mirror of https://github.com/yt-dlp/yt-dlp
Merge remote-tracking branch 'origin/master'
commit
850555c484
@ -1,56 +1,59 @@
|
|||||||
import re
|
import re
|
||||||
import xml.etree.ElementTree
|
import json
|
||||||
|
|
||||||
from .common import InfoExtractor
|
from .common import InfoExtractor
|
||||||
from ..utils import (
|
from ..utils import (
|
||||||
unified_strdate,
|
|
||||||
compat_urllib_parse,
|
compat_urllib_parse,
|
||||||
|
compat_urlparse,
|
||||||
|
unescapeHTML,
|
||||||
|
get_meta_content,
|
||||||
)
|
)
|
||||||
|
|
||||||
|
|
||||||
class GameSpotIE(InfoExtractor):
|
class GameSpotIE(InfoExtractor):
|
||||||
_WORKING = False
|
|
||||||
_VALID_URL = r'(?:http://)?(?:www\.)?gamespot\.com/.*-(?P<page_id>\d+)/?'
|
_VALID_URL = r'(?:http://)?(?:www\.)?gamespot\.com/.*-(?P<page_id>\d+)/?'
|
||||||
_TEST = {
|
_TEST = {
|
||||||
u"url": u"http://www.gamespot.com/arma-iii/videos/arma-iii-community-guide-sitrep-i-6410818/",
|
u"url": u"http://www.gamespot.com/arma-iii/videos/arma-iii-community-guide-sitrep-i-6410818/",
|
||||||
u"file": u"6410818.mp4",
|
u"file": u"gs-2300-6410818.mp4",
|
||||||
u"md5": u"b2a30deaa8654fcccd43713a6b6a4825",
|
u"md5": u"b2a30deaa8654fcccd43713a6b6a4825",
|
||||||
u"info_dict": {
|
u"info_dict": {
|
||||||
u"title": u"Arma 3 - Community Guide: SITREP I",
|
u"title": u"Arma 3 - Community Guide: SITREP I",
|
||||||
u"upload_date": u"20130627",
|
u'description': u'Check out this video where some of the basics of Arma 3 is explained.',
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
def _real_extract(self, url):
|
def _real_extract(self, url):
|
||||||
mobj = re.match(self._VALID_URL, url)
|
mobj = re.match(self._VALID_URL, url)
|
||||||
page_id = mobj.group('page_id')
|
page_id = video_id = mobj.group('page_id')
|
||||||
webpage = self._download_webpage(url, page_id)
|
webpage = self._download_webpage(url, page_id)
|
||||||
video_id = self._html_search_regex([r'"og:video" content=".*?\?id=(\d+)"',
|
data_video_json = self._search_regex(r'data-video=\'(.*?)\'', webpage, u'data video')
|
||||||
r'http://www\.gamespot\.com/videoembed/(\d+)'],
|
data_video = json.loads(unescapeHTML(data_video_json))
|
||||||
webpage, 'video id')
|
|
||||||
data = compat_urllib_parse.urlencode({'id': video_id, 'newplayer': '1'})
|
|
||||||
info_url = 'http://www.gamespot.com/pages/video_player/xml.php?' + data
|
|
||||||
info_xml = self._download_webpage(info_url, video_id)
|
|
||||||
doc = xml.etree.ElementTree.fromstring(info_xml)
|
|
||||||
clip_el = doc.find('./playList/clip')
|
|
||||||
|
|
||||||
http_urls = [{'url': node.find('filePath').text,
|
# Transform the manifest url to a link to the mp4 files
|
||||||
'rate': int(node.find('rate').text)}
|
# they are used in mobile devices.
|
||||||
for node in clip_el.find('./httpURI')]
|
f4m_url = data_video['videoStreams']['f4m_stream']
|
||||||
best_quality = sorted(http_urls, key=lambda f: f['rate'])[-1]
|
f4m_path = compat_urlparse.urlparse(f4m_url).path
|
||||||
video_url = best_quality['url']
|
QUALITIES_RE = r'((,\d+)+,?)'
|
||||||
title = clip_el.find('./title').text
|
qualities = self._search_regex(QUALITIES_RE, f4m_path, u'qualities').strip(',').split(',')
|
||||||
ext = video_url.rpartition('.')[2]
|
http_path = f4m_path[1:].split('/', 1)[1]
|
||||||
thumbnail_url = clip_el.find('./screenGrabURI').text
|
http_template = re.sub(QUALITIES_RE, r'%s', http_path)
|
||||||
view_count = int(clip_el.find('./views').text)
|
http_template = http_template.replace('.csmil/manifest.f4m', '')
|
||||||
upload_date = unified_strdate(clip_el.find('./postDate').text)
|
http_template = compat_urlparse.urljoin('http://video.gamespotcdn.com/', http_template)
|
||||||
|
formats = []
|
||||||
|
for q in qualities:
|
||||||
|
formats.append({
|
||||||
|
'url': http_template % q,
|
||||||
|
'ext': 'mp4',
|
||||||
|
'format_id': q,
|
||||||
|
})
|
||||||
|
|
||||||
return [{
|
info = {
|
||||||
'id' : video_id,
|
'id': data_video['guid'],
|
||||||
'url' : video_url,
|
'title': compat_urllib_parse.unquote(data_video['title']),
|
||||||
'ext' : ext,
|
'formats': formats,
|
||||||
'title' : title,
|
'description': get_meta_content('description', webpage),
|
||||||
'thumbnail' : thumbnail_url,
|
'thumbnail': self._og_search_thumbnail(webpage),
|
||||||
'upload_date' : upload_date,
|
}
|
||||||
'view_count' : view_count,
|
# TODO: Remove when #980 has been merged
|
||||||
}]
|
info.update(formats[-1])
|
||||||
|
return info
|
||||||
|
Loading…
Reference in New Issue