Browse Source
[howstuffworks] Add extractor (#3500)
[howstuffworks] Add extractor (#3500)
Content-length is invalid for final download links.master
Sergey M․
10 years ago
2 changed files with 135 additions and 0 deletions
Split View
Diff Options
@ -0,0 +1,134 @@ |
|||
from __future__ import unicode_literals |
|||
|
|||
import re |
|||
import json |
|||
import random |
|||
import string |
|||
|
|||
from .common import InfoExtractor |
|||
from ..utils import find_xpath_attr |
|||
|
|||
|
|||
class HowStuffWorksIE(InfoExtractor): |
|||
_VALID_URL = r'https?://[\da-z-]+\.howstuffworks\.com/(?:[^/]+/)*\d+-(?P<id>.+?)-video\.htm' |
|||
_TESTS = [ |
|||
{ |
|||
'url': 'http://adventure.howstuffworks.com/5266-cool-jobs-iditarod-musher-video.htm', |
|||
'info_dict': { |
|||
'id': '450221', |
|||
'display_id': 'cool-jobs-iditarod-musher', |
|||
'ext': 'flv', |
|||
'title': 'Cool Jobs - Iditarod Musher', |
|||
'description': 'md5:82bb58438a88027b8186a1fccb365f90', |
|||
'thumbnail': 're:^https?://.*\.jpg$', |
|||
}, |
|||
'params': { |
|||
# md5 is not consistent |
|||
'skip_download': True |
|||
} |
|||
}, |
|||
{ |
|||
'url': 'http://adventure.howstuffworks.com/39516-deadliest-catch-jakes-farewell-pots-video.htm', |
|||
'info_dict': { |
|||
'id': '553470', |
|||
'display_id': 'deadliest-catch-jakes-farewell-pots', |
|||
'ext': 'mp4', |
|||
'title': 'Deadliest Catch: Jake\'s Farewell Pots', |
|||
'description': 'md5:9632c346d5e43ee238028c9cefd8dbbc', |
|||
'thumbnail': 're:^https?://.*\.jpg$', |
|||
}, |
|||
'params': { |
|||
# md5 is not consistent |
|||
'skip_download': True |
|||
} |
|||
}, |
|||
{ |
|||
'url': 'http://entertainment.howstuffworks.com/arts/2706-sword-swallowing-1-by-dan-meyer-video.htm', |
|||
'info_dict': { |
|||
'id': '440011', |
|||
'display_id': 'sword-swallowing-1-by-dan-meyer', |
|||
'ext': 'flv', |
|||
'title': 'Sword Swallowing #1 by Dan Meyer', |
|||
'description': 'md5:b2409e88172913e2e7d3d1159b0ef735', |
|||
'thumbnail': 're:^https?://.*\.jpg$', |
|||
}, |
|||
'params': { |
|||
# md5 is not consistent |
|||
'skip_download': True |
|||
} |
|||
}, |
|||
] |
|||
|
|||
def _real_extract(self, url): |
|||
mobj = re.match(self._VALID_URL, url) |
|||
display_id = mobj.group('id') |
|||
webpage = self._download_webpage(url, display_id) |
|||
|
|||
content_id = self._search_regex(r'var siteSectionId="(\d+)";', webpage, 'content id') |
|||
|
|||
mp4 = self._search_regex( |
|||
r'''(?xs)var\s+clip\s*=\s*{\s* |
|||
.+?\s* |
|||
content_id\s*:\s*%s\s*,\s* |
|||
.+?\s* |
|||
mp4\s*:\s*\[(.*?),?\]\s* |
|||
};\s* |
|||
videoData\.push\(clip\);''' % content_id, |
|||
webpage, 'mp4', fatal=False, default=None) |
|||
|
|||
smil = self._download_xml( |
|||
'http://services.media.howstuffworks.com/videos/%s/smil-service.smil' % content_id, |
|||
content_id, 'Downloading video SMIL') |
|||
|
|||
http_base = find_xpath_attr( |
|||
smil, |
|||
'./{0}head/{0}meta'.format('{http://www.w3.org/2001/SMIL20/Language}'), |
|||
'name', |
|||
'httpBase').get('content') |
|||
|
|||
def random_string(str_len=0): |
|||
return ''.join([random.choice(string.ascii_uppercase) for _ in range(str_len)]) |
|||
|
|||
URL_SUFFIX = '?v=2.11.3&fp=LNX 11,2,202,356&r=%s&g=%s' % (random_string(5), random_string(12)) |
|||
|
|||
formats = [] |
|||
|
|||
if mp4: |
|||
for video in json.loads('[%s]' % mp4): |
|||
bitrate = video['bitrate'] |
|||
fmt = { |
|||
'url': video['src'].replace('http://pmd.video.howstuffworks.com', http_base) + URL_SUFFIX, |
|||
'format_id': bitrate, |
|||
} |
|||
m = re.search(r'(?P<vbr>\d+)[Kk]', bitrate) |
|||
if m: |
|||
fmt['vbr'] = int(m.group('vbr')) |
|||
formats.append(fmt) |
|||
else: |
|||
for video in smil.findall( |
|||
'.//{0}body/{0}switch/{0}video'.format('{http://www.w3.org/2001/SMIL20/Language}')): |
|||
vbr = int(video.attrib['system-bitrate']) / 1000 |
|||
formats.append({ |
|||
'url': '%s/%s%s' % (http_base, video.attrib['src'], URL_SUFFIX), |
|||
'format_id': '%dk' % vbr, |
|||
'vbr': vbr, |
|||
}) |
|||
|
|||
self._sort_formats(formats) |
|||
|
|||
title = self._og_search_title(webpage) |
|||
TITLE_SUFFIX = ' : HowStuffWorks' |
|||
if title.endswith(TITLE_SUFFIX): |
|||
title = title[:-len(TITLE_SUFFIX)] |
|||
|
|||
description = self._og_search_description(webpage) |
|||
thumbnail = self._og_search_thumbnail(webpage) |
|||
|
|||
return { |
|||
'id': content_id, |
|||
'display_id': display_id, |
|||
'title': title, |
|||
'description': description, |
|||
'thumbnail': thumbnail, |
|||
'formats': formats, |
|||
} |
Write
Preview
Loading…
Cancel
Save