mirror of
https://github.com/blackjack4494/yt-dlc.git
synced 2024-12-26 17:58:09 +00:00
9c15e9de84
There's no need to use two different methods. Now we can also download videos over http if possible. Also run the test for rtmp videos, but skip the download.
119 lines
4.1 KiB
Python
119 lines
4.1 KiB
Python
import itertools
|
|
import json
|
|
import re
|
|
|
|
from .common import InfoExtractor, SearchInfoExtractor
|
|
from ..utils import (
|
|
compat_urllib_parse,
|
|
compat_urlparse,
|
|
determine_ext,
|
|
clean_html,
|
|
)
|
|
|
|
|
|
class YahooIE(InfoExtractor):
|
|
IE_DESC = u'Yahoo screen'
|
|
_VALID_URL = r'http://screen\.yahoo\.com/.*?-(?P<id>\d*?)\.html'
|
|
_TESTS = [
|
|
{
|
|
u'url': u'http://screen.yahoo.com/julian-smith-travis-legg-watch-214727115.html',
|
|
u'file': u'214727115.mp4',
|
|
u'info_dict': {
|
|
u'title': u'Julian Smith & Travis Legg Watch Julian Smith',
|
|
u'description': u'Julian and Travis watch Julian Smith',
|
|
},
|
|
},
|
|
{
|
|
u'url': u'http://screen.yahoo.com/wired/codefellas-s1-ep12-cougar-lies-103000935.html',
|
|
u'file': u'103000935.flv',
|
|
u'info_dict': {
|
|
u'title': u'The Cougar Lies with Spanish Moss',
|
|
u'description': u'Agent Topple\'s mustache does its dirty work, and Nicole brokers a deal for peace. But why is the NSA collecting millions of Instagram brunch photos? And if your waffles have nothing to hide, what are they so worried about?',
|
|
},
|
|
u'params': {
|
|
# Requires rtmpdump
|
|
u'skip_download': True,
|
|
},
|
|
},
|
|
]
|
|
|
|
def _real_extract(self, url):
|
|
mobj = re.match(self._VALID_URL, url)
|
|
video_id = mobj.group('id')
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
items_json = self._search_regex(r'YVIDEO_INIT_ITEMS = ({.*?});$',
|
|
webpage, u'items', flags=re.MULTILINE)
|
|
items = json.loads(items_json)
|
|
info = items['mediaItems']['query']['results']['mediaObj'][0]
|
|
meta = info['meta']
|
|
|
|
formats = []
|
|
for s in info['streams']:
|
|
format_info = {
|
|
'width': s.get('width'),
|
|
'height': s.get('height'),
|
|
'bitrate': s.get('bitrate'),
|
|
}
|
|
|
|
host = s['host']
|
|
path = s['path']
|
|
if host.startswith('rtmp'):
|
|
format_info.update({
|
|
'url': host,
|
|
'play_path': path,
|
|
'ext': 'flv',
|
|
})
|
|
else:
|
|
format_url = compat_urlparse.urljoin(host, path)
|
|
format_info['url'] = format_url
|
|
format_info['ext'] = determine_ext(format_url)
|
|
|
|
formats.append(format_info)
|
|
formats = sorted(formats, key=lambda f:(f['height'], f['width']))
|
|
|
|
info = {
|
|
'id': video_id,
|
|
'title': meta['title'],
|
|
'formats': formats,
|
|
'description': clean_html(meta['description']),
|
|
'thumbnail': meta['thumbnail'],
|
|
}
|
|
# TODO: Remove when #980 has been merged
|
|
info.update(formats[-1])
|
|
|
|
return info
|
|
|
|
|
|
class YahooSearchIE(SearchInfoExtractor):
|
|
IE_DESC = u'Yahoo screen search'
|
|
_MAX_RESULTS = 1000
|
|
IE_NAME = u'screen.yahoo:search'
|
|
_SEARCH_KEY = 'yvsearch'
|
|
|
|
def _get_n_results(self, query, n):
|
|
"""Get a specified number of results for a query"""
|
|
|
|
res = {
|
|
'_type': 'playlist',
|
|
'id': query,
|
|
'entries': []
|
|
}
|
|
for pagenum in itertools.count(0):
|
|
result_url = u'http://video.search.yahoo.com/search/?p=%s&fr=screen&o=js&gs=0&b=%d' % (compat_urllib_parse.quote_plus(query), pagenum * 30)
|
|
webpage = self._download_webpage(result_url, query,
|
|
note='Downloading results page '+str(pagenum+1))
|
|
info = json.loads(webpage)
|
|
m = info[u'm']
|
|
results = info[u'results']
|
|
|
|
for (i, r) in enumerate(results):
|
|
if (pagenum * 30) +i >= n:
|
|
break
|
|
mobj = re.search(r'(?P<url>screen\.yahoo\.com/.*?-\d*?\.html)"', r)
|
|
e = self.url_result('http://' + mobj.group('url'), 'Yahoo')
|
|
res['entries'].append(e)
|
|
if (pagenum * 30 +i >= n) or (m[u'last'] >= (m[u'total'] -1 )):
|
|
break
|
|
|
|
return res
|