[cbsnews] extract all formats

This commit is contained in:
remitamine 2016-01-27 19:10:34 +01:00
parent a9d5f12fec
commit f125d9115b
1 changed files with 15 additions and 31 deletions

View File

@ -4,11 +4,11 @@ from __future__ import unicode_literals
import re import re
import json import json
from .common import InfoExtractor from .theplatform import ThePlatformIE
from ..utils import remove_start from ..utils import remove_start
class CBSNewsIE(InfoExtractor): class CBSNewsIE(ThePlatformIE):
IE_DESC = 'CBS News' IE_DESC = 'CBS News'
_VALID_URL = r'http://(?:www\.)?cbsnews\.com/(?:[^/]+/)+(?P<id>[\da-z_-]+)' _VALID_URL = r'http://(?:www\.)?cbsnews\.com/(?:[^/]+/)+(?P<id>[\da-z_-]+)'
@ -31,7 +31,7 @@ class CBSNewsIE(InfoExtractor):
'url': 'http://www.cbsnews.com/videos/fort-hood-shooting-army-downplays-mental-illness-as-cause-of-attack/', 'url': 'http://www.cbsnews.com/videos/fort-hood-shooting-army-downplays-mental-illness-as-cause-of-attack/',
'info_dict': { 'info_dict': {
'id': 'fort-hood-shooting-army-downplays-mental-illness-as-cause-of-attack', 'id': 'fort-hood-shooting-army-downplays-mental-illness-as-cause-of-attack',
'ext': 'flv', 'ext': 'mp4',
'title': 'Fort Hood shooting: Army downplays mental illness as cause of attack', 'title': 'Fort Hood shooting: Army downplays mental illness as cause of attack',
'thumbnail': 're:^https?://.*\.jpg$', 'thumbnail': 're:^https?://.*\.jpg$',
'duration': 205, 'duration': 205,
@ -42,7 +42,7 @@ class CBSNewsIE(InfoExtractor):
}, },
}, },
'params': { 'params': {
# rtmp download # m3u8 download
'skip_download': True, 'skip_download': True,
}, },
}, },
@ -63,33 +63,6 @@ class CBSNewsIE(InfoExtractor):
duration = item.get('duration') duration = item.get('duration')
thumbnail = item.get('mediaImage') or item.get('thumbnail') thumbnail = item.get('mediaImage') or item.get('thumbnail')
formats = []
for format_id in ['RtmpMobileLow', 'RtmpMobileHigh', 'Hls', 'RtmpDesktop']:
uri = item.get('media' + format_id + 'URI')
if not uri:
continue
uri = remove_start(uri, '{manifest:none}')
fmt = {
'url': uri,
'format_id': format_id,
}
if uri.startswith('rtmp'):
play_path = re.sub(
r'{slistFilePath}', '',
uri.split('<break>')[-1].split('{break}')[-1])
play_path = re.sub(
r'{manifest:.+}.*$', '', play_path)
fmt.update({
'app': 'ondemand?auth=cbs',
'play_path': 'mp4:' + play_path,
'player_url': 'http://www.cbsnews.com/[[IMPORT]]/vidtech.cbsinteractive.com/player/3_3_0/CBSI_PLAYER_HD.swf',
'page_url': 'http://www.cbsnews.com',
'ext': 'flv',
})
elif uri.endswith('.m3u8'):
fmt['ext'] = 'mp4'
formats.append(fmt)
subtitles = {} subtitles = {}
if 'mpxRefId' in video_info: if 'mpxRefId' in video_info:
subtitles['en'] = [{ subtitles['en'] = [{
@ -97,6 +70,17 @@ class CBSNewsIE(InfoExtractor):
'url': 'http://www.cbsnews.com/videos/captions/%s.adb_xml' % video_info['mpxRefId'], 'url': 'http://www.cbsnews.com/videos/captions/%s.adb_xml' % video_info['mpxRefId'],
}] }]
formats = []
for format_id in ['RtmpMobileLow', 'RtmpMobileHigh', 'Hls', 'RtmpDesktop']:
pid = item.get('media' + format_id)
if not pid:
continue
release_url = 'http://link.theplatform.com/s/dJ5BDC/%s?format=SMIL&mbr=true' % pid
tp_formats, tp_subtitles = self._extract_theplatform_smil(release_url, video_id, 'Downloading %s SMIL data' % pid)
formats.extend(tp_formats)
subtitles = self._merge_subtitles(subtitles, tp_subtitles)
self._sort_formats(formats)
return { return {
'id': video_id, 'id': video_id,
'title': title, 'title': title,