You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
bazarr/list_subtitles.py

229 lines
9.5 KiB

import gc
import os
import enzyme
import babelfish
from subliminal import *
import pycountry
import sqlite3
import ast
import langdetect
from bs4 import UnicodeDammit
from itertools import islice
from get_general_settings import *
gc.enable()
def store_subtitles(file):
languages = []
actual_subtitles = []
if os.path.exists(file):
if os.path.splitext(file)[1] == '.mkv':
try:
with open(file, 'rb') as f:
mkv = enzyme.MKV(f)
for subtitle_track in mkv.subtitle_tracks:
try:
actual_subtitles.append([str(pycountry.languages.lookup(subtitle_track.language).alpha_2),None])
except:
pass
except:
pass
subtitles = core.search_external_subtitles(file)
for subtitle, language in subtitles.iteritems():
if str(language) != 'und':
actual_subtitles.append([str(language), path_replace_reverse(os.path.join(os.path.dirname(file), subtitle))])
else:
with open(path_replace(os.path.join(os.path.dirname(file), subtitle)), 'r') as f:
text = list(islice(f, 100))
text = ' '.join(text)
encoding = UnicodeDammit(text)
try:
text = text.decode(encoding.original_encoding)
except Exception as e:
logging.exception('Error trying to detect character encoding for this subtitles file: ' + path_replace(os.path.join(os.path.dirname(file), subtitle)) + ' You should try to delete this subtitles file manually and ask Bazarr to download it again.')
else:
detected_language = langdetect.detect(text)
if len(detected_language) > 0:
actual_subtitles.append([str(detected_language), path_replace_reverse(os.path.join(os.path.dirname(file), subtitle))])
conn_db = sqlite3.connect(os.path.join(os.path.dirname(__file__), 'data/db/bazarr.db'), timeout=30)
c_db = conn_db.cursor()
c_db.execute("UPDATE table_episodes SET subtitles = ? WHERE path = ?", (str(actual_subtitles), path_replace_reverse(file)))
conn_db.commit()
c_db.close()
return actual_subtitles
def store_subtitles_movie(file):
languages = []
actual_subtitles = []
if os.path.exists(file):
if os.path.splitext(file)[1] == '.mkv':
try:
with open(file, 'rb') as f:
mkv = enzyme.MKV(f)
for subtitle_track in mkv.subtitle_tracks:
try:
actual_subtitles.append([str(pycountry.languages.lookup(subtitle_track.language).alpha_2), None])
except:
pass
except:
pass
subtitles = core.search_external_subtitles(file)
for subtitle, language in subtitles.iteritems():
if str(language) != 'und':
actual_subtitles.append([str(language), path_replace_reverse_movie(os.path.join(os.path.dirname(file), subtitle))])
else:
if os.path.splitext(subtitle)[1] != ".sub":
with open(path_replace_movie(os.path.join(os.path.dirname(file), subtitle)), 'r') as f:
text = list(islice(f, 100))
text = ' '.join(text)
encoding = UnicodeDammit(text)
try:
text = text.decode(encoding.original_encoding)
except Exception as e:
logging.exception('Error trying to detect character encoding for this subtitles file: ' + path_replace_movie(os.path.join(os.path.dirname(file), subtitle)) + ' You should try to delete this subtitles file manually and ask Bazarr to download it again.')
else:
detected_language = langdetect.detect(text)
if len(detected_language) > 0:
actual_subtitles.append([str(detected_language), path_replace_reverse_movie(os.path.join(os.path.dirname(file), subtitle))])
conn_db = sqlite3.connect(os.path.join(os.path.dirname(__file__), 'data/db/bazarr.db'), timeout=30)
c_db = conn_db.cursor()
c_db.execute("UPDATE table_movies SET subtitles = ? WHERE path = ?", (str(actual_subtitles), path_replace_reverse_movie(file)))
conn_db.commit()
c_db.close()
return actual_subtitles
def list_missing_subtitles(*no):
query_string = ''
try:
query_string = " WHERE table_shows.sonarrSeriesId = " + str(no[0])
except:
pass
conn_db = sqlite3.connect(os.path.join(os.path.dirname(__file__), 'data/db/bazarr.db'), timeout=30)
c_db = conn_db.cursor()
episodes_subtitles = c_db.execute("SELECT table_episodes.sonarrEpisodeId, table_episodes.subtitles, table_shows.languages FROM table_episodes INNER JOIN table_shows on table_episodes.sonarrSeriesId = table_shows.sonarrSeriesId" + query_string).fetchall()
c_db.close()
missing_subtitles_global = []
for episode_subtitles in episodes_subtitles:
actual_subtitles = []
desired_subtitles = []
missing_subtitles = []
if episode_subtitles[1] != None:
actual_subtitles = ast.literal_eval(episode_subtitles[1])
if episode_subtitles[2] != None:
desired_subtitles = ast.literal_eval(episode_subtitles[2])
actual_subtitles_list = []
if desired_subtitles == None:
missing_subtitles_global.append(tuple(['[]', episode_subtitles[0]]))
else:
for item in actual_subtitles:
actual_subtitles_list.append(item[0])
missing_subtitles = list(set(desired_subtitles) - set(actual_subtitles_list))
missing_subtitles_global.append(tuple([str(missing_subtitles), episode_subtitles[0]]))
conn_db = sqlite3.connect(os.path.join(os.path.dirname(__file__), 'data/db/bazarr.db'), timeout=30)
c_db = conn_db.cursor()
c_db.executemany("UPDATE table_episodes SET missing_subtitles = ? WHERE sonarrEpisodeId = ?", (missing_subtitles_global))
conn_db.commit()
c_db.close()
def list_missing_subtitles_movies(*no):
query_string = ''
try:
query_string = " WHERE table_movies.radarrId = " + str(no[0])
except:
pass
conn_db = sqlite3.connect(os.path.join(os.path.dirname(__file__), 'data/db/bazarr.db'), timeout=30)
c_db = conn_db.cursor()
movies_subtitles = c_db.execute("SELECT radarrId, subtitles, languages FROM table_movies" + query_string).fetchall()
c_db.close()
missing_subtitles_global = []
for movie_subtitles in movies_subtitles:
actual_subtitles = []
desired_subtitles = []
missing_subtitles = []
if movie_subtitles[1] != None:
actual_subtitles = ast.literal_eval(movie_subtitles[1])
if movie_subtitles[2] != None:
desired_subtitles = ast.literal_eval(movie_subtitles[2])
actual_subtitles_list = []
if desired_subtitles == None:
missing_subtitles_global.append(tuple(['[]', movie_subtitles[0]]))
else:
for item in actual_subtitles:
actual_subtitles_list.append(item[0])
missing_subtitles = list(set(desired_subtitles) - set(actual_subtitles_list))
missing_subtitles_global.append(tuple([str(missing_subtitles), movie_subtitles[0]]))
conn_db = sqlite3.connect(os.path.join(os.path.dirname(__file__), 'data/db/bazarr.db'), timeout=30)
c_db = conn_db.cursor()
c_db.executemany("UPDATE table_movies SET missing_subtitles = ? WHERE radarrId = ?", (missing_subtitles_global))
conn_db.commit()
c_db.close()
def series_full_scan_subtitles():
conn_db = sqlite3.connect(os.path.join(os.path.dirname(__file__), 'data/db/bazarr.db'), timeout=30)
c_db = conn_db.cursor()
episodes = c_db.execute("SELECT path FROM table_episodes").fetchall()
c_db.close()
for episode in episodes:
store_subtitles(path_replace(episode[0]))
gc.collect()
def movies_full_scan_subtitles():
conn_db = sqlite3.connect(os.path.join(os.path.dirname(__file__), 'data/db/bazarr.db'), timeout=30)
c_db = conn_db.cursor()
movies = c_db.execute("SELECT path FROM table_movies").fetchall()
c_db.close()
for movie in movies:
store_subtitles_movie(path_replace_movie(movie[0]))
gc.collect()
def series_scan_subtitles(no):
conn_db = sqlite3.connect(os.path.join(os.path.dirname(__file__), 'data/db/bazarr.db'), timeout=30)
c_db = conn_db.cursor()
episodes = c_db.execute("SELECT path FROM table_episodes WHERE sonarrSeriesId = ?", (no,)).fetchall()
c_db.close()
for episode in episodes:
store_subtitles(path_replace(episode[0]))
list_missing_subtitles(no)
def movies_scan_subtitles(no):
conn_db = sqlite3.connect(os.path.join(os.path.dirname(__file__), 'data/db/bazarr.db'), timeout=30)
c_db = conn_db.cursor()
movies = c_db.execute("SELECT path FROM table_movies WHERE radarrId = ?", (no,)).fetchall()
c_db.close()
for movie in movies:
store_subtitles_movie(path_replace_movie(movie[0]))
list_missing_subtitles_movies(no)