import base64 import youtube.common as common from youtube.common import default_multi_get, URL_ORIGIN, get_thumbnail_url, video_id import urllib import json from string import Template import youtube.proto as proto import html import math import gevent import re import functools with open("yt_channel_items_template.html", "r") as file: yt_channel_items_template = Template(file.read()) with open("yt_channel_about_template.html", "r") as file: yt_channel_about_template = Template(file.read()) '''continuation = Proto( Field('optional', 'continuation', 80226972, Proto( Field('optional', 'browse_id', 2, String), Field('optional', 'params', 3, Base64(Proto( Field('optional', 'channel_tab', 2, String), Field('optional', 'sort', 3, ENUM Field('optional', 'page', 15, String), ))) )) )''' '''channel_continuation = Proto( Field('optional', 'pointless_nest', 80226972, Proto( Field('optional', 'channel_id', 2, String), Field('optional', 'continuation_info', 3, Base64(Proto( Field('optional', 'channel_tab', 2, String), Field('optional', 'sort', 3, ENUM Field('optional', 'page', 15, String), ))) )) )''' headers_1 = ( ('Accept', '*/*'), ('Accept-Language', 'en-US,en;q=0.5'), ('X-YouTube-Client-Name', '1'), ('X-YouTube-Client-Version', '2.20180830'), ) # https://www.youtube.com/browse_ajax?action_continuation=1&direct_render=1&continuation=4qmFsgJAEhhVQzdVY3M0MkZaeTN1WXpqcnF6T0lIc3caJEVnWjJhV1JsYjNNZ0FEZ0JZQUZxQUhvQk1yZ0JBQSUzRCUzRA%3D%3D # https://www.youtube.com/browse_ajax?ctoken=4qmFsgJAEhhVQzdVY3M0MkZaeTN1WXpqcnF6T0lIc3caJEVnWjJhV1JsYjNNZ0FEZ0JZQUZxQUhvQk1yZ0JBQSUzRCUzRA%3D%3D&continuation=4qmFsgJAEhhVQzdVY3M0MkZaeTN1WXpqcnF6T0lIc3caJEVnWjJhV1JsYjNNZ0FEZ0JZQUZxQUhvQk1yZ0JBQSUzRCUzRA%3D%3D&itct=CDsQybcCIhMIhZi1krTc2wIVjMicCh2HXQnhKJsc # grid view: 4qmFsgJAEhhVQzdVY3M0MkZaeTN1WXpqcnF6T0lIc3caJEVnWjJhV1JsYjNNZ0FEZ0JZQUZxQUhvQk1yZ0JBQSUzRCUzRA # list view: 4qmFsgJCEhhVQzdVY3M0MkZaeTN1WXpqcnF6T0lIc3caJkVnWjJhV1JsYjNNWUF5QUFNQUk0QVdBQmFnQjZBVEs0QVFBJTNE # SORT: # Popular - 1 # Oldest - 2 # Newest - 3 # view: # grid: 0 or 1 # list: 2 def channel_ctoken(channel_id, page, sort, tab, view=1): tab = proto.string(2, tab ) sort = proto.uint(3, int(sort)) 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) ) 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): ctoken = channel_ctoken(channel_id, page, sort, tab, view).replace('=', '%3D') url = "https://www.youtube.com/browse_ajax?ctoken=" + ctoken print("Sending channel tab ajax request") content = common.fetch_url(url, common.desktop_ua + headers_1) print("Finished recieving channel tab response") '''with open('debug/channel_debug', 'wb') as f: f.write(content)''' info = json.loads(content) return info def get_number_of_videos(channel_id): # Uploads playlist playlist_id = 'UU' + channel_id[2:] url = 'https://m.youtube.com/playlist?list=' + playlist_id + '&ajax=1&disable_polymer=true' print("Getting number of videos") response = common.fetch_url(url, common.mobile_ua + headers_1) '''with open('debug/playlist_debug_metadata', 'wb') as f: f.write(response)''' response = response.decode('utf-8') print("Got response for number of videos") match = re.search(r'"num_videos_text":\s*{(?:"item_type":\s*"formatted_string",)?\s*"runs":\s*\[{"text":\s*"([\d,]*) videos"', response) if match: return int(match.group(1).replace(',','')) else: return 0 @functools.lru_cache(maxsize=128) def get_channel_id(username): # method that gives the smallest possible response at ~10 kb # needs to be as fast as possible url = 'https://m.youtube.com/user/' + username + '/about?ajax=1&disable_polymer=true' response = common.fetch_url(url, common.mobile_ua + headers_1).decode('utf-8') return re.search(r'"channel_id":\s*"([a-zA-Z0-9_-]*)"', response).group(1) def grid_items_html(items, additional_info={}): result = ''' ''' return result def list_items_html(items, additional_info={}): result = ''' ''' return result channel_tab_template = Template('''\n$tab_name''') channel_search_template = Template(''' ''') tabs = ('Videos', 'Playlists', 'About') def channel_tabs_html(channel_id, current_tab, search_box_value=''): result = '' for tab_name in tabs: if tab_name == current_tab: result += channel_tab_template.substitute( href_attribute = '', tab_name = tab_name, ) else: result += channel_tab_template.substitute( href_attribute = ' href="' + URL_ORIGIN + '/channel/' + channel_id + '/' + tab_name.lower() + '"', tab_name = tab_name, ) result += channel_search_template.substitute( action = URL_ORIGIN + "/channel/" + channel_id + "/search", search_box_value = html.escape(search_box_value), ) return result channel_sort_button_template = Template('''\n$text''') sorts = {1: 'views', 2: 'oldest', 3: 'newest'} def channel_sort_buttons_html(channel_id, current_sort): result = '' current_sort = int(current_sort) for i in range(1,4): if i == current_sort: result += channel_sort_button_template.substitute( href_attribute='', text = 'Sorted by ' + sorts[current_sort] ) else: result += channel_sort_button_template.substitute( href_attribute=' href="' + URL_ORIGIN + '/channel/' + channel_id + '/videos?sort=' + str(i) + '"', text = 'Sort by ' + sorts[i] ) return result def channel_videos_html(polymer_json, current_page=1, current_sort=3, number_of_videos = 1000, current_query_string=''): microformat = polymer_json[1]['response']['microformat']['microformatDataRenderer'] channel_url = microformat['urlCanonical'].rstrip('/') channel_id = channel_url[channel_url.rfind('/')+1:] try: items = polymer_json[1]['response']['continuationContents']['gridContinuation']['items'] except KeyError: response = polymer_json[1]['response'] try: contents = response['contents'] except KeyError: items = [] else: items = tab_with_content(contents['twoColumnBrowseResultsRenderer']['tabs'])['sectionListRenderer']['contents'][0]['itemSectionRenderer']['contents'][0]['gridRenderer']['items'] items_html = grid_items_html(items, {'author': microformat['title']}) return yt_channel_items_template.substitute( header = common.get_header(), channel_title = microformat['title'], channel_tabs = channel_tabs_html(channel_id, 'Videos'), sort_buttons = channel_sort_buttons_html(channel_id, current_sort), avatar = '/' + microformat['thumbnail']['thumbnails'][0]['url'], page_title = microformat['title'] + ' - Channel', items = items_html, page_buttons = common.page_buttons_html(current_page, math.ceil(number_of_videos/30), URL_ORIGIN + "/channel/" + channel_id + "/videos", current_query_string), number_of_results = '{:,}'.format(number_of_videos) + " videos", ) def channel_playlists_html(polymer_json): microformat = polymer_json[1]['response']['microformat']['microformatDataRenderer'] channel_url = microformat['urlCanonical'].rstrip('/') channel_id = channel_url[channel_url.rfind('/')+1:] try: items = polymer_json[1]['response']['continuationContents']['gridContinuation']['items'] except KeyError: response = polymer_json[1]['response'] try: contents = response['contents'] except KeyError: items = [] else: item_section = tab_with_content(contents['twoColumnBrowseResultsRenderer']['tabs'])['sectionListRenderer']['contents'][0]['itemSectionRenderer']['contents'][0] try: items = item_section['gridRenderer']['items'] except KeyError: if "messageRenderer" in item_section: items = [] else: raise items_html = grid_items_html(items, {'author': microformat['title']}) return yt_channel_items_template.substitute( header = common.get_header(), channel_title = microformat['title'], channel_tabs = channel_tabs_html(channel_id, 'Playlists'), avatar = '/' + microformat['thumbnail']['thumbnails'][0]['url'], page_title = microformat['title'] + ' - Channel', items = items_html, page_buttons = '', number_of_results = '', sort_buttons = '', ) # Example channel where tabs do not have definite index: https://www.youtube.com/channel/UC4gQ8i3FD7YbhOgqUkeQEJg def tab_with_content(tabs): for tab in tabs: try: renderer = tab['tabRenderer'] except KeyError: renderer = tab['expandableTabRenderer'] try: return renderer['content'] except KeyError: pass raise Exception("No tabs found with content") channel_link_template = Template('''
  • $text
  • ''') stat_template = Template('''
  • $stat_value
  • ''') def channel_about_page(polymer_json): avatar = '/' + polymer_json[1]['response']['microformat']['microformatDataRenderer']['thumbnail']['thumbnails'][0]['url'] # my goodness... channel_metadata = tab_with_content(polymer_json[1]['response']['contents']['twoColumnBrowseResultsRenderer']['tabs'])['sectionListRenderer']['contents'][0]['itemSectionRenderer']['contents'][0]['channelAboutFullMetadataRenderer'] channel_links = '' for link_json in channel_metadata.get('primaryLinks', ()): url = link_json['navigationEndpoint']['urlEndpoint']['url'] if url.startswith("/redirect"): query_string = url[url.find('?')+1: ] url = urllib.parse.parse_qs(query_string)['q'][0] channel_links += channel_link_template.substitute( url = html.escape(url), text = common.get_plain_text(link_json['title']), ) stats = '' for stat_name in ('subscriberCountText', 'joinedDateText', 'viewCountText', 'country'): try: stat_value = common.get_plain_text(channel_metadata[stat_name]) except KeyError: continue else: stats += stat_template.substitute(stat_value=stat_value) try: description = common.format_text_runs(common.get_formatted_text(channel_metadata['description'])) except KeyError: description = '' return yt_channel_about_template.substitute( header = common.get_header(), page_title = common.get_plain_text(channel_metadata['title']) + ' - About', channel_title = common.get_plain_text(channel_metadata['title']), avatar = html.escape(avatar), description = description, links = channel_links, stats = stats, channel_tabs = channel_tabs_html(channel_metadata['channelId'], 'About'), ) def channel_search_page(polymer_json, query, current_page=1, number_of_videos = 1000, current_query_string=''): microformat = polymer_json[1]['response']['microformat']['microformatDataRenderer'] channel_url = microformat['urlCanonical'].rstrip('/') channel_id = channel_url[channel_url.rfind('/')+1:] response = polymer_json[1]['response'] try: items = tab_with_content(response['contents']['twoColumnBrowseResultsRenderer']['tabs'])['sectionListRenderer']['contents'] except KeyError: items = response['continuationContents']['sectionListContinuation']['contents'] items_html = list_items_html(items) return yt_channel_items_template.substitute( header = common.get_header(), channel_title = html.escape(microformat['title']), channel_tabs = channel_tabs_html(channel_id, '', query), avatar = '/' + microformat['thumbnail']['thumbnails'][0]['url'], page_title = html.escape(query + ' - Channel search'), items = items_html, page_buttons = common.page_buttons_html(current_page, math.ceil(number_of_videos/29), URL_ORIGIN + "/channel/" + channel_id + "/search", current_query_string), number_of_results = '', sort_buttons = '', ) def get_channel_search_json(channel_id, query, page): params = proto.string(2, 'search') + proto.string(15, str(page)) params = proto.percent_b64encode(params) ctoken = proto.string(2, channel_id) + proto.string(3, params) + proto.string(11, query) ctoken = base64.urlsafe_b64encode(proto.nested(80226972, ctoken)).decode('ascii') polymer_json = common.fetch_url("https://www.youtube.com/browse_ajax?ctoken=" + ctoken, common.desktop_ua + headers_1) '''with open('debug/channel_search_debug', 'wb') as f: f.write(polymer_json)''' polymer_json = json.loads(polymer_json) return polymer_json def get_channel_page(url, query_string=''): path_components = url.rstrip('/').lstrip('/').split('/') channel_id = path_components[0] try: tab = path_components[1] except IndexError: tab = 'videos' parameters = urllib.parse.parse_qs(query_string) page_number = int(common.default_multi_get(parameters, 'page', 0, default='1')) sort = common.default_multi_get(parameters, 'sort', 0, default='3') view = common.default_multi_get(parameters, 'view', 0, default='1') query = common.default_multi_get(parameters, 'query', 0, default='') if tab == 'videos': tasks = ( gevent.spawn(get_number_of_videos, channel_id ), gevent.spawn(get_channel_tab, channel_id, page_number, sort, 'videos', view) ) gevent.joinall(tasks) number_of_videos, polymer_json = tasks[0].value, tasks[1].value return channel_videos_html(polymer_json, page_number, sort, number_of_videos, query_string) elif tab == 'about': polymer_json = common.fetch_url('https://www.youtube.com/channel/' + channel_id + '/about?pbj=1', common.desktop_ua + headers_1) polymer_json = json.loads(polymer_json) return channel_about_page(polymer_json) elif tab == 'playlists': polymer_json = common.fetch_url('https://www.youtube.com/channel/' + channel_id + '/playlists?pbj=1&view=1', common.desktop_ua + headers_1) '''with open('debug/channel_playlists_debug', 'wb') as f: f.write(polymer_json)''' polymer_json = json.loads(polymer_json) return channel_playlists_html(polymer_json) elif tab == 'search': tasks = ( gevent.spawn(get_number_of_videos, channel_id ), gevent.spawn(get_channel_search_json, channel_id, query, page_number) ) gevent.joinall(tasks) number_of_videos, polymer_json = tasks[0].value, tasks[1].value return channel_search_page(polymer_json, query, page_number, number_of_videos, query_string) else: raise ValueError('Unknown channel tab: ' + tab) # youtube.com/user/[username]/[page] # youtube.com/c/[custom]/[page] # youtube.com/[custom]/[page] def get_channel_page_general_url(url, query_string=''): path_components = url.rstrip('/').lstrip('/').split('/') is_toplevel = not path_components[0] in ('user', 'c') if len(path_components) + int(is_toplevel) == 3: # has /[page] after it page = path_components[2] base_url = 'https://www.youtube.com/' + '/'.join(path_components[0:-1]) elif len(path_components) + int(is_toplevel) == 2: # does not have /[page] after it, use /videos by default page = 'videos' base_url = 'https://www.youtube.com/' + '/'.join(path_components) else: raise ValueError() if page == 'videos': polymer_json = common.fetch_url(base_url + '/videos?pbj=1&view=0', common.desktop_ua + headers_1) polymer_json = json.loads(polymer_json) return channel_videos_html(polymer_json) elif page == 'about': polymer_json = common.fetch_url(base_url + '/about?pbj=1', common.desktop_ua + headers_1) polymer_json = json.loads(polymer_json) return channel_about_page(polymer_json) elif page == 'playlists': polymer_json = common.fetch_url(base_url+ '/playlists?pbj=1&view=1', common.desktop_ua + headers_1) polymer_json = json.loads(polymer_json) return channel_playlists_html(polymer_json) elif page == 'search': raise NotImplementedError() '''polymer_json = common.fetch_url('https://www.youtube.com/user' + username + '/search?pbj=1&' + query_string, common.desktop_ua + headers_1) polymer_json = json.loads(polymer_json) return channel_search_page(''' else: raise ValueError('Unknown channel page: ' + page)