[rai] Support videos with embedded content item ID (#8551)

This commit is contained in:
Yen Chi Hsuan 2016-07-03 19:44:58 +08:00
parent 7507fc98cb
commit 2b28b892d8
No known key found for this signature in database
GPG Key ID: 3FDDD575826C5C30

View File

@ -60,6 +60,57 @@ class RaiBaseIE(InfoExtractor):
return formats
def _extract_from_content_id(self, content_id, base_url):
media = self._download_json(
'http://www.rai.tv/dl/RaiTV/programmi/media/ContentItem-%s.html?json' % content_id,
content_id, 'Downloading video JSON')
thumbnails = []
for image_type in ('image', 'image_medium', 'image_300'):
thumbnail_url = media.get(image_type)
if thumbnail_url:
thumbnails.append({
'url': compat_urlparse.urljoin(base_url, thumbnail_url),
})
formats = []
media_type = media['type']
if 'Audio' in media_type:
formats.append({
'format_id': media.get('formatoAudio'),
'url': media['audioUrl'],
'ext': media.get('formatoAudio'),
})
elif 'Video' in media_type:
formats.extend(self._extract_relinker_formats(media['mediaUri'], content_id))
self._sort_formats(formats)
else:
raise ExtractorError('not a media file')
subtitles = {}
captions = media.get('subtitlesUrl')
if captions:
STL_EXT = '.stl'
SRT_EXT = '.srt'
if captions.endswith(STL_EXT):
captions = captions[:-len(STL_EXT)] + SRT_EXT
subtitles['it'] = [{
'ext': 'srt',
'url': captions,
}]
return {
'id': content_id,
'title': media['name'],
'description': media.get('desc'),
'thumbnails': thumbnails,
'uploader': media.get('author'),
'upload_date': unified_strdate(media.get('date')),
'duration': parse_duration(media.get('length')),
'formats': formats,
'subtitles': subtitles,
}
class RaiTVIE(RaiBaseIE):
_VALID_URL = r'https?://(?:.+?\.)?(?:rai\.it|rai\.tv|rainews\.it)/dl/(?:[^/]+/)+(?:media|ondemand)/.+?-(?P<id>[\da-f]{8}-[\da-f]{4}-[\da-f]{4}-[\da-f]{4}-[\da-f]{12})(?:-.+?)?\.html'
@ -131,55 +182,7 @@ class RaiTVIE(RaiBaseIE):
def _real_extract(self, url):
video_id = self._match_id(url)
media = self._download_json(
'http://www.rai.tv/dl/RaiTV/programmi/media/ContentItem-%s.html?json' % video_id,
video_id, 'Downloading video JSON')
thumbnails = []
for image_type in ('image', 'image_medium', 'image_300'):
thumbnail_url = media.get(image_type)
if thumbnail_url:
thumbnails.append({
'url': compat_urlparse.urljoin(url, thumbnail_url),
})
formats = []
media_type = media['type']
if 'Audio' in media_type:
formats.append({
'format_id': media.get('formatoAudio'),
'url': media['audioUrl'],
'ext': media.get('formatoAudio'),
})
elif 'Video' in media_type:
formats.extend(self._extract_relinker_formats(media['mediaUri'], video_id))
self._sort_formats(formats)
else:
raise ExtractorError('not a media file')
subtitles = {}
captions = media.get('subtitlesUrl')
if captions:
STL_EXT = '.stl'
SRT_EXT = '.srt'
if captions.endswith(STL_EXT):
captions = captions[:-len(STL_EXT)] + SRT_EXT
subtitles['it'] = [{
'ext': 'srt',
'url': captions,
}]
return {
'id': video_id,
'title': media['name'],
'description': media.get('desc'),
'thumbnails': thumbnails,
'uploader': media.get('author'),
'upload_date': unified_strdate(media.get('date')),
'duration': parse_duration(media.get('length')),
'formats': formats,
'subtitles': subtitles,
}
return self._extract_from_content_id(video_id, url)
class RaiIE(RaiBaseIE):
@ -197,6 +200,7 @@ class RaiIE(RaiBaseIE):
},
},
{
# Direct relinker URL
'url': 'http://www.rai.tv/dl/RaiTV/dirette/PublishingBlock-1912dbbf-3f96-44c3-b4cf-523681fbacbc.html?channel=EuroNews',
# HDS live stream, MD5 is unstable
'info_dict': {
@ -205,7 +209,18 @@ class RaiIE(RaiBaseIE):
'title': 'EuroNews',
},
'skip': 'Geo-restricted to Italy',
}
},
{
# Embedded content item ID
'url': 'http://www.tg1.rai.it/dl/tg1/2010/edizioni/ContentSet-9b6e0cba-4bef-4aef-8cf0-9f7f665b7dfb-tg1.html?item=undefined',
'md5': '84c1135ce960e8822ae63cec34441d63',
'info_dict': {
'id': '0960e765-62c8-474a-ac4b-7eb3e2be39c8',
'ext': 'mp4',
'title': 'TG1 ore 20:00 del 02/07/2016',
'upload_date': '20160702',
},
},
]
@classmethod
@ -225,6 +240,12 @@ class RaiIE(RaiBaseIE):
iframe_url = compat_urlparse.urljoin(url, iframe_url)
return self.url_result(iframe_url)
content_item_id = self._search_regex(
r'initEdizione\((?P<q1>[\'"])ContentItem-(?P<content_id>[^\'"]+)(?P=q1)',
webpage, 'content item ID', group='content_id', default=None)
if content_item_id:
return self._extract_from_content_id(content_item_id, url)
relinker_url = compat_urlparse.urljoin(url, self._search_regex(
r'var\s+videoURL\s*=\s*(?P<q1>[\'"])(?P<url>(https?:)?//mediapolis\.rai\.it/relinker/relinkerServlet\.htm\?cont=\d+)(?P=q1)',
webpage, 'relinker URL', group='url'))