
|
# noinspection PyProtectedMember
|
|
sickgear.update_software_scheduler.action.updater._find_installed_version()
|
|
return t.respond()
|
|
|
|
@staticmethod
|
|
def update_alt():
|
|
""" Load scene exceptions """
|
|
|
|
changed_exceptions, cnt_updated_numbers, min_remain_iv = scene_exceptions.ReleaseMap().fetch_exceptions()
|
|
|
|
return json_dumps(dict(names=int(changed_exceptions), numbers=cnt_updated_numbers, min_remain_iv=min_remain_iv))
|
|
|
|
@staticmethod
|
|
def export_alt(tvid_prodid=None):
|
|
""" Return alternative release names and numbering as json text"""
|
|
|
|
# alternative release names and numbers
|
|
alt_names = scene_exceptions.ReleaseMap().get_show_exceptions(tvid_prodid)
|
|
alt_numbers = get_scene_numbering_for_show(*TVidProdid(tvid_prodid).tuple) # arbitrary order
|
|
ui_output = 'No alternative names or numbers to export'
|
|
|
|
# combine all possible season numbers into a sorted desc list
|
|
seasons = sorted(set(list(set([s for (s, e) in alt_numbers])) + [s for s in alt_names]), reverse=True)
|
|
if seasons:
|
|
if -1 == seasons[-1]:
|
|
seasons = [-1] + seasons[0:-1] # bubble -1
|
|
|
|
# prepare a seasonal ordered dict for output
|
|
alts = dict([(season, {}) for season in seasons])
|
|
|
|
# add original show name
|
|
show_obj = sickgear.helpers.find_show_by_id(tvid_prodid, no_mapped_ids=True)
|
|
first_key = next(iteritems(alts))[0]
|
|
alts[first_key].update(dict({'#': show_obj.name}))
|
|
|
|
# process alternative release names
|
|
for (season, names) in iteritems(alt_names):
|
|
alts[season].update(dict(n=names))
|
|
|
|
# process alternative release numbers
|
|
for_target_group = {}
|
|
# uses a sorted list of (for seasons, for episodes) as a method
|
|
# to group (for, target) seasons with lists of target episodes
|
|
for f_se in sorted(alt_numbers): # sort season list (and therefore, implicitly asc/desc of targets)
|
|
t_se = alt_numbers[f_se]
|
|
for_target_group.setdefault((f_se[0], t_se[0]), []) # f_se[0] = for_season, t_se[0] = target_season
|
|
for_target_group[(f_se[0], t_se[0])] += [(f_se[1], t_se[1])] # f_se[1] = for_ep, t_se[1] = target_ep
|
|
|
|
# minimise episode lists into ranges e.g. 1x1, 2x2, ... 5x5 => 1x1-5
|
|
minimal = {}
|
|
for ft_s, ft_e_range in iteritems(for_target_group):
|
|
minimal.setdefault(ft_s, [])
|
|
last_f_e = None
|
|
for (f_e, t_e) in ft_e_range:
|
|
add_new = True
|
|
if minimal[ft_s]:
|
|
last = minimal[ft_s][-1]
|
|
last_t_e = last[-1]
|
|
if (f_e, t_e) in ((last_f_e + 1, last_t_e + 1), (last_f_e - 1, last_t_e - 1)):
|
|
add_new = False
|
|
if 2 == len(last):
|
|
minimal[ft_s][-1] += [t_e] # create range
|
|
else:
|
|
minimal[ft_s][-1][-1] += (-1, 1)[t_e == last_t_e + 1] # adjust range
|
|
last_f_e = f_e
|
|
if add_new:
|
|
minimal[ft_s] += [[f_e, t_e]] # singular
|
|
|
|
for (f_s, t_s), ft_list in iteritems(minimal):
|
|
alts[f_s].setdefault('se', [])
|
|
for fe_te in ft_list:
|
|
alts[f_s]['se'] += [dict({fe_te[0]: '%sx%s' % (t_s, '-'.join(['%s' % x for x in fe_te[1:]]))})]
|
|
|
|
ui_output = json_dumps(dict({tvid_prodid: alts}), indent=2, separators=(',', ': '))
|
|
return json_dumps(dict(text='%s\n\n' % ui_output))
|
|
|
|
@staticmethod
|
|
def generate_key(*args, **kwargs):
|
|
""" Return a new randomized API_KEY
|
|
"""
|
|
# Create some values to seed md5
|
|
seed = str(time.time()) + str(random.random())
|
|
|
|
result = hashlib.new('md5', decode_bytes(seed)).hexdigest()
|
|
|
|
# Return a hex digest of the md5, e.g. 49f68a5c8493ec2c0bf489821c21fc3b
|
|
app_name = kwargs.get('app_name')
|
|
app_name = '' if not app_name else ' for [%s]' % app_name
|
|
logger.log(f'New API generated{app_name}')
|
|
|
|
return result
|
|
|
|
@staticmethod
|
|
def save_root_dirs(root_dir_string=None):
|
|
|
|
sickgear.ROOT_DIRS = root_dir_string
|
|
|
|
@staticmethod
|
|
def save_result_prefs(ui_results_sortby=None):
|
|
|
|
if ui_results_sortby in ('az', 'za', 'newest', 'oldest', 'rel', 'notop', 'ontop', 'nogroup', 'ingroup'):
|
|
is_notop = ('', ' notop')['notop' in sickgear.RESULTS_SORTBY]
|
|
is_nogrp = ('', ' nogroup')['nogroup' in sickgear.RESULTS_SORTBY]
|
|
if 'top' == ui_results_sortby[-3:] or 'group' == ui_results_sortby[-5:]:
|
|
maybe_ontop = (is_notop, ('', ' notop')[not is_notop])['top' == ui_results_sortby[-3:]]
|
|
maybe_ingroup = (is_nogrp, ('', ' nogroup')[not is_nogrp])['group' == ui_results_sortby[-5:]]
|
|
sortby = sickgear.RESULTS_SORTBY.replace(' notop', '').replace(' nogroup', '')
|
|
sickgear.RESULTS_SORTBY = '%s%s%s' % (('rel', sortby)[any([sortby])], maybe_ontop, maybe_ingroup)
|
|
else:
|
|
sickgear.RESULTS_SORTBY = '%s%s%s' % (ui_results_sortby, is_notop, is_nogrp)
|
|
|
|
sickgear.save_config()
|
|
|
|
@staticmethod
|
|
def save_add_show_defaults(default_status, any_qualities='', best_qualities='', default_wanted_begin=None,
|
|
default_wanted_latest=None, default_flatten_folders=False, default_scene=False,
|
|
default_subs=False, default_anime=False, default_pause=False, default_tag=''):
|
|
|
|
any_qualities = ([], any_qualities.split(','))[any(any_qualities)]
|
|
best_qualities = ([], best_qualities.split(','))[any(best_qualities)]
|
|
|
|
sickgear.QUALITY_DEFAULT = int(Quality.combine_qualities(list(map(int, any_qualities)),
|
|
list(map(int, best_qualities))))
|
|
sickgear.WANTED_BEGIN_DEFAULT = config.minimax(default_wanted_begin, 0, -1, 10)
|
|
sickgear.WANTED_LATEST_DEFAULT = config.minimax(default_wanted_latest, 0, -1, 10)
|
|
sickgear.SHOW_TAG_DEFAULT = default_tag
|
|
sickgear.PAUSE_DEFAULT = config.checkbox_to_value(default_pause)
|
|
sickgear.STATUS_DEFAULT = int(default_status)
|
|
sickgear.SCENE_DEFAULT = config.checkbox_to_value(default_scene)
|
|
sickgear.SUBTITLES_DEFAULT = config.checkbox_to_value(default_subs)
|
|
sickgear.FLATTEN_FOLDERS_DEFAULT = config.checkbox_to_value(default_flatten_folders)
|
|
sickgear.ANIME_DEFAULT = config.checkbox_to_value(default_anime)
|
|
|
|
sickgear.save_config()
|
|
|
|
def create_apikey(self, app_name):
|
|
result = dict()
|
|
if not app_name:
|
|
result['result'] = 'Failed: no name given'
|
|
elif app_name in [k[0] for k in sickgear.API_KEYS if k[0]]:
|
|
result['result'] = 'Failed: name is not unique'
|
|
else:
|
|
api_key = self.generate_key(app_name=app_name)
|
|
if api_key in [k[1] for k in sickgear.API_KEYS if k[0]]:
|
|
result['result'] = 'Failed: apikey already exists, try again'
|
|
else:
|
|
sickgear.API_KEYS.append([app_name, api_key])
|
|
logger.debug('Created apikey for [%s]' % app_name)
|
|
result.update(dict(result='Success: apikey added', added=api_key))
|
|
sickgear.USE_API = 1
|
|
sickgear.save_config()
|
|
ui.notifications.message('Configuration Saved', os.path.join(sickgear.CONFIG_FILE))
|
|
|
|
return json_dumps(result)
|
|
|
|
@staticmethod
|
|
def revoke_apikey(app_name, api_key):
|
|
result = dict()
|
|
if not app_name:
|
|
result['result'] = 'Failed: no name given'
|
|
elif not api_key or 32 != len(re.sub('(?i)[^0-9a-f]', '', api_key)):
|
|
result['result'] = 'Failed: key not valid'
|
|
elif api_key not in [k[1] for k in sickgear.API_KEYS if k[0]]:
|
|
result['result'] = 'Failed: key doesn\'t exist'
|
|
else:
|
|
sickgear.API_KEYS = [ak for ak in sickgear.API_KEYS if ak[0] and api_key != ak[1]]
|
|
logger.debug('Revoked [%s] apikey [%s]' % (app_name, api_key))
|
|
result.update(dict(result='Success: apikey removed', removed=True))
|
|
sickgear.save_config()
|
|
ui.notifications.message('Configuration Saved', os.path.join(sickgear.CONFIG_FILE))
|
|
|
|
return json_dumps(result)
|
|
|
|
def save_general(self, launch_browser=None, update_shows_on_start=None, show_update_hour=None,
|
|
trash_remove_show=None, trash_rotate_logs=None,
|
|
log_dir=None, web_log=None,
|
|
indexer_default=None, indexer_timeout=None,
|
|
show_dirs_with_dots=None,
|
|
update_notify=None, update_auto=None, update_interval=None, notify_on_update=None,
|
|
update_packages_notify=None, update_packages_auto=None, update_packages_menu=None,
|
|
update_packages_interval=None,
|
|
update_frequency=None, # deprecated 2020.11.07
|
|
theme_name=None, default_home=None, fanart_limit=None, showlist_tagview=None, show_tags=None,
|
|
home_search_focus=None, use_imdb_info=None, display_freespace=None, sort_article=None,
|
|
fuzzy_dating=None, trim_zero=None, date_preset=None, time_preset=None,
|
|
timezone_display=None,
|
|
web_username=None, web_password=None,
|
|
calendar_unprotected=None, use_api=None, web_port=None,
|
|
enable_https=None, https_cert=None, https_key=None,
|
|
web_ipv6=None, web_ipv64=None,
|
|
handle_reverse_proxy=None, send_security_headers=None, allowed_hosts=None, allow_anyip=None,
|
|
git_remote=None,
|
|
git_path=None, cpu_preset=None, anon_redirect=None, encryption_version=None,
|
|
proxy_setting=None, proxy_indexers=None, file_logging_preset=None, backup_db_oneday=None):
|
|
|
|
# 2020.11.07 prevent deprecated var issues from existing ui, delete in future, added
|
|
if None is update_interval and None is not update_frequency:
|
|
update_interval = update_frequency
|
|
|
|
results = []
|
|
|
|
# Misc
|
|
sickgear.LAUNCH_BROWSER = config.checkbox_to_value(launch_browser)
|
|
sickgear.UPDATE_SHOWS_ON_START = config.checkbox_to_value(update_shows_on_start)
|
|
sickgear.SHOW_UPDATE_HOUR = config.minimax(show_update_hour, 3, 0, 23)
|
|
try:
|
|
with sickgear.update_show_scheduler.lock:
|
|
sickgear.update_show_scheduler.start_time = dt_time(hour=sickgear.SHOW_UPDATE_HOUR)
|
|
except (BaseException, Exception) as e:
|
|
logger.error('Could not change Show Update Scheduler time: %s' % ex(e))
|
|
sickgear.TRASH_REMOVE_SHOW = config.checkbox_to_value(trash_remove_show)
|
|
sg_helpers.TRASH_REMOVE_SHOW = sickgear.TRASH_REMOVE_SHOW
|
|
sickgear.TRASH_ROTATE_LOGS = config.checkbox_to_value(trash_rotate_logs)
|
|
if not config.change_log_dir(log_dir, web_log):
|
|
results += ['Unable to create directory ' + os.path.normpath(log_dir) + ', log directory not changed.']
|
|
if indexer_default:
|
|
sickgear.TVINFO_DEFAULT = config.to_int(indexer_default)
|
|
if 0 != sickgear.TVINFO_DEFAULT and not sickgear.TVInfoAPI(sickgear.TVINFO_DEFAULT).config.get('active'):
|
|
sickgear.TVINFO_DEFAULT = TVINFO_TVDB
|
|
if indexer_timeout:
|
|
sickgear.TVINFO_TIMEOUT = config.to_int(indexer_timeout)
|
|
sickgear.SHOW_DIRS_WITH_DOTS = config.checkbox_to_value(show_dirs_with_dots)
|
|
|
|
# Updates
|
|
config.schedule_update_software_notify(config.checkbox_to_value(update_notify))
|
|
sickgear.UPDATE_AUTO = config.checkbox_to_value(update_auto)
|
|
config.schedule_update_software(update_interval)
|
|
sickgear.NOTIFY_ON_UPDATE = config.checkbox_to_value(notify_on_update)
|
|
|
|
config.schedule_update_packages_notify(config.checkbox_to_value(update_packages_notify))
|
|
sickgear.UPDATE_PACKAGES_AUTO = config.checkbox_to_value(update_packages_auto)
|
|
sickgear.UPDATE_PACKAGES_MENU = config.checkbox_to_value(update_packages_menu)
|
|
config.schedule_update_packages(update_packages_interval)
|
|
|
|
# Interface
|
|
sickgear.THEME_NAME = theme_name
|
|
sickgear.DEFAULT_HOME = default_home
|
|
sickgear.FANART_LIMIT = config.minimax(fanart_limit, 3, 0, 500)
|
|
sickgear.SHOWLIST_TAGVIEW = showlist_tagview
|
|
|
|
# 'Show List' is the must-have default fallback. Tags in use that are removed from config ui are restored,
|
|
# not deleted. De-duped list order preservation is key to feature function.
|
|
my_db = db.DBConnection()
|
|
sql_result = my_db.select('SELECT DISTINCT tag FROM tv_shows')
|
|
new_names = [v.strip() for v in (show_tags.split(','), [])[None is show_tags] if v.strip()]
|
|
orphans = [item for item in [v['tag'] for v in sql_result or []] if item not in new_names]
|
|
cleanser = []
|
|
if 0 < len(orphans):
|
|
cleanser = [item for item in sickgear.SHOW_TAGS if item in orphans or item in new_names]
|
|
results += ['An attempt was prevented to remove a show list group name still in use']
|
|
dedupe = {}
|
|
sickgear.SHOW_TAGS = [dedupe.setdefault(item, item) for item in (cleanser + new_names + ['Show List'])
|
|
if item not in dedupe]
|
|
|
|
sickgear.HOME_SEARCH_FOCUS = config.checkbox_to_value(home_search_focus)
|
|
sickgear.USE_IMDB_INFO = config.checkbox_to_value(use_imdb_info)
|
|
sickgear.DISPLAY_FREESPACE = config.checkbox_to_value(display_freespace)
|
|
sickgear.SORT_ARTICLE = config.checkbox_to_value(sort_article)
|
|
sickgear.FUZZY_DATING = config.checkbox_to_value(fuzzy_dating)
|
|
sickgear.TRIM_ZERO = config.checkbox_to_value(trim_zero)
|
|
if date_preset:
|
|
sickgear.DATE_PRESET = date_preset
|
|
if time_preset:
|
|
sickgear.TIME_PRESET_W_SECONDS = time_preset
|
|
sickgear.TIME_PRESET = sickgear.TIME_PRESET_W_SECONDS.replace(':%S', '')
|
|
sickgear.TIMEZONE_DISPLAY = timezone_display
|
|
|
|
# Web interface
|
|
restart = False
|
|
reload_page = False
|
|
if sickgear.WEB_USERNAME != web_username:
|
|
sickgear.WEB_USERNAME = web_username
|
|
reload_page = True
|
|
if set('*') != set(web_password):
|
|
sickgear.WEB_PASSWORD = web_password
|
|
reload_page = True
|
|
|
|
sickgear.CALENDAR_UNPROTECTED = config.checkbox_to_value(calendar_unprotected)
|
|
sickgear.USE_API = config.checkbox_to_value(use_api)
|
|
sickgear.WEB_PORT = config.to_int(web_port)
|
|
# sickgear.WEB_LOG is set in config.change_log_dir()
|
|
|
|
restart |= sickgear.ENABLE_HTTPS != config.checkbox_to_value(enable_https)
|
|
sickgear.ENABLE_HTTPS = config.checkbox_to_value(enable_https)
|
|
if not config.change_https_cert(https_cert):
|
|
results += [
|
|
'Unable to create directory ' + os.path.normpath(https_cert) + ', https cert directory not changed.']
|
|
if not config.change_https_key(https_key):
|
|
results += [
|
|
'Unable to create directory ' + os.path.normpath(https_key) + ', https key directory not changed.']
|
|
|
|
sickgear.WEB_IPV6 = config.checkbox_to_value(web_ipv6)
|
|
sickgear.WEB_IPV64 = config.checkbox_to_value(web_ipv64)
|
|
sickgear.HANDLE_REVERSE_PROXY = config.checkbox_to_value(handle_reverse_proxy)
|
|
sickgear.SEND_SECURITY_HEADERS = config.checkbox_to_value(send_security_headers)
|
|
hosts = ','.join(filter(lambda name: not helpers.re_valid_hostname(with_allowed=False).match(name),
|
|
config.clean_hosts(allowed_hosts).split(',')))
|
|
if not hosts or self.request.host_name in hosts:
|
|
sickgear.ALLOWED_HOSTS = hosts
|
|
sickgear.ALLOW_ANYIP = config.checkbox_to_value(allow_anyip)
|
|
|
|
# Advanced
|
|
sickgear.GIT_REMOTE = git_remote
|
|
sickgear.GIT_PATH = git_path
|
|
sickgear.CPU_PRESET = cpu_preset
|
|
sickgear.ANON_REDIRECT = anon_redirect
|
|
sickgear.ENCRYPTION_VERSION = config.checkbox_to_value(encryption_version)
|
|
sickgear.PROXY_SETTING = proxy_setting
|
|
sg_helpers.PROXY_SETTING = proxy_setting
|
|
sickgear.PROXY_INDEXERS = config.checkbox_to_value(proxy_indexers)
|
|
sickgear.FILE_LOGGING_PRESET = file_logging_preset
|
|
# sickgear.LOG_DIR is set in config.change_log_dir()
|
|
sickgear.BACKUP_DB_ONEDAY = bool(config.checkbox_to_value(backup_db_oneday))
|
|
|
|
logger.log_set_level()
|
|
|
|
sickgear.save_config()
|
|
|
|
if 0 < len(results):
|
|
for v in results:
|
|
logger.error(v)
|
|
ui.notifications.error('Error(s) Saving Configuration',
|
|
'<br>\n'.join(results))
|
|
else:
|
|
ui.notifications.message('Configuration Saved', os.path.join(sickgear.CONFIG_FILE))
|
|
|
|
if restart:
|
|
self.clear_cookie('sickgear-session-%s' % helpers.md5_for_text(sickgear.WEB_PORT))
|
|
self.write('restart')
|
|
reload_page = False
|
|
|
|
if reload_page:
|
|
self.clear_cookie('sickgear-session-%s' % helpers.md5_for_text(sickgear.WEB_PORT))
|
|
self.write('reload')
|
|
|
|
@staticmethod
|
|
def fetch_pullrequests():
|
|
if 'main' == sickgear.BRANCH:
|
|
return json_dumps({'result': 'success', 'pulls': []})
|
|
else:
|
|
try:
|
|
pulls = sickgear.update_software_scheduler.action.list_remote_pulls()
|
|
return json_dumps({'result': 'success', 'pulls': pulls})
|
|
except (BaseException, Exception) as e:
|
|
logger.debug(f'exception msg: {ex(e)}')
|
|
return json_dumps({'result': 'fail'})
|
|
|
|
@staticmethod
|
|
def fetch_branches():
|
|
try:
|
|
branches = sickgear.update_software_scheduler.action.list_remote_branches()
|
|
return json_dumps({'result': 'success', 'branches': branches, 'current': sickgear.BRANCH or 'main'})
|
|
except (BaseException, Exception) as e:
|
|
logger.debug(f'exception msg: {ex(e)}')
|
|
return json_dumps({'result': 'fail'})
|
|
|
|
|
|
class ConfigSearch(Config):
|
|
|
|
def index(self):
|
|
|
|
t = PageTemplate(web_handler=self, file='config_search.tmpl')
|
|
t.submenu = self.config_menu('Search')
|
|
t.using_rls_ignore_words = [(cur_so.tvid_prodid, cur_so.name) for cur_so in sickgear.showList
|
|
if cur_so.rls_ignore_words and cur_so.rls_ignore_words]
|
|
t.using_rls_ignore_words.sort(key=lambda x: x[1], reverse=False)
|
|
t.using_rls_require_words = [(cur_so.tvid_prodid, cur_so.name) for cur_so in sickgear.showList
|
|
if cur_so.rls_require_words and cur_so.rls_require_words]
|
|
t.using_rls_require_words.sort(key=lambda x: x[1], reverse=False)
|
|
t.using_exclude_ignore_words = [(cur_so.tvid_prodid, cur_so.name)
|
|
for cur_so in sickgear.showList if cur_so.rls_global_exclude_ignore]
|
|
t.using_exclude_ignore_words.sort(key=lambda x: x[1], reverse=False)
|
|
t.using_exclude_require_words = [(cur_so.tvid_prodid, cur_so.name)
|
|
for cur_so in sickgear.showList if cur_so.rls_global_exclude_require]
|
|
t.using_exclude_require_words.sort(key=lambda x: x[1], reverse=False)
|
|
t.using_regex = False
|
|
try:
|
|
from sickgear.name_parser.parser import regex
|
|
t.using_regex = None is not regex
|
|
except (BaseException, Exception):
|
|
pass
|
|
return t.respond()
|
|
|
|
def save_search(self, nzb_dir=None, torrent_dir=None,
|
|
recentsearch_interval=None, backlog_period=None, backlog_limited_period=None, backlog_nofull=None,
|
|
recentsearch_frequency=None, backlog_frequency=None, backlog_days=None,
|
|
use_nzbs=None, use_torrents=None, nzb_method=None, torrent_method=None,
|
|
usenet_retention=None, ignore_words=None, require_words=None,
|
|
download_propers=None, propers_webdl_onegrp=None,
|
|
search_unaired=None, unaired_recent_search_only=None, flaresolverr_host=None,
|
|
allow_high_priority=None,
|
|
sab_username=None, sab_password=None, sab_apikey=None, sab_category=None, sab_host=None,
|
|
nzbget_username=None, nzbget_password=None, nzbget_category=None, nzbget_host=None,
|
|
nzbget_use_https=None, nzbget_priority=None, nzbget_parent_map=None,
|
|
torrent_username=None, torrent_password=None, torrent_label=None, torrent_label_var=None,
|
|
torrent_verify_cert=None, torrent_path=None, torrent_seed_time=None, torrent_paused=None,
|
|
torrent_high_bandwidth=None, torrent_host=None):
|
|
|
|
# prevent deprecated var issues from existing ui, delete in future, added 2020.11.07
|
|
if None is recentsearch_interval and None is not recentsearch_frequency:
|
|
recentsearch_interval = recentsearch_frequency
|
|
if None is backlog_period and None is not backlog_frequency:
|
|
backlog_period = backlog_frequency
|
|
if None is backlog_limited_period and None is not backlog_days:
|
|
backlog_limited_period = backlog_days
|
|
|
|
results = []
|
|
|
|
if not config.change_nzb_dir(nzb_dir):
|
|
results += ['Unable to create directory ' + os.path.normpath(nzb_dir) + ', dir not changed.']
|
|
|
|
if not config.change_torrent_dir(torrent_dir):
|
|
results += ['Unable to create directory ' + os.path.normpath(torrent_dir) + ', dir not changed.']
|
|
|
|
config.schedule_recentsearch(recentsearch_interval)
|
|
|
|
old_backlog_period = sickgear.BACKLOG_PERIOD
|
|
config.schedule_backlog(backlog_period)
|
|
sickgear.search_backlog.BacklogSearcher.change_backlog_parts(
|
|
old_backlog_period, sickgear.BACKLOG_PERIOD)
|
|
sickgear.BACKLOG_LIMITED_PERIOD = config.to_int(backlog_limited_period, default=7)
|
|
|
|
sickgear.BACKLOG_NOFULL = bool(config.checkbox_to_value(backlog_nofull))
|
|
if sickgear.BACKLOG_NOFULL:
|
|
my_db = db.DBConnection('cache.db')
|
|
# noinspection SqlConstantCondition
|
|
my_db.action('DELETE FROM backlogparts WHERE 1=1')
|
|
|
|
sickgear.USE_NZBS = config.checkbox_to_value(use_nzbs)
|
|
sickgear.USE_TORRENTS = config.checkbox_to_value(use_torrents)
|
|
|
|
sickgear.NZB_METHOD = nzb_method
|
|
sickgear.TORRENT_METHOD = torrent_method
|
|
sickgear.USENET_RETENTION = config.to_int(usenet_retention, default=500)
|
|
|
|
sickgear.IGNORE_WORDS, sickgear.IGNORE_WORDS_REGEX = helpers.split_word_str(ignore_words
|
|
if ignore_words else '')
|
|
sickgear.REQUIRE_WORDS, sickgear.REQUIRE_WORDS_REGEX = helpers.split_word_str(require_words
|
|
if require_words else '')
|
|
|
|
clean_ignore_require_words()
|
|
|
|
config.schedule_download_propers(config.checkbox_to_value(download_propers))
|
|
sickgear.PROPERS_WEBDL_ONEGRP = config.checkbox_to_value(propers_webdl_onegrp)
|
|
|
|
sickgear.SEARCH_UNAIRED = bool(config.checkbox_to_value(search_unaired))
|
|
sickgear.UNAIRED_RECENT_SEARCH_ONLY = bool(config.checkbox_to_value(unaired_recent_search_only,
|
|
value_off=1, value_on=0))
|
|
|
|
sickgear.FLARESOLVERR_HOST = config.clean_url(flaresolverr_host)
|
|
sg_helpers.FLARESOLVERR_HOST = sickgear.FLARESOLVERR_HOST
|
|
|
|
sickgear.ALLOW_HIGH_PRIORITY = config.checkbox_to_value(allow_high_priority)
|
|
|
|
sickgear.SAB_USERNAME = sab_username
|
|
if set('*') != set(sab_password):
|
|
sickgear.SAB_PASSWORD = sab_password
|
|
key = sab_apikey.strip()
|
|
if not starify(key, True):
|
|
sickgear.SAB_APIKEY = key
|
|
sickgear.SAB_CATEGORY = sab_category
|
|
sickgear.SAB_HOST = config.clean_url(sab_host)
|
|
|
|
sickgear.NZBGET_USERNAME = nzbget_username
|
|
if set('*') != set(nzbget_password):
|
|
sickgear.NZBGET_PASSWORD = nzbget_password
|
|
sickgear.NZBGET_CATEGORY = nzbget_category
|
|
sickgear.NZBGET_HOST = config.clean_host(nzbget_host)
|
|
sickgear.NZBGET_USE_HTTPS = config.checkbox_to_value(nzbget_use_https)
|
|
sickgear.NZBGET_PRIORITY = config.to_int(nzbget_priority, default=100)
|
|
sickgear.NZBGET_MAP = config.kv_csv(nzbget_parent_map)
|
|
|
|
sickgear.TORRENT_USERNAME = torrent_username
|
|
if set('*') != set(torrent_password):
|
|
sickgear.TORRENT_PASSWORD = torrent_password
|
|
sickgear.TORRENT_LABEL = torrent_label
|
|
sickgear.TORRENT_LABEL_VAR = config.to_int((0, torrent_label_var)['rtorrent' == torrent_method], 1)
|
|
if not (0 <= sickgear.TORRENT_LABEL_VAR <= 5):
|
|
logger.debug('Setting rTorrent custom%s is not 0-5, defaulting to custom1' % torrent_label_var)
|
|
sickgear.TORRENT_LABEL_VAR = 1
|
|
sickgear.TORRENT_VERIFY_CERT = config.checkbox_to_value(torrent_verify_cert)
|
|
sickgear.TORRENT_PATH = torrent_path
|
|
sickgear.TORRENT_SEED_TIME = config.to_int(torrent_seed_time, 0)
|
|
sickgear.TORRENT_PAUSED = config.checkbox_to_value(torrent_paused)
|
|
sickgear.TORRENT_HIGH_BANDWIDTH = config.checkbox_to_value(torrent_high_bandwidth)
|
|
sickgear.TORRENT_HOST = config.clean_url(torrent_host)
|
|
|
|
sickgear.save_config()
|
|
|
|
if 0 < len(results):
|
|
for x in results:
|
|
logger.error(x)
|
|
ui.notifications.error('Error(s) Saving Configuration',
|
|
'<br>\n'.join(results))
|
|
else:
|
|
ui.notifications.message('Configuration Saved', os.path.join(sickgear.CONFIG_FILE))
|
|
|
|
self.redirect('/config/search/')
|
|
|
|
|
|
class ConfigMediaProcess(Config):
|
|
|
|
def index(self):
|
|
|
|
t = PageTemplate(web_handler=self, file='config_postProcessing.tmpl')
|
|
t.submenu = self.config_menu('Processing')
|
|
return t.respond()
|
|
|
|
def save_post_processing(
|
|
self, tv_download_dir=None, process_method=None, process_automatically=None, mediaprocess_interval=None,
|
|
postpone_if_sync_files=None, process_positive_log=None, extra_scripts='', sg_extra_scripts='',
|
|
unpack=None, skip_removed_files=None, move_associated_files=None, nfo_rename=None,
|
|
rename_episodes=None, rename_tba_episodes=None, rename_name_changed_episodes=None,
|
|
airdate_episodes=None, use_failed_downloads=None, delete_failed=None,
|
|
naming_pattern=None, naming_multi_ep=None, naming_strip_year=None,
|
|
naming_custom_abd=None, naming_abd_pattern=None,
|
|
naming_custom_sports=None, naming_sports_pattern=None,
|
|
naming_custom_anime=None, naming_anime_pattern=None,naming_anime_multi_ep=None, naming_anime=None,
|
|
kodi_data=None, mede8er_data=None, xbmc_data=None, mediabrowser_data=None,
|
|
sony_ps3_data=None, tivo_data=None, wdtv_data=None, xbmc_12plus_data=None,
|
|
keep_processed_dir=None,
|
|
**kwargs): # kwargs picks up deprecated vars sent from legacy UIs
|
|
|
|
results = []
|
|
|
|
if not config.change_tv_download_dir(tv_download_dir):
|
|
results += ['Unable to create directory ' + os.path.normpath(tv_download_dir) + ', dir not changed.']
|
|
|
|
new_val = config.checkbox_to_value(process_automatically)
|
|
sickgear.PROCESS_AUTOMATICALLY = new_val
|
|
config.schedule_mediaprocess(mediaprocess_interval)
|
|
|
|
if unpack:
|
|
if 'not supported' != self.is_rar_supported():
|
|
sickgear.UNPACK = config.checkbox_to_value(unpack)
|
|
else:
|
|
sickgear.UNPACK = 0
|
|
results.append('Unpacking Not Supported, disabling unpack setting')
|
|
else:
|
|
sickgear.UNPACK = config.checkbox_to_value(unpack)
|
|
|
|
sickgear.KEEP_PROCESSED_DIR = config.checkbox_to_value(keep_processed_dir)
|
|
sickgear.PROCESS_METHOD = process_method
|
|
sickgear.EXTRA_SCRIPTS = [x.strip() for x in extra_scripts.split('|') if x.strip()]
|
|
sickgear.SG_EXTRA_SCRIPTS = [x.strip() for x in sg_extra_scripts.split('|') if x.strip()]
|
|
sickgear.RENAME_EPISODES = config.checkbox_to_value(rename_episodes)
|
|
sickgear.RENAME_TBA_EPISODES = config.checkbox_to_value(rename_tba_episodes)
|
|
sickgear.RENAME_NAME_CHANGED_EPISODES = config.checkbox_to_value(rename_name_changed_episodes)
|
|
sickgear.AIRDATE_EPISODES = config.checkbox_to_value(airdate_episodes)
|
|
sickgear.MOVE_ASSOCIATED_FILES = config.checkbox_to_value(move_associated_files)
|
|
sickgear.POSTPONE_IF_SYNC_FILES = config.checkbox_to_value(postpone_if_sync_files)
|
|
sickgear.PROCESS_POSITIVE_LOG = config.checkbox_to_value(process_positive_log)
|
|
sickgear.NAMING_CUSTOM_ABD = config.checkbox_to_value(naming_custom_abd)
|
|
sickgear.NAMING_CUSTOM_SPORTS = config.checkbox_to_value(naming_custom_sports)
|
|
sickgear.NAMING_CUSTOM_ANIME = config.checkbox_to_value(naming_custom_anime)
|
|
sickgear.NAMING_STRIP_YEAR = config.checkbox_to_value(naming_strip_year)
|
|
sickgear.USE_FAILED_DOWNLOADS = config.checkbox_to_value(use_failed_downloads)
|
|
sickgear.DELETE_FAILED = config.checkbox_to_value(delete_failed)
|
|
sickgear.SKIP_REMOVED_FILES = config.minimax(skip_removed_files, IGNORED, 1, IGNORED)
|
|
sickgear.NFO_RENAME = config.checkbox_to_value(nfo_rename)
|
|
|
|
sickgear.METADATA_XBMC = xbmc_data
|
|
sickgear.METADATA_XBMC_12PLUS = xbmc_12plus_data
|
|
sickgear.METADATA_MEDIABROWSER = mediabrowser_data
|
|
sickgear.METADATA_PS3 = sony_ps3_data
|
|
sickgear.METADATA_WDTV = wdtv_data
|
|
sickgear.METADATA_TIVO = tivo_data
|
|
sickgear.METADATA_MEDE8ER = mede8er_data
|
|
sickgear.METADATA_KODI = kodi_data
|
|
|
|
sickgear.metadata_provider_dict['XBMC'].set_config(sickgear.METADATA_XBMC)
|
|
sickgear.metadata_provider_dict['XBMC 12+'].set_config(sickgear.METADATA_XBMC_12PLUS)
|
|
sickgear.metadata_provider_dict['MediaBrowser'].set_config(sickgear.METADATA_MEDIABROWSER)
|
|
sickgear.metadata_provider_dict['Sony PS3'].set_config(sickgear.METADATA_PS3)
|
|
sickgear.metadata_provider_dict['WDTV'].set_config(sickgear.METADATA_WDTV)
|
|
sickgear.metadata_provider_dict['TIVO'].set_config(sickgear.METADATA_TIVO)
|
|
sickgear.metadata_provider_dict['Mede8er'].set_config(sickgear.METADATA_MEDE8ER)
|
|
sickgear.metadata_provider_dict['Kodi'].set_config(sickgear.METADATA_KODI)
|
|
|
|
if 'invalid' != self.is_naming_valid(naming_pattern, naming_multi_ep, anime_type=naming_anime):
|
|
sickgear.NAMING_PATTERN = naming_pattern
|
|
sickgear.NAMING_MULTI_EP = int(naming_multi_ep)
|
|
sickgear.NAMING_ANIME = int(naming_anime)
|
|
sickgear.NAMING_FORCE_FOLDERS = naming.check_force_season_folders()
|
|
else:
|
|
if int(naming_anime) in [1, 2]:
|
|
results.append('You tried saving an invalid anime naming config, not saving your naming settings')
|
|
else:
|
|
results.append('You tried saving an invalid naming config, not saving your naming settings')
|
|
|
|
if 'invalid' != self.is_naming_valid(naming_anime_pattern, naming_anime_multi_ep, anime_type=naming_anime):
|
|
sickgear.NAMING_ANIME_PATTERN = naming_anime_pattern
|
|
sickgear.NAMING_ANIME_MULTI_EP = int(naming_anime_multi_ep)
|
|
sickgear.NAMING_ANIME = int(naming_anime)
|
|
sickgear.NAMING_FORCE_FOLDERS = naming.check_force_season_folders()
|
|
else:
|
|
if int(naming_anime) in [1, 2]:
|
|
results.append('You tried saving an invalid anime naming config, not saving your naming settings')
|
|
else:
|
|
results.append('You tried saving an invalid naming config, not saving your naming settings')
|
|
|
|
if 'invalid' != self.is_naming_valid(naming_abd_pattern, abd=True):
|
|
sickgear.NAMING_ABD_PATTERN = naming_abd_pattern
|
|
else:
|
|
results.append(
|
|
'You tried saving an invalid air-by-date naming config, not saving your air-by-date settings')
|
|
|
|
if 'invalid' != self.is_naming_valid(naming_sports_pattern, sports=True):
|
|
sickgear.NAMING_SPORTS_PATTERN = naming_sports_pattern
|
|
else:
|
|
results.append(
|
|
'You tried saving an invalid sports naming config, not saving your sports settings')
|
|
|
|
sickgear.save_config()
|
|
|
|
if 0 < len(results):
|
|
for x in results:
|
|
logger.error(x)
|
|
ui.notifications.error('Error(s) Saving Configuration',
|
|
'<br>\n'.join(results))
|
|
else:
|
|
ui.notifications.message('Configuration Saved', os.path.join(sickgear.CONFIG_FILE))
|
|
|
|
self.redirect('/config/media-process/')
|
|
|
|
@staticmethod
|
|
def test_naming(pattern=None, multi=None, abd=False, sports=False, anime=False, anime_type=None):
|
|
|
|
if None is not multi:
|
|
multi = int(multi)
|
|
|
|
if None is not anime_type:
|
|
anime_type = int(anime_type)
|
|
|
|
result = naming.test_name(pattern, multi, abd, sports, anime, anime_type)
|
|
|
|
result = os.path.join(result['dir'], result['name'])
|
|
|
|
return result
|
|
|
|
@staticmethod
|
|
def is_naming_valid(pattern=None, multi=None, abd=False, sports=False, anime_type=None):
|
|
if None is pattern:
|
|
return 'invalid'
|
|
|
|
if None is not multi:
|
|
multi = int(multi)
|
|
|
|
if None is not anime_type:
|
|
anime_type = int(anime_type)
|
|
|
|
# air by date shows just need one check, we don't need to worry about season folders
|
|
if abd:
|
|
is_valid = naming.check_valid_abd_naming(pattern)
|
|
require_season_folders = False
|
|
|
|
# sport shows just need one check, we don't need to worry about season folders
|
|
elif sports:
|
|
is_valid = naming.check_valid_sports_naming(pattern)
|
|
require_season_folders = False
|
|
|
|
else:
|
|
# check validity of single and multi ep cases for the whole path
|
|
is_valid = naming.check_valid_naming(pattern, multi, anime_type)
|
|
|
|
# check validity of single and multi ep cases for only the file name
|
|
require_season_folders = naming.check_force_season_folders(pattern, multi, anime_type)
|
|
|
|
if is_valid and not require_season_folders:
|
|
return 'valid'
|
|
elif is_valid and require_season_folders:
|
|
return 'seasonfolders'
|
|
|
|
return 'invalid'
|
|
|
|
@staticmethod
|
|
def is_rar_supported():
|
|
"""
|
|
Test Packing Support:
|
|
"""
|
|
|
|
try:
|
|
if 'win32' == sys.platform:
|
|
rarfile.UNRAR_TOOL = os.path.join(sickgear.PROG_DIR, 'lib', 'rarfile', 'UnRAR.exe')
|
|
rar_path = os.path.join(sickgear.PROG_DIR, 'lib', 'rarfile', 'test.rar')
|
|
if 'This is only a test.' == decode_str(rarfile.RarFile(rar_path).read(r'test/test.txt')):
|
|
return 'supported'
|
|
msg = 'Could not read test file content'
|
|
except (BaseException, Exception) as e:
|
|
msg = ex(e)
|
|
|
|
logger.error(f'Rar Not Supported: {msg}')
|
|
return 'not supported'
|
|
|
|
|
|
class ConfigProviders(Config):
|
|
|
|
def index(self):
|
|
t = PageTemplate(web_handler=self, file='config_providers.tmpl')
|
|
t.submenu = self.config_menu('Providers')
|
|
return t.respond()
|
|
|
|
@staticmethod
|
|
def can_add_newznab_provider(name, url):
|
|
if not name or not url:
|
|
return json_dumps({'error': 'No Provider Name or url specified'})
|
|
|
|
provider_dict = dict(zip([sickgear.providers.generic_provider_name(x.get_id())
|
|
for x in sickgear.newznab_providers], sickgear.newznab_providers))
|
|
provider_url_dict = dict(zip([sickgear.providers.generic_provider_url(x.url)
|
|
for x in sickgear.newznab_providers], sickgear.newznab_providers))
|
|
|
|
temp_provider = newznab.NewznabProvider(name, config.clean_url(url))
|
|
|
|
e_p = provider_dict.get(sickgear.providers.generic_provider_name(temp_provider.get_id()), None) or \
|
|
provider_url_dict.get(sickgear.providers.generic_provider_url(temp_provider.url), None)
|
|
|
|
if e_p:
|
|
return json_dumps({'error': 'Provider already exists as %s' % e_p.name})
|
|
|
|
return json_dumps({'success': temp_provider.get_id()})
|
|
|
|
@staticmethod
|
|
def get_newznab_categories(name, url, key):
|
|
"""
|
|
Retrieves a list of possible categories with category id's
|
|
Using the default url/api?cat
|
|
https://yournewznaburl.com/api?t=caps&apikey=yourapikey
|
|
"""
|
|
error = not name and 'Name' or not url and 'Url' or not key and 'Apikey' or ''
|
|
if error:
|
|
error = '\nNo provider %s specified' % error
|
|
return json_dumps({'success': False, 'error': error})
|
|
|
|
if name in [n.name for n in sickgear.newznab_providers if n.url == url]:
|
|
provider = [n for n in sickgear.newznab_providers if n.name == name][0]
|
|
tv_categories = provider.clean_newznab_categories(provider.all_cats)
|
|
state = provider.is_enabled()
|
|
else:
|
|
providers = dict(zip([x.get_id() for x in sickgear.newznab_providers], sickgear.newznab_providers))
|
|
temp_provider = newznab.NewznabProvider(name, url, key)
|
|
if None is not key and starify(key, True):
|
|
temp_provider.key = providers[temp_provider.get_id()].key
|
|
|
|
tv_categories = temp_provider.clean_newznab_categories(temp_provider.all_cats)
|
|
state = False
|
|
|
|
return json_dumps({'success': True, 'tv_categories': tv_categories, 'state': state, 'error': ''})
|
|
|
|
@staticmethod
|
|
def can_add_torrent_rss_provider(name, url, cookies):
|
|
if not name:
|
|
return json_dumps({'error': 'Invalid name specified'})
|
|
|
|
provider_dict = dict(
|
|
zip([x.get_id() for x in sickgear.torrent_rss_providers], sickgear.torrent_rss_providers))
|
|
|
|
temp_provider = rsstorrent.TorrentRssProvider(name, url, cookies)
|
|
|
|
if temp_provider.get_id() in provider_dict:
|
|
return json_dumps({'error': 'A provider exists as [%s]' % provider_dict[temp_provider.get_id()].name})
|
|
else:
|
|
(succ, errMsg) = temp_provider.validate_feed()
|
|
if succ:
|
|
return json_dumps({'success': temp_provider.get_id()})
|
|
|
|
return json_dumps({'error': errMsg})
|
|
|
|
@staticmethod
|
|
def check_providers_ping():
|
|
for p in sickgear.providers.sorted_sources():
|
|
if getattr(p, 'ping_iv', None):
|
|
if p.is_active() and (p.get_id() not in sickgear.provider_ping_thread_pool
|
|
or not sickgear.provider_ping_thread_pool[p.get_id()].is_alive()):
|
|
# noinspection PyProtectedMember
|
|
sickgear.provider_ping_thread_pool[p.get_id()] = threading.Thread(
|
|
name='PING-PROVIDER %s' % p.name, target=p._ping)
|
|
sickgear.provider_ping_thread_pool[p.get_id()].start()
|
|
elif not p.is_active() and p.get_id() in sickgear.provider_ping_thread_pool:
|
|
sickgear.provider_ping_thread_pool[p.get_id()].stop = True
|
|
try:
|
|
sickgear.provider_ping_thread_pool[p.get_id()].join(120)
|
|
if not sickgear.provider_ping_thread_pool[p.get_id()].is_alive():
|
|
sickgear.provider_ping_thread_pool.pop(p.get_id())
|
|
except RuntimeError:
|
|
pass
|
|
|
|
# stop removed providers
|
|
prov = [n.get_id() for n in sickgear.providers.sorted_sources()]
|
|
for p in [x for x in sickgear.provider_ping_thread_pool if x not in prov]:
|
|
sickgear.provider_ping_thread_pool[p].stop = True
|
|
try:
|
|
sickgear.provider_ping_thread_pool[p].join(120)
|
|
if not sickgear.provider_ping_thread_pool[p].is_alive():
|
|
sickgear.provider_ping_thread_pool.pop(p)
|
|
except RuntimeError:
|
|
pass
|
|
|
|
def save_providers(self, newznab_string='', torrentrss_string='', provider_order=None, **kwargs):
|
|
|
|
results = []
|
|
provider_list = []
|
|
|
|
# add all the newznab info we have into our list
|
|
newznab_sources = dict(zip([x.get_id() for x in sickgear.newznab_providers], sickgear.newznab_providers))
|
|
active_ids = []
|
|
reload_page = False
|
|
if newznab_string:
|
|
for curNewznabProviderStr in newznab_string.split('!!!'):
|
|
|
|
if not curNewznabProviderStr:
|
|
continue
|
|
|
|
cur_name, cur_url, cur_key, cur_cat = curNewznabProviderStr.split('|')
|
|
cur_url = config.clean_url(cur_url)
|
|
cur_key = cur_key.strip()
|
|
|
|
if starify(cur_key, True):
|
|
cur_key = ''
|
|
|
|
# correct user entry mistakes
|
|
test_url = cur_url.lower()
|
|
if 'nzbs2go' in test_url and test_url.endswith('.com/') or 'api/v1/api' in test_url:
|
|
cur_url = 'https://nzbs2go.com/api/v1/'
|
|
|
|
new_provider = newznab.NewznabProvider(cur_name, cur_url, key=cur_key)
|
|
|
|
cur_id = new_provider.get_id()
|
|
|
|
# if it already exists then update it
|
|
if cur_id in newznab_sources:
|
|
nzb_src = newznab_sources[cur_id]
|
|
|
|
nzb_src.name, nzb_src.url, nzb_src.cat_ids = cur_name, cur_url, cur_cat
|
|
|
|
if cur_key:
|
|
nzb_src.key = cur_key
|
|
|
|
# a 0 in the key spot indicates that no key is needed
|
|
nzb_src.needs_auth = '0' != cur_key
|
|
|
|
attr = 'filter'
|
|
if hasattr(nzb_src, attr):
|
|
setattr(nzb_src, attr,
|
|
[k for k in nzb_src.may_filter
|
|
if config.checkbox_to_value(kwargs.get('%s_filter_%s' % (cur_id, k)))])
|
|
|
|
for attr in filter(lambda a: hasattr(nzb_src, a), [
|
|
'search_fallback', 'enable_recentsearch', 'enable_backlog', 'enable_scheduled_backlog',
|
|
'scene_only', 'scene_loose', 'scene_loose_active', 'scene_rej_nuked', 'scene_nuked_active'
|
|
]):
|
|
setattr(nzb_src, attr, config.checkbox_to_value(kwargs.get(cur_id + '_' + attr)))
|
|
|
|
for attr in ['scene_or_contain', 'search_mode']:
|
|
attr_check = '%s_%s' % (cur_id, attr)
|
|
if attr_check in kwargs:
|
|
setattr(nzb_src, attr, str(kwargs.get(attr_check) or '').strip())
|
|
else:
|
|
new_provider.enabled = True
|
|
_ = new_provider.caps # when adding a custom, trigger server_type update
|
|
new_provider.enabled = False
|
|
sickgear.newznab_providers.append(new_provider)
|
|
|
|
active_ids.append(cur_id)
|
|
|
|
# delete anything that is missing
|
|
if sickgear.USE_NZBS:
|
|
for source in [x for x in sickgear.newznab_providers if x.get_id() not in active_ids]:
|
|
sickgear.newznab_providers.remove(source)
|
|
|
|
# add all the torrent RSS info we have into our list
|
|
torrent_rss_sources = dict(zip([x.get_id() for x in sickgear.torrent_rss_providers],
|
|
sickgear.torrent_rss_providers))
|
|
active_ids = []
|
|
if torrentrss_string:
|
|
for curTorrentRssProviderStr in torrentrss_string.split('!!!'):
|
|
|
|
if not curTorrentRssProviderStr:
|
|
continue
|
|
|
|
cur_name, cur_url, cur_cookies = curTorrentRssProviderStr.split('|')
|
|
cur_url = config.clean_url(cur_url, False)
|
|
|
|
if starify(cur_cookies, True):
|
|
cur_cookies = ''
|
|
|
|
new_provider = rsstorrent.TorrentRssProvider(cur_name, cur_url, cur_cookies)
|
|
|
|
cur_id = new_provider.get_id()
|
|
|
|
# if it already exists then update it
|
|
if cur_id in torrent_rss_sources:
|
|
torrss_src = torrent_rss_sources[cur_id]
|
|
|
|
torrss_src.name = cur_name
|
|
torrss_src.url = cur_url
|
|
if cur_cookies:
|
|
torrss_src.cookies = cur_cookies
|
|
|
|
for attr in ['scene_only', 'scene_loose', 'scene_loose_active',
|
|
'scene_rej_nuked', 'scene_nuked_active']:
|
|
setattr(torrss_src, attr, config.checkbox_to_value(kwargs.get(cur_id + '_' + attr)))
|
|
|
|
for attr in ['scene_or_contain']:
|
|
attr_check = '%s_%s' % (cur_id, attr)
|
|
if attr_check in kwargs:
|
|
setattr(torrss_src, attr, str(kwargs.get(attr_check) or '').strip())
|
|
else:
|
|
sickgear.torrent_rss_providers.append(new_provider)
|
|
|
|
active_ids.append(cur_id)
|
|
|
|
# delete anything that is missing
|
|
if sickgear.USE_TORRENTS:
|
|
for source in [x for x in sickgear.torrent_rss_providers if x.get_id() not in active_ids]:
|
|
sickgear.torrent_rss_providers.remove(source)
|
|
|
|
# enable/disable states of source providers
|
|
provider_str_list = provider_order.split()
|
|
sources = dict(zip([x.get_id() for x in sickgear.providers.sorted_sources()],
|
|
sickgear.providers.sorted_sources()))
|
|
for cur_src_str in provider_str_list:
|
|
src_name, src_enabled = cur_src_str.split(':')
|
|
|
|
provider_list.append(src_name)
|
|
src_enabled = bool(config.to_int(src_enabled))
|
|
|
|
if src_name in sources and '' != getattr(sources[src_name], 'enabled', '') \
|
|
and sources[src_name].is_enabled() != src_enabled:
|
|
if isinstance(sources[src_name], sickgear.providers.newznab.NewznabProvider) and \
|
|
not sources[src_name].enabled and src_enabled:
|
|
reload_page = True
|
|
sources[src_name].enabled = src_enabled
|
|
if not reload_page and sickgear.GenericProvider.TORRENT == sources[src_name].providerType:
|
|
reload_page = True
|
|
|
|
if src_name in newznab_sources:
|
|
if not newznab_sources[src_name].enabled and src_enabled:
|
|
reload_page = True
|
|
newznab_sources[src_name].enabled = src_enabled
|
|
elif src_name in torrent_rss_sources:
|
|
torrent_rss_sources[src_name].enabled = src_enabled
|
|
|
|
# update torrent source settings
|
|
for torrent_src in [src for src in sickgear.providers.sorted_sources()
|
|
if sickgear.GenericProvider.TORRENT == src.providerType]: # type: TorrentProvider
|
|
src_id_prefix = torrent_src.get_id() + '_'
|
|
|
|
attr = 'url_edit'
|
|
if getattr(torrent_src, attr, None):
|
|
url_edit = ','.join(set(['%s' % url.strip() for url in kwargs.get(
|
|
src_id_prefix + attr, '').split(',')]))
|
|
torrent_src.url_home = ([url_edit], [])[not url_edit]
|
|
|
|
for attr in [x for x in ['password', 'api_key', 'passkey', 'digest', 'hash'] if hasattr(torrent_src, x)]:
|
|
key = str(kwargs.get(src_id_prefix + attr, '')).strip()
|
|
if 'password' == attr:
|
|
set('*') != set(key) and setattr(torrent_src, attr, key)
|
|
elif not starify(key, True):
|
|
setattr(torrent_src, attr, key)
|
|
|
|
for attr in filter(lambda a: hasattr(torrent_src, a), [
|
|
'username', 'uid', '_seed_ratio', 'scene_or_contain'
|
|
]):
|
|
setattr(torrent_src, attr, str(kwargs.get(src_id_prefix + attr.replace('_seed_', ''), '')).strip())
|
|
|
|
for attr in filter(lambda a: hasattr(torrent_src, a), [
|
|
'minseed', 'minleech', 'seed_time'
|
|
]):
|
|
setattr(torrent_src, attr, config.to_int(str(kwargs.get(src_id_prefix + attr, '')).strip()))
|
|
|
|
attr = 'filter'
|
|
if hasattr(torrent_src, attr) and torrent_src.may_filter:
|
|
setattr(torrent_src, attr,
|
|
[k for k in getattr(torrent_src, 'may_filter', 'nop')
|
|
if config.checkbox_to_value(kwargs.get('%sfilter_%s' % (src_id_prefix, k)))])
|
|
|
|
for attr in filter(lambda a: hasattr(torrent_src, a), [
|
|
'confirmed', 'freeleech', 'reject_m2ts', 'use_after_get_data', 'enable_recentsearch',
|
|
'enable_backlog', 'search_fallback', 'enable_scheduled_backlog',
|
|
'scene_only', 'scene_loose', 'scene_loose_active',
|
|
'scene_rej_nuked', 'scene_nuked_active'
|
|
]):
|
|
setattr(torrent_src, attr, config.checkbox_to_value(kwargs.get(src_id_prefix + attr)))
|
|
|
|
for attr, default in filter(lambda arg: hasattr(torrent_src, arg[0]), [
|
|
('search_mode', 'eponly'),
|
|
]):
|
|
setattr(torrent_src, attr, str(kwargs.get(src_id_prefix + attr) or default).strip())
|
|
|
|
# update nzb source settings
|
|
for nzb_src in [src for src in sickgear.providers.sorted_sources() if
|
|
sickgear.GenericProvider.NZB == src.providerType]:
|
|
src_id_prefix = nzb_src.get_id() + '_'
|
|
|
|
for attr in [x for x in ['api_key', 'digest'] if hasattr(nzb_src, x)]:
|
|
key = str(kwargs.get(src_id_prefix + attr, '')).strip()
|
|
if not starify(key, True):
|
|
setattr(nzb_src, attr, key)
|
|
|
|
attr = 'username'
|
|
if hasattr(nzb_src, attr):
|
|
setattr(nzb_src, attr, str(kwargs.get(src_id_prefix + attr, '')).strip() or None)
|
|
|
|
attr = 'enable_recentsearch'
|
|
if hasattr(nzb_src, attr):
|
|
setattr(nzb_src, attr, config.checkbox_to_value(kwargs.get(src_id_prefix + attr)) or
|
|
not getattr(nzb_src, 'supports_backlog', True))
|
|
|
|
for attr in filter(lambda a: hasattr(nzb_src, a),
|
|
['search_fallback', 'enable_backlog', 'enable_scheduled_backlog',
|
|
'scene_only', 'scene_loose', 'scene_loose_active',
|
|
'scene_rej_nuked', 'scene_nuked_active']):
|
|
setattr(nzb_src, attr, config.checkbox_to_value(kwargs.get(src_id_prefix + attr)))
|
|
|
|
for (attr, default) in [('scene_or_contain', ''), ('search_mode', 'eponly')]:
|
|
if hasattr(nzb_src, attr):
|
|
setattr(nzb_src, attr, str(kwargs.get(src_id_prefix + attr) or default).strip())
|
|
|
|
sickgear.NEWZNAB_DATA = '!!!'.join([x.config_str() for x in sickgear.newznab_providers])
|
|
sickgear.PROVIDER_ORDER = provider_list
|
|
|
|
helpers.clear_unused_providers()
|
|
|
|
sickgear.save_config()
|
|
|
|
cp = threading.Thread(name='Check-Ping-Providers', target=self.check_providers_ping)
|
|
cp.start()
|
|
|
|
if 0 < len(results):
|
|
for x in results:
|
|
logger.error(x)
|
|
ui.notifications.error('Error(s) Saving Configuration', '<br>\n'.join(results))
|
|
else:
|
|
ui.notifications.message('Configuration Saved', os.path.join(sickgear.CONFIG_FILE))
|
|
|
|
if reload_page:
|
|
self.write('reload')
|
|
else:
|
|
self.redirect('/config/providers/')
|
|
|
|
|
|
class ConfigNotifications(Config):
|
|
|
|
def index(self):
|
|
t = PageTemplate(web_handler=self, file='config_notifications.tmpl')
|
|
t.submenu = self.config_menu('Notifications')
|
|
t.root_dirs = []
|
|
if sickgear.ROOT_DIRS:
|
|
root_pieces = sickgear.ROOT_DIRS.split('|')
|
|
root_default = helpers.try_int(root_pieces[0], None)
|
|
for i, location in enumerate(root_pieces[1:]):
|
|
t.root_dirs.append({'root_def': root_default and i == root_default,
|
|
'loc': location,
|
|
'b64': decode_str(base64.urlsafe_b64encode(decode_bytes(location)))})
|
|
return t.respond()
|
|
|
|
def save_notifications(
|
|
self,
|
|
use_emby=None, emby_update_library=None, emby_watched_interval=None, emby_parent_maps=None,
|
|
emby_host=None, emby_apikey=None,
|
|
use_kodi=None, kodi_always_on=None, kodi_update_library=None, kodi_update_full=None,
|
|
kodi_update_onlyfirst=None, kodi_parent_maps=None, kodi_host=None, kodi_username=None, kodi_password=None,
|
|
kodi_notify_onsnatch=None, kodi_notify_ondownload=None, kodi_notify_onsubtitledownload=None,
|
|
use_plex=None, plex_update_library=None, plex_watched_interval=None, plex_parent_maps=None,
|
|
plex_username=None, plex_password=None, plex_server_host=None,
|
|
plex_notify_onsnatch=None, plex_notify_ondownload=None, plex_notify_onsubtitledownload=None, plex_host=None,
|
|
use_nmj=None, nmj_host=None, nmj_database=None, nmj_mount=None,
|
|
use_nmjv2=None, nmjv2_host=None, nmjv2_dbloc=None, nmjv2_database=None,
|
|
use_synoindex=None, use_synologynotifier=None, synologynotifier_notify_onsnatch=None,
|
|
synologynotifier_notify_ondownload=None, synologynotifier_notify_onsubtitledownload=None,
|
|
use_pytivo=None, pytivo_host=None, pytivo_share_name=None, pytivo_tivo_name=None,
|
|
|
|
use_boxcar2=None, boxcar2_notify_onsnatch=None, boxcar2_notify_ondownload=None,
|
|
boxcar2_notify_onsubtitledownload=None, boxcar2_access_token=None, boxcar2_sound=None,
|
|
use_pushbullet=None, pushbullet_notify_onsnatch=None, pushbullet_notify_ondownload=None,
|
|
pushbullet_notify_onsubtitledownload=None, pushbullet_access_token=None, pushbullet_device_iden=None,
|
|
use_pushover=None, pushover_notify_onsnatch=None, pushover_notify_ondownload=None,
|
|
pushover_notify_onsubtitledownload=None, pushover_userkey=None, pushover_apikey=None,
|
|
pushover_priority=None, pushover_device=None, pushover_sound=None,
|
|
use_growl=None, growl_notify_onsnatch=None, growl_notify_ondownload=None,
|
|
growl_notify_onsubtitledownload=None, growl_host=None,
|
|
use_prowl=None, prowl_notify_onsnatch=None, prowl_notify_ondownload=None,
|
|
prowl_notify_onsubtitledownload=None, prowl_api=None, prowl_priority=0,
|
|
use_libnotify=None, libnotify_notify_onsnatch=None, libnotify_notify_ondownload=None,
|
|
libnotify_notify_onsubtitledownload=None,
|
|
|
|
use_trakt=None,
|
|
# trakt_pin=None, trakt_remove_watchlist=None, trakt_use_watchlist=None, trakt_method_add=None,
|
|
# trakt_start_paused=None, trakt_sync=None, trakt_default_indexer=None, trakt_remove_serieslist=None,
|
|
# trakt_collection=None, trakt_accounts=None,
|
|
use_slack=None, slack_notify_onsnatch=None, slack_notify_ondownload=None,
|
|
slack_notify_onsubtitledownload=None, slack_access_token=None, slack_channel=None,
|
|
slack_as_authed=None, slack_bot_name=None, slack_icon_url=None,
|
|
use_discord=None, discord_notify_onsnatch=None, discord_notify_ondownload=None,
|
|
discord_notify_onsubtitledownload=None, discord_access_token=None,
|
|
discord_as_authed=None, discord_username=None, discord_icon_url=None,
|
|
discord_as_tts=None,
|
|
use_gitter=None, gitter_notify_onsnatch=None, gitter_notify_ondownload=None,
|
|
gitter_notify_onsubtitledownload=None, gitter_access_token=None, gitter_room=None,
|
|
use_telegram=None, telegram_notify_onsnatch=None, telegram_notify_ondownload=None,
|
|
telegram_notify_onsubtitledownload=None, telegram_access_token=None, telegram_chatid=None,
|
|
telegram_send_image=None, telegram_quiet=None,
|
|
use_email=None, email_notify_onsnatch=None, email_notify_ondownload=None,
|
|
email_notify_onsubtitledownload=None, email_host=None, email_port=25, email_from=None,
|
|
email_tls=None, email_user=None, email_password=None, email_list=None,
|
|
# email_show_list=None, email_show=None,
|
|
**kwargs):
|
|
|
|
results = []
|
|
|
|
sickgear.USE_EMBY = config.checkbox_to_value(use_emby)
|
|
sickgear.EMBY_UPDATE_LIBRARY = config.checkbox_to_value(emby_update_library)
|
|
sickgear.EMBY_PARENT_MAPS = config.kv_csv(emby_parent_maps)
|
|
sickgear.EMBY_HOST = config.clean_hosts(emby_host, allow_base=True)
|
|
keys_changed = False
|
|
all_keys = []
|
|
old_keys = [x.strip() for x in sickgear.EMBY_APIKEY.split(',') if x.strip()]
|
|
new_keys = [x.strip() for x in emby_apikey.split(',') if x.strip()]
|
|
for key in new_keys:
|
|
if not starify(key, True):
|
|
keys_changed = True
|
|
all_keys += [key]
|
|
continue
|
|
for x in old_keys:
|
|
if key.startswith(x[0:3]) and key.endswith(x[-4:]):
|
|
all_keys += [x]
|
|
break
|
|
if keys_changed or (len(all_keys) != len(old_keys)):
|
|
sickgear.EMBY_APIKEY = ','.join(all_keys)
|
|
|
|
sickgear.USE_KODI = config.checkbox_to_value(use_kodi)
|
|
sickgear.KODI_ALWAYS_ON = config.checkbox_to_value(kodi_always_on)
|
|
sickgear.KODI_NOTIFY_ONSNATCH = config.checkbox_to_value(kodi_notify_onsnatch)
|
|
sickgear.KODI_NOTIFY_ONDOWNLOAD = config.checkbox_to_value(kodi_notify_ondownload)
|
|
sickgear.KODI_NOTIFY_ONSUBTITLEDOWNLOAD = config.checkbox_to_value(kodi_notify_onsubtitledownload)
|
|
sickgear.KODI_UPDATE_LIBRARY = config.checkbox_to_value(kodi_update_library)
|
|
sickgear.KODI_UPDATE_FULL = config.checkbox_to_value(kodi_update_full)
|
|
sickgear.KODI_UPDATE_ONLYFIRST = config.checkbox_to_value(kodi_update_onlyfirst)
|
|
sickgear.KODI_PARENT_MAPS = config.kv_csv(kodi_parent_maps)
|
|
sickgear.KODI_HOST = config.clean_hosts(kodi_host)
|
|
sickgear.KODI_USERNAME = kodi_username
|
|
if set('*') != set(kodi_password):
|
|
sickgear.KODI_PASSWORD = kodi_password
|
|
|
|
sickgear.USE_PLEX = config.checkbox_to_value(use_plex)
|
|
sickgear.PLEX_NOTIFY_ONSNATCH = config.checkbox_to_value(plex_notify_onsnatch)
|
|
sickgear.PLEX_NOTIFY_ONDOWNLOAD = config.checkbox_to_value(plex_notify_ondownload)
|
|
sickgear.PLEX_NOTIFY_ONSUBTITLEDOWNLOAD = config.checkbox_to_value(plex_notify_onsubtitledownload)
|
|
sickgear.PLEX_UPDATE_LIBRARY = config.checkbox_to_value(plex_update_library)
|
|
sickgear.PLEX_PARENT_MAPS = config.kv_csv(plex_parent_maps)
|
|
sickgear.PLEX_HOST = config.clean_hosts(plex_host)
|
|
sickgear.PLEX_SERVER_HOST = config.clean_hosts(plex_server_host)
|
|
sickgear.PLEX_USERNAME = plex_username
|
|
if set('*') != set(plex_password):
|
|
sickgear.PLEX_PASSWORD = plex_password
|
|
config.schedule_emby_watched(emby_watched_interval)
|
|
config.schedule_plex_watched(plex_watched_interval)
|
|
|
|
sickgear.USE_GROWL = config.checkbox_to_value(use_growl)
|
|
sickgear.GROWL_NOTIFY_ONSNATCH = config.checkbox_to_value(growl_notify_onsnatch)
|
|
sickgear.GROWL_NOTIFY_ONDOWNLOAD = config.checkbox_to_value(growl_notify_ondownload)
|
|
sickgear.GROWL_NOTIFY_ONSUBTITLEDOWNLOAD = config.checkbox_to_value(growl_notify_onsubtitledownload)
|
|
sickgear.GROWL_HOST = config.clean_hosts(growl_host, default_port=23053)
|
|
|
|
sickgear.USE_PROWL = config.checkbox_to_value(use_prowl)
|
|
sickgear.PROWL_NOTIFY_ONSNATCH = config.checkbox_to_value(prowl_notify_onsnatch)
|
|
sickgear.PROWL_NOTIFY_ONDOWNLOAD = config.checkbox_to_value(prowl_notify_ondownload)
|
|
sickgear.PROWL_NOTIFY_ONSUBTITLEDOWNLOAD = config.checkbox_to_value(prowl_notify_onsubtitledownload)
|
|
key = prowl_api.strip()
|
|
if not starify(key, True):
|
|
sickgear.PROWL_API = key
|
|
sickgear.PROWL_PRIORITY = prowl_priority
|
|
|
|
sickgear.USE_BOXCAR2 = config.checkbox_to_value(use_boxcar2)
|
|
sickgear.BOXCAR2_NOTIFY_ONSNATCH = config.checkbox_to_value(boxcar2_notify_onsnatch)
|
|
sickgear.BOXCAR2_NOTIFY_ONDOWNLOAD = config.checkbox_to_value(boxcar2_notify_ondownload)
|
|
sickgear.BOXCAR2_NOTIFY_ONSUBTITLEDOWNLOAD = config.checkbox_to_value(boxcar2_notify_onsubtitledownload)
|
|
key = boxcar2_access_token.strip()
|
|
if not starify(key, True):
|
|
sickgear.BOXCAR2_ACCESSTOKEN = key
|
|
sickgear.BOXCAR2_SOUND = boxcar2_sound
|
|
|
|
sickgear.USE_PUSHOVER = config.checkbox_to_value(use_pushover)
|
|
sickgear.PUSHOVER_NOTIFY_ONSNATCH = config.checkbox_to_value(pushover_notify_onsnatch)
|
|
sickgear.PUSHOVER_NOTIFY_ONDOWNLOAD = config.checkbox_to_value(pushover_notify_ondownload)
|
|
sickgear.PUSHOVER_NOTIFY_ONSUBTITLEDOWNLOAD = config.checkbox_to_value(pushover_notify_onsubtitledownload)
|
|
key = pushover_userkey.strip()
|
|
if not starify(key, True):
|
|
sickgear.PUSHOVER_USERKEY = key
|
|
key = pushover_apikey.strip()
|
|
if not starify(key, True):
|
|
sickgear.PUSHOVER_APIKEY = key
|
|
sickgear.PUSHOVER_PRIORITY = pushover_priority
|
|
sickgear.PUSHOVER_DEVICE = pushover_device
|
|
sickgear.PUSHOVER_SOUND = pushover_sound
|
|
|
|
sickgear.USE_LIBNOTIFY = config.checkbox_to_value(use_libnotify)
|
|
sickgear.LIBNOTIFY_NOTIFY_ONSNATCH = config.checkbox_to_value(libnotify_notify_onsnatch)
|
|
sickgear.LIBNOTIFY_NOTIFY_ONDOWNLOAD = config.checkbox_to_value(libnotify_notify_ondownload)
|
|
sickgear.LIBNOTIFY_NOTIFY_ONSUBTITLEDOWNLOAD = config.checkbox_to_value(libnotify_notify_onsubtitledownload)
|
|
|
|
sickgear.USE_NMJ = config.checkbox_to_value(use_nmj)
|
|
sickgear.NMJ_HOST = config.clean_host(nmj_host)
|
|
sickgear.NMJ_DATABASE = nmj_database
|
|
sickgear.NMJ_MOUNT = nmj_mount
|
|
|
|
sickgear.USE_NMJv2 = config.checkbox_to_value(use_nmjv2)
|
|
sickgear.NMJv2_HOST = config.clean_host(nmjv2_host)
|
|
sickgear.NMJv2_DATABASE = nmjv2_database
|
|
sickgear.NMJv2_DBLOC = nmjv2_dbloc
|
|
|
|
sickgear.USE_SYNOINDEX = config.checkbox_to_value(use_synoindex)
|
|
|
|
sickgear.USE_SYNOLOGYNOTIFIER = config.checkbox_to_value(use_synologynotifier)
|
|
sickgear.SYNOLOGYNOTIFIER_NOTIFY_ONSNATCH = config.checkbox_to_value(synologynotifier_notify_onsnatch)
|
|
sickgear.SYNOLOGYNOTIFIER_NOTIFY_ONDOWNLOAD = config.checkbox_to_value(synologynotifier_notify_ondownload)
|
|
sickgear.SYNOLOGYNOTIFIER_NOTIFY_ONSUBTITLEDOWNLOAD = config.checkbox_to_value(
|
|
synologynotifier_notify_onsubtitledownload)
|
|
|
|
sickgear.USE_TRAKT = config.checkbox_to_value(use_trakt)
|
|
sickgear.TRAKT_UPDATE_COLLECTION = build_config(**kwargs)
|
|
# sickgear.trakt_checker_scheduler.silent = not sickgear.USE_TRAKT
|
|
# sickgear.TRAKT_DEFAULT_INDEXER = int(trakt_default_indexer)
|
|
# sickgear.TRAKT_SYNC = config.checkbox_to_value(trakt_sync)
|
|
# sickgear.TRAKT_USE_WATCHLIST = config.checkbox_to_value(trakt_use_watchlist)
|
|
# sickgear.TRAKT_METHOD_ADD = int(trakt_method_add)
|
|
# sickgear.TRAKT_REMOVE_WATCHLIST = config.checkbox_to_value(trakt_remove_watchlist)
|
|
# sickgear.TRAKT_REMOVE_SERIESLIST = config.checkbox_to_value(trakt_remove_serieslist)
|
|
# sickgear.TRAKT_START_PAUSED = config.checkbox_to_value(trakt_start_paused)
|
|
|
|
sickgear.USE_SLACK = config.checkbox_to_value(use_slack)
|
|
sickgear.SLACK_NOTIFY_ONSNATCH = config.checkbox_to_value(slack_notify_onsnatch)
|
|
sickgear.SLACK_NOTIFY_ONDOWNLOAD = config.checkbox_to_value(slack_notify_ondownload)
|
|
sickgear.SLACK_NOTIFY_ONSUBTITLEDOWNLOAD = config.checkbox_to_value(slack_notify_onsubtitledownload)
|
|
sickgear.SLACK_ACCESS_TOKEN = slack_access_token
|
|
sickgear.SLACK_CHANNEL = slack_channel
|
|
sickgear.SLACK_AS_AUTHED = config.checkbox_to_value(slack_as_authed)
|
|
sickgear.SLACK_BOT_NAME = slack_bot_name
|
|
sickgear.SLACK_ICON_URL = slack_icon_url
|
|
|
|
sickgear.USE_DISCORD = config.checkbox_to_value(use_discord)
|
|
sickgear.DISCORD_NOTIFY_ONSNATCH = config.checkbox_to_value(discord_notify_onsnatch)
|
|
sickgear.DISCORD_NOTIFY_ONDOWNLOAD = config.checkbox_to_value(discord_notify_ondownload)
|
|
sickgear.DISCORD_NOTIFY_ONSUBTITLEDOWNLOAD = config.checkbox_to_value(discord_notify_onsubtitledownload)
|
|
sickgear.DISCORD_ACCESS_TOKEN = discord_access_token
|
|
sickgear.DISCORD_AS_AUTHED = config.checkbox_to_value(discord_as_authed)
|
|
sickgear.DISCORD_USERNAME = discord_username
|
|
sickgear.DISCORD_ICON_URL = discord_icon_url
|
|
sickgear.DISCORD_AS_TTS = config.checkbox_to_value(discord_as_tts)
|
|
|
|
sickgear.USE_GITTER = config.checkbox_to_value(use_gitter)
|
|
sickgear.GITTER_NOTIFY_ONSNATCH = config.checkbox_to_value(gitter_notify_onsnatch)
|
|
sickgear.GITTER_NOTIFY_ONDOWNLOAD = config.checkbox_to_value(gitter_notify_ondownload)
|
|
sickgear.GITTER_NOTIFY_ONSUBTITLEDOWNLOAD = config.checkbox_to_value(gitter_notify_onsubtitledownload)
|
|
sickgear.GITTER_ACCESS_TOKEN = gitter_access_token
|
|
sickgear.GITTER_ROOM = gitter_room
|
|
|
|
sickgear.USE_TELEGRAM = config.checkbox_to_value(use_telegram)
|
|
sickgear.TELEGRAM_NOTIFY_ONSNATCH = config.checkbox_to_value(telegram_notify_onsnatch)
|
|
sickgear.TELEGRAM_NOTIFY_ONDOWNLOAD = config.checkbox_to_value(telegram_notify_ondownload)
|
|
sickgear.TELEGRAM_NOTIFY_ONSUBTITLEDOWNLOAD = config.checkbox_to_value(telegram_notify_onsubtitledownload)
|
|
sickgear.TELEGRAM_ACCESS_TOKEN = telegram_access_token
|
|
sickgear.TELEGRAM_CHATID = telegram_chatid
|
|
sickgear.TELEGRAM_SEND_IMAGE = config.checkbox_to_value(telegram_send_image)
|
|
sickgear.TELEGRAM_QUIET = config.checkbox_to_value(telegram_quiet)
|
|
|
|
sickgear.USE_EMAIL = config.checkbox_to_value(use_email)
|
|
sickgear.EMAIL_NOTIFY_ONSNATCH = config.checkbox_to_value(email_notify_onsnatch)
|
|
sickgear.EMAIL_NOTIFY_ONDOWNLOAD = config.checkbox_to_value(email_notify_ondownload)
|
|
sickgear.EMAIL_NOTIFY_ONSUBTITLEDOWNLOAD = config.checkbox_to_value(email_notify_onsubtitledownload)
|
|
sickgear.EMAIL_HOST = config.clean_host(email_host)
|
|
sickgear.EMAIL_PORT = config.to_int(email_port, default=25)
|
|
sickgear.EMAIL_FROM = email_from
|
|
sickgear.EMAIL_TLS = config.checkbox_to_value(email_tls)
|
|
sickgear.EMAIL_USER = email_user
|
|
if set('*') != set(email_password):
|
|
sickgear.EMAIL_PASSWORD = email_password
|
|
sickgear.EMAIL_LIST = email_list
|
|
|
|
sickgear.USE_PYTIVO = config.checkbox_to_value(use_pytivo)
|
|
sickgear.PYTIVO_HOST = config.clean_host(pytivo_host)
|
|
sickgear.PYTIVO_SHARE_NAME = pytivo_share_name
|
|
sickgear.PYTIVO_TIVO_NAME = pytivo_tivo_name
|
|
|
|
sickgear.USE_PUSHBULLET = config.checkbox_to_value(use_pushbullet)
|
|
sickgear.PUSHBULLET_NOTIFY_ONSNATCH = config.checkbox_to_value(pushbullet_notify_onsnatch)
|
|
sickgear.PUSHBULLET_NOTIFY_ONDOWNLOAD = config.checkbox_to_value(pushbullet_notify_ondownload)
|
|
sickgear.PUSHBULLET_NOTIFY_ONSUBTITLEDOWNLOAD = config.checkbox_to_value(pushbullet_notify_onsubtitledownload)
|
|
key = pushbullet_access_token.strip()
|
|
if not starify(key, True):
|
|
sickgear.PUSHBULLET_ACCESS_TOKEN = key
|
|
sickgear.PUSHBULLET_DEVICE_IDEN = pushbullet_device_iden
|
|
|
|
sickgear.save_config()
|
|
|
|
if 0 < len(results):
|
|
for x in results:
|
|
logger.error(x)
|
|
ui.notifications.error('Error(s) Saving Configuration',
|
|
'<br>\n'.join(results))
|
|
else:
|
|
ui.notifications.message('Configuration Saved', os.path.join(sickgear.CONFIG_FILE))
|
|
|
|
self.redirect('/config/notifications/')
|
|
|
|
|
|
class ConfigSubtitles(Config):
|
|
|
|
def index(self):
|
|
t = PageTemplate(web_handler=self, file='config_subtitles.tmpl')
|
|
t.submenu = self.config_menu('Subtitle')
|
|
return t.respond()
|
|
|
|
def save_subtitles(self, use_subtitles=None, subtitles_languages=None, subtitles_dir=None,
|
|
service_order=None, subtitles_history=None, subtitles_finder_interval=None,
|
|
subtitles_finder_frequency=None,
|
|
os_hash=None, os_user='', os_pass=''):
|
|
|
|
# prevent deprecated var issues from existing ui, delete in future, added 2020.11.07
|
|
if None is subtitles_finder_interval and None is not subtitles_finder_frequency:
|
|
subtitles_finder_interval = subtitles_finder_frequency
|
|
|
|
results = []
|
|
|
|
if '' == subtitles_finder_interval or None is subtitles_finder_interval:
|
|
subtitles_finder_interval = 1
|
|
|
|
config.schedule_subtitles(config.checkbox_to_value(use_subtitles))
|
|
sickgear.SUBTITLES_LANGUAGES = [lang.alpha2 for lang in subtitles.is_valid_language(
|
|
subtitles_languages.replace(' ', '').split(','))] if '' != subtitles_languages else ''
|
|
sickgear.SUBTITLES_DIR = subtitles_dir
|
|
sickgear.SUBTITLES_HISTORY = config.checkbox_to_value(subtitles_history)
|
|
sickgear.SUBTITLES_FINDER_INTERVAL = config.to_int(subtitles_finder_interval, default=1)
|
|
sickgear.SUBTITLES_OS_HASH = config.checkbox_to_value(os_hash)
|
|
|
|
# Subtitles services
|
|
services_str_list = service_order.split()
|
|
subtitles_services_list = []
|
|
subtitles_services_enabled = []
|
|
for cur_service in services_str_list:
|
|
service, enabled = cur_service.split(':')
|
|
subtitles_services_list.append(service)
|
|
subtitles_services_enabled.append(int(enabled))
|
|
|
|
sickgear.SUBTITLES_SERVICES_LIST = subtitles_services_list
|
|
sickgear.SUBTITLES_SERVICES_ENABLED = subtitles_services_enabled
|
|
sickgear.SUBTITLES_SERVICES_AUTH = [[os_user, os_pass]]
|
|
|
|
sickgear.save_config()
|
|
|
|
if 0 < len(results):
|
|
for x in results:
|
|
logger.error(x)
|
|
ui.notifications.error('Error(s) Saving Configuration',
|
|
'<br>\n'.join(results))
|
|
else:
|
|
ui.notifications.message('Configuration Saved', os.path.join(sickgear.CONFIG_FILE))
|
|
|
|
self.redirect('/config/subtitles/')
|
|
|
|
|
|
class ConfigAnime(Config):
|
|
|
|
def index(self):
|
|
|
|
t = PageTemplate(web_handler=self, file='config_anime.tmpl')
|
|
t.submenu = self.config_menu('Anime')
|
|
return t.respond()
|
|
|
|
def save_anime(self, use_anidb=None, anidb_username=None, anidb_password=None, anidb_use_mylist=None,
|
|
anime_treat_as_hdtv=None):
|
|
|
|
results = []
|
|
|
|
sickgear.USE_ANIDB = config.checkbox_to_value(use_anidb)
|
|
sickgear.ANIDB_USERNAME = anidb_username
|
|
if set('*') != set(anidb_password):
|
|
sickgear.ANIDB_PASSWORD = anidb_password
|
|
sickgear.ANIDB_USE_MYLIST = config.checkbox_to_value(anidb_use_mylist)
|
|
sickgear.ANIME_TREAT_AS_HDTV = config.checkbox_to_value(anime_treat_as_hdtv)
|
|
|
|
sickgear.save_config()
|
|
|
|
if 0 < len(results):
|
|
for x in results:
|
|
logger.error(x)
|
|
ui.notifications.error('Error(s) Saving Configuration',
|
|
'<br>\n'.join(results))
|
|
else:
|
|
ui.notifications.message('Configuration Saved', os.path.join(sickgear.CONFIG_FILE))
|
|
|
|
self.redirect('/config/anime/')
|
|
|
|
|
|
class UI(MainHandler):
|
|
|
|
@staticmethod
|
|
def add_message():
|
|
ui.notifications.message('Test 1', 'This is test number 1')
|
|
ui.notifications.error('Test 2', 'This is test number 2')
|
|
|
|
return 'ok'
|
|
|
|
def get_messages(self):
|
|
messages = {}
|
|
cur_notification_num = 1
|
|
for cur_notification in ui.notifications.get_notifications(self.request.remote_ip):
|
|
messages['notification-' + str(cur_notification_num)] = {'title': cur_notification.title,
|
|
'message': cur_notification.message,
|
|
'type': cur_notification.type}
|
|
cur_notification_num += 1
|
|
|
|
return json_dumps(messages)
|
|
|
|
|
|
class EventLogs(MainHandler):
|
|
|
|
@staticmethod
|
|
def error_logs_menu():
|
|
menu = [{'title': 'Download Log', 'path': 'events/download-log/'}]
|
|
if len(classes.ErrorViewer.errors):
|
|
menu += [{'title': 'Clear Errors', 'path': 'errors/clear-log/'}]
|
|
return menu
|
|
|
|
def index(self):
|
|
|
|
t = PageTemplate(web_handler=self, file='errorlogs.tmpl')
|
|
t.submenu = self.error_logs_menu
|
|
|
|
return t.respond()
|
|
|
|
def clear_log(self):
|
|
classes.ErrorViewer.clear()
|
|
self.redirect('/events/')
|
|
|
|
def download_log(self):
|
|
self.redirect('/logfile/sickgear.log')
|
|
|
|
def view_log(self, min_level=logger.MESSAGE, max_lines=500):
|
|
|
|
t = PageTemplate(web_handler=self, file='viewlogs.tmpl')
|
|
t.submenu = self.error_logs_menu
|
|
|
|
min_level = int(min_level)
|
|
|
|
regex = re.compile(r'^\d{4}-\d{2}-\d{2}\s*\d{2}:\d{2}:\d{2}\s*([A-Z]+)\s*(\S+)\s+:{2}\s*(.*\r?\n)$')
|
|
|
|
final_data = []
|
|
normal_data = []
|
|
truncate = []
|
|
repeated = None
|
|
num_lines = 0
|
|
if os.path.isfile(logger.sb_log_instance.log_file_path):
|
|
auths = sickgear.GenericProvider.dedupe_auths(True)
|
|
rxc_auths = re.compile('(?i)%s' % '|'.join([(re.escape(_a)) for _a in auths]))
|
|
replacements = dict([(_a, starify(_a)) for _a in auths])
|
|
for cur_line in logger.sb_log_instance.reverse_readline(logger.sb_log_instance.log_file_path):
|
|
|
|
cur_line = helpers.xhtml_escape(decode_str(cur_line, errors='replace'), False)
|
|
try:
|
|
match = regex.findall(cur_line)[0]
|
|
except(BaseException, Exception):
|
|
if not any(normal_data) and not any([cur_line.strip()]):
|
|
continue
|
|
|
|
normal_data.append(re.sub(r'\r?\n', '<br>', cur_line))
|
|
else:
|
|
level, log = match[0], ' '.join(match[1:])
|
|
if level not in logger.reverseNames:
|
|
normal_data = []
|
|
continue
|
|
|
|
if logger.reverseNames[level] < min_level:
|
|
normal_data = []
|
|
continue
|
|
else:
|
|
if truncate and not normal_data and truncate[0] == log:
|
|
truncate += [log]
|
|
repeated = cur_line
|
|
continue
|
|
|
|
if 1 < len(truncate):
|
|
data = repeated.strip() + \
|
|
' <span class="grey-text">(...%s repeat lines)</span>\n' % len(truncate)
|
|
if not final_data:
|
|
final_data = [data]
|
|
else:
|
|
final_data[-1] = data
|
|
|
|
truncate = [log]
|
|
|
|
# noinspection HttpUrlsUsage
|
|
if 'https://' in cur_line or 'http://' in cur_line:
|
|
for cur_change in rxc_auths.finditer(cur_line):
|
|
cur_line = '%s%s%s' % (cur_line[:cur_change.start()],
|
|
replacements[cur_line[cur_change.start():cur_change.end()]],
|
|
cur_line[cur_change.end():])
|
|
|
|
final_data.append(cur_line)
|
|
if 'Starting SickGear' in cur_line:
|
|
final_data[-1].replace(' Starting SickGear',
|
|
' <span class="prelight2">Starting SickGear</span>')
|
|
if any(normal_data):
|
|
final_data += ['<code><span class="prelight">'] + \
|
|
['<span class="prelight-num">%02s)</span> %s' % (n + 1, x)
|
|
for n, x in enumerate(normal_data[::-1])] + \
|
|
['</span></code><br>']
|
|
num_lines += len(normal_data)
|
|
normal_data = []
|
|
|
|
num_lines += 1
|
|
|
|
if num_lines >= max_lines:
|
|
break
|
|
|
|
result = ''.join(final_data)
|
|
|
|
t.logLines = result
|
|
t.min_level = min_level
|
|
|
|
return t.respond()
|
|
|
|
|
|
class WebFileBrowser(MainHandler):
|
|
|
|
def index(self, path='', include_files=False, **kwargs):
|
|
|
|
self.set_header('Content-Type', 'application/json')
|
|
return json_dumps(folders_at_path(path, True, bool(int(include_files))))
|
|
|
|
def complete(self, term, include_files=0, **kwargs):
|
|
|
|
self.set_header('Content-Type', 'application/json')
|
|
return json_dumps([entry['path'] for entry in folders_at_path(
|
|
os.path.dirname(term), include_files=bool(int(include_files))) if 'path' in entry])
|
|
|
|
|
|
class ApiBuilder(MainHandler):
|
|
|
|
def index(self):
|
|
""" expose the api-builder template """
|
|
t = PageTemplate(web_handler=self, file='apiBuilder.tmpl')
|
|
|
|
def titler(x):
|
|
return (remove_article(x), x)[not x or sickgear.SORT_ARTICLE].lower()
|
|
|
|
t.sortedShowList = sorted(sickgear.showList, key=lambda x: titler(x.name))
|
|
|
|
season_sql_result = {}
|
|
episode_sql_result = {}
|
|
|
|
my_db = db.DBConnection(row_type='dict')
|
|
for cur_show_obj in t.sortedShowList:
|
|
season_sql_result[cur_show_obj.tvid_prodid] = my_db.select(
|
|
'SELECT DISTINCT season'
|
|
' FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' ORDER BY season DESC',
|
|
[cur_show_obj.tvid, cur_show_obj.prodid])
|
|
|
|
for cur_show_obj in t.sortedShowList:
|
|
episode_sql_result[cur_show_obj.tvid_prodid] = my_db.select(
|
|
'SELECT DISTINCT season,episode'
|
|
' FROM tv_episodes'
|
|
' WHERE indexer = ? AND showid = ?'
|
|
' ORDER BY season DESC, episode DESC',
|
|
[cur_show_obj.tvid, cur_show_obj.prodid])
|
|
|
|
t.seasonSQLResults = season_sql_result
|
|
t.episodeSQLResults = episode_sql_result
|
|
t.indexers = sickgear.TVInfoAPI().all_sources
|
|
t.searchindexers = sickgear.TVInfoAPI().search_sources
|
|
|
|
if len(sickgear.API_KEYS):
|
|
# use first APIKEY for apibuilder tests
|
|
t.apikey = sickgear.API_KEYS[0][1]
|
|
else:
|
|
t.apikey = 'api key not generated'
|
|
|
|
return t.respond()
|
|
|
|
|
|
class Cache(MainHandler):
|
|
|
|
def index(self):
|
|
my_db = db.DBConnection('cache.db')
|
|
sql_result = my_db.select('SELECT * FROM provider_cache')
|
|
if not sql_result:
|
|
sql_result = []
|
|
|
|
t = PageTemplate(web_handler=self, file='cache.tmpl')
|
|
t.cacheResults = sql_result
|
|
|
|
return t.respond()
|
|
|
|
|
|
class CachedImages(MainHandler):
|
|
def set_default_headers(self):
|
|
super(CachedImages, self).set_default_headers()
|
|
self.set_header('Cache-Control', 'no-cache, max-age=0')
|
|
self.set_header('Pragma', 'no-cache')
|
|
self.set_header('Expires', '0')
|
|
|
|
@staticmethod
|
|
def should_try_image(filename, source, days=1, minutes=0):
|
|
result = True
|
|
try:
|
|
dummy_file = '%s.%s.dummy' % (os.path.splitext(filename)[0], source)
|
|
if os.path.isfile(dummy_file):
|
|
if os.stat(dummy_file).st_mtime \
|
|
< (SGDatetime.timestamp_near(datetime.now() - timedelta(days=days, minutes=minutes))):
|
|
CachedImages.delete_dummy_image(dummy_file)
|
|
else:
|
|
result = False
|
|
except (BaseException, Exception):
|
|
pass
|
|
return result
|
|
|
|
@staticmethod
|
|
def create_dummy_image(filename, source):
|
|
dummy_file = '%s.%s.dummy' % (os.path.splitext(filename)[0], source)
|
|
CachedImages.delete_dummy_image(dummy_file)
|
|
try:
|
|
with open(dummy_file, 'w'):
|
|
pass
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
@staticmethod
|
|
def delete_dummy_image(dummy_file):
|
|
try:
|
|
if os.path.isfile(dummy_file):
|
|
os.remove(dummy_file)
|
|
except (BaseException, Exception):
|
|
pass
|
|
|
|
@staticmethod
|
|
def delete_all_dummy_images(filename):
|
|
for f in ['tmdb', 'tvdb', 'tvmaze']:
|
|
CachedImages.delete_dummy_image('%s.%s.dummy' % (os.path.splitext(filename)[0], f))
|
|
|
|
def index(self, path='', source=None, filename=None, tmdbid=None, tvdbid=None, trans=True, tvmazeid=None):
|
|
|
|
path = path.strip('/')
|
|
file_name = ''
|
|
if None is not source:
|
|
file_name = os.path.basename(source)
|
|
elif filename not in [None, 0, '0']:
|
|
file_name = filename
|
|
image_file = os.path.join(sickgear.CACHE_DIR, 'images', path, file_name)
|
|
image_file = os.path.abspath(image_file.replace('\\', '/'))
|
|
if not os.path.isfile(image_file) and has_image_ext(file_name):
|
|
basepath = os.path.dirname(image_file)
|
|
helpers.make_path(basepath)
|
|
poster_url = ''
|
|
tmdb_image = False
|
|
tvmaze_image = False
|
|
if None is not source and source in sickgear.CACHE_IMAGE_URL_LIST:
|
|
poster_url = source
|
|
if None is source and tvmazeid not in [None, 'None', 0, '0'] \
|
|
and self.should_try_image(image_file, 'tvmaze'):
|
|
tvmaze_image = True
|
|
try:
|
|
tvinfo_config = sickgear.TVInfoAPI(TVINFO_TVMAZE).api_params.copy()
|
|
t = sickgear.TVInfoAPI(TVINFO_TVMAZE).setup(**tvinfo_config)
|
|
show_obj = t.get_show(tvmazeid, load_episodes=False, posters=True)
|
|
if show_obj and show_obj.poster:
|
|
poster_url = show_obj.poster
|
|
except (BaseException, Exception):
|
|
poster_url = ''
|
|
if poster_url:
|
|
sg_helpers.download_file(poster_url, image_file, nocache=True)
|
|
if tvmaze_image and not os.path.isfile(image_file):
|
|
self.create_dummy_image(image_file, 'tvmaze')
|
|
|
|
if None is source and tmdbid not in [None, 'None', 0, '0'] \
|
|
and self.should_try_image(image_file, 'tmdb'):
|
|
tmdb_image = True
|
|
try:
|
|
tvinfo_config = sickgear.TVInfoAPI(TVINFO_TMDB).api_params.copy()
|
|
t = sickgear.TVInfoAPI(TVINFO_TMDB).setup(**tvinfo_config)
|
|
show_obj = t.get_show(tmdbid, load_episodes=False, posters=True)
|
|
if show_obj and show_obj.poster:
|
|
poster_url = show_obj.poster
|
|
except (BaseException, Exception):
|
|
poster_url = ''
|
|
if poster_url \
|
|
and not sg_helpers.download_file(poster_url, image_file, nocache=True) \
|
|
and poster_url.find('trakt.us'):
|
|
sg_helpers.download_file(poster_url.replace('trakt.us', 'trakt.tv'), image_file, nocache=True)
|
|
if tmdb_image and not os.path.isfile(image_file):
|
|
self.create_dummy_image(image_file, 'tmdb')
|
|
|
|
if None is source and tvdbid not in [None, 'None', 0, '0'] \
|
|
and not os.path.isfile(image_file) \
|
|
and self.should_try_image(image_file, 'tvdb'):
|
|
try:
|
|
tvinfo_config = sickgear.TVInfoAPI(TVINFO_TVDB).api_params.copy()
|
|
tvinfo_config['posters'] = True
|
|
t = sickgear.TVInfoAPI(TVINFO_TVDB).setup(**tvinfo_config).get_show(
|
|
helpers.try_int(tvdbid), load_episodes=False, posters=True)
|
|
if hasattr(t, 'data') and 'poster' in t.data:
|
|
poster_url = t.data['poster']
|
|
except (BaseException, Exception):
|
|
poster_url = ''
|
|
if poster_url:
|
|
sg_helpers.download_file(poster_url, image_file, nocache=True)
|
|
if not os.path.isfile(image_file):
|
|
self.create_dummy_image(image_file, 'tvdb')
|
|
|
|
if os.path.isfile(image_file):
|
|
self.delete_all_dummy_images(image_file)
|
|
|
|
if not os.path.isfile(image_file):
|
|
image_file = os.path.join(sickgear.PROG_DIR, 'gui', 'slick', 'images',
|
|
('image-light.png', 'trans.png')[bool(int(trans))])
|
|
else:
|
|
helpers.set_file_timestamp(image_file, min_age=3, new_time=None)
|
|
|
|
return self.image_data(image_file)
|
|
|
|
@staticmethod
|
|
def should_load_image(filename, days=7):
|
|
# type: (AnyStr, integer_types) -> bool
|
|
"""
|
|
should image be (re-)loaded
|
|
|
|
:param filename: image file name with path
|
|
:param days: max age to trigger reload of image
|
|
"""
|
|
if not os.path.isfile(filename) or \
|
|
os.stat(filename).st_mtime < \
|
|
SGDatetime.timestamp_near(td=timedelta(days=days)):
|
|
return True
|
|
return False
|
|
|
|
@staticmethod
|
|
def find_cast_by_id(ref_id, cast_list):
|
|
for cur_item in cast_list:
|
|
if cur_item.has_ref_id(ref_id):
|
|
return cur_item
|
|
|
|
def character(self, rid=None, tvid_prodid=None, thumb=True, pid=None, prefer_person=False, **kwargs):
|
|
"""
|
|
|
|
:param rid:
|
|
:param tvid_prodid:
|
|
:param thumb: return thumb or normal as fallback
|
|
:param pid: optional person_id
|
|
:param prefer_person: prefer person image if person_id is set and character has more than 1 person assigned
|
|
"""
|
|
_ = kwargs.get('oid') # suppress pyc non used var highlight, oid (original id) is a visual ui key
|
|
show_obj = tvid_prodid and helpers.find_show_by_id(tvid_prodid)
|
|
char_id = usable_id(rid)
|
|
person_id = usable_id(pid)
|
|
if not show_obj or not char_id:
|
|
return
|
|
char_obj = self.find_cast_by_id(char_id, show_obj.cast_list)
|
|
if not char_obj:
|
|
return
|
|
if person_id:
|
|
person_obj = TVPerson(sid=person_id)
|
|
person_id = person_obj.id # a reference could be passed in, replace it with id for later use
|
|
if not char_obj.person or person_obj not in char_obj.person:
|
|
person_obj = None
|
|
else:
|
|
person_obj = None
|
|
thumb = thumb in (True, '1', 'true', 'True')
|
|
prefer_person = prefer_person in (True, '1', 'true', 'True') and char_obj.person and 1 < len(char_obj.person) \
|
|
and bool(person_obj)
|
|
|
|
image_file = None
|
|
if not prefer_person and (char_obj.thumb_url or char_obj.image_url):
|
|
image_cache_obj = image_cache.ImageCache()
|
|
image_normal, image_thumb = image_cache_obj.character_both_path(char_obj, show_obj, person_obj=person_obj)
|
|
sg_helpers.make_path(image_cache_obj.characters_dir)
|
|
if self.should_load_image(image_normal) and char_obj.image_url:
|
|
sg_helpers.download_file(char_obj.image_url, image_normal, nocache=True)
|
|
if self.should_load_image(image_thumb) and char_obj.thumb_url:
|
|
sg_helpers.download_file(char_obj.thumb_url, image_thumb, nocache=True)
|
|
|
|
primary, fallback = ((image_normal, image_thumb), (image_thumb, image_normal))[thumb]
|
|
if os.path.isfile(primary):
|
|
image_file = primary
|
|
elif os.path.isfile(fallback):
|
|
image_file = fallback
|
|
|
|
elif person_id:
|
|
return self.person(rid=char_id, pid=person_id, show_obj=show_obj, thumb=thumb)
|
|
elif char_obj.person and (char_obj.person[0].thumb_url or char_obj.person[0].image_url):
|
|
return self.person(rid=char_id, pid=char_obj.person[0].id, show_obj=show_obj, thumb=thumb)
|
|
|
|
return self.image_data(image_file, cast_default=True)
|
|
|
|
def person(self, rid=None, pid=None, tvid_prodid=None, show_obj=None, thumb=True, **kwargs):
|
|
_ = kwargs.get('oid') # suppress pyc non used var highlight, oid (original id) is a visual ui key
|
|
show_obj = show_obj or tvid_prodid and helpers.find_show_by_id(tvid_prodid)
|
|
char_id = usable_id(rid)
|
|
person_id = usable_id(pid)
|
|
if not person_id:
|
|
return
|
|
person_obj = TVPerson(sid=person_id)
|
|
if char_id and show_obj and not person_obj:
|
|
char_obj = self.find_cast_by_id(char_id, show_obj.cast_list)
|
|
person_obj = char_obj.person and char_obj.person[0]
|
|
if not person_obj:
|
|
return
|
|
thumb = thumb in (True, '1', 'true', 'True')
|
|
|
|
image_file = None
|
|
if person_obj.thumb_url or person_obj.image_url:
|
|
image_cache_obj = image_cache.ImageCache()
|
|
image_normal, image_thumb = image_cache_obj.person_both_paths(person_obj)
|
|
sg_helpers.make_path(image_cache_obj.characters_dir)
|
|
if self.should_load_image(image_normal) and person_obj.image_url:
|
|
sg_helpers.download_file(person_obj.image_url, image_normal, nocache=True)
|
|
if self.should_load_image(image_thumb) and person_obj.thumb_url:
|
|
sg_helpers.download_file(person_obj.thumb_url, image_thumb, nocache=True)
|
|
|
|
primary, fallback = ((image_normal, image_thumb), (image_thumb, image_normal))[thumb]
|
|
if os.path.isfile(primary):
|
|
image_file = primary
|
|
elif os.path.isfile(fallback):
|
|
image_file = fallback
|
|
|
|
return self.image_data(image_file, cast_default=True)
|
|
|
|
def image_data(self, image_file, cast_default=False):
|
|
# type: (Optional[AnyStr], bool) -> Optional[Any]
|
|
"""
|
|
return image file binary data
|
|
|
|
:param image_file: file path
|
|
:param cast_default: if required, use default cast file path if None is image_file
|
|
:return: binary image data or None
|
|
"""
|
|
if cast_default and None is image_file:
|
|
image_file = os.path.join(sickgear.PROG_DIR, 'gui', 'slick', 'images', 'poster-person.jpg')
|
|
|
|
mime_type, encoding = MimeTypes().guess_type(image_file)
|
|
self.set_header('Content-Type', mime_type)
|
|
with open(image_file, 'rb') as io_stream:
|
|
return io_stream.read()
|