[pbs] Extract all formats (Closes #8538)

This commit is contained in:
Sergey M․ 2016-02-22 01:23:27 +06:00
parent f5bdb44443
commit 5d6e1011df

View file

@ -337,6 +337,21 @@ class PBSIE(InfoExtractor):
'skip_download': True, # requires ffmpeg
},
},
{
# Serves hd only via wigget/partnerplayer page
'url': 'http://www.pbs.org/video/2365641075/',
'info_dict': {
'id': '2365641075',
'ext': 'mp4',
'title': 'FRONTLINE - Netanyahu at War',
'duration': 6852,
'thumbnail': 're:^https?://.*\.jpg$',
'formats': 'mincount:8',
},
'params': {
'skip_download': True, # requires ffmpeg
},
},
{
'url': 'http://player.pbs.org/widget/partnerplayer/2365297708/?start=0&end=0&chapterbar=false&endscreen=false&topbar=true',
'only_matching': True,
@ -437,34 +452,54 @@ class PBSIE(InfoExtractor):
for vid_id in video_id]
return self.playlist_result(entries, display_id)
try:
info = self._download_json(
'http://player.pbs.org/videoInfo/%s?format=json&type=partner' % video_id,
display_id, 'Downloading video info JSON')
except ExtractorError as e:
if not isinstance(e.cause, compat_HTTPError) or e.cause.code != 404:
raise
# videoInfo API may not work for some videos, fallback to portalplayer API
player = self._download_webpage(
'http://player.pbs.org/portalplayer/%s' % video_id, display_id)
info = self._parse_json(
self._search_regex(
r'(?s)PBS\.videoData\s*=\s*({.+?});\n',
player, 'video data', default='{}'),
display_id, transform_source=js_to_json, fatal=False)
info = None
redirects = []
redirect_urls = set()
formats = []
def extract_redirect_urls(info):
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
if redirect_url and redirect_url not in redirect_urls:
redirects.append(redirect)
redirect_urls.add(redirect_url)
try:
video_info = self._download_json(
'http://player.pbs.org/videoInfo/%s?format=json&type=partner' % video_id,
display_id, 'Downloading video info JSON')
extract_redirect_urls(video_info)
info = video_info
except ExtractorError as e:
# videoInfo API may not work for some videos
if not isinstance(e.cause, compat_HTTPError) or e.cause.code != 404:
raise
# Player pages may also serve different qualities
for page in ('widget/partnerplayer', 'portalplayer'):
player = self._download_webpage(
'http://player.pbs.org/%s/%s' % (page, video_id),
display_id, 'Downloading %s page' % page, fatal=False)
if player:
video_info = self._parse_json(
self._search_regex(
r'(?s)PBS\.videoData\s*=\s*({.+?});\n',
player, '%s video data' % page, default='{}'),
display_id, transform_source=js_to_json, fatal=False)
if video_info:
extract_redirect_urls(video_info)
if not info:
info = video_info
formats = []
for num, redirect in enumerate(redirects):
redirect_id = redirect.get('eeid')
redirect_info = self._download_json(
redirect_url + '?format=json', display_id,
'Downloading %s video url info' % encoding_name)
'%s?format=json' % redirect['url'], display_id,
'Downloading %s video url info' % (redirect_id or num))
if redirect_info['status'] == 'error':
raise ExtractorError(
@ -483,8 +518,9 @@ class PBSIE(InfoExtractor):
else:
formats.append({
'url': format_url,
'format_id': redirect.get('eeid'),
'format_id': redirect_id,
})
self._remove_duplicate_formats(formats)
self._sort_formats(formats)
rating_str = info.get('rating')