[nerdist] Add new extractor (Fixes #4851)

This commit is contained in:
Philipp Hagemeister 2015-02-02 23:38:35 +01:00
parent 531572578e
commit 995029a142
3 changed files with 86 additions and 2 deletions

View File

@ -285,6 +285,7 @@ from .ndr import NDRIE
from .ndtv import NDTVIE
from .netzkino import NetzkinoIE
from .nerdcubed import NerdCubedFeedIE
from .nerdist import NerdistIE
from .newgrounds import NewgroundsIE
from .newstube import NewstubeIE
from .nextmedia import (

View File

@ -860,10 +860,13 @@ class InfoExtractor(object):
return formats
# TODO: improve extraction
def _extract_smil_formats(self, smil_url, video_id):
def _extract_smil_formats(self, smil_url, video_id, fatal=True):
smil = self._download_xml(
smil_url, video_id, 'Downloading SMIL file',
'Unable to download SMIL file')
'Unable to download SMIL file', fatal=fatal)
if smil is False:
assert not fatal
return []
base = smil.find('./head/meta').get('base')

View File

@ -0,0 +1,80 @@
# encoding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
determine_ext,
parse_iso8601,
xpath_text,
)
class NerdistIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?nerdist\.com/vepisode/(?P<id>[^/?#]+)'
_TEST = {
'url': 'http://www.nerdist.com/vepisode/exclusive-which-dc-characters-w',
'md5': '3698ed582931b90d9e81e02e26e89f23',
'info_dict': {
'display_id': 'exclusive-which-dc-characters-w',
'id': 'RPHpvJyr',
'ext': 'mp4',
'title': 'Your TEEN TITANS Revealed! Who\'s on the show?',
'thumbnail': 're:^https?://.*/thumbs/.*\.jpg$',
'description': 'Exclusive: Find out which DC Comics superheroes will star in TEEN TITANS Live-Action TV Show on Nerdist News with Jessica Chobot!',
'uploader': 'Eric Diaz',
'upload_date': '20150202',
'timestamp': 1422892808,
}
}
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
video_id = self._search_regex(
r'''(?x)<script\s+(?:type="text/javascript"\s+)?
src="https?://content\.nerdist\.com/players/([a-zA-Z0-9_]+)-''',
webpage, 'video ID')
timestamp = parse_iso8601(self._html_search_meta(
'shareaholic:article_published_time', webpage, 'upload date'))
uploader = self._html_search_meta(
'shareaholic:article_author_name', webpage, 'article author')
doc = self._download_xml(
'http://content.nerdist.com/jw6/%s.xml' % video_id, video_id)
video_info = doc.find('.//item')
title = xpath_text(video_info, './title', fatal=True)
description = xpath_text(video_info, './description')
thumbnail = xpath_text(
video_info, './{http://rss.jwpcdn.com/}image', 'thumbnail')
formats = []
for source in video_info.findall('./{http://rss.jwpcdn.com/}source'):
vurl = source.attrib['file']
ext = determine_ext(vurl)
if ext == 'm3u8':
formats.extend(self._extract_m3u8_formats(
vurl, video_id, entry_protocol='m3u8_native', ext='mp4',
preference=0))
elif ext == 'smil':
formats.extend(self._extract_smil_formats(
vurl, video_id, fatal=False
))
else:
formats.append({
'format_id': ext,
'url': vurl,
})
self._sort_formats(formats)
return {
'id': video_id,
'display_id': display_id,
'title': title,
'description': description,
'thumbnail': thumbnail,
'timestamp': timestamp,
'formats': formats,
'uploader': uploader,
}