X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/7df4343e553a9942a290e5e1dffe931261ab844e..dd50658396011c8e98b7201f035408b7959d56e1:/youtube_dl/extractor/ted.py diff --git a/youtube_dl/extractor/ted.py b/youtube_dl/extractor/ted.py index a48d77c..d3e4205 100644 --- a/youtube_dl/extractor/ted.py +++ b/youtube_dl/extractor/ted.py @@ -6,7 +6,12 @@ import re from .common import InfoExtractor from ..compat import compat_str -from ..utils import int_or_none +from ..utils import ( + float_or_none, + int_or_none, + try_get, + url_or_none, +) class TEDIE(InfoExtractor): @@ -27,7 +32,7 @@ class TEDIE(InfoExtractor): ''' _TESTS = [{ 'url': 'http://www.ted.com/talks/dan_dennett_on_our_consciousness.html', - 'md5': 'fc94ac279feebbce69f21c0c6ee82810', + 'md5': 'b0ce2b05ca215042124fbc9e3886493a', 'info_dict': { 'id': '102', 'ext': 'mp4', @@ -37,21 +42,32 @@ class TEDIE(InfoExtractor): 'consciousness, but that half the time our brains are ' 'actively fooling us.'), 'uploader': 'Dan Dennett', - 'width': 854, + 'width': 853, 'duration': 1308, - } + 'view_count': int, + 'comment_count': int, + 'tags': list, + }, + 'params': { + 'skip_download': True, + }, }, { - 'url': 'http://www.ted.com/watch/ted-institute/ted-bcg/vishal-sikka-the-beauty-and-power-of-algorithms', - 'md5': '226f4fb9c62380d11b7995efa4c87994', + # missing HTTP bitrates + 'url': 'https://www.ted.com/talks/vishal_sikka_the_beauty_and_power_of_algorithms', 'info_dict': { - 'id': 'vishal-sikka-the-beauty-and-power-of-algorithms', + 'id': '6069', 'ext': 'mp4', - 'title': 'Vishal Sikka: The beauty and power of algorithms', - 'thumbnail': 're:^https?://.+\.jpg', - 'description': 'Adaptive, intelligent, and consistent, algorithms are emerging as the ultimate app for everything from matching consumers to products to assessing medical diagnoses. Vishal Sikka shares his appreciation for the algorithm, charting both its inherent beauty and its growing power.', - } + 'title': 'The beauty and power of algorithms', + 'thumbnail': r're:^https?://.+\.jpg', + 'description': 'md5:734e352710fb00d840ab87ae31aaf688', + 'uploader': 'Vishal Sikka', + }, + 'params': { + 'skip_download': True, + }, }, { 'url': 'http://www.ted.com/talks/gabby_giffords_and_mark_kelly_be_passionate_be_courageous_be_your_best', + 'md5': 'e6b9617c01a7970ceac8bb2c92c346c0', 'info_dict': { 'id': '1972', 'ext': 'mp4', @@ -60,6 +76,9 @@ class TEDIE(InfoExtractor): 'description': 'md5:5174aed4d0f16021b704120360f72b92', 'duration': 1128, }, + 'params': { + 'skip_download': True, + }, }, { 'url': 'http://www.ted.com/playlists/who_are_the_hackers', 'info_dict': { @@ -73,7 +92,7 @@ class TEDIE(InfoExtractor): 'add_ie': ['Youtube'], 'info_dict': { 'id': '_ZG8HBuDjgc', - 'ext': 'mp4', + 'ext': 'webm', 'title': 'Douglas Adams: Parrots the Universe and Everything', 'description': 'md5:01ad1e199c49ac640cb1196c0e9016af', 'uploader': 'University of California Television (UCTV)', @@ -84,17 +103,17 @@ class TEDIE(InfoExtractor): 'skip_download': True, }, }, { - # YouTube video - 'url': 'http://www.ted.com/talks/jeffrey_kluger_the_sibling_bond', - 'add_ie': ['Youtube'], + # no nativeDownloads + 'url': 'https://www.ted.com/talks/tom_thum_the_orchestra_in_my_mouth', 'info_dict': { - 'id': 'aFBIPO-P7LM', + 'id': '1792', 'ext': 'mp4', - 'title': 'The hidden power of siblings: Jeff Kluger at TEDxAsheville', - 'description': 'md5:3d7a4f50d95ca5dd67104e2a20f43fe1', - 'uploader': 'TEDx Talks', - 'uploader_id': 'TEDxTalks', - 'upload_date': '20111216', + 'title': 'The orchestra in my mouth', + 'description': 'md5:5d1d78650e2f8dfcbb8ebee2951ac29a', + 'uploader': 'Tom Thum', + 'view_count': int, + 'comment_count': int, + 'tags': list, }, 'params': { 'skip_download': True, @@ -102,14 +121,15 @@ class TEDIE(InfoExtractor): }] _NATIVE_FORMATS = { - 'low': {'preference': 1, 'width': 320, 'height': 180}, - 'medium': {'preference': 2, 'width': 512, 'height': 288}, - 'high': {'preference': 3, 'width': 854, 'height': 480}, + 'low': {'width': 320, 'height': 180}, + 'medium': {'width': 512, 'height': 288}, + 'high': {'width': 854, 'height': 480}, } def _extract_info(self, webpage): - info_json = self._search_regex(r'q\("\w+.init",({.+})\)', - webpage, 'info json') + info_json = self._search_regex( + r'(?s)q\(\s*"\w+.init"\s*,\s*({.+})\)\s*', + webpage, 'info json') return json.loads(info_json) def _real_extract(self, url): @@ -131,11 +151,16 @@ class TEDIE(InfoExtractor): webpage = self._download_webpage(url, name, 'Downloading playlist webpage') info = self._extract_info(webpage) - playlist_info = info['playlist'] + + playlist_info = try_get( + info, lambda x: x['__INITIAL_DATA__']['playlist'], + dict) or info['playlist'] playlist_entries = [ self.url_result('http://www.ted.com/talks/' + talk['slug'], self.ie_key()) - for talk in info['talks'] + for talk in try_get( + info, lambda x: x['__INITIAL_DATA__']['talks'], + dict) or info['talks'] ] return self.playlist_result( playlist_entries, @@ -144,42 +169,60 @@ class TEDIE(InfoExtractor): def _talk_info(self, url, video_name): webpage = self._download_webpage(url, video_name) - self.report_extraction(video_name) - - talk_info = self._extract_info(webpage)['talks'][0] - - external = talk_info.get('external') - if external: - service = external['service'] - self.to_screen('Found video from %s' % service) - ext_url = None - if service.lower() == 'youtube': - ext_url = external.get('code') - return { - '_type': 'url', - 'url': ext_url or external['uri'], - } + + info = self._extract_info(webpage) + + data = try_get(info, lambda x: x['__INITIAL_DATA__'], dict) or info + talk_info = data['talks'][0] + + title = talk_info['title'].strip() + + native_downloads = try_get( + talk_info, + (lambda x: x['downloads']['nativeDownloads'], + lambda x: x['nativeDownloads']), + dict) or {} formats = [{ 'url': format_url, 'format_id': format_id, 'format': format_id, - } for (format_id, format_url) in talk_info['nativeDownloads'].items() if format_url is not None] + } for (format_id, format_url) in native_downloads.items() if format_url is not None] if formats: for f in formats: finfo = self._NATIVE_FORMATS.get(f['format_id']) if finfo: f.update(finfo) - for format_id, resources in talk_info['resources'].items(): + player_talk = talk_info['player_talks'][0] + + external = player_talk.get('external') + if isinstance(external, dict): + service = external.get('service') + if isinstance(service, compat_str): + ext_url = None + if service.lower() == 'youtube': + ext_url = external.get('code') + + return self.url_result(ext_url or external['uri']) + + resources_ = player_talk.get('resources') or talk_info.get('resources') + + http_url = None + for format_id, resources in resources_.items(): if format_id == 'h264': for resource in resources: + h264_url = resource.get('file') + if not h264_url: + continue bitrate = int_or_none(resource.get('bitrate')) formats.append({ - 'url': resource['file'], + 'url': h264_url, 'format_id': '%s-%sk' % (format_id, bitrate), 'tbr': bitrate, }) + if re.search(r'\d+k', h264_url): + http_url = h264_url elif format_id == 'rtmp': streamer = talk_info.get('streamer') if not streamer: @@ -195,16 +238,34 @@ class TEDIE(InfoExtractor): 'tbr': int_or_none(resource.get('bitrate')), }) elif format_id == 'hls': - hls_formats = self._extract_m3u8_formats( - resources.get('stream'), video_name, 'mp4', m3u8_id=format_id) - for f in hls_formats: - if f.get('format_id') == 'hls-meta': - continue - if not f.get('height'): - f['vcodec'] = 'none' - else: - f['acodec'] = 'none' - formats.extend(hls_formats) + if not isinstance(resources, dict): + continue + stream_url = url_or_none(resources.get('stream')) + if not stream_url: + continue + formats.extend(self._extract_m3u8_formats( + stream_url, video_name, 'mp4', m3u8_id=format_id, + fatal=False)) + + m3u8_formats = list(filter( + lambda f: f.get('protocol') == 'm3u8' and f.get('vcodec') != 'none', + formats)) + if http_url: + for m3u8_format in m3u8_formats: + bitrate = self._search_regex(r'(\d+k)', m3u8_format['url'], 'bitrate', default=None) + if not bitrate: + continue + bitrate_url = re.sub(r'\d+k', bitrate, http_url) + if not self._is_valid_url( + bitrate_url, video_name, '%s bitrate' % bitrate): + continue + f = m3u8_format.copy() + f.update({ + 'url': bitrate_url, + 'format_id': m3u8_format['format_id'].replace('hls', 'http'), + 'protocol': 'http', + }) + formats.append(f) audio_download = talk_info.get('audioDownload') if audio_download: @@ -212,49 +273,55 @@ class TEDIE(InfoExtractor): 'url': audio_download, 'format_id': 'audio', 'vcodec': 'none', - 'preference': -0.5, }) self._sort_formats(formats) video_id = compat_str(talk_info['id']) - thumbnail = talk_info['thumb'] - if not thumbnail.startswith('http'): - thumbnail = 'http://' + thumbnail return { 'id': video_id, - 'title': talk_info['title'].strip(), - 'uploader': talk_info['speaker'], - 'thumbnail': thumbnail, + 'title': title, + 'uploader': player_talk.get('speaker') or talk_info.get('speaker'), + 'thumbnail': player_talk.get('thumb') or talk_info.get('thumb'), 'description': self._og_search_description(webpage), 'subtitles': self._get_subtitles(video_id, talk_info), 'formats': formats, - 'duration': talk_info.get('duration'), + 'duration': float_or_none(talk_info.get('duration')), + 'view_count': int_or_none(data.get('viewed_count')), + 'comment_count': int_or_none( + try_get(data, lambda x: x['comments']['count'])), + 'tags': try_get(talk_info, lambda x: x['tags'], list), } def _get_subtitles(self, video_id, talk_info): - languages = [lang['languageCode'] for lang in talk_info.get('languages', [])] - if languages: - sub_lang_list = {} - for l in languages: - sub_lang_list[l] = [ - { - 'url': 'http://www.ted.com/talks/subtitles/id/%s/lang/%s/format/%s' % (video_id, l, ext), - 'ext': ext, - } - for ext in ['ted', 'srt'] - ] - return sub_lang_list - else: - return {} + sub_lang_list = {} + for language in try_get( + talk_info, + (lambda x: x['downloads']['languages'], + lambda x: x['languages']), list): + lang_code = language.get('languageCode') or language.get('ianaCode') + if not lang_code: + continue + sub_lang_list[lang_code] = [ + { + 'url': 'http://www.ted.com/talks/subtitles/id/%s/lang/%s/format/%s' % (video_id, lang_code, ext), + 'ext': ext, + } + for ext in ['ted', 'srt'] + ] + return sub_lang_list def _watch_info(self, url, name): webpage = self._download_webpage(url, name) config_json = self._html_search_regex( r'"pages\.jwplayer"\s*,\s*({.+?})\s*\)\s*', - webpage, 'config') + webpage, 'config', default=None) + if not config_json: + embed_url = self._search_regex( + r"]+class='pages-video-embed__video__object'[^>]+src='([^']+)'", webpage, 'embed url') + return self.url_result(self._proto_relative_url(embed_url)) config = json.loads(config_json)['config'] video_url = config['video']['url'] thumbnail = config.get('image', {}).get('url')