diff --git a/yt_dlp/extractor/nitter.py b/yt_dlp/extractor/nitter.py index a0546cda0f..8bb709cd7e 100644 --- a/yt_dlp/extractor/nitter.py +++ b/yt_dlp/extractor/nitter.py @@ -5,7 +5,6 @@ from ..compat import compat_urlparse from ..utils import ( parse_count, - unified_strdate, unified_timestamp, remove_end, determine_ext, @@ -25,6 +24,16 @@ class NitterIE(InfoExtractor): 'nitter.v6vgyqpa7yefkorazmg5d5fimstmvm2vtbirt6676mt7qmllrcnwycqd.onion', 'i23nv6w3juvzlw32xzoxcqzktegd4i4fu3nmnc2ewv4ggiu4ledwklad.onion', '26oq3gioiwcmfojub37nz5gzbkdiqp7fue5kvye7d4txv4ny6fb4wwid.onion', + 'vfaomgh4jxphpbdfizkm5gbtjahmei234giqj4facbwhrfjtcldauqad.onion', + 'iwgu3cv7ywf3gssed5iqtavmrlszgsxazkmwwnt4h2kdait75thdyrqd.onion', + 'erpnncl5nhyji3c32dcfmztujtl3xaddqb457jsbkulq24zqq7ifdgad.onion', + 'ckzuw5misyahmg7j5t5xwwuj3bwy62jfolxyux4brfflramzsvvd3syd.onion', + 'jebqj47jgxleaiosfcxfibx2xdahjettuydlxbg64azd4khsxv6kawid.onion', + 'nttr2iupbb6fazdpr2rgbooon2tzbbsvvkagkgkwohhodjzj43stxhad.onion', + 'nitraeju2mipeziu2wtcrqsxg7h62v5y4eqgwi75uprynkj74gevvuqd.onion', + 'nitter.lqs5fjmajyp7rvp4qvyubwofzi6d4imua7vs237rkc4m5qogitqwrgyd.onion', + 'ibsboeui2im5o7dxnik3s5yghufumgy5abevtij5nbizequfpu4qi4ad.onion', + 'ec5nvbycpfa5k6ro77blxgkyrzbkv7uy6r5cngcbkadtjj2733nm3uyd.onion', 'nitter.i2p', 'u6ikd6zndl3c4dsdq4mmujpntgeevdk5qzkfb57r4tnfeccrn2qa.b32.i2p', @@ -36,28 +45,55 @@ class NitterIE(InfoExtractor): 'nitter.42l.fr', 'nitter.pussthecat.org', 'nitter.nixnet.services', - 'nitter.mastodont.cat', - 'nitter.tedomum.net', 'nitter.fdn.fr', 'nitter.1d4.us', 'nitter.kavin.rocks', - 'tweet.lambda.dance', - 'nitter.cc', - 'nitter.vxempire.xyz', 'nitter.unixfox.eu', 'nitter.domain.glass', - 'nitter.himiko.cloud', 'nitter.eu', 'nitter.namazso.eu', - 'nitter.mailstation.de', 'nitter.actionsack.com', - 'nitter.cattube.org', - 'nitter.dark.fail', 'birdsite.xanny.family', - 'nitter.40two.app', - 'nitter.skrep.in', + 'nitter.hu', + 'twitr.gq', + 'nitter.moomoo.me', + 'nittereu.moomoo.me', + 'bird.from.tf', + 'nitter.it', + 'twitter.censors.us', + 'twitter.grimneko.de', + 'nitter.alefvanoon.xyz', + 'n.hyperborea.cloud', + 'nitter.ca', + 'twitter.076.ne.jp', + 'twitter.mstdn.social', + 'nitter.fly.dev', + 'notabird.site', + 'nitter.weiler.rocks', + 'nitter.silkky.cloud', + 'nitter.sethforprivacy.com', + 'nttr.stream', + 'nitter.cutelab.space', + 'nitter.nl', + 'nitter.mint.lgbt', + 'nitter.bus-hit.me', + 'fuckthesacklers.network', + 'nitter.govt.land', + 'nitter.datatunnel.xyz', + 'nitter.esmailelbob.xyz', + 'tw.artemislena.eu', + 'de.nttr.stream', + 'nitter.winscloud.net', + 'nitter.tiekoetter.com', + 'nitter.spaceint.fr', + 'twtr.bch.bar', + 'nitter.exonip.de', + 'nitter.mastodon.pro', + 'nitter.notraxx.ch', + # not in the list anymore + 'nitter.skrep.in', 'nitter.snopyta.org', ) @@ -68,96 +104,121 @@ class NitterIE(InfoExtractor): # official, rate limited 'nitter.net', # offline + 'is-nitter.resolv.ee', + 'lu-nitter.resolv.ee', 'nitter.13ad.de', + 'nitter.40two.app', + 'nitter.cattube.org', + 'nitter.cc', + 'nitter.dark.fail', + 'nitter.himiko.cloud', + 'nitter.koyu.space', + 'nitter.mailstation.de', + 'nitter.mastodont.cat', + 'nitter.tedomum.net', + 'nitter.tokhmi.xyz', 'nitter.weaponizedhumiliation.com', + 'nitter.vxempire.xyz', + 'tweet.lambda.dance', ) INSTANCES = NON_HTTP_INSTANCES + HTTP_INSTANCES + DEAD_INSTANCES - _INSTANCES_RE = '(?:' + '|'.join([re.escape(instance) for instance in INSTANCES]) + ')' - _VALID_URL = r'https?://%(instance)s/(?P.+)/status/(?P[0-9]+)(#.)?' % {'instance': _INSTANCES_RE} + _INSTANCES_RE = f'(?:{"|".join(map(re.escape, INSTANCES))})' + _VALID_URL = fr'https?://{_INSTANCES_RE}/(?P.+)/status/(?P[0-9]+)(#.)?' current_instance = random.choice(HTTP_INSTANCES) _TESTS = [ { # GIF (wrapped in mp4) - 'url': 'https://%s/firefox/status/1314279897502629888#m' % current_instance, + 'url': f'https://{current_instance}/firefox/status/1314279897502629888#m', 'info_dict': { 'id': '1314279897502629888', 'ext': 'mp4', - 'title': 'Firefox 🔥 - You know the old saying, if you see something say something. Now you actually can with the YouTube regrets extension. \n\nReport harmful YouTube recommendations so others can avoid watching them. ➡️ https://mzl.la/3iFIiyg\n\n#UnfckTheInternet', - 'description': 'You know the old saying, if you see something say something. Now you actually can with the YouTube regrets extension. \n\nReport harmful YouTube recommendations so others can avoid watching them. ➡️ https://mzl.la/3iFIiyg\n\n#UnfckTheInternet', + 'title': 'md5:7890a9277da4639ab624dd899424c5d8', + 'description': 'md5:5fea96a4d3716c350f8b95b21b3111fe', 'thumbnail': r're:^https?://.*\.jpg$', 'uploader': 'Firefox 🔥', 'uploader_id': 'firefox', - 'uploader_url': 'https://%s/firefox' % current_instance, + 'uploader_url': f'https://{current_instance}/firefox', 'upload_date': '20201008', 'timestamp': 1602183720, + 'like_count': int, + 'repost_count': int, + 'comment_count': int, }, }, { # normal video - 'url': 'https://%s/Le___Doc/status/1299715685392756737#m' % current_instance, + 'url': f'https://{current_instance}/Le___Doc/status/1299715685392756737#m', 'info_dict': { 'id': '1299715685392756737', 'ext': 'mp4', - 'title': 'Le Doc - "Je ne prédis jamais rien"\nD Raoult, Août 2020...', + 'title': 're:^.* - "Je ne prédis jamais rien"\nD Raoult, Août 2020...', 'description': '"Je ne prédis jamais rien"\nD Raoult, Août 2020...', 'thumbnail': r're:^https?://.*\.jpg$', - 'uploader': 'Le Doc', + 'uploader': 're:^Le *Doc', 'uploader_id': 'Le___Doc', - 'uploader_url': 'https://%s/Le___Doc' % current_instance, + 'uploader_url': f'https://{current_instance}/Le___Doc', 'upload_date': '20200829', - 'timestamp': 1598711341, + 'timestamp': 1598711340, 'view_count': int, 'like_count': int, 'repost_count': int, 'comment_count': int, }, }, { # video embed in a "Streaming Political Ads" box - 'url': 'https://%s/mozilla/status/1321147074491092994#m' % current_instance, + 'url': f'https://{current_instance}/mozilla/status/1321147074491092994#m', 'info_dict': { 'id': '1321147074491092994', 'ext': 'mp4', - 'title': "Mozilla - Are you being targeted with weird, ominous or just plain annoying political ads while streaming your favorite shows?\n\nThis isn't a real political ad, but if you're watching streaming TV in the U.S., chances are you've seen quite a few. \n\nLearn more ➡️ https://mzl.la/StreamingAds", - 'description': "Are you being targeted with weird, ominous or just plain annoying political ads while streaming your favorite shows?\n\nThis isn't a real political ad, but if you're watching streaming TV in the U.S., chances are you've seen quite a few. \n\nLearn more ➡️ https://mzl.la/StreamingAds", + 'title': 'md5:8290664aabb43b9189145c008386bf12', + 'description': 'md5:9cf2762d49674bc416a191a689fb2aaa', 'thumbnail': r're:^https?://.*\.jpg$', 'uploader': 'Mozilla', 'uploader_id': 'mozilla', - 'uploader_url': 'https://%s/mozilla' % current_instance, + 'uploader_url': f'https://{current_instance}/mozilla', 'upload_date': '20201027', - 'timestamp': 1603820982 + 'timestamp': 1603820940, + 'view_count': int, + 'like_count': int, + 'repost_count': int, + 'comment_count': int, }, + 'expected_warnings': ['Ignoring subtitle tracks found in the HLS manifest'], }, { # not the first tweet but main-tweet - 'url': 'https://%s/TheNaturalNu/status/1379050895539724290#m' % current_instance, + 'url': f'https://{current_instance}/firefox/status/1354848277481414657#m', 'info_dict': { - 'id': '1379050895539724290', + 'id': '1354848277481414657', 'ext': 'mp4', - 'title': 'Dorothy Zbornak - This had me hollering!!', - 'description': 'This had me hollering!!', + 'title': 'md5:bef647f03bd1c6b15b687ea70dfc9700', + 'description': 'md5:5efba25e2f9dac85ebcd21160cb4341f', 'thumbnail': r're:^https?://.*\.jpg$', - 'uploader': 'Dorothy Zbornak', - 'uploader_id': 'TheNaturalNu', - 'uploader_url': 'https://%s/TheNaturalNu' % current_instance, - 'timestamp': 1617626329, - 'upload_date': '20210405' + 'uploader': 'Firefox 🔥', + 'uploader_id': 'firefox', + 'uploader_url': f'https://{current_instance}/firefox', + 'upload_date': '20210128', + 'timestamp': 1611855960, + 'view_count': int, + 'like_count': int, + 'repost_count': int, + 'comment_count': int, } } ] def _real_extract(self, url): - video_id = self._match_id(url) + video_id, uploader_id = self._match_valid_url(url).group('id', 'uploader_id') parsed_url = compat_urlparse.urlparse(url) - base_url = '%s://%s' % (parsed_url.scheme, parsed_url.netloc) + base_url = f'{parsed_url.scheme}://{parsed_url.netloc}' self._set_cookie(parsed_url.netloc, 'hlsPlayback', 'on') - full_webpage = self._download_webpage(url, video_id) + full_webpage = webpage = self._download_webpage(url, video_id) main_tweet_start = full_webpage.find('class="main-tweet"') if main_tweet_start > 0: webpage = full_webpage[main_tweet_start:] - if not webpage: - webpage = full_webpage - video_url = '%s%s' % (base_url, self._html_search_regex(r'(?:]+data-url|]+src)="([^"]+)"', webpage, 'video url')) + video_url = '%s%s' % (base_url, self._html_search_regex( + r'(?:]+data-url|]+src)="([^"]+)"', webpage, 'video url')) ext = determine_ext(video_url) if ext == 'unknown_video': @@ -168,61 +229,49 @@ def _real_extract(self, url): 'ext': ext }] - title = self._og_search_description(full_webpage) - if not title: - title = self._html_search_regex(r'
]+>([^<]+)
', webpage, 'title', fatal=False) - mobj = self._match_valid_url(url) - uploader_id = ( - mobj.group('uploader_id') - or self._html_search_regex(r']+title="([^"]+)"', webpage, 'uploader name', fatal=False) - ) - - if uploader_id: - uploader_url = '%s/%s' % (base_url, uploader_id) - - uploader = self._html_search_regex(r']+title="([^"]+)"', webpage, 'uploader name', fatal=False) + uploader_id = self._html_search_regex( + r']+title="@([^"]+)"', webpage, 'uploader id', fatal=False) or uploader_id + uploader = self._html_search_regex( + r']+title="([^"]+)"', webpage, 'uploader name', fatal=False) if uploader: - title = '%s - %s' % (uploader, title) + title = f'{uploader} - {title}' - view_count = parse_count(self._html_search_regex(r']+class="icon-play[^>]*>\s([^<]+)', webpage, 'view count', fatal=False)) - like_count = parse_count(self._html_search_regex(r']+class="icon-heart[^>]*>\s([^<]+)', webpage, 'like count', fatal=False)) - repost_count = parse_count(self._html_search_regex(r']+class="icon-retweet[^>]*>\s([^<]+)', webpage, 'repost count', fatal=False)) - comment_count = parse_count(self._html_search_regex(r']+class="icon-comment[^>]*>\s([^<]+)', webpage, 'repost count', fatal=False)) + counts = { + f'{x[0]}_count': self._html_search_regex( + fr']+class="icon-{x[1]}[^>]*>([^<]*)', + webpage, f'{x[0]} count', fatal=False) + for x in (('view', 'play'), ('like', 'heart'), ('repost', 'retweet'), ('comment', 'comment')) + } + counts = {field: 0 if count == '' else parse_count(count) for field, count in counts.items()} - thumbnail = self._html_search_meta('og:image', full_webpage, 'thumbnail url') - if not thumbnail: - thumbnail = '%s%s' % (base_url, self._html_search_regex(r']+poster="([^"]+)"', webpage, 'thumbnail url', fatal=False)) - thumbnail = remove_end(thumbnail, '%3Asmall') + thumbnail = ( + self._html_search_meta('og:image', full_webpage, 'thumbnail url') + or remove_end('%s%s' % (base_url, self._html_search_regex( + r']+poster="([^"]+)"', webpage, 'thumbnail url', fatal=False)), '%3Asmall')) - thumbnails = [] - thumbnail_ids = ('thumb', 'small', 'large', 'medium', 'orig') - for id in thumbnail_ids: - thumbnails.append({ - 'id': id, - 'url': thumbnail + '%3A' + id, - }) + thumbnails = [ + {'id': id, 'url': f'{thumbnail}%3A{id}'} + for id in ('thumb', 'small', 'large', 'medium', 'orig') + ] - date = self._html_search_regex(r']+class="tweet-date"[^>]*>]+title="([^"]+)"', webpage, 'upload date', fatal=False) - upload_date = unified_strdate(date) - timestamp = unified_timestamp(date) + date = self._html_search_regex( + r']+class="tweet-date"[^>]*>]+title="([^"]+)"', + webpage, 'upload date', default='').replace('·', '') return { 'id': video_id, 'title': title, 'description': description, 'uploader': uploader, - 'timestamp': timestamp, + 'timestamp': unified_timestamp(date), 'uploader_id': uploader_id, - 'uploader_url': uploader_url, - 'view_count': view_count, - 'like_count': like_count, - 'repost_count': repost_count, - 'comment_count': comment_count, + 'uploader_url': f'{base_url}/{uploader_id}', 'formats': formats, 'thumbnails': thumbnails, 'thumbnail': thumbnail, - 'upload_date': upload_date, + **counts, }