youtube/yt_data_extract/watch_extraction.py

from .common import (get, multi_get, deep_get, multi_deep_get,
    liberal_update, conservative_update, remove_redirect, normalize_url,
    extract_str, extract_formatted_text, extract_int, extract_approx_int,
    extract_date, check_missing_keys, extract_item_info, extract_items,
    extract_response, concat_or_none)

import json
import urllib.parse
import traceback
import re

# from https://github.com/ytdl-org/youtube-dl/blob/master/youtube_dl/extractor/youtube.py
_formats = {
    '5': {'ext': 'flv', 'width': 400, 'height': 240, 'acodec': 'mp3', 'audio_bitrate': 64, 'vcodec': 'h263'},
    '6': {'ext': 'flv', 'width': 450, 'height': 270, 'acodec': 'mp3', 'audio_bitrate': 64, 'vcodec': 'h263'},
    '13': {'ext': '3gp', 'acodec': 'aac', 'vcodec': 'mp4v'},
    '17': {'ext': '3gp', 'width': 176, 'height': 144, 'acodec': 'aac', 'audio_bitrate': 24, 'vcodec': 'mp4v'},
    '18': {'ext': 'mp4', 'width': 640, 'height': 360, 'acodec': 'aac', 'audio_bitrate': 96, 'vcodec': 'h264'},
    '22': {'ext': 'mp4', 'width': 1280, 'height': 720, 'acodec': 'aac', 'audio_bitrate': 192, 'vcodec': 'h264'},
    '34': {'ext': 'flv', 'width': 640, 'height': 360, 'acodec': 'aac', 'audio_bitrate': 128, 'vcodec': 'h264'},
    '35': {'ext': 'flv', 'width': 854, 'height': 480, 'acodec': 'aac', 'audio_bitrate': 128, 'vcodec': 'h264'},
    # itag 36 videos are either 320x180 (BaW_jenozKc) or 320x240 (__2ABJjxzNo), audio_bitrate varies as well
    '36': {'ext': '3gp', 'width': 320, 'acodec': 'aac', 'vcodec': 'mp4v'},
    '37': {'ext': 'mp4', 'width': 1920, 'height': 1080, 'acodec': 'aac', 'audio_bitrate': 192, 'vcodec': 'h264'},
    '38': {'ext': 'mp4', 'width': 4096, 'height': 3072, 'acodec': 'aac', 'audio_bitrate': 192, 'vcodec': 'h264'},
    '43': {'ext': 'webm', 'width': 640, 'height': 360, 'acodec': 'vorbis', 'audio_bitrate': 128, 'vcodec': 'vp8'},
    '44': {'ext': 'webm', 'width': 854, 'height': 480, 'acodec': 'vorbis', 'audio_bitrate': 128, 'vcodec': 'vp8'},
    '45': {'ext': 'webm', 'width': 1280, 'height': 720, 'acodec': 'vorbis', 'audio_bitrate': 192, 'vcodec': 'vp8'},
    '46': {'ext': 'webm', 'width': 1920, 'height': 1080, 'acodec': 'vorbis', 'audio_bitrate': 192, 'vcodec': 'vp8'},
    '59': {'ext': 'mp4', 'width': 854, 'height': 480, 'acodec': 'aac', 'audio_bitrate': 128, 'vcodec': 'h264'},
    '78': {'ext': 'mp4', 'width': 854, 'height': 480, 'acodec': 'aac', 'audio_bitrate': 128, 'vcodec': 'h264'},


    # 3D videos
    '82': {'ext': 'mp4', 'height': 360, 'format_note': '3D', 'acodec': 'aac', 'audio_bitrate': 128, 'vcodec': 'h264'},
    '83': {'ext': 'mp4', 'height': 480, 'format_note': '3D', 'acodec': 'aac', 'audio_bitrate': 128, 'vcodec': 'h264'},
    '84': {'ext': 'mp4', 'height': 720, 'format_note': '3D', 'acodec': 'aac', 'audio_bitrate': 192, 'vcodec': 'h264'},
    '85': {'ext': 'mp4', 'height': 1080, 'format_note': '3D', 'acodec': 'aac', 'audio_bitrate': 192, 'vcodec': 'h264'},
    '100': {'ext': 'webm', 'height': 360, 'format_note': '3D', 'acodec': 'vorbis', 'audio_bitrate': 128, 'vcodec': 'vp8'},
    '101': {'ext': 'webm', 'height': 480, 'format_note': '3D', 'acodec': 'vorbis', 'audio_bitrate': 192, 'vcodec': 'vp8'},
    '102': {'ext': 'webm', 'height': 720, 'format_note': '3D', 'acodec': 'vorbis', 'audio_bitrate': 192, 'vcodec': 'vp8'},

    # Apple HTTP Live Streaming
    '91': {'ext': 'mp4', 'height': 144, 'format_note': 'HLS', 'acodec': 'aac', 'audio_bitrate': 48, 'vcodec': 'h264'},
    '92': {'ext': 'mp4', 'height': 240, 'format_note': 'HLS', 'acodec': 'aac', 'audio_bitrate': 48, 'vcodec': 'h264'},
    '93': {'ext': 'mp4', 'height': 360, 'format_note': 'HLS', 'acodec': 'aac', 'audio_bitrate': 128, 'vcodec': 'h264'},
    '94': {'ext': 'mp4', 'height': 480, 'format_note': 'HLS', 'acodec': 'aac', 'audio_bitrate': 128, 'vcodec': 'h264'},
    '95': {'ext': 'mp4', 'height': 720, 'format_note': 'HLS', 'acodec': 'aac', 'audio_bitrate': 256, 'vcodec': 'h264'},
    '96': {'ext': 'mp4', 'height': 1080, 'format_note': 'HLS', 'acodec': 'aac', 'audio_bitrate': 256, 'vcodec': 'h264'},
    '132': {'ext': 'mp4', 'height': 240, 'format_note': 'HLS', 'acodec': 'aac', 'audio_bitrate': 48, 'vcodec': 'h264'},
    '151': {'ext': 'mp4', 'height': 72, 'format_note': 'HLS', 'acodec': 'aac', 'audio_bitrate': 24, 'vcodec': 'h264'},

    # DASH mp4 video
    '133': {'ext': 'mp4', 'height': 240, 'format_note': 'DASH video', 'vcodec': 'h264'},
    '134': {'ext': 'mp4', 'height': 360, 'format_note': 'DASH video', 'vcodec': 'h264'},
    '135': {'ext': 'mp4', 'height': 480, 'format_note': 'DASH video', 'vcodec': 'h264'},
    '136': {'ext': 'mp4', 'height': 720, 'format_note': 'DASH video', 'vcodec': 'h264'},
    '137': {'ext': 'mp4', 'height': 1080, 'format_note': 'DASH video', 'vcodec': 'h264'},
    '138': {'ext': 'mp4', 'format_note': 'DASH video', 'vcodec': 'h264'},  # Height can vary (https://github.com/ytdl-org/youtube-dl/issues/4559)
    '160': {'ext': 'mp4', 'height': 144, 'format_note': 'DASH video', 'vcodec': 'h264'},
    '212': {'ext': 'mp4', 'height': 480, 'format_note': 'DASH video', 'vcodec': 'h264'},
    '264': {'ext': 'mp4', 'height': 1440, 'format_note': 'DASH video', 'vcodec': 'h264'},
    '298': {'ext': 'mp4', 'height': 720, 'format_note': 'DASH video', 'vcodec': 'h264', 'fps': 60},
    '299': {'ext': 'mp4', 'height': 1080, 'format_note': 'DASH video', 'vcodec': 'h264', 'fps': 60},
    '266': {'ext': 'mp4', 'height': 2160, 'format_note': 'DASH video', 'vcodec': 'h264'},

    # Dash mp4 audio
    '139': {'ext': 'm4a', 'format_note': 'DASH audio', 'acodec': 'aac', 'audio_bitrate': 48, 'container': 'm4a_dash'},
    '140': {'ext': 'm4a', 'format_note': 'DASH audio', 'acodec': 'aac', 'audio_bitrate': 128, 'container': 'm4a_dash'},
    '141': {'ext': 'm4a', 'format_note': 'DASH audio', 'acodec': 'aac', 'audio_bitrate': 256, 'container': 'm4a_dash'},
    '256': {'ext': 'm4a', 'format_note': 'DASH audio', 'acodec': 'aac', 'container': 'm4a_dash'},
    '258': {'ext': 'm4a', 'format_note': 'DASH audio', 'acodec': 'aac', 'container': 'm4a_dash'},
    '325': {'ext': 'm4a', 'format_note': 'DASH audio', 'acodec': 'dtse', 'container': 'm4a_dash'},
    '328': {'ext': 'm4a', 'format_note': 'DASH audio', 'acodec': 'ec-3', 'container': 'm4a_dash'},

    # Dash webm
    '167': {'ext': 'webm', 'height': 360, 'width': 640, 'format_note': 'DASH video', 'container': 'webm', 'vcodec': 'vp8'},
    '168': {'ext': 'webm', 'height': 480, 'width': 854, 'format_note': 'DASH video', 'container': 'webm', 'vcodec': 'vp8'},
    '169': {'ext': 'webm', 'height': 720, 'width': 1280, 'format_note': 'DASH video', 'container': 'webm', 'vcodec': 'vp8'},
    '170': {'ext': 'webm', 'height': 1080, 'width': 1920, 'format_note': 'DASH video', 'container': 'webm', 'vcodec': 'vp8'},
    '218': {'ext': 'webm', 'height': 480, 'width': 854, 'format_note': 'DASH video', 'container': 'webm', 'vcodec': 'vp8'},
    '219': {'ext': 'webm', 'height': 480, 'width': 854, 'format_note': 'DASH video', 'container': 'webm', 'vcodec': 'vp8'},
    '278': {'ext': 'webm', 'height': 144, 'format_note': 'DASH video', 'container': 'webm', 'vcodec': 'vp9'},
    '242': {'ext': 'webm', 'height': 240, 'format_note': 'DASH video', 'vcodec': 'vp9'},
    '243': {'ext': 'webm', 'height': 360, 'format_note': 'DASH video', 'vcodec': 'vp9'},
    '244': {'ext': 'webm', 'height': 480, 'format_note': 'DASH video', 'vcodec': 'vp9'},
    '245': {'ext': 'webm', 'height': 480, 'format_note': 'DASH video', 'vcodec': 'vp9'},
    '246': {'ext': 'webm', 'height': 480, 'format_note': 'DASH video', 'vcodec': 'vp9'},
    '247': {'ext': 'webm', 'height': 720, 'format_note': 'DASH video', 'vcodec': 'vp9'},
    '248': {'ext': 'webm', 'height': 1080, 'format_note': 'DASH video', 'vcodec': 'vp9'},
    '271': {'ext': 'webm', 'height': 1440, 'format_note': 'DASH video', 'vcodec': 'vp9'},
    # itag 272 videos are either 3840x2160 (e.g. RtoitU2A-3E) or 7680x4320 (sLprVF6d7Ug)
    '272': {'ext': 'webm', 'height': 2160, 'format_note': 'DASH video', 'vcodec': 'vp9'},
    '302': {'ext': 'webm', 'height': 720, 'format_note': 'DASH video', 'vcodec': 'vp9', 'fps': 60},
    '303': {'ext': 'webm', 'height': 1080, 'format_note': 'DASH video', 'vcodec': 'vp9', 'fps': 60},
    '308': {'ext': 'webm', 'height': 1440, 'format_note': 'DASH video', 'vcodec': 'vp9', 'fps': 60},
    '313': {'ext': 'webm', 'height': 2160, 'format_note': 'DASH video', 'vcodec': 'vp9'},
    '315': {'ext': 'webm', 'height': 2160, 'format_note': 'DASH video', 'vcodec': 'vp9', 'fps': 60},

    # Dash webm audio
    '171': {'ext': 'webm', 'acodec': 'vorbis', 'format_note': 'DASH audio', 'audio_bitrate': 128},
    '172': {'ext': 'webm', 'acodec': 'vorbis', 'format_note': 'DASH audio', 'audio_bitrate': 256},

    # Dash webm audio with opus inside
    '249': {'ext': 'webm', 'format_note': 'DASH audio', 'acodec': 'opus', 'audio_bitrate': 50},
    '250': {'ext': 'webm', 'format_note': 'DASH audio', 'acodec': 'opus', 'audio_bitrate': 70},
    '251': {'ext': 'webm', 'format_note': 'DASH audio', 'acodec': 'opus', 'audio_bitrate': 160},

    # RTMP (unnamed)
    '_rtmp': {'protocol': 'rtmp'},

    # av01 video only formats sometimes served with "unknown" codecs
    '394': {'vcodec': 'av01.0.05M.08'},
    '395': {'vcodec': 'av01.0.05M.08'},
    '396': {'vcodec': 'av01.0.05M.08'},
    '397': {'vcodec': 'av01.0.05M.08'},
}

def _extract_metadata_row_info(video_renderer_info):
    # extract category and music list
    info = {
        'category': None,
        'music_list': [],
    }

    current_song = {}
    for row in deep_get(video_renderer_info, 'metadataRowContainer', 'metadataRowContainerRenderer', 'rows', default=[]):
        row_title = extract_str(deep_get(row, 'metadataRowRenderer', 'title'), default='')
        row_content = extract_str(deep_get(row, 'metadataRowRenderer', 'contents', 0))
        if row_title == 'Category':
            info['category'] = row_content
        elif row_title in ('Song', 'Music'):
            if current_song:
                info['music_list'].append(current_song)
            current_song = {'title': row_content}
        elif row_title == 'Artist':
            current_song['artist'] = row_content
        elif row_title == 'Album':
            current_song['album'] = row_content
        elif row_title == 'Writers':
            current_song['writers'] = row_content
        elif row_title.startswith('Licensed'):
            current_song['licensor'] = row_content
    if current_song:
        info['music_list'].append(current_song)

    return info

def _extract_watch_info_mobile(top_level):
    info = {}
    microformat = deep_get(top_level, 'playerResponse', 'microformat', 'playerMicroformatRenderer', default={})

    family_safe = microformat.get('isFamilySafe')
    if family_safe is None:
        info['age_restricted'] = None
    else:
        info['age_restricted'] = not family_safe
    info['allowed_countries'] = microformat.get('availableCountries', [])
    info['time_published'] = microformat.get('publishDate')

    response = top_level.get('response', {})

    # this renderer has the stuff visible on the page
    # check for playlist
    items, _ = extract_items(response,
        item_types={'singleColumnWatchNextResults'})
    if items:
        watch_next_results = items[0]['singleColumnWatchNextResults']
        playlist = deep_get(watch_next_results, 'playlist', 'playlist')
        if playlist is None:
            info['playlist'] = None
        else:
            info['playlist'] = {}
            info['playlist']['title'] = playlist.get('title')
            info['playlist']['author'] = extract_str(multi_get(playlist,
                'ownerName', 'longBylineText', 'shortBylineText', 'ownerText'))
            author_id = deep_get(playlist, 'longBylineText', 'runs', 0,
                'navigationEndpoint', 'browseEndpoint', 'browseId')
            info['playlist']['author_id'] = author_id
            info['playlist']['author_url'] = concat_or_none(
                'https://www.youtube.com/channel/', author_id)
            info['playlist']['id'] = playlist.get('playlistId')
            info['playlist']['url'] = concat_or_none(
                'https://www.youtube.com/playlist?list=',
                info['playlist']['id'])
            info['playlist']['video_count'] = playlist.get('totalVideos')
            info['playlist']['current_index'] = playlist.get('currentIndex')
            info['playlist']['items'] = [
                extract_item_info(i) for i in playlist.get('contents', ())]
    else:
        info['playlist'] = None

    # Holds the visible video info. It is inside singleColumnWatchNextResults
    # but use our convenience function instead
    items, _ = extract_items(response, item_types={'slimVideoMetadataRenderer'})
    if items:
        video_info = items[0]['slimVideoMetadataRenderer']
    else:
        print('Failed to extract video metadata')
        video_info = {}

    info.update(_extract_metadata_row_info(video_info))
    info['description'] = extract_str(video_info.get('description'), recover_urls=True)
    info['view_count'] = extract_int(extract_str(video_info.get('expandedSubtitle')))
    info['author'] = extract_str(deep_get(video_info, 'owner', 'slimOwnerRenderer', 'title'))
    info['author_id'] = deep_get(video_info, 'owner', 'slimOwnerRenderer', 'navigationEndpoint', 'browseEndpoint', 'browseId')
    info['title'] = extract_str(video_info.get('title'))
    info['live'] = 'watching' in extract_str(video_info.get('expandedSubtitle'), default='')
    info['unlisted'] = False
    for badge in video_info.get('badges', []):
        if deep_get(badge, 'metadataBadgeRenderer', 'label') == 'Unlisted':
            info['unlisted'] = True
    info['like_count'] = None
    info['dislike_count'] = None
    if not info['time_published']:
        info['time_published'] = extract_date(extract_str(video_info.get('dateText', None)))
    for button in video_info.get('buttons', ()):
        button_renderer = button.get('slimMetadataToggleButtonRenderer', {})

        # all the digits can be found in the accessibility data
        count = extract_int(deep_get(button_renderer, 'button', 'toggleButtonRenderer', 'defaultText', 'accessibility', 'accessibilityData', 'label'))

        # this count doesn't have all the digits, it's like 53K for instance
        dumb_count = extract_int(extract_str(deep_get(button_renderer, 'button', 'toggleButtonRenderer', 'defaultText')))

        # the accessibility text will be "No likes" or "No dislikes" or something like that, but dumb count will be 0
        if dumb_count == 0:
            count = 0

        if 'isLike' in button_renderer:
            info['like_count'] = count
        elif 'isDislike' in button_renderer:
            info['dislike_count'] = count

    # comment section info
    items, _ = extract_items(response, item_types={
        'commentSectionRenderer', 'commentsEntryPointHeaderRenderer'})
    if items:
        header_type = list(items[0])[0]
        comment_info = items[0][header_type]
        # This seems to be some kind of A/B test being done on mobile, where
        # this is present instead of the normal commentSectionRenderer. It can
        # be seen here:
        # https://www.androidpolice.com/2019/10/31/google-youtube-app-comment-section-below-videos/
        # https://www.youtube.com/watch?v=bR5Q-wD-6qo
        if header_type == 'commentsEntryPointHeaderRenderer':
            comment_count_text = extract_str(comment_info.get('headerText'))
        else:
            comment_count_text = extract_str(deep_get(comment_info,
                'header', 'commentSectionHeaderRenderer', 'countText'))
        if comment_count_text == 'Comments':    # just this with no number, means 0 comments
            info['comment_count'] = 0
        else:
            info['comment_count'] = extract_int(comment_count_text)
        info['comments_disabled'] = False
    else:   # no comment section present means comments are disabled
        info['comment_count'] = 0
        info['comments_disabled'] = True

    # check for limited state
    items, _ = extract_items(response, item_types={'limitedStateMessageRenderer'})
    if items:
        info['limited_state'] = True
    else:
        info['limited_state'] = False

    # related videos
    related, _ = extract_items(response)
    info['related_videos'] = [extract_item_info(renderer) for renderer in related]

    return info

def _extract_watch_info_desktop(top_level):
    info = {
        'comment_count': None,
        'comments_disabled': None,
        'allowed_countries': [],
        'limited_state': None,
        'playlist': None,
    }

    video_info = {}
    for renderer in deep_get(top_level, 'response', 'contents', 'twoColumnWatchNextResults', 'results', 'results', 'contents', default=()):
        if renderer and list(renderer.keys())[0] in ('videoPrimaryInfoRenderer', 'videoSecondaryInfoRenderer'):
            video_info.update(list(renderer.values())[0])

    info.update(_extract_metadata_row_info(video_info))
    info['description'] = extract_str(video_info.get('description', None), recover_urls=True)
    info['time_published'] = extract_date(extract_str(video_info.get('dateText', None)))

    likes_dislikes = deep_get(video_info, 'sentimentBar', 'sentimentBarRenderer', 'tooltip', default='').split('/')
    if len(likes_dislikes) == 2:
        info['like_count'] = extract_int(likes_dislikes[0])
        info['dislike_count'] = extract_int(likes_dislikes[1])
    else:
        info['like_count'] = None
        info['dislike_count'] = None

    info['title'] = extract_str(video_info.get('title', None))
    info['author'] = extract_str(deep_get(video_info, 'owner', 'videoOwnerRenderer', 'title'))
    info['author_id'] = deep_get(video_info, 'owner', 'videoOwnerRenderer', 'navigationEndpoint', 'browseEndpoint', 'browseId')
    info['view_count'] = extract_int(extract_str(deep_get(video_info, 'viewCount', 'videoViewCountRenderer', 'viewCount')))

    related = deep_get(top_level, 'response', 'contents', 'twoColumnWatchNextResults', 'secondaryResults', 'secondaryResults', 'results', default=[])
    info['related_videos'] = [extract_item_info(renderer) for renderer in related]

    return info

def update_format_with_codec_info(fmt, codec):
    if (codec.startswith('av')
            or codec in ('vp9', 'vp8', 'vp8.0', 'h263', 'h264', 'mp4v')):
        if codec == 'vp8.0':
            codec = 'vp8'
        conservative_update(fmt, 'vcodec', codec)
    elif (codec.startswith('mp4a')
            or codec in ('opus', 'mp3', 'aac', 'dtse', 'ec-3', 'vorbis')):
        conservative_update(fmt, 'acodec', codec)
    else:
        print('Warning: unrecognized codec: ' + codec)

fmt_type_re = re.compile(
    r'(text|audio|video)/([\w0-9]+); codecs="([\w0-9\.]+(?:, [\w0-9\.]+)*)"')
def update_format_with_type_info(fmt, yt_fmt):
    # 'type' for invidious api format
    mime_type = multi_get(yt_fmt, 'mimeType', 'type')
    if mime_type is None:
        return
    match = re.fullmatch(fmt_type_re, mime_type)

    type, fmt['ext'], codecs = match.groups()
    codecs = codecs.split(', ')
    for codec in codecs:
        update_format_with_codec_info(fmt, codec)
    if type == 'audio':
        assert len(codecs) == 1

def _extract_formats(info, player_response):
    streaming_data = player_response.get('streamingData', {})
    yt_formats = streaming_data.get('formats', []) + streaming_data.get('adaptiveFormats', [])

    info['formats'] = []
    # because we may retry the extract_formats with a different player_response
    # so keep what we have
    conservative_update(info, 'hls_manifest_url',
        streaming_data.get('hlsManifestUrl'))
    conservative_update(info, 'dash_manifest_url',
        streaming_data.get('dash_manifest_url'))

    for yt_fmt in yt_formats:
        itag = yt_fmt.get('itag')

        fmt = {}
        fmt['itag'] = itag
        fmt['ext'] = None
        fmt['audio_bitrate'] = None
        fmt['acodec'] = None
        fmt['vcodec'] = None
        fmt['width'] = yt_fmt.get('width')
        fmt['height'] = yt_fmt.get('height')
        fmt['file_size'] = yt_fmt.get('contentLength')
        fmt['audio_sample_rate'] = yt_fmt.get('audioSampleRate')
        fmt['fps'] = yt_fmt.get('fps')
        update_format_with_type_info(fmt, yt_fmt)
        cipher = dict(urllib.parse.parse_qsl(multi_get(yt_fmt,
            'cipher', 'signatureCipher', default='')))
        if cipher:
            fmt['url'] = cipher.get('url')
        else:
            fmt['url'] = yt_fmt.get('url')
        fmt['s'] = cipher.get('s')
        fmt['sp'] = cipher.get('sp')

        # update with information from big table
        hardcoded_itag_info = _formats.get(str(itag), {})
        for key, value in hardcoded_itag_info.items():
            conservative_update(fmt, key, value) # prefer info from Youtube
        fmt['quality'] = hardcoded_itag_info.get('height')

        info['formats'].append(fmt)

    # get ip address
    if info['formats']:
        query_string = (info['formats'][0].get('url') or '?').split('?')[1]
        info['ip_address'] = deep_get(
            urllib.parse.parse_qs(query_string), 'ip', 0)
    else:
        info['ip_address'] = None

hls_regex = re.compile(r'[\w_-]+=(?:"[^"]+"|[^",]+),')
def extract_hls_formats(hls_manifest):
    '''returns hls_formats, err'''
    hls_formats = []
    try:
        lines = hls_manifest.splitlines()
        i = 0
        while i < len(lines):
            if lines[i].startswith('#EXT-X-STREAM-INF'):
                fmt = {'acodec': None, 'vcodec': None, 'height': None,
                    'width': None, 'fps': None, 'audio_bitrate': None,
                    'itag': None, 'file_size': None,
                    'audio_sample_rate': None, 'url': None}
                properties = lines[i].split(':')[1]
                properties += ',' # make regex work for last key-value pair

                for pair in hls_regex.findall(properties):
                    key, value = pair.rstrip(',').split('=')
                    if key == 'CODECS':
                        for codec in value.strip('"').split(','):
                            update_format_with_codec_info(fmt, codec)
                    elif key == 'RESOLUTION':
                        fmt['width'], fmt['height'] = map(int, value.split('x'))
                        fmt['resolution'] = value
                    elif key == 'FRAME-RATE':
                        fmt['fps'] = int(value)
                i += 1
                fmt['url'] = lines[i]
                assert fmt['url'].startswith('http')
                fmt['ext'] = 'm3u8'
                hls_formats.append(fmt)
            i += 1
    except Exception as e:
        traceback.print_exc()
        return [], str(e)
    return hls_formats, None


def _extract_playability_error(info, player_response, error_prefix=''):
    if info['formats']:
        info['playability_status'] = None
        info['playability_error'] = None
        return

    playability_status = deep_get(player_response, 'playabilityStatus', 'status', default=None)
    info['playability_status'] = playability_status

    playability_reason = extract_str(multi_deep_get(player_response,
        ['playabilityStatus', 'reason'],
        ['playabilityStatus', 'errorScreen', 'playerErrorMessageRenderer', 'reason'],
        default='Could not find playability error')
    )

    if playability_status not in (None, 'OK'):
        info['playability_error'] = error_prefix + playability_reason
    elif not info['playability_error']: # do not override
        info['playability_error'] = error_prefix + 'Unknown playability error'

SUBTITLE_FORMATS = ('srv1', 'srv2', 'srv3', 'ttml', 'vtt')
def extract_watch_info(polymer_json):
    info = {'playability_error': None, 'error': None,
        'player_response_missing': None}

    if isinstance(polymer_json, dict):
        top_level = polymer_json
    elif isinstance(polymer_json, (list, tuple)):
        top_level = {}
        for page_part in polymer_json:
            if not isinstance(page_part, dict):
                return {'error': 'Invalid page part'}
            top_level.update(page_part)
    else:
        return {'error': 'Invalid top level polymer data'}

    error = check_missing_keys(top_level,
        ['player', 'args'],
        ['player', 'assets', 'js'],
        ['playerResponse'],
    )
    if error:
        info['playability_error'] = error

    player_response = top_level.get('playerResponse', {})

    # usually, only the embedded one has the urls
    player_args = deep_get(top_level, 'player', 'args', default={})
    if 'player_response' in player_args:
        embedded_player_response = json.loads(player_args['player_response'])
    else:
        embedded_player_response = {}

    # captions
    info['automatic_caption_languages'] = []
    info['manual_caption_languages'] = []
    info['_manual_caption_language_names'] = {}     # language name written in that language, needed in some cases to create the url
    info['translation_languages'] = []
    captions_info = player_response.get('captions', {})
    info['_captions_base_url'] = normalize_url(deep_get(captions_info, 'playerCaptionsRenderer', 'baseUrl'))
    for caption_track in deep_get(captions_info, 'playerCaptionsTracklistRenderer', 'captionTracks', default=()):
        lang_code = caption_track.get('languageCode')
        if not lang_code:
            continue
        if caption_track.get('kind') == 'asr':
            info['automatic_caption_languages'].append(lang_code)
        else:
            info['manual_caption_languages'].append(lang_code)
        base_url = caption_track.get('baseUrl', '')
        lang_name = deep_get(urllib.parse.parse_qs(urllib.parse.urlparse(base_url).query), 'name', 0)
        if lang_name:
            info['_manual_caption_language_names'][lang_code] = lang_name

    for translation_lang_info in deep_get(captions_info, 'playerCaptionsTracklistRenderer', 'translationLanguages', default=()):
        lang_code = translation_lang_info.get('languageCode')
        if lang_code:
            info['translation_languages'].append(lang_code)
        if translation_lang_info.get('isTranslatable') == False:
            print('WARNING: Found non-translatable caption language')

    # formats
    _extract_formats(info, embedded_player_response)
    if not info['formats']:
        _extract_formats(info, player_response)

    # see https://github.com/user234683/youtube-local/issues/22#issuecomment-706395160
    info['player_urls_missing'] = (
        not info['formats'] and not embedded_player_response)

    # playability errors
    _extract_playability_error(info, player_response)

    # check age-restriction
    info['age_restricted'] = (info['playability_status'] == 'LOGIN_REQUIRED' and info['playability_error'] and ' age' in info['playability_error'])

    # base_js (for decryption of signatures)
    info['base_js'] = deep_get(top_level, 'player', 'assets', 'js')
    if info['base_js']:
        info['base_js'] = normalize_url(info['base_js'])
        # must uniquely identify url
        info['player_name'] = urllib.parse.urlparse(info['base_js']).path
    else:
        info['player_name'] = None

    # extract stuff from visible parts of page
    mobile = 'singleColumnWatchNextResults' in deep_get(top_level, 'response', 'contents', default={})
    if mobile:
        info.update(_extract_watch_info_mobile(top_level))
    else:
        info.update(_extract_watch_info_desktop(top_level))

    # stuff from videoDetails. Use liberal_update to prioritize info from videoDetails over existing info
    vd = deep_get(top_level, 'playerResponse', 'videoDetails', default={})
    liberal_update(info, 'title',      extract_str(vd.get('title')))
    liberal_update(info, 'duration',   extract_int(vd.get('lengthSeconds')))
    liberal_update(info, 'view_count', extract_int(vd.get('viewCount')))
    # videos with no description have a blank string
    liberal_update(info, 'description', vd.get('shortDescription'))
    liberal_update(info, 'id',          vd.get('videoId'))
    liberal_update(info, 'author',      vd.get('author'))
    liberal_update(info, 'author_id',   vd.get('channelId'))
    info['was_live'] =                  vd.get('isLiveContent')
    conservative_update(info, 'unlisted', not vd.get('isCrawlable', True))  #isCrawlable is false on limited state videos even if they aren't unlisted
    liberal_update(info, 'tags',        vd.get('keywords', []))

    # fallback stuff from microformat
    mf = deep_get(top_level, 'playerResponse', 'microformat', 'playerMicroformatRenderer', default={})
    conservative_update(info, 'title',      extract_str(mf.get('title')))
    conservative_update(info, 'duration', extract_int(mf.get('lengthSeconds')))
    # this gives the view count for limited state videos
    conservative_update(info, 'view_count', extract_int(mf.get('viewCount')))
    conservative_update(info, 'description', extract_str(mf.get('description'), recover_urls=True))
    conservative_update(info, 'author', mf.get('ownerChannelName'))
    conservative_update(info, 'author_id', mf.get('externalChannelId'))
    conservative_update(info, 'live', deep_get(mf, 'liveBroadcastDetails',
        'isLiveNow'))
    liberal_update(info, 'unlisted', mf.get('isUnlisted'))
    liberal_update(info, 'category', mf.get('category'))
    liberal_update(info, 'time_published', mf.get('publishDate'))
    liberal_update(info, 'time_uploaded', mf.get('uploadDate'))

    # other stuff
    info['author_url'] = 'https://www.youtube.com/channel/' + info['author_id'] if info['author_id'] else None
    return info

single_char_codes = {
    'n': '\n',
    '\\': '\\',
    '"': '"',
    "'": "'",
    'b': '\b',
    'f': '\f',
    'n': '\n',
    'r': '\r',
    't': '\t',
    'v': '\x0b',
    '0': '\x00',
    '\n': '', # backslash followed by literal newline joins lines
}
def js_escape_replace(match):
    r'''Resolves javascript string escape sequences such as \x..'''
    # some js-strings in the watch page html include them for no reason
    # https://mathiasbynens.be/notes/javascript-escapes
    escaped_sequence = match.group(1)
    if escaped_sequence[0] in ('x', 'u'):
        return chr(int(escaped_sequence[1:], base=16))

    # In javascript, if it's not one of those escape codes, it's just the
    # literal character. e.g., "\a" = "a"
    return single_char_codes.get(escaped_sequence, escaped_sequence)

# works but complicated and unsafe:
#PLAYER_RESPONSE_RE = re.compile(r'<script[^>]*?>[^<]*?var ytInitialPlayerResponse = ({(?:"(?:[^"\\]|\\.)*?"|[^"])+?});')

# Because there are sometimes additional statements after the json object
# so we just capture all of those until end of script and tell json decoder
# to ignore extra stuff after the json object
PLAYER_RESPONSE_RE = re.compile(r'<script[^>]*?>[^<]*?var ytInitialPlayerResponse = ({.*?)</script>')
INITIAL_DATA_RE = re.compile(r"<script[^>]*?>var ytInitialData = '(.+?[^\\])';")
BASE_JS_RE = re.compile(r'jsUrl":\s*"([\w\-\./]+?/base.js)"')
JS_STRING_ESCAPE_RE = re.compile(r'\\([^xu]|x..|u....)')
def extract_watch_info_from_html(watch_html):
    base_js_match = BASE_JS_RE.search(watch_html)
    player_response_match = PLAYER_RESPONSE_RE.search(watch_html)
    initial_data_match = INITIAL_DATA_RE.search(watch_html)

    if base_js_match is not None:
        base_js_url = base_js_match.group(1)
    else:
        base_js_url = None

    if player_response_match is not None:
        decoder = json.JSONDecoder()
        # this will make it ignore extra stuff after end of object
        player_response = decoder.raw_decode(player_response_match.group(1))[0]
    else:
        return {'error': 'Could not find ytInitialPlayerResponse'}
        player_response = None

    if initial_data_match is not None:
        initial_data = initial_data_match.group(1)
        initial_data = JS_STRING_ESCAPE_RE.sub(js_escape_replace, initial_data)
        initial_data = json.loads(initial_data)
    else:
        print('extract_watch_info_from_html: failed to find initialData')
        initial_data = None

    # imitate old format expected by extract_watch_info
    fake_polymer_json = {
        'player': {
            'args': {},
            'assets': {
                'js': base_js_url
            }
        },
        'playerResponse': player_response,
        'response': initial_data,
    }

    return extract_watch_info(fake_polymer_json)



def get_caption_url(info, language, format, automatic=False, translation_language=None):
    '''Gets the url for captions with the given language and format. If automatic is True, get the automatic captions for that language. If translation_language is given, translate the captions from `language` to `translation_language`. If automatic is true and translation_language is given, the automatic captions will be translated.'''
    url = info['_captions_base_url']
    url += '&lang=' + language
    url += '&fmt=' + format
    if automatic:
        url += '&kind=asr'
    elif language in info['_manual_caption_language_names']:
        url += '&name=' + urllib.parse.quote(info['_manual_caption_language_names'][language], safe='')

    if translation_language:
        url += '&tlang=' + translation_language
    return url

def update_with_age_restricted_info(info, video_info_page):
    '''Inserts urls from 'player_response' in get_video_info page'''
    ERROR_PREFIX = 'Error getting missing player or bypassing age-restriction: '

    video_info = urllib.parse.parse_qs(video_info_page)
    player_response = deep_get(video_info, 'player_response', 0)
    if player_response is None:
        info['playability_error'] = ERROR_PREFIX + 'Could not find player_response in video_info_page'
        return
    try:
        player_response = json.loads(player_response)
    except json.decoder.JSONDecodeError:
        traceback.print_exc()
        info['playability_error'] = ERROR_PREFIX + 'Failed to parse json response'
        return

    _extract_formats(info, player_response)
    _extract_playability_error(info, player_response, error_prefix=ERROR_PREFIX)

def requires_decryption(info):
    return ('formats' in info) and info['formats'] and info['formats'][0]['s']

# adapted from youtube-dl and invidious:
# https://github.com/omarroth/invidious/blob/master/src/invidious/helpers/signatures.cr
decrypt_function_re = re.compile(r'function\(a\)\{(a=a\.split\(""\)[^\}{]+)return a\.join\(""\)\}')
# gives us e.g. rt, .xK, 5 from rt.xK(a,5) or rt, ["xK"], 5 from rt["xK"](a,5)
# (var, operation, argument)
var_op_arg_re = re.compile(r'(\w+)(\.\w+|\["[^"]+"\])\(a,(\d+)\)')
def extract_decryption_function(info, base_js):
    '''Insert decryption function into info. Return error string if not successful.

    Decryption function is a list of list[2] of numbers.

    It is advisable to cache the decryption function (uniquely identified by info['player_name']) so base.js (1 MB) doesn't need to be redownloaded each time'''
    info['decryption_function'] = None
    decrypt_function_match = decrypt_function_re.search(base_js)
    if decrypt_function_match is None:
        return 'Could not find decryption function in base.js'

    function_body = decrypt_function_match.group(1).split(';')[1:-1]
    if not function_body:
        return 'Empty decryption function body'

    var_with_operation_match = var_op_arg_re.fullmatch(function_body[0])
    if var_with_operation_match is None:
        return 'Could not find var_name'

    var_name = var_with_operation_match.group(1)
    var_body_match = re.search(r'var ' + re.escape(var_name) + r'=\{(.*?)\};', base_js, flags=re.DOTALL)
    if var_body_match is None:
        return 'Could not find var_body'

    operations = var_body_match.group(1).replace('\n', '').split('},')
    if not operations:
        return 'Did not find any definitions in var_body'
    operations[-1] = operations[-1][:-1]    # remove the trailing '}' since we split by '},' on the others
    operation_definitions = {}
    for op in operations:
        colon_index = op.find(':')
        opening_brace_index = op.find('{')

        if colon_index == -1 or opening_brace_index == -1:
            return 'Could not parse operation'
        op_name = op[:colon_index]
        op_body = op[opening_brace_index+1:]
        if op_body == 'a.reverse()':
            operation_definitions[op_name] = 0
        elif op_body == 'a.splice(0,b)':
            operation_definitions[op_name] = 1
        elif op_body.startswith('var c=a[0]'):
            operation_definitions[op_name] = 2
        else:
            return 'Unknown op_body: ' + op_body

    decryption_function = []
    for op_with_arg in function_body:
        match = var_op_arg_re.fullmatch(op_with_arg)
        if match is None:
            return 'Could not parse operation with arg'
        op_name = match.group(2).strip('[].')
        if op_name not in operation_definitions:
            return 'Unknown op_name: ' + str(op_name)
        op_argument = match.group(3)
        decryption_function.append([operation_definitions[op_name], int(op_argument)])

    info['decryption_function'] = decryption_function
    return False

def _operation_2(a, b):
    c = a[0]
    a[0] = a[b % len(a)]
    a[b % len(a)] = c

def decrypt_signatures(info):
    '''Applies info['decryption_function'] to decrypt all the signatures. Return err.'''
    if not info.get('decryption_function'):
        return 'decryption_function not in info'
    for format in info['formats']:
        if not format['s'] or not format['sp'] or not format['url']:
            print('Warning: s, sp, or url not in format')
            continue

        a = list(format['s'])
        for op, argument in info['decryption_function']:
            if op == 0:
                a.reverse()
            elif op == 1:
                a = a[argument:]
            else:
                _operation_2(a, argument)

        signature = ''.join(a)
        format['url'] += '&' + format['sp'] + '=' + signature
    return False