+# coding: utf-8
+from __future__ import unicode_literals
+
import re
-import json
from .common import InfoExtractor
+from ..utils import (
+ ExtractorError,
+ determine_ext,
+ int_or_none,
+ strip_jsonp,
+ unified_strdate,
+ US_RATINGS,
+)
class PBSIE(InfoExtractor):
- _VALID_URL = r'https?://video.pbs.org/video/(?P<id>\d+)/?'
+ _VALID_URL = r'''(?x)https?://
+ (?:
+ # Direct video URL
+ video\.pbs\.org/(?:viralplayer|video)/(?P<id>[0-9]+)/? |
+ # Article with embedded player (or direct video)
+ (?:www\.)?pbs\.org/(?:[^/]+/){2,5}(?P<presumptive_id>[^/]+?)(?:\.html)?/?(?:$|[?\#]) |
+ # Player
+ video\.pbs\.org/(?:widget/)?partnerplayer/(?P<player_id>[^/]+)/
+ )
+ '''
- _TEST = {
- u'url': u'http://video.pbs.org/video/2365006249/',
- u'file': u'2365006249.mp4',
- u'md5': 'ce1888486f0908d555a8093cac9a7362',
- u'info_dict': {
- u'title': u'A More Perfect Union',
- u'description': u'md5:ba0c207295339c8d6eced00b7c363c6a',
- u'duration': 3190,
+ _TESTS = [
+ {
+ 'url': 'http://www.pbs.org/tpt/constitution-usa-peter-sagal/watch/a-more-perfect-union/',
+ 'md5': 'ce1888486f0908d555a8093cac9a7362',
+ 'info_dict': {
+ 'id': '2365006249',
+ 'ext': 'mp4',
+ 'title': 'Constitution USA with Peter Sagal - A More Perfect Union',
+ 'description': 'md5:ba0c207295339c8d6eced00b7c363c6a',
+ 'duration': 3190,
+ },
+ 'params': {
+ 'skip_download': True, # requires ffmpeg
+ },
+ },
+ {
+ 'url': 'http://www.pbs.org/wgbh/pages/frontline/losing-iraq/',
+ 'md5': '143c98aa54a346738a3d78f54c925321',
+ 'info_dict': {
+ 'id': '2365297690',
+ 'ext': 'mp4',
+ 'title': 'FRONTLINE - Losing Iraq',
+ 'description': 'md5:f5bfbefadf421e8bb8647602011caf8e',
+ 'duration': 5050,
+ },
+ 'params': {
+ 'skip_download': True, # requires ffmpeg
+ }
+ },
+ {
+ 'url': 'http://www.pbs.org/newshour/bb/education-jan-june12-cyberschools_02-23/',
+ 'md5': 'b19856d7f5351b17a5ab1dc6a64be633',
+ 'info_dict': {
+ 'id': '2201174722',
+ 'ext': 'mp4',
+ 'title': 'PBS NewsHour - Cyber Schools Gain Popularity, but Quality Questions Persist',
+ 'description': 'md5:5871c15cba347c1b3d28ac47a73c7c28',
+ 'duration': 801,
+ },
+ },
+ {
+ 'url': 'http://www.pbs.org/wnet/gperf/dudamel-conducts-verdi-requiem-hollywood-bowl-full-episode/3374/',
+ 'md5': 'c62859342be2a0358d6c9eb306595978',
+ 'info_dict': {
+ 'id': '2365297708',
+ 'ext': 'mp4',
+ 'description': 'md5:68d87ef760660eb564455eb30ca464fe',
+ 'title': 'Great Performances - Dudamel Conducts Verdi Requiem at the Hollywood Bowl - Full',
+ 'duration': 6559,
+ 'thumbnail': 're:^https?://.*\.jpg$',
+ },
+ 'params': {
+ 'skip_download': True, # requires ffmpeg
+ },
+ },
+ {
+ 'url': 'http://www.pbs.org/wgbh/nova/earth/killer-typhoon.html',
+ 'md5': '908f3e5473a693b266b84e25e1cf9703',
+ 'info_dict': {
+ 'id': '2365160389',
+ 'display_id': 'killer-typhoon',
+ 'ext': 'mp4',
+ 'description': 'md5:c741d14e979fc53228c575894094f157',
+ 'title': 'NOVA - Killer Typhoon',
+ 'duration': 3172,
+ 'thumbnail': 're:^https?://.*\.jpg$',
+ 'upload_date': '20140122',
+ 'age_limit': 10,
+ },
+ 'params': {
+ 'skip_download': True, # requires ffmpeg
+ },
+ },
+ {
+ 'url': 'http://www.pbs.org/wgbh/pages/frontline/united-states-of-secrets/',
+ 'info_dict': {
+ 'id': 'united-states-of-secrets',
+ },
+ 'playlist_count': 2,
},
+ {
+ 'url': 'http://www.pbs.org/wgbh/americanexperience/films/death/player/',
+ 'info_dict': {
+ 'id': '2276541483',
+ 'display_id': 'player',
+ 'ext': 'mp4',
+ 'title': 'American Experience - Death and the Civil War, Chapter 1',
+ 'description': 'American Experience, TV’s most-watched history series, brings to life the compelling stories from our past that inform our understanding of the world today.',
+ 'duration': 682,
+ 'thumbnail': 're:^https?://.*\.jpg$',
+ },
+ 'params': {
+ 'skip_download': True, # requires ffmpeg
+ },
+ },
+ {
+ 'url': 'http://video.pbs.org/video/2365367186/',
+ 'info_dict': {
+ 'id': '2365367186',
+ 'display_id': '2365367186',
+ 'ext': 'mp4',
+ 'title': 'To Catch A Comet - Full Episode',
+ 'description': 'On November 12, 2014, billions of kilometers from Earth, spacecraft orbiter Rosetta and lander Philae did what no other had dared to attempt \u2014 land on the volatile surface of a comet as it zooms around the sun at 67,000 km/hr. The European Space Agency hopes this mission can help peer into our past and unlock secrets of our origins.',
+ 'duration': 3342,
+ 'thumbnail': 're:^https?://.*\.jpg$',
+ },
+ 'params': {
+ 'skip_download': True, # requires ffmpeg
+ },
+ 'skip': 'Expired',
+ },
+ {
+ # Video embedded in iframe containing angle brackets as attribute's value (e.g.
+ # "<iframe style='position: absolute;<br />\ntop: 0; left: 0;' ...", see
+ # https://github.com/rg3/youtube-dl/issues/7059)
+ 'url': 'http://www.pbs.org/food/features/a-chefs-life-season-3-episode-5-prickly-business/',
+ 'info_dict': {
+ 'id': '2365546844',
+ 'display_id': 'a-chefs-life-season-3-episode-5-prickly-business',
+ 'ext': 'mp4',
+ 'title': "A Chef's Life - Season 3, Ep. 5: Prickly Business",
+ 'description': 'md5:61db2ddf27c9912f09c241014b118ed1',
+ 'duration': 1480,
+ 'thumbnail': 're:^https?://.*\.jpg$',
+ },
+ 'params': {
+ 'skip_download': True, # requires ffmpeg
+ },
+ },
+ {
+ # Frontline video embedded via flp2012.js
+ 'url': 'http://www.pbs.org/wgbh/pages/frontline/the-atomic-artists',
+ 'info_dict': {
+ 'id': '2070868960',
+ 'display_id': 'the-atomic-artists',
+ 'ext': 'mp4',
+ 'title': 'FRONTLINE - The Atomic Artists',
+ 'description': 'md5:f5bfbefadf421e8bb8647602011caf8e',
+ 'duration': 723,
+ 'thumbnail': 're:^https?://.*\.jpg$',
+ },
+ 'params': {
+ 'skip_download': True, # requires ffmpeg
+ },
+ }
+ ]
+ _ERRORS = {
+ 101: 'We\'re sorry, but this video is not yet available.',
+ 403: 'We\'re sorry, but this video is not available in your region due to right restrictions.',
+ 404: 'We are experiencing technical difficulties that are preventing us from playing the video at this time. Please check back again soon.',
+ 410: 'This video has expired and is no longer available for online streaming.',
}
- def _real_extract(self, url):
+ def _extract_webpage(self, url):
mobj = re.match(self._VALID_URL, url)
- video_id = mobj.group('id')
- info_url = 'http://video.pbs.org/videoInfo/%s?format=json' % video_id
- info_page = self._download_webpage(info_url, video_id)
- info =json.loads(info_page)
- return {'id': video_id,
- 'title': info['title'],
- 'url': info['alternate_encoding']['url'],
- 'ext': 'mp4',
- 'description': info['program'].get('description'),
- 'thumbnail': info.get('image_url'),
- 'duration': info.get('duration'),
- }
+
+ presumptive_id = mobj.group('presumptive_id')
+ display_id = presumptive_id
+ if presumptive_id:
+ webpage = self._download_webpage(url, display_id)
+
+ upload_date = unified_strdate(self._search_regex(
+ r'<input type="hidden" id="air_date_[0-9]+" value="([^"]+)"',
+ webpage, 'upload date', default=None))
+
+ # tabbed frontline videos
+ tabbed_videos = re.findall(
+ r'<div[^>]+class="videotab[^"]*"[^>]+vid="(\d+)"', webpage)
+ if tabbed_videos:
+ return tabbed_videos, presumptive_id, upload_date
+
+ MEDIA_ID_REGEXES = [
+ r"div\s*:\s*'videoembed'\s*,\s*mediaid\s*:\s*'(\d+)'", # frontline video embed
+ r'class="coveplayerid">([^<]+)<', # coveplayer
+ r'<input type="hidden" id="pbs_video_id_[0-9]+" value="([0-9]+)"/>', # jwplayer
+ ]
+
+ media_id = self._search_regex(
+ MEDIA_ID_REGEXES, webpage, 'media ID', fatal=False, default=None)
+ if media_id:
+ return media_id, presumptive_id, upload_date
+
+ # Fronline video embedded via flp
+ video_id = self._search_regex(
+ r'videoid\s*:\s*"([\d+a-z]{7,})"', webpage, 'videoid', default=None)
+ if video_id:
+ # pkg_id calculation is reverse engineered from
+ # http://www.pbs.org/wgbh/pages/frontline/js/flp2012.js
+ prg_id = self._search_regex(
+ r'videoid\s*:\s*"([\d+a-z]{7,})"', webpage, 'videoid')[7:]
+ if 'q' in prg_id:
+ prg_id = prg_id.split('q')[1]
+ prg_id = int(prg_id, 16)
+ getdir = self._download_json(
+ 'http://www.pbs.org/wgbh/pages/frontline/.json/getdir/getdir%d.json' % prg_id,
+ presumptive_id, 'Downloading getdir JSON',
+ transform_source=strip_jsonp)
+ return getdir['mid'], presumptive_id, upload_date
+
+ for iframe in re.findall(r'(?s)<iframe(.+?)></iframe>', webpage):
+ url = self._search_regex(
+ r'src=(["\'])(?P<url>.+?partnerplayer.+?)\1', iframe,
+ 'player URL', default=None, group='url')
+ if url:
+ break
+
+ mobj = re.match(self._VALID_URL, url)
+
+ player_id = mobj.group('player_id')
+ if not display_id:
+ display_id = player_id
+ if player_id:
+ player_page = self._download_webpage(
+ url, display_id, note='Downloading player page',
+ errnote='Could not download player page')
+ video_id = self._search_regex(
+ r'<div\s+id="video_([0-9]+)"', player_page, 'video ID')
+ else:
+ video_id = mobj.group('id')
+ display_id = video_id
+
+ return video_id, display_id, None
+
+ def _real_extract(self, url):
+ video_id, display_id, upload_date = self._extract_webpage(url)
+
+ if isinstance(video_id, list):
+ entries = [self.url_result(
+ 'http://video.pbs.org/video/%s' % vid_id, 'PBS', vid_id)
+ for vid_id in video_id]
+ return self.playlist_result(entries, display_id)
+
+ info = self._download_json(
+ 'http://video.pbs.org/videoInfo/%s?format=json&type=partner' % video_id,
+ display_id)
+
+ formats = []
+ for encoding_name in ('recommended_encoding', 'alternate_encoding'):
+ redirect = info.get(encoding_name)
+ if not redirect:
+ continue
+ redirect_url = redirect.get('url')
+ if not redirect_url:
+ continue
+
+ redirect_info = self._download_json(
+ redirect_url + '?format=json', display_id,
+ 'Downloading %s video url info' % encoding_name)
+
+ if redirect_info['status'] == 'error':
+ raise ExtractorError(
+ '%s said: %s' % (
+ self.IE_NAME,
+ self._ERRORS.get(redirect_info['http_code'], redirect_info['message'])),
+ expected=True)
+
+ format_url = redirect_info.get('url')
+ if not format_url:
+ continue
+
+ if determine_ext(format_url) == 'm3u8':
+ formats.extend(self._extract_m3u8_formats(
+ format_url, display_id, 'mp4', preference=1, m3u8_id='hls'))
+ else:
+ formats.append({
+ 'url': format_url,
+ 'format_id': redirect.get('eeid'),
+ })
+ self._sort_formats(formats)
+
+ rating_str = info.get('rating')
+ if rating_str is not None:
+ rating_str = rating_str.rpartition('-')[2]
+ age_limit = US_RATINGS.get(rating_str)
+
+ subtitles = {}
+ closed_captions_url = info.get('closed_captions_url')
+ if closed_captions_url:
+ subtitles['en'] = [{
+ 'ext': 'ttml',
+ 'url': closed_captions_url,
+ }]
+
+ # info['title'] is often incomplete (e.g. 'Full Episode', 'Episode 5', etc)
+ # Try turning it to 'program - title' naming scheme if possible
+ alt_title = info.get('program', {}).get('title')
+ if alt_title:
+ info['title'] = alt_title + ' - ' + re.sub(r'^' + alt_title + '[\s\-:]+', '', info['title'])
+
+ return {
+ 'id': video_id,
+ 'display_id': display_id,
+ 'title': info['title'],
+ 'description': info['program'].get('description'),
+ 'thumbnail': info.get('image_url'),
+ 'duration': int_or_none(info.get('duration')),
+ 'age_limit': age_limit,
+ 'upload_date': upload_date,
+ 'formats': formats,
+ 'subtitles': subtitles,
+ }