[cspan] correct the clip info extraction

This commit is contained in:
remitamine 2015-10-03 19:28:48 +01:00
parent 85557f635a
commit 30787f7259
1 changed files with 27 additions and 31 deletions

View File

@ -18,22 +18,21 @@ class CSpanIE(InfoExtractor):
IE_DESC = 'C-SPAN'
_TESTS = [{
'url': 'http://www.c-span.org/video/?313572-1/HolderonV',
'md5': '8e44ce11f0f725527daccc453f553eb0',
'md5': '067803f994e049b455a58b16e5aab442',
'info_dict': {
'id': '315139',
'ext': 'mp4',
'title': 'Attorney General Eric Holder on Voting Rights Act Decision',
'description': 'Attorney General Eric Holder spoke to reporters following the Supreme Court decision in Shelby County v. Holder in which the court ruled that the preclearance provisions of the Voting Rights Act could not be enforced until Congress established new guidelines for review.',
'description': 'Attorney General Eric Holder speaks to reporters following the Supreme Court decision in [Shelby County v. Holder], in which the court ruled that the preclearance provisions of the Voting Rights Act could not be enforced.',
},
'skip': 'Regularly fails on travis, for unknown reasons',
}, {
'url': 'http://www.c-span.org/video/?c4486943/cspan-international-health-care-models',
# For whatever reason, the served video alternates between
# two different ones
'md5': '4eafd1e91a75d2b1e6a3cbd0995816a2',
'info_dict': {
'id': '340723',
'id': 'c4486943',
'ext': 'mp4',
'title': 'International Health Care Models',
'title': 'CSPAN - International Health Care Models',
'description': 'md5:7a985a2d595dba00af3d9c9f0783c967',
}
}, {
@ -44,7 +43,7 @@ class CSpanIE(InfoExtractor):
'ext': 'mp4',
'title': 'General Motors Ignition Switch Recall',
'duration': 14848,
'description': 'md5:70c7c3b8fa63fa60d42772440596034c'
'description': 'md5:118081aedd24bf1d3b68b3803344e7f3'
},
}, {
# Video from senate.gov
@ -57,36 +56,33 @@ class CSpanIE(InfoExtractor):
}]
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
page_id = mobj.group('id')
webpage = self._download_webpage(url, page_id)
video_id = self._search_regex(r'progid=\'?([0-9]+)\'?>', webpage, 'video id')
video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id)
matches = re.search(r'data-(prog|clip)id=\'([0-9]+)\'', webpage)
if matches:
video_type, video_id = matches.groups()
if video_type == 'prog':
video_type = 'program'
else:
senate_isvp_url = SenateISVPIE._search_iframe_url(webpage)
if senate_isvp_url:
title = self._og_search_title(webpage)
surl = smuggle_url(senate_isvp_url, {'force_title': title})
return self.url_result(surl, 'SenateISVP', video_id, title)
description = self._html_search_regex(
[
# The full description
r'<div class=\'expandable\'>(.*?)<a href=\'#\'',
# If the description is small enough the other div is not
# present, otherwise this is a stripped version
r'<p class=\'initial\'>(.*?)</p>'
],
webpage, 'description', flags=re.DOTALL, default=None)
info_url = 'http://c-spanvideo.org/videoLibrary/assets/player/ajax-player.php?os=android&html5=program&id=' + video_id
data = self._download_json(info_url, video_id)
data = self._download_json(
'http://c-spanvideo.org/videoLibrary/assets/player/ajax-player.php?os=android&html5=%s&id=%s' % (video_type, video_id),
video_id)
doc = self._download_xml(
'http://www.c-span.org/common/services/flashXml.php?programid=' + video_id,
'http://www.c-span.org/common/services/flashXml.php?%sid=%s' % (video_type, video_id),
video_id)
description = self._html_search_meta('description', webpage)
title = find_xpath_attr(doc, './/string', 'name', 'title').text
thumbnail = find_xpath_attr(doc, './/string', 'name', 'poster').text
senate_isvp_url = SenateISVPIE._search_iframe_url(webpage)
if senate_isvp_url:
surl = smuggle_url(senate_isvp_url, {'force_title': title})
return self.url_result(surl, 'SenateISVP', video_id, title)
files = data['video']['files']
try:
capfile = data['video']['capfile']['#text']
@ -112,12 +108,12 @@ class CSpanIE(InfoExtractor):
if len(entries) == 1:
entry = dict(entries[0])
entry['id'] = video_id
entry['id'] = 'c' + video_id if video_type == 'clip' else video_id
return entry
else:
return {
'_type': 'playlist',
'entries': entries,
'title': title,
'id': video_id,
'id': 'c' + video_id if video_type == 'clip' else video_id,
}