[sohu] Handle encoding, and fix tests

This commit is contained in:
Philipp Hagemeister 2013-08-28 13:59:08 +02:00
parent f8b362739e
commit f143d86ad2
2 changed files with 72 additions and 70 deletions

View File

@ -145,12 +145,17 @@ class InfoExtractor(object):
urlh = self._request_webpage(url_or_request, video_id, note, errnote) urlh = self._request_webpage(url_or_request, video_id, note, errnote)
content_type = urlh.headers.get('Content-Type', '') content_type = urlh.headers.get('Content-Type', '')
webpage_bytes = urlh.read()
m = re.match(r'[a-zA-Z0-9_.-]+/[a-zA-Z0-9_.-]+\s*;\s*charset=(.+)', content_type) m = re.match(r'[a-zA-Z0-9_.-]+/[a-zA-Z0-9_.-]+\s*;\s*charset=(.+)', content_type)
if m: if m:
encoding = m.group(1) encoding = m.group(1)
else: else:
encoding = 'utf-8' m = re.search(br'<meta[^>]+charset="?([^"]+)[ /">]',
webpage_bytes = urlh.read() webpage_bytes[:1024])
if m:
encoding = m.group(1).decode('ascii')
else:
encoding = 'utf-8'
if self._downloader.params.get('dump_intermediate_pages', False): if self._downloader.params.get('dump_intermediate_pages', False):
try: try:
url = url_or_request.get_full_url() url = url_or_request.get_full_url()

View File

@ -1,13 +1,10 @@
# encoding: utf-8 # encoding: utf-8
import re
import json import json
import time import re
import logging
import urllib2
from .common import InfoExtractor from .common import InfoExtractor
from ..utils import compat_urllib_request, clean_html from ..utils import ExtractorError
class SohuIE(InfoExtractor): class SohuIE(InfoExtractor):
@ -15,79 +12,79 @@ class SohuIE(InfoExtractor):
_TEST = { _TEST = {
u'url': u'http://tv.sohu.com/20130724/n382479172.shtml#super', u'url': u'http://tv.sohu.com/20130724/n382479172.shtml#super',
u'file': u'382479172.flv', u'file': u'382479172.mp4',
u'md5': u'cc84eed6b6fbf0f2f9a8d3cb9da1939b', u'md5': u'bde8d9a6ffd82c63a1eefaef4eeefec7',
u'info_dict': { u'info_dict': {
u'title': u'The Illest - Far East Movement Riff Raff', u'title': u'MVFar East Movement《The Illest》',
}, },
} }
def _real_extract(self, url): def _real_extract(self, url):
def _fetch_data(vid_id):
base_data_url = u'http://hot.vrs.sohu.com/vrs_flash.action?vid='
data_url = base_data_url + str(vid_id)
data_json = self._download_webpage(
data_url, video_id,
note=u'Downloading JSON data for ' + str(vid_id))
return json.loads(data_json)
mobj = re.match(self._VALID_URL, url) mobj = re.match(self._VALID_URL, url)
video_id = mobj.group('id') video_id = mobj.group('id')
webpage = self._download_webpage(url, video_id)
pattern = r'<title>(.+?)</title>'
compiled = re.compile(pattern, re.DOTALL)
title = self._search_regex(compiled, webpage, u'video title')
title = clean_html(title).split('-')[0].strip()
self.to_screen('Title: %s' % title)
pattern = re.compile(r'var vid="(\d+)"')
result = re.search(pattern, webpage)
if not result:
logging.info('[Sohu] could not get vid')
return None
vid = result.group(1)
logging.info('vid: %s' % vid)
base_url_1 = 'http://hot.vrs.sohu.com/vrs_flash.action?vid='
url_1 = base_url_1 + vid
logging.info('json url: %s' % url_1)
webpage = self._download_webpage(url_1, vid)
json_1 = json.loads(webpage)
# get the highest definition video vid and json infomation.
vids = []
qualities = ('oriVid', 'superVid', 'highVid', 'norVid')
for vid_name in qualities:
vids.append(json_1['data'][vid_name])
clearest_vid = 0
for i, v in enumerate(vids):
if v != 0:
clearest_vid = v
logging.info('quality definition: %s' % qualities[i][:-3])
break
if not clearest_vid:
logging.warning('could not find valid clearest_vid')
return None
if vid != clearest_vid:
url_1 = '%s%d' % (base_url_1, clearest_vid)
logging.info('highest definition json url: %s' % url_1)
json_1 = json.loads(urllib2.urlopen(url_1).read())
allot = json_1['allot']
prot = json_1['prot']
clipsURL = json_1['data']['clipsURL']
su = json_1['data']['su']
num_of_parts = json_1['data']['totalBlocks']
logging.info('Total parts: %d' % num_of_parts)
base_url_3 = 'http://allot/?prot=prot&file=clipsURL[i]&new=su[i]'
files_info = []
for i in range(num_of_parts):
self.to_screen('Geting json infomation of part %s/%s' % (i + 1, num_of_parts))
middle_url = 'http://%s/?prot=%s&file=%s&new=%s' % (allot, prot, clipsURL[i], su[i])
logging.info('middle url part %d: %s' % (i, middle_url))
middle_info = urllib2.urlopen(middle_url).read().split('|')
middle_part_1 = middle_info[0]
download_url = '%s%s?key=%s' % (middle_info[0], su[i], middle_info[3])
info = { webpage = self._download_webpage(url, video_id)
raw_title = self._html_search_regex(r'(?s)<title>(.+?)</title>',
webpage, u'video title')
title = raw_title.partition('-')[0].strip()
vid = self._html_search_regex(r'var vid="(\d+)"', webpage,
u'video path')
data = _fetch_data(vid)
QUALITIES = ('ori', 'super', 'high', 'nor')
vid_ids = [data['data'][q + 'Vid']
for q in QUALITIES
if data['data'][q + 'Vid'] != 0]
if not vid_ids:
raise ExtractorError(u'No formats available for this video')
# For now, we just pick the highest available quality
vid_id = vid_ids[-1]
format_data = data if vid == vid_id else _fetch_data(vid_id)
part_count = format_data['data']['totalBlocks']
allot = format_data['allot']
prot = format_data['prot']
clipsURL = format_data['data']['clipsURL']
su = format_data['data']['su']
playlist = []
for i in range(part_count):
part_url = ('http://%s/?prot=%s&file=%s&new=%s' %
(allot, prot, clipsURL[i], su[i]))
part_str = self._download_webpage(
part_url, video_id,
note=u'Downloading part %d of %d' % (i+1, part_count))
part_info = part_str.split('|')
video_url = '%s%s?key=%s' % (part_info[0], su[i], part_info[3])
video_info = {
'id': '%s_part%02d' % (video_id, i + 1), 'id': '%s_part%02d' % (video_id, i + 1),
'title': title, 'title': title,
'url': download_url, 'url': video_url,
'ext': 'mp4', 'ext': 'mp4',
} }
files_info.append(info) playlist.append(video_info)
time.sleep(1)
if num_of_parts == 1: if len(playlist) == 1:
info = files_info[0] info = playlist[0]
info['id'] = video_id info['id'] = video_id
return info else:
return files_info info = {
'_type': 'playlist',
'entries': playlist,
'id': video_id,
}
return info