aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorJesús <heckyel@hyperbola.info>2020-12-15 21:52:04 -0500
committerJesús <heckyel@hyperbola.info>2020-12-15 21:52:04 -0500
commitb9a3082e7c654d97cecc5410c086e13a7b046909 (patch)
treedc093ccb2246e7df0beaca5deb8b7467620b74a2
parentf4b36a220d085080a881dbe2f63e51b2fb28a003 (diff)
downloadyt-local-b9a3082e7c654d97cecc5410c086e13a7b046909.tar.lz
yt-local-b9a3082e7c654d97cecc5410c086e13a7b046909.tar.xz
yt-local-b9a3082e7c654d97cecc5410c086e13a7b046909.zip
pep8
-rw-r--r--youtube/__init__.py17
-rw-r--r--youtube/channel.py65
-rw-r--r--youtube/comments.py57
-rw-r--r--youtube/local_playlist.py23
-rw-r--r--youtube/playlist.py27
-rw-r--r--youtube/proto.py18
-rw-r--r--youtube/search.py20
-rw-r--r--youtube/subscriptions.py121
-rw-r--r--youtube/util.py66
9 files changed, 248 insertions, 166 deletions
diff --git a/youtube/__init__.py b/youtube/__init__.py
index a8ca227..d9edbc6 100644
--- a/youtube/__init__.py
+++ b/youtube/__init__.py
@@ -12,10 +12,9 @@ yt_app.url_map.strict_slashes = False
# yt_app.jinja_env.lstrip_blocks = True
-
-
yt_app.add_url_rule('/settings', 'settings_page', settings.settings_page, methods=['POST', 'GET'])
+
@yt_app.route('/')
def homepage():
return flask.render_template('home.html', title="Youtube local")
@@ -27,6 +26,7 @@ theme_names = {
2: 'dark_theme',
}
+
@yt_app.context_processor
def inject_theme_preference():
return {
@@ -34,6 +34,7 @@ def inject_theme_preference():
'settings': settings,
}
+
@yt_app.template_filter('commatize')
def commatize(num):
if num is None:
@@ -42,6 +43,7 @@ def commatize(num):
num = int(num)
return '{:,}'.format(num)
+
def timestamp_replacement(match):
time_seconds = 0
for part in match.group(0).split(':'):
@@ -53,11 +55,15 @@ def timestamp_replacement(match):
+ '</a>'
)
+
TIMESTAMP_RE = re.compile(r'\b(\d?\d:)?\d?\d:\d\d\b')
+
+
@yt_app.template_filter('timestamps')
def timestamps(text):
return TIMESTAMP_RE.sub(timestamp_replacement, text)
+
@yt_app.errorhandler(500)
def error_page(e):
slim = request.args.get('slim', False) # whether it was an ajax request
@@ -75,6 +81,7 @@ def error_page(e):
return flask.render_template('error.html', error_message=error_message, slim=slim), 502
return flask.render_template('error.html', traceback=traceback.format_exc(), slim=slim), 500
+
font_choices = {
0: 'initial',
1: 'arial, "liberation sans", sans-serif',
@@ -83,11 +90,13 @@ font_choices = {
4: 'tahoma, sans-serif',
}
+
@yt_app.route('/shared.css')
def get_css():
return flask.Response(
- flask.render_template('shared.css',
- font_family = font_choices[settings.font]
+ flask.render_template(
+ 'shared.css',
+ font_family=font_choices[settings.font]
),
mimetype='text/css',
)
diff --git a/youtube/channel.py b/youtube/channel.py
index e9cc87b..6f40965 100644
--- a/youtube/channel.py
+++ b/youtube/channel.py
@@ -51,7 +51,7 @@ def channel_ctoken_v3(channel_id, page, sort, tab, view=1):
proto.string(1, proto.unpadded_b64encode(proto.uint(1,offset)))
))
- tab = proto.string(2, tab )
+ tab = proto.string(2, tab)
sort = proto.uint(3, int(sort))
shelf_view = proto.uint(4, 0)
@@ -60,11 +60,12 @@ def channel_ctoken_v3(channel_id, page, sort, tab, view=1):
proto.percent_b64encode(tab + sort + shelf_view + view + page_token)
)
- channel_id = proto.string(2, channel_id )
+ channel_id = proto.string(2, channel_id)
pointless_nest = proto.string(80226972, channel_id + continuation_info)
return base64.urlsafe_b64encode(pointless_nest).decode('ascii')
+
def channel_ctoken_v2(channel_id, page, sort, tab, view=1):
# see https://github.com/iv-org/invidious/issues/1319#issuecomment-671732646
# page > 1 doesn't work when sorting by oldest
@@ -74,41 +75,44 @@ def channel_ctoken_v2(channel_id, page, sort, tab, view=1):
2: 17254859483345278706,
1: 16570086088270825023,
}[int(sort)]
- page_token = proto.string(61, proto.unpadded_b64encode(proto.string(1,
- proto.uint(1, schema_number) + proto.string(2,
- proto.string(1, proto.unpadded_b64encode(proto.uint(1,offset)))
- )
- )))
+ page_token = proto.string(61, proto.unpadded_b64encode(
+ proto.string(1, proto.uint(1, schema_number) + proto.string(
+ 2,
+ proto.string(1, proto.unpadded_b64encode(proto.uint(1, offset)))
+ ))))
- tab = proto.string(2, tab )
+ tab = proto.string(2, tab)
sort = proto.uint(3, int(sort))
- #page = proto.string(15, str(page) )
+ # page = proto.string(15, str(page) )
shelf_view = proto.uint(4, 0)
view = proto.uint(6, int(view))
- continuation_info = proto.string(3,
+ continuation_info = proto.string(
+ 3,
proto.percent_b64encode(tab + sort + shelf_view + view + page_token)
)
- channel_id = proto.string(2, channel_id )
+ channel_id = proto.string(2, channel_id)
pointless_nest = proto.string(80226972, channel_id + continuation_info)
return base64.urlsafe_b64encode(pointless_nest).decode('ascii')
+
def channel_ctoken_v1(channel_id, page, sort, tab, view=1):
- tab = proto.string(2, tab )
+ tab = proto.string(2, tab)
sort = proto.uint(3, int(sort))
- page = proto.string(15, str(page) )
+ page = proto.string(15, str(page))
# example with shelves in videos tab: https://www.youtube.com/channel/UCNL1ZadSjHpjm4q9j2sVtOA/videos
shelf_view = proto.uint(4, 0)
view = proto.uint(6, int(view))
continuation_info = proto.string(3, proto.percent_b64encode(tab + view + sort + shelf_view + page + proto.uint(23, 0)) )
- channel_id = proto.string(2, channel_id )
+ channel_id = proto.string(2, channel_id)
pointless_nest = proto.string(80226972, channel_id + continuation_info)
return base64.urlsafe_b64encode(pointless_nest).decode('ascii')
+
def get_channel_tab(channel_id, page="1", sort=3, tab='videos', view=1, print_status=True):
message = 'Got channel tab' if print_status else None
@@ -118,18 +122,21 @@ def get_channel_tab(channel_id, page="1", sort=3, tab='videos', view=1, print_st
url = ('https://www.youtube.com/channel/' + channel_id + '/' + tab
+ '?action_continuation=1&continuation=' + ctoken
+ '&pbj=1')
- content = util.fetch_url(url, headers_desktop + real_cookie,
+ content = util.fetch_url(
+ url, headers_desktop + real_cookie,
debug_name='channel_tab', report_text=message)
else:
ctoken = channel_ctoken_v3(channel_id, page, sort, tab, view)
ctoken = ctoken.replace('=', '%3D')
url = 'https://www.youtube.com/browse_ajax?ctoken=' + ctoken
- content = util.fetch_url(url,
+ content = util.fetch_url(
+ url,
headers_desktop + generic_cookie,
debug_name='channel_tab', report_text=message)
return content
+
# cache entries expire after 30 minutes
@cachetools.func.ttl_cache(maxsize=128, ttl=30*60)
def get_number_of_videos_channel(channel_id):
@@ -157,22 +164,28 @@ def get_number_of_videos_channel(channel_id):
else:
return 0
+
channel_id_re = re.compile(r'videos\.xml\?channel_id=([a-zA-Z0-9_-]{24})"')
+
+
@cachetools.func.lru_cache(maxsize=128)
def get_channel_id(base_url):
# method that gives the smallest possible response at ~4 kb
# needs to be as fast as possible
base_url = base_url.replace('https://www', 'https://m') # avoid redirect
- response = util.fetch_url(base_url + '/about?pbj=1', headers_mobile,
+ response = util.fetch_url(
+ base_url + '/about?pbj=1', headers_mobile,
debug_name='get_channel_id', report_text='Got channel id').decode('utf-8')
match = channel_id_re.search(response)
if match:
return match.group(1)
return None
+
def get_number_of_videos_general(base_url):
return get_number_of_videos_channel(get_channel_id(base_url))
+
def get_channel_search_json(channel_id, query, page):
params = proto.string(2, 'search') + proto.string(15, str(page))
params = proto.percent_b64encode(params)
@@ -192,15 +205,14 @@ def post_process_channel_info(info):
util.add_extra_html_info(item)
-
-
-
playlist_sort_codes = {'2': "da", '3': "dd", '4': "lad"}
# youtube.com/[channel_id]/[tab]
# youtube.com/user/[username]/[tab]
# youtube.com/c/[custom]/[tab]
# youtube.com/[custom]/[tab]
+
+
def get_channel_page_general_url(base_url, tab, request, channel_id=None):
page_number = int(request.args.get('page', 1))
@@ -236,10 +248,9 @@ def get_channel_page_general_url(base_url, tab, request, channel_id=None):
else:
flask.abort(404, 'Unknown channel tab: ' + tab)
-
info = yt_data_extract.extract_channel_info(json.loads(polymer_json), tab)
if info['error'] is not None:
- return flask.render_template('error.html', error_message = info['error'])
+ return flask.render_template('error.html', error_message=info['error'])
post_process_channel_info(info)
if tab == 'videos':
@@ -254,28 +265,32 @@ def get_channel_page_general_url(base_url, tab, request, channel_id=None):
info['page_number'] = page_number
info['subscribed'] = subscriptions.is_subscribed(info['channel_id'])
- return flask.render_template('channel.html',
- parameters_dictionary = request.args,
+ return flask.render_template(
+ 'channel.html',
+ parameters_dictionary=request.args,
**info
)
+
@yt_app.route('/channel/<channel_id>/')
@yt_app.route('/channel/<channel_id>/<tab>')
def get_channel_page(channel_id, tab='videos'):
return get_channel_page_general_url('https://www.youtube.com/channel/' + channel_id, tab, request, channel_id)
+
@yt_app.route('/user/<username>/')
@yt_app.route('/user/<username>/<tab>')
def get_user_page(username, tab='videos'):
return get_channel_page_general_url('https://www.youtube.com/user/' + username, tab, request)
+
@yt_app.route('/c/<custom>/')
@yt_app.route('/c/<custom>/<tab>')
def get_custom_c_page(custom, tab='videos'):
return get_channel_page_general_url('https://www.youtube.com/c/' + custom, tab, request)
+
@yt_app.route('/<custom>')
@yt_app.route('/<custom>/<tab>')
def get_toplevel_custom_page(custom, tab='videos'):
return get_channel_page_general_url('https://www.youtube.com/' + custom, tab, request)
-
diff --git a/youtube/comments.py b/youtube/comments.py
index 2fb1fa2..fc353f9 100644
--- a/youtube/comments.py
+++ b/youtube/comments.py
@@ -25,12 +25,13 @@ from flask import request
# *Old ASJN's continue to work, and start at the same comment even if new comments have been posted since
# *The ASJN has no relation with any of the data in the response it came from
+
def make_comment_ctoken(video_id, sort=0, offset=0, lc='', secret_key=''):
video_id = proto.as_bytes(video_id)
secret_key = proto.as_bytes(secret_key)
+ page_info = proto.string(4, video_id) + proto.uint(6, sort)
- page_info = proto.string(4,video_id) + proto.uint(6, sort)
offset_information = proto.nested(4, page_info) + proto.uint(5, offset)
if secret_key:
offset_information = proto.string(1, secret_key) + offset_information
@@ -39,19 +40,19 @@ def make_comment_ctoken(video_id, sort=0, offset=0, lc='', secret_key=''):
if lc:
page_params += proto.string(6, proto.percent_b64encode(proto.string(15, lc)))
- result = proto.nested(2, page_params) + proto.uint(3,6) + proto.nested(6, offset_information)
+ result = proto.nested(2, page_params) + proto.uint(3, 6) + proto.nested(6, offset_information)
return base64.urlsafe_b64encode(result).decode('ascii')
+
def comment_replies_ctoken(video_id, comment_id, max_results=500):
params = proto.string(2, comment_id) + proto.uint(9, max_results)
params = proto.nested(3, params)
- result = proto.nested(2, proto.string(2, video_id)) + proto.uint(3,6) + proto.nested(6, params)
+ result = proto.nested(2, proto.string(2, video_id)) + proto.uint(3, 6) + proto.nested(6, params)
return base64.urlsafe_b64encode(result).decode('ascii')
-
mobile_headers = {
'User-Agent': 'Mozilla/5.0 (iPhone; CPU iPhone OS 10_3_1 like Mac OS X) AppleWebKit/603.1.30 (KHTML, like Gecko) Version/10.0 Mobile/14E304 Safari/602.1',
'Accept': '*/*',
@@ -59,6 +60,8 @@ mobile_headers = {
'X-YouTube-Client-Name': '2',
'X-YouTube-Client-Version': '2.20180823',
}
+
+
def request_comments(ctoken, replies=False):
if replies: # let's make it use different urls for no reason despite all the data being encoded
base_url = "https://m.youtube.com/watch_comment?action_get_comment_replies=1&ctoken="
@@ -66,7 +69,7 @@ def request_comments(ctoken, replies=False):
base_url = "https://m.youtube.com/watch_comment?action_get_comments=1&ctoken="
url = base_url + ctoken.replace("=", "%3D") + "&pbj=1"
- for i in range(0,8): # don't retry more than 8 times
+ for i in range(0, 8): # don't retry more than 8 times
content = util.fetch_url(url, headers=mobile_headers, report_text="Retrieved comments", debug_name='request_comments')
if content[0:4] == b")]}'": # random closing characters included at beginning of response for some reason
content = content[4:]
@@ -81,13 +84,13 @@ def request_comments(ctoken, replies=False):
def single_comment_ctoken(video_id, comment_id):
- page_params = proto.string(2, video_id) + proto.string(6, proto.percent_b64encode(proto.string(15, comment_id)))
+ page_params = proto.string(2, video_id) + proto.string(
+ 6, proto.percent_b64encode(proto.string(15, comment_id)))
- result = proto.nested(2, page_params) + proto.uint(3,6)
+ result = proto.nested(2, page_params) + proto.uint(3, 6)
return base64.urlsafe_b64encode(result).decode('ascii')
-
def post_process_comments_info(comments_info):
for comment in comments_info['comments']:
comment['author_url'] = concat_or_none(
@@ -95,15 +98,17 @@ def post_process_comments_info(comments_info):
comment['author_avatar'] = concat_or_none(
settings.img_prefix, comment['author_avatar'])
- comment['permalink'] = concat_or_none(util.URL_ORIGIN, '/watch?v=',
+ comment['permalink'] = concat_or_none(
+ util.URL_ORIGIN, '/watch?v=',
comments_info['video_id'], '&lc=', comment['id'])
-
reply_count = comment['reply_count']
+
if reply_count == 0:
comment['replies_url'] = None
else:
- comment['replies_url'] = concat_or_none(util.URL_ORIGIN,
+ comment['replies_url'] = concat_or_none(
+ util.URL_ORIGIN,
'/comments?parent_id=', comment['id'],
'&video_id=', comments_info['video_id'])
@@ -122,18 +127,25 @@ def post_process_comments_info(comments_info):
comments_info['include_avatars'] = settings.enable_comment_avatars
if comments_info['ctoken']:
- comments_info['more_comments_url'] = concat_or_none(util.URL_ORIGIN,
- '/comments?ctoken=', comments_info['ctoken'])
+ comments_info['more_comments_url'] = concat_or_none(
+ util.URL_ORIGIN,
+ '/comments?ctoken=',
+ comments_info['ctoken']
+ )
comments_info['page_number'] = page_number = str(int(comments_info['offset']/20) + 1)
if not comments_info['is_replies']:
comments_info['sort_text'] = 'top' if comments_info['sort'] == 0 else 'newest'
+ comments_info['video_url'] = concat_or_none(
+ util.URL_ORIGIN,
+ '/watch?v=',
+ comments_info['video_id']
+ )
- comments_info['video_url'] = concat_or_none(util.URL_ORIGIN,
- '/watch?v=', comments_info['video_id'])
- comments_info['video_thumbnail'] = concat_or_none(settings.img_prefix, 'https://i.ytimg.com/vi/',
+ comments_info['video_thumbnail'] = concat_or_none(
+ settings.img_prefix, 'https://i.ytimg.com/vi/',
comments_info['video_id'], '/mqdefault.jpg')
@@ -183,7 +195,6 @@ def video_comments(video_id, sort=0, offset=0, lc='', secret_key=''):
return comments_info
-
@yt_app.route('/comments')
def get_comments_page():
ctoken = request.args.get('ctoken', '')
@@ -195,7 +206,9 @@ def get_comments_page():
ctoken = comment_replies_ctoken(video_id, parent_id)
replies = True
- comments_info = yt_data_extract.extract_comments_info(request_comments(ctoken, replies))
+ comments_info = yt_data_extract.extract_comments_info(
+ request_comments(ctoken, replies))
+
post_process_comments_info(comments_info)
if not replies:
@@ -203,8 +216,8 @@ def get_comments_page():
other_sort_text = 'Sort by ' + ('newest' if comments_info['sort'] == 0 else 'top')
comments_info['comment_links'] = [(other_sort_text, other_sort_url)]
- return flask.render_template('comments_page.html',
- comments_info = comments_info,
- slim = request.args.get('slim', False)
+ return flask.render_template(
+ 'comments_page.html',
+ comments_info=comments_info,
+ slim=request.args.get('slim', False)
)
-
diff --git a/youtube/local_playlist.py b/youtube/local_playlist.py
index 891bb76..776e992 100644
--- a/youtube/local_playlist.py
+++ b/youtube/local_playlist.py
@@ -15,6 +15,7 @@ from flask import request
playlists_directory = os.path.join(settings.data_dir, "playlists")
thumbnails_directory = os.path.join(settings.data_dir, "playlist_thumbnails")
+
def video_ids_in_playlist(name):
try:
with open(os.path.join(playlists_directory, name + ".txt"), 'r', encoding='utf-8') as file:
@@ -23,6 +24,7 @@ def video_ids_in_playlist(name):
except FileNotFoundError:
return set()
+
def add_to_playlist(name, video_info_list):
if not os.path.exists(playlists_directory):
os.makedirs(playlists_directory)
@@ -65,6 +67,7 @@ def get_local_playlist_videos(name, offset=0, amount=50):
gevent.spawn(util.download_thumbnails, os.path.join(thumbnails_directory, name), missing_thumbnails)
return videos[offset:offset+amount], len(videos)
+
def get_playlist_names():
try:
items = os.listdir(playlists_directory)
@@ -75,6 +78,7 @@ def get_playlist_names():
if ext == '.txt':
yield name
+
def remove_from_playlist(name, video_info_list):
ids = [json.loads(video)['id'] for video in video_info_list]
with open(os.path.join(playlists_directory, name + ".txt"), 'r', encoding='utf-8') as file:
@@ -109,14 +113,16 @@ def get_local_playlist_page(playlist_name=None):
page = int(request.args.get('page', 1))
offset = 50*(page - 1)
videos, num_videos = get_local_playlist_videos(playlist_name, offset=offset, amount=50)
- return flask.render_template('local_playlist.html',
- header_playlist_names = get_playlist_names(),
- playlist_name = playlist_name,
- videos = videos,
- num_pages = math.ceil(num_videos/50),
- parameters_dictionary = request.args,
+ return flask.render_template(
+ 'local_playlist.html',
+ header_playlist_names=get_playlist_names(),
+ playlist_name=playlist_name,
+ videos=videos,
+ num_pages=math.ceil(num_videos/50),
+ parameters_dictionary=request.args,
)
+
@yt_app.route('/playlists/<playlist_name>', methods=['POST'])
def path_edit_playlist(playlist_name):
'''Called when making changes to the playlist from that playlist's page'''
@@ -128,6 +134,7 @@ def path_edit_playlist(playlist_name):
else:
flask.abort(400)
+
@yt_app.route('/edit_playlist', methods=['POST'])
def edit_playlist():
'''Called when adding videos to a playlist from elsewhere'''
@@ -137,7 +144,9 @@ def edit_playlist():
else:
flask.abort(400)
+
@yt_app.route('/data/playlist_thumbnails/<playlist_name>/<thumbnail>')
def serve_thumbnail(playlist_name, thumbnail):
# .. is necessary because flask always uses the application directory at ./youtube, not the working directory
- return flask.send_from_directory(os.path.join('..', thumbnails_directory, playlist_name), thumbnail)
+ return flask.send_from_directory(
+ os.path.join('..', thumbnails_directory, playlist_name), thumbnail)
diff --git a/youtube/playlist.py b/youtube/playlist.py
index e596eae..64c717e 100644
--- a/youtube/playlist.py
+++ b/youtube/playlist.py
@@ -12,9 +12,6 @@ from flask import request
import flask
-
-
-
def playlist_ctoken(playlist_id, offset):
offset = proto.uint(1, offset)
@@ -22,9 +19,9 @@ def playlist_ctoken(playlist_id, offset):
offset = b'PT:' + proto.unpadded_b64encode(offset)
offset = proto.string(15, offset)
- continuation_info = proto.string( 3, proto.percent_b64encode(offset) )
+ continuation_info = proto.string(3, proto.percent_b64encode(offset))
- playlist_id = proto.string(2, 'VL' + playlist_id )
+ playlist_id = proto.string(2, 'VL' + playlist_id)
pointless_nest = proto.string(80226972, playlist_id + continuation_info)
return base64.urlsafe_b64encode(pointless_nest).decode('ascii')
@@ -46,7 +43,8 @@ headers_1 = (
('X-YouTube-Client-Version', '2.20180614'),
)
-def playlist_first_page(playlist_id, report_text = "Retrieved playlist"):
+
+def playlist_first_page(playlist_id, report_text="Retrieved playlist"):
url = 'https://m.youtube.com/playlist?list=' + playlist_id + '&pbj=1'
content = util.fetch_url(url, util.mobile_ua + headers_1, report_text=report_text, debug_name='playlist_first_page')
content = json.loads(util.uppercase_escape(content.decode('utf-8')))
@@ -66,7 +64,9 @@ def get_videos(playlist_id, page):
'X-YouTube-Client-Version': '2.20180508',
}
- content = util.fetch_url(url, headers, report_text="Retrieved playlist", debug_name='playlist_videos')
+ content = util.fetch_url(
+ url, headers,
+ report_text="Retrieved playlist", debug_name='playlist_videos')
info = json.loads(util.uppercase_escape(content.decode('utf-8')))
return info
@@ -94,7 +94,7 @@ def get_playlist_page():
info = yt_data_extract.extract_playlist_info(this_page_json)
if info['error']:
- return flask.render_template('error.html', error_message = info['error'])
+ return flask.render_template('error.html', error_message=info['error'])
if page != '1':
info['metadata'] = yt_data_extract.extract_playlist_metadata(first_page_json)
@@ -114,11 +114,12 @@ def get_playlist_page():
if video_count is None:
video_count = 40
- return flask.render_template('playlist.html',
- header_playlist_names = local_playlist.get_playlist_names(),
- video_list = info.get('items', []),
- num_pages = math.ceil(video_count/20),
- parameters_dictionary = request.args,
+ return flask.render_template(
+ 'playlist.html',
+ header_playlist_names=local_playlist.get_playlist_names(),
+ video_list=info.get('items', []),
+ num_pages=math.ceil(video_count/20),
+ parameters_dictionary=request.args,
**info['metadata']
).encode('utf-8')
diff --git a/youtube/proto.py b/youtube/proto.py
index 5fd16d5..ff59eac 100644
--- a/youtube/proto.py
+++ b/youtube/proto.py
@@ -2,6 +2,7 @@ from math import ceil
import base64
import io
+
def byte(n):
return bytes((n,))
@@ -19,7 +20,7 @@ def varint_encode(offset):
for i in range(0, needed_bytes - 1):
encoded_bytes[i] = (offset & 127) | 128 # 7 least significant bits
offset = offset >> 7
- encoded_bytes[-1] = offset & 127 # leave first bit as zero for last byte
+ encoded_bytes[-1] = offset & 127 # leave first bit as zero for last byte
return bytes(encoded_bytes)
@@ -37,18 +38,18 @@ def varint_decode(encoded):
def string(field_number, data):
data = as_bytes(data)
return _proto_field(2, field_number, varint_encode(len(data)) + data)
+
+
nested = string
+
def uint(field_number, value):
return _proto_field(0, field_number, varint_encode(value))
-
-
def _proto_field(wire_type, field_number, data):
''' See https://developers.google.com/protocol-buffers/docs/encoding#structure '''
- return varint_encode( (field_number << 3) | wire_type) + data
-
+ return varint_encode((field_number << 3) | wire_type) + data
def percent_b64encode(data):
@@ -58,6 +59,7 @@ def percent_b64encode(data):
def unpadded_b64encode(data):
return base64.urlsafe_b64encode(data).replace(b'=', b'')
+
def as_bytes(value):
if isinstance(value, str):
return value.encode('utf-8')
@@ -90,6 +92,7 @@ def read_group(data, end_sequence):
data.seek(index + len(end_sequence))
return data.original[start:index]
+
def read_protobuf(data):
data_original = data
data = io.BytesIO(data)
@@ -118,12 +121,13 @@ def read_protobuf(data):
raise Exception("Unknown wire type: " + str(wire_type) + ", Tag: " + bytes_to_hex(succinct_encode(tag)) + ", at position " + str(data.tell()))
yield (wire_type, field_number, value)
+
def parse(data):
return {field_number: value for _, field_number, value in read_protobuf(data)}
+
def b64_to_bytes(data):
if isinstance(data, bytes):
data = data.decode('ascii')
data = data.replace("%3D", "=")
- return base64.urlsafe_b64decode(data + "="*((4 - len(data)%4)%4) )
-
+ return base64.urlsafe_b64decode(data + "="*((4 - len(data)%4)%4))
diff --git a/youtube/search.py b/youtube/search.py
index 34df76f..da8cdab 100644
--- a/youtube/search.py
+++ b/youtube/search.py
@@ -78,7 +78,7 @@ def get_search_page():
search_info = yt_data_extract.extract_search_info(polymer_json)
if search_info['error']:
- return flask.render_template('error.html', error_message = search_info['error'])
+ return flask.render_template('error.html', error_message=search_info['error'])
for extract_item_info in search_info['items']:
util.prefix_urls(extract_item_info)
@@ -95,16 +95,18 @@ def get_search_page():
no_autocorrect_query_url = util.URL_ORIGIN + '/search?' + urllib.parse.urlencode(no_autocorrect_query_string, doseq=True)
corrections['original_query_url'] = no_autocorrect_query_url
- return flask.render_template('search.html',
- header_playlist_names = local_playlist.get_playlist_names(),
- query = query,
- estimated_results = search_info['estimated_results'],
- estimated_pages = search_info['estimated_pages'],
- corrections = search_info['corrections'],
- results = search_info['items'],
- parameters_dictionary = request.args,
+ return flask.render_template(
+ 'search.html',
+ header_playlist_names=local_playlist.get_playlist_names(),
+ query=query,
+ estimated_results=search_info['estimated_results'],
+ estimated_pages=search_info['estimated_pages'],
+ corrections=search_info['corrections'],
+ results=search_info['items'],
+ parameters_dictionary=request.args,
)
+
@yt_app.route('/opensearch.xml')
def get_search_engine_xml():
with open(os.path.join(settings.program_directory, 'youtube/opensearch.xml'), 'rb') as f:
diff --git a/youtube/subscriptions.py b/youtube/subscriptions.py
index 6f75578..b841f5d 100644
--- a/youtube/subscriptions.py
+++ b/youtube/subscriptions.py
@@ -26,6 +26,7 @@ thumbnails_directory = os.path.join(settings.data_dir, "subscription_thumbnails"
database_path = os.path.join(settings.data_dir, "subscriptions.sqlite")
+
def open_database():
if not os.path.exists(settings.data_dir):
os.makedirs(settings.data_dir)
@@ -74,11 +75,13 @@ def open_database():
# https://stackoverflow.com/questions/19522505/using-sqlite3-in-python-with-with-keyword
return contextlib.closing(connection)
+
def with_open_db(function, *args, **kwargs):
with open_database() as connection:
with connection as cursor:
return function(cursor, *args, **kwargs)
+
def _is_subscribed(cursor, channel_id):
result = cursor.execute('''SELECT EXISTS(
SELECT 1
@@ -88,12 +91,14 @@ def _is_subscribed(cursor, channel_id):
)''', [channel_id]).fetchone()
return bool(result[0])
+
def is_subscribed(channel_id):
if not os.path.exists(database_path):
return False
return with_open_db(_is_subscribed, channel_id)
+
def _subscribe(channels):
''' channels is a list of (channel_id, channel_name) '''
channels = list(channels)
@@ -101,7 +106,8 @@ def _subscribe(channels):
with connection as cursor:
channel_ids_to_check = [channel[0] for channel in channels if not _is_subscribed(cursor, channel[0])]
- rows = ( (channel_id, channel_name, 0, 0) for channel_id, channel_name in channels)
+ rows = ((channel_id, channel_name, 0, 0) for channel_id,
+ channel_name in channels)
cursor.executemany('''INSERT OR IGNORE INTO subscribed_channels (yt_channel_id, channel_name, time_last_checked, next_check_time)
VALUES (?, ?, ?, ?)''', rows)
@@ -111,6 +117,7 @@ def _subscribe(channels):
channel_names.update(channels)
check_channels_if_necessary(channel_ids_to_check)
+
def delete_thumbnails(to_delete):
for thumbnail in to_delete:
try:
@@ -122,6 +129,7 @@ def delete_thumbnails(to_delete):
print('Failed to delete thumbnail: ' + thumbnail)
traceback.print_exc()
+
def _unsubscribe(cursor, channel_ids):
''' channel_ids is a list of channel_ids '''
to_delete = []
@@ -138,7 +146,8 @@ def _unsubscribe(cursor, channel_ids):
gevent.spawn(delete_thumbnails, to_delete)
cursor.executemany("DELETE FROM subscribed_channels WHERE yt_channel_id=?", ((channel_id, ) for channel_id in channel_ids))
-def _get_videos(cursor, number_per_page, offset, tag = None):
+
+def _get_videos(cursor, number_per_page, offset, tag=None):
'''Returns a full page of videos with an offset, and a value good enough to be used as the total number of videos'''
# We ask for the next 9 pages from the database
# Then the actual length of the results tell us if there are more than 9 pages left, and if not, how many there actually are
@@ -181,8 +190,6 @@ def _get_videos(cursor, number_per_page, offset, tag = None):
return videos, pseudo_number_of_videos
-
-
def _get_subscribed_channels(cursor):
for item in cursor.execute('''SELECT channel_name, yt_channel_id, muted
FROM subscribed_channels
@@ -204,7 +211,6 @@ def _remove_tags(cursor, channel_ids, tags):
)''', pairs)
-
def _get_tags(cursor, channel_id):
return [row[0] for row in cursor.execute('''SELECT tag
FROM tag_associations
@@ -212,9 +218,11 @@ def _get_tags(cursor, channel_id):
SELECT id FROM subscribed_channels WHERE yt_channel_id = ?
)''', (channel_id,))]
+
def _get_all_tags(cursor):
return [row[0] for row in cursor.execute('''SELECT DISTINCT tag FROM tag_associations''')]
+
def _get_channel_names(cursor, channel_ids):
''' returns list of (channel_id, channel_name) '''
result = []
@@ -222,11 +230,12 @@ def _get_channel_names(cursor, channel_ids):
row = cursor.execute('''SELECT channel_name
FROM subscribed_channels
WHERE yt_channel_id = ?''', (channel_id,)).fetchone()
- result.append( (channel_id, row[0]) )
+ result.append((channel_id, row[0]))
return result
-def _channels_with_tag(cursor, tag, order=False, exclude_muted=False, include_muted_status=False):
+def _channels_with_tag(cursor, tag, order=False, exclude_muted=False,
+ include_muted_status=False):
''' returns list of (channel_id, channel_name) '''
statement = '''SELECT yt_channel_id, channel_name'''
@@ -247,12 +256,15 @@ def _channels_with_tag(cursor, tag, order=False, exclude_muted=False, include_mu
return cursor.execute(statement, [tag]).fetchall()
+
def _schedule_checking(cursor, channel_id, next_check_time):
cursor.execute('''UPDATE subscribed_channels SET next_check_time = ? WHERE yt_channel_id = ?''', [int(next_check_time), channel_id])
+
def _is_muted(cursor, channel_id):
return bool(cursor.execute('''SELECT muted FROM subscribed_channels WHERE yt_channel_id=?''', [channel_id]).fetchone()[0])
+
units = collections.OrderedDict([
('year', 31536000), # 365*24*3600
('month', 2592000), # 30*24*3600
@@ -262,6 +274,8 @@ units = collections.OrderedDict([
('minute', 60),
('second', 1),
])
+
+
def youtube_timestamp_to_posix(dumb_timestamp):
''' Given a dumbed down timestamp such as 1 year ago, 3 hours ago,
approximates the unix time (seconds since 1/1/1970) '''
@@ -275,6 +289,7 @@ def youtube_timestamp_to_posix(dumb_timestamp):
unit = unit[:-1] # remove s from end
return now - quantifier*units[unit]
+
def posix_to_dumbed_down(posix_time):
'''Inverse of youtube_timestamp_to_posix.'''
delta = int(time.time() - posix_time)
@@ -293,12 +308,14 @@ def posix_to_dumbed_down(posix_time):
else:
raise Exception()
+
def exact_timestamp(posix_time):
result = time.strftime('%I:%M %p %m/%d/%y', time.localtime(posix_time))
if result[0] == '0': # remove 0 infront of hour (like 01:00 PM)
return result[1:]
return result
+
try:
existing_thumbnails = set(os.path.splitext(name)[0] for name in os.listdir(thumbnails_directory))
except FileNotFoundError:
@@ -314,6 +331,7 @@ checking_channels = set()
# Just to use for printing channel checking status to console without opening database
channel_names = dict()
+
def check_channel_worker():
while True:
channel_id = check_channels_queue.get()
@@ -324,12 +342,12 @@ def check_channel_worker():
finally:
checking_channels.remove(channel_id)
-for i in range(0,5):
+
+for i in range(0, 5):
gevent.spawn(check_channel_worker)
# ----------------------------
-
# --- Auto checking system - Spaghetti code ---
def autocheck_dispatcher():
'''Scans the auto_check_list. Sleeps until the earliest job is due, then adds that channel to the checking queue above. Can be sent a new job through autocheck_job_application'''
@@ -356,7 +374,7 @@ def autocheck_dispatcher():
if time_until_earliest_job > 0: # it can become less than zero (in the past) when it's set to go off while the dispatcher is doing something else at that moment
try:
- new_job = autocheck_job_application.get(timeout = time_until_earliest_job) # sleep for time_until_earliest_job time, but allow to be interrupted by new jobs
+ new_job = autocheck_job_application.get(timeout=time_until_earliest_job) # sleep for time_until_earliest_job time, but allow to be interrupted by new jobs
except gevent.queue.Empty: # no new jobs
pass
else: # new job, add it to the list
@@ -369,7 +387,10 @@ def autocheck_dispatcher():
check_channels_queue.put(earliest_job['channel_id'])
del autocheck_jobs[earliest_job_index]
+
dispatcher_greenlet = None
+
+
def start_autocheck_system():
global autocheck_job_application
global autocheck_jobs
@@ -398,30 +419,34 @@ def start_autocheck_system():
autocheck_jobs.append({'channel_id': row[0], 'channel_name': row[1], 'next_check_time': next_check_time})
dispatcher_greenlet = gevent.spawn(autocheck_dispatcher)
+
def stop_autocheck_system():
if dispatcher_greenlet is not None:
dispatcher_greenlet.kill()
+
def autocheck_setting_changed(old_value, new_value):
if new_value:
start_autocheck_system()
else:
stop_autocheck_system()
-settings.add_setting_changed_hook('autocheck_subscriptions',
+
+settings.add_setting_changed_hook(
+ 'autocheck_subscriptions',
autocheck_setting_changed)
if settings.autocheck_subscriptions:
start_autocheck_system()
# ----------------------------
-
def check_channels_if_necessary(channel_ids):
for channel_id in channel_ids:
if channel_id not in checking_channels:
checking_channels.add(channel_id)
check_channels_queue.put(channel_id)
+
def _get_atoma_feed(channel_id):
url = 'https://www.youtube.com/feeds/videos.xml?channel_id=' + channel_id
try:
@@ -432,6 +457,7 @@ def _get_atoma_feed(channel_id):
return ''
raise
+
def _get_channel_tab(channel_id, channel_status_name):
try:
return channel.get_channel_tab(channel_id, print_status=False)
@@ -447,6 +473,7 @@ def _get_channel_tab(channel_id, channel_status_name):
return None
raise
+
def _get_upstream_videos(channel_id):
try:
channel_status_name = channel_names[channel_id]
@@ -527,9 +554,8 @@ def _get_upstream_videos(channel_id):
video_item['channel_id'] = channel_id
-
if len(videos) == 0:
- average_upload_period = 4*7*24*3600 # assume 1 month for channel with no videos
+ average_upload_period = 4*7*24*3600 # assume 1 month for channel with no videos
elif len(videos) < 5:
average_upload_period = int((time.time() - videos[len(videos)-1]['time_published'])/len(videos))
else:
@@ -591,7 +617,6 @@ def _get_upstream_videos(channel_id):
video_item['description'],
))
-
cursor.executemany('''INSERT OR IGNORE INTO videos (
sql_channel_id,
video_id,
@@ -619,7 +644,6 @@ def _get_upstream_videos(channel_id):
print(str(number_of_new_videos) + ' new videos from ' + channel_status_name)
-
def check_all_channels():
with open_database() as connection:
with connection as cursor:
@@ -654,22 +678,20 @@ def check_specific_channels(channel_ids):
check_channels_if_necessary(channel_ids)
-
@yt_app.route('/import_subscriptions', methods=['POST'])
def import_subscriptions():
# check if the post request has the file part
if 'subscriptions_file' not in request.files:
- #flash('No file part')
+ # flash('No file part')
return flask.redirect(util.URL_ORIGIN + request.full_path)
file = request.files['subscriptions_file']
# if user does not select file, browser also
# submit an empty part without filename
if file.filename == '':
- #flash('No selected file')
+ # flash('No selected file')
return flask.redirect(util.URL_ORIGIN + request.full_path)
-
mime_type = file.mimetype
if mime_type == 'application/json':
@@ -681,7 +703,7 @@ def import_subscriptions():
return '400 Bad Request: Invalid json file', 400
try:
- channels = ( (item['snippet']['resourceId']['channelId'], item['snippet']['title']) for item in file)
+ channels = ((item['snippet']['resourceId']['channelId'], item['snippet']['title']) for item in file)
except (KeyError, IndexError):
traceback.print_exc()
return '400 Bad Request: Unknown json structure', 400
@@ -695,11 +717,10 @@ def import_subscriptions():
if (outline_element.tag != 'outline') or ('xmlUrl' not in outline_element.attrib):
continue
-
channel_name = outline_element.attrib['text']
channel_rss_url = outline_element.attrib['xmlUrl']
channel_id = channel_rss_url[channel_rss_url.find('channel_id=')+11:].strip()
- channels.append( (channel_id, channel_name) )
+ channels.append((channel_id, channel_name))
except (AssertionError, IndexError, defusedxml.ElementTree.ParseError) as e:
return '400 Bad Request: Unable to read opml xml file, or the file is not the expected format', 400
@@ -711,7 +732,6 @@ def import_subscriptions():
return flask.redirect(util.URL_ORIGIN + '/subscription_manager', 303)
-
@yt_app.route('/subscription_manager', methods=['GET'])
def get_subscription_manager_page():
group_by_tags = request.args.get('group_by_tags', '0') == '1'
@@ -731,7 +751,7 @@ def get_subscription_manager_page():
'tags': [t for t in _get_tags(cursor, channel_id) if t != tag],
})
- tag_groups.append( (tag, sub_list) )
+ tag_groups.append((tag, sub_list))
# Channels with no tags
channel_list = cursor.execute('''SELECT yt_channel_id, channel_name, muted
@@ -751,7 +771,7 @@ def get_subscription_manager_page():
'tags': [],
})
- tag_groups.append( ('No tags', sub_list) )
+ tag_groups.append(('No tags', sub_list))
else:
sub_list = []
for channel_name, channel_id, muted in _get_subscribed_channels(cursor):
@@ -763,20 +783,20 @@ def get_subscription_manager_page():
'tags': _get_tags(cursor, channel_id),
})
-
-
-
if group_by_tags:
- return flask.render_template('subscription_manager.html',
- group_by_tags = True,
- tag_groups = tag_groups,
+ return flask.render_template(
+ 'subscription_manager.html',
+ group_by_tags=True,
+ tag_groups=tag_groups,
)
else:
- return flask.render_template('subscription_manager.html',
- group_by_tags = False,
- sub_list = sub_list,
+ return flask.render_template(
+ 'subscription_manager.html',
+ group_by_tags=False,
+ sub_list=sub_list,
)
+
def list_from_comma_separated_tags(string):
return [tag.strip() for tag in string.split(',') if tag.strip()]
@@ -795,7 +815,7 @@ def post_subscription_manager_page():
_unsubscribe(cursor, request.values.getlist('channel_ids'))
elif action == 'unsubscribe_verify':
unsubscribe_list = _get_channel_names(cursor, request.values.getlist('channel_ids'))
- return flask.render_template('unsubscribe_verify.html', unsubscribe_list = unsubscribe_list)
+ return flask.render_template('unsubscribe_verify.html', unsubscribe_list=unsubscribe_list)
elif action == 'mute':
cursor.executemany('''UPDATE subscribed_channels
@@ -810,6 +830,7 @@ def post_subscription_manager_page():
return flask.redirect(util.URL_ORIGIN + request.full_path, 303)
+
@yt_app.route('/subscriptions', methods=['GET'])
@yt_app.route('/feed/subscriptions', methods=['GET'])
def get_subscriptions_page():
@@ -826,7 +847,6 @@ def get_subscriptions_page():
tags = _get_all_tags(cursor)
-
subscription_list = []
for channel_name, channel_id, muted in _get_subscribed_channels(cursor):
subscription_list.append({
@@ -836,16 +856,18 @@ def get_subscriptions_page():
'muted': muted,
})
- return flask.render_template('subscriptions.html',
- header_playlist_names = local_playlist.get_playlist_names(),
- videos = videos,
- num_pages = math.ceil(number_of_videos_in_db/60),
- parameters_dictionary = request.args,
- tags = tags,
- current_tag = tag,
- subscription_list = subscription_list,
+ return flask.render_template(
+ 'subscriptions.html',
+ header_playlist_names=local_playlist.get_playlist_names(),
+ videos=videos,
+ num_pages=math.ceil(number_of_videos_in_db/60),
+ parameters_dictionary=request.args,
+ tags=tags,
+ current_tag=tag,
+ subscription_list=subscription_list,
)
+
@yt_app.route('/subscriptions', methods=['POST'])
@yt_app.route('/feed/subscriptions', methods=['POST'])
def post_subscriptions_page():
@@ -900,17 +922,10 @@ def serve_subscription_thumbnail(thumbnail):
try:
f = open(thumbnail_path, 'wb')
except FileNotFoundError:
- os.makedirs(thumbnails_directory, exist_ok = True)
+ os.makedirs(thumbnails_directory, exist_ok=True)
f = open(thumbnail_path, 'wb')
f.write(image)
f.close()
existing_thumbnails.add(video_id)
return flask.Response(image, mimetype='image/jpeg')
-
-
-
-
-
-
-
diff --git a/youtube/util.py b/youtube/util.py
index 8945b9f..df4759e 100644
--- a/youtube/util.py
+++ b/youtube/util.py
@@ -1,6 +1,7 @@
import settings
from youtube import yt_data_extract
-import socks, sockshandler
+import socks
+import sockshandler
import gzip
try:
import brotli
@@ -55,14 +56,15 @@ import urllib3.contrib.socks
URL_ORIGIN = "/https://www.youtube.com"
-connection_pool = urllib3.PoolManager(cert_reqs = 'CERT_REQUIRED')
+connection_pool = urllib3.PoolManager(cert_reqs='CERT_REQUIRED')
+
class TorManager:
def __init__(self):
self.old_tor_connection_pool = None
self.tor_connection_pool = urllib3.contrib.socks.SOCKSProxyManager(
'socks5h://127.0.0.1:' + str(settings.tor_port) + '/',
- cert_reqs = 'CERT_REQUIRED')
+ cert_reqs='CERT_REQUIRED')
self.tor_pool_refresh_time = time.monotonic()
self.new_identity_lock = gevent.lock.BoundedSemaphore(1)
@@ -77,7 +79,7 @@ class TorManager:
self.tor_connection_pool = urllib3.contrib.socks.SOCKSProxyManager(
'socks5h://127.0.0.1:' + str(settings.tor_port) + '/',
- cert_reqs = 'CERT_REQUIRED')
+ cert_reqs='CERT_REQUIRED')
self.tor_pool_refresh_time = time.monotonic()
def get_tor_connection_pool(self):
@@ -125,6 +127,7 @@ class TorManager:
finally:
self.new_identity_lock.release()
+
tor_manager = TorManager()
@@ -154,6 +157,7 @@ class HTTPAsymmetricCookieProcessor(urllib.request.BaseHandler):
https_request = http_request
https_response = http_response
+
class FetchError(Exception):
def __init__(self, code, reason='', ip=None, error_message=None):
Exception.__init__(self, 'HTTP error during request: ' + code + ' ' + reason)
@@ -162,6 +166,7 @@ class FetchError(Exception):
self.ip = ip
self.error_message = error_message
+
def decode_content(content, encoding_header):
encodings = encoding_header.replace(' ', '').split(',')
for encoding in reversed(encodings):
@@ -173,6 +178,7 @@ def decode_content(content, encoding_header):
content = gzip.decompress(content)
return content
+
def fetch_url_response(url, headers=(), timeout=15, data=None,
cookiejar_send=None, cookiejar_receive=None,
use_tor=True, max_redirects=None):
@@ -234,6 +240,7 @@ def fetch_url_response(url, headers=(), timeout=15, data=None,
return response, cleanup_func
+
def fetch_url(url, headers=(), timeout=15, report_text=None, data=None,
cookiejar_send=None, cookiejar_receive=None, use_tor=True,
debug_name=None):
@@ -284,7 +291,7 @@ def fetch_url(url, headers=(), timeout=15, report_text=None, data=None,
break
if report_text:
- print(report_text, ' Latency:', round(response_time - start_time,3), ' Read time:', round(read_finish - response_time,3))
+ print(report_text, ' Latency:', round(response_time - start_time, 3), ' Read time:', round(read_finish - response_time,3))
if settings.debugging_save_responses and debug_name is not None:
save_dir = os.path.join(settings.data_dir, 'debug')
@@ -296,6 +303,7 @@ def fetch_url(url, headers=(), timeout=15, report_text=None, data=None,
return content
+
def head(url, use_tor=False, report_text=None, max_redirects=10):
pool = get_pool(use_tor and settings.route_tor)
start_time = time.monotonic()
@@ -305,7 +313,9 @@ def head(url, use_tor=False, report_text=None, max_redirects=10):
# According to the documentation for urlopen, a redirect counts as a retry
# So there are 3 redirects max by default. Let's change that
# to 10 since googlevideo redirects a lot.
- retries = urllib3.Retry(3+max_redirects, redirect=max_redirects,
+ retries = urllib3.Retry(
+ 3+max_redirects,
+ redirect=max_redirects,
raise_on_redirect=False)
headers = {'User-Agent': 'Python-urllib'}
response = pool.request('HEAD', url, headers=headers, retries=retries)
@@ -313,19 +323,16 @@ def head(url, use_tor=False, report_text=None, max_redirects=10):
print(
report_text,
' Latency:',
- round(time.monotonic() - start_time,3))
+ round(time.monotonic() - start_time, 3))
return response
+
mobile_user_agent = 'Mozilla/5.0 (Linux; Android 7.0; Redmi Note 4 Build/NRD90M) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/69.0.3497.100 Mobile Safari/537.36'
mobile_ua = (('User-Agent', mobile_user_agent),)
desktop_user_agent = 'Mozilla/5.0 (Windows NT 6.1; rv:52.0) Gecko/20100101 Firefox/52.0'
desktop_ua = (('User-Agent', desktop_user_agent),)
-
-
-
-
class RateLimitedQueue(gevent.queue.Queue):
''' Does initial_burst (def. 30) at first, then alternates between waiting waiting_period (def. 5) seconds and doing subsequent_bursts (def. 10) queries. After 5 seconds with nothing left in the queue, resets rate limiting. '''
@@ -342,7 +349,6 @@ class RateLimitedQueue(gevent.queue.Queue):
self.empty_start = 0
gevent.queue.Queue.__init__(self)
-
def get(self):
self.lock.acquire() # blocks if another greenlet currently has the lock
if self.count_since_last_wait >= self.subsequent_bursts and self.surpassed_initial:
@@ -374,7 +380,6 @@ class RateLimitedQueue(gevent.queue.Queue):
return item
-
def download_thumbnail(save_directory, video_id):
url = "https://i.ytimg.com/vi/" + video_id + "/mqdefault.jpg"
save_location = os.path.join(save_directory, video_id + ".jpg")
@@ -386,12 +391,13 @@ def download_thumbnail(save_directory, video_id):
try:
f = open(save_location, 'wb')
except FileNotFoundError:
- os.makedirs(save_directory, exist_ok = True)
+ os.makedirs(save_directory, exist_ok=True)
f = open(save_location, 'wb')
f.write(thumbnail)
f.close()
return True
+
def download_thumbnails(save_directory, ids):
if not isinstance(ids, (list, tuple)):
ids = list(ids)
@@ -404,15 +410,12 @@ def download_thumbnails(save_directory, ids):
gevent.joinall([gevent.spawn(download_thumbnail, save_directory, ids[j]) for j in range(i*5 + 5, len(ids))])
-
-
-
-
def dict_add(*dicts):
for dictionary in dicts[1:]:
dicts[0].update(dictionary)
return dicts[0]
+
def video_id(url):
url_parts = urllib.parse.urlparse(url)
return urllib.parse.parse_qs(url_parts.query)['v'][0]
@@ -422,10 +425,11 @@ def video_id(url):
def get_thumbnail_url(video_id):
return settings.img_prefix + "https://i.ytimg.com/vi/" + video_id + "/mqdefault.jpg"
+
def seconds_to_timestamp(seconds):
seconds = int(seconds)
- hours, seconds = divmod(seconds,3600)
- minutes, seconds = divmod(seconds,60)
+ hours, seconds = divmod(seconds, 3600)
+ minutes, seconds = divmod(seconds, 60)
if hours != 0:
timestamp = str(hours) + ":"
timestamp += str(minutes).zfill(2) # zfill pads with zeros
@@ -436,18 +440,17 @@ def seconds_to_timestamp(seconds):
return timestamp
-
def update_query_string(query_string, items):
parameters = urllib.parse.parse_qs(query_string)
parameters.update(items)
return urllib.parse.urlencode(parameters, doseq=True)
-
def uppercase_escape(s):
- return re.sub(
- r'\\U([0-9a-fA-F]{8})',
- lambda m: chr(int(m.group(1), base=16)), s)
+ return re.sub(
+ r'\\U([0-9a-fA-F]{8})',
+ lambda m: chr(int(m.group(1), base=16)), s)
+
def prefix_url(url):
if url is None:
@@ -455,12 +458,14 @@ def prefix_url(url):
url = url.lstrip('/') # some urls have // before them, which has a special meaning
return '/' + url
+
def left_remove(string, substring):
'''removes substring from the start of string, if present'''
if string.startswith(substring):
return string[len(substring):]
return string
+
def concat_or_none(*strings):
'''Concatenates strings. Returns None if any of the arguments are None'''
result = ''
@@ -483,6 +488,7 @@ def prefix_urls(item):
except KeyError:
pass
+
def add_extra_html_info(item):
if item['type'] == 'video':
item['url'] = (URL_ORIGIN + '/watch?v=' + item['id']) if item.get('id') else None
@@ -501,6 +507,7 @@ def add_extra_html_info(item):
elif item['type'] == 'channel':
item['url'] = (URL_ORIGIN + "/channel/" + item['id']) if item.get('id') else None
+
def parse_info_prepare_for_html(renderer, additional_info={}):
item = yt_data_extract.extract_item_info(renderer, additional_info)
prefix_urls(item)
@@ -508,6 +515,7 @@ def parse_info_prepare_for_html(renderer, additional_info={}):
return item
+
def check_gevent_exceptions(*tasks):
for task in tasks:
if task.exception:
@@ -528,7 +536,13 @@ replacement_map = collections.OrderedDict([
('*', '_'),
('\t', ' '),
])
-DOS_names = {'con', 'prn', 'aux', 'nul', 'com0', 'com1', 'com2', 'com3', 'com4', 'com5', 'com6', 'com7', 'com8', 'com9', 'lpt0', 'lpt1', 'lpt2', 'lpt3', 'lpt4', 'lpt5', 'lpt6', 'lpt7', 'lpt8', 'lpt9'}
+
+DOS_names = {'con', 'prn', 'aux', 'nul', 'com0', 'com1', 'com2', 'com3',
+ 'com4', 'com5', 'com6', 'com7', 'com8', 'com9', 'lpt0',
+ 'lpt1', 'lpt2', 'lpt3', 'lpt4', 'lpt5', 'lpt6', 'lpt7',
+ 'lpt8', 'lpt9'}
+
+
def to_valid_filename(name):
'''Changes the name so it's valid on Windows, Linux, and Mac'''
# See https://docs.microsoft.com/en-us/windows/win32/fileio/naming-a-file