[YouPorn] Improve extraction
* detect unwatchable videos * improve duration extraction * fix count extraction and support large values * detect and remove SEO spam boilerplate description
This commit is contained in:
parent
c2766cb80e
commit
0b2ce3685e
@ -5,10 +5,13 @@ import re
|
|||||||
|
|
||||||
from .common import InfoExtractor
|
from .common import InfoExtractor
|
||||||
from ..utils import (
|
from ..utils import (
|
||||||
|
clean_html,
|
||||||
extract_attributes,
|
extract_attributes,
|
||||||
|
ExtractorError,
|
||||||
|
get_element_by_id,
|
||||||
int_or_none,
|
int_or_none,
|
||||||
str_to_int,
|
|
||||||
merge_dicts,
|
merge_dicts,
|
||||||
|
parse_count,
|
||||||
T,
|
T,
|
||||||
traverse_obj,
|
traverse_obj,
|
||||||
unified_strdate,
|
unified_strdate,
|
||||||
@ -17,7 +20,13 @@ from ..utils import (
|
|||||||
|
|
||||||
|
|
||||||
class YouPornIE(InfoExtractor):
|
class YouPornIE(InfoExtractor):
|
||||||
_VALID_URL = r'https?://(?:www\.)?youporn\.com/(?:watch|embed)/(?P<id>\d+)(?:/(?P<display_id>[^/?#&]+))?'
|
_VALID_URL = (
|
||||||
|
r'youporn:(?P<id>\d+)',
|
||||||
|
r'''(?x)
|
||||||
|
https?://(?:www\.)?youporn\.com/(?:watch|embed)/(?P<id>\d+)
|
||||||
|
(?:/(?:(?P<display_id>[^/?#&]+)/?)?)?(?:[#?]|$)
|
||||||
|
'''
|
||||||
|
)
|
||||||
_EMBED_REGEX = [r'<iframe[^>]+\bsrc=["\'](?P<url>(?:https?:)?//(?:www\.)?youporn\.com/embed/\d+)']
|
_EMBED_REGEX = [r'<iframe[^>]+\bsrc=["\'](?P<url>(?:https?:)?//(?:www\.)?youporn\.com/embed/\d+)']
|
||||||
_TESTS = [{
|
_TESTS = [{
|
||||||
'url': 'http://www.youporn.com/watch/505835/sex-ed-is-it-safe-to-masturbate-daily/',
|
'url': 'http://www.youporn.com/watch/505835/sex-ed-is-it-safe-to-masturbate-daily/',
|
||||||
@ -38,7 +47,7 @@ class YouPornIE(InfoExtractor):
|
|||||||
'tags': list,
|
'tags': list,
|
||||||
'age_limit': 18,
|
'age_limit': 18,
|
||||||
},
|
},
|
||||||
'skip': 'This video has been disabled',
|
'skip': 'This video has been deactivated',
|
||||||
}, {
|
}, {
|
||||||
# Unknown uploader
|
# Unknown uploader
|
||||||
'url': 'http://www.youporn.com/watch/561726/big-tits-awesome-brunette-on-amazing-webcam-show/?from=related3&al=2&from_id=561726&pos=4',
|
'url': 'http://www.youporn.com/watch/561726/big-tits-awesome-brunette-on-amazing-webcam-show/?from=related3&al=2&from_id=561726&pos=4',
|
||||||
@ -73,7 +82,7 @@ class YouPornIE(InfoExtractor):
|
|||||||
}, {
|
}, {
|
||||||
'url': 'https://www.youporn.com/watch/16290308/tinderspecial-trailer1/',
|
'url': 'https://www.youporn.com/watch/16290308/tinderspecial-trailer1/',
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
'id': '46949121',
|
'id': '16290308',
|
||||||
'age_limit': 18,
|
'age_limit': 18,
|
||||||
'categories': [],
|
'categories': [],
|
||||||
'description': None, # SEO spam using title removed
|
'description': None, # SEO spam using title removed
|
||||||
@ -100,15 +109,28 @@ class YouPornIE(InfoExtractor):
|
|||||||
return list(yield_urls())
|
return list(yield_urls())
|
||||||
|
|
||||||
def _real_extract(self, url):
|
def _real_extract(self, url):
|
||||||
display_id = self._match_valid_url(url).group('id', 'display_id')
|
# A different video ID (data-video-id) is hidden in the page but
|
||||||
url = 'http://www.youporn.com/watch/%s' % (display_id[0],)
|
# never seems to be used
|
||||||
display_id = display_id[1] or display_id[0]
|
video_id, display_id = self._match_valid_url(url).group('id', 'display_id')
|
||||||
|
url = 'http://www.youporn.com/watch/%s' % (video_id,)
|
||||||
webpage = self._download_webpage(
|
webpage = self._download_webpage(
|
||||||
url, display_id, headers={'Cookie': 'age_verified=1'})
|
url, video_id, headers={'Cookie': 'age_verified=1'})
|
||||||
video_id = display_id[0]
|
|
||||||
|
watchable = self._search_regex(
|
||||||
|
r'''(<div\s[^>]*\bid\s*=\s*('|")?watch-container(?(2)\2|(?!-)\b)[^>]*>)''',
|
||||||
|
webpage, 'watchability', default=None)
|
||||||
|
if not watchable:
|
||||||
|
msg = re.split(r'\s{4}', clean_html(get_element_by_id(
|
||||||
|
'mainContent', webpage)) or '')[0]
|
||||||
|
raise ExtractorError(
|
||||||
|
('%s says: %s' % (self.IE_NAME, msg))
|
||||||
|
if msg else 'Video unavailable: no reason found',
|
||||||
|
expected=True)
|
||||||
|
# internal ID ?
|
||||||
|
# video_id = extract_attributes(watchable).get('data-video-id')
|
||||||
|
|
||||||
playervars = self._search_json(
|
playervars = self._search_json(
|
||||||
r'\bplayervars\s*:', webpage, 'playervars', display_id)
|
r'\bplayervars\s*:', webpage, 'playervars', video_id)
|
||||||
|
|
||||||
def get_fmt(x):
|
def get_fmt(x):
|
||||||
v_url = url_or_none(x.get('videoUrl'))
|
v_url = url_or_none(x.get('videoUrl'))
|
||||||
@ -123,7 +145,7 @@ class YouPornIE(InfoExtractor):
|
|||||||
if f not in defs_by_format:
|
if f not in defs_by_format:
|
||||||
return []
|
return []
|
||||||
return self._download_json(
|
return self._download_json(
|
||||||
defs_by_format[f]['videoUrl'], display_id, '{0}-formats'.format(f))
|
defs_by_format[f]['videoUrl'], video_id, '{0}-formats'.format(f))
|
||||||
|
|
||||||
formats = []
|
formats = []
|
||||||
# Try to extract only the actual master m3u8 first, avoiding the duplicate single resolution "master" m3u8s
|
# Try to extract only the actual master m3u8 first, avoiding the duplicate single resolution "master" m3u8s
|
||||||
@ -169,8 +191,10 @@ class YouPornIE(InfoExtractor):
|
|||||||
thumbnail = self._search_regex(
|
thumbnail = self._search_regex(
|
||||||
r'(?:imageurl\s*=|poster\s*:)\s*(["\'])(?P<thumbnail>.+?)\1',
|
r'(?:imageurl\s*=|poster\s*:)\s*(["\'])(?P<thumbnail>.+?)\1',
|
||||||
webpage, 'thumbnail', fatal=False, group='thumbnail')
|
webpage, 'thumbnail', fatal=False, group='thumbnail')
|
||||||
duration = int_or_none(self._html_search_meta(
|
duration = traverse_obj(playervars, ('duration', T(int_or_none)))
|
||||||
'video:duration', webpage, 'duration', fatal=False))
|
if duration is None:
|
||||||
|
duration = int_or_none(self._html_search_meta(
|
||||||
|
'video:duration', webpage, 'duration', fatal=False))
|
||||||
|
|
||||||
uploader = self._html_search_regex(
|
uploader = self._html_search_regex(
|
||||||
r'(?s)<div[^>]+class=["\']submitByLink["\'][^>]*>(.+?)</div>',
|
r'(?s)<div[^>]+class=["\']submitByLink["\'][^>]*>(.+?)</div>',
|
||||||
@ -186,11 +210,11 @@ class YouPornIE(InfoExtractor):
|
|||||||
|
|
||||||
view_count = None
|
view_count = None
|
||||||
views = self._search_regex(
|
views = self._search_regex(
|
||||||
r'(<div[^>]+\bclass=["\']js_videoInfoViews["\']>)', webpage,
|
r'(<div\s[^>]*\bdata-value\s*=[^>]+>)\s*<label>Views:</label>',
|
||||||
'views', default=None)
|
webpage, 'views', default=None)
|
||||||
if views:
|
if views:
|
||||||
view_count = str_to_int(extract_attributes(views).get('data-value'))
|
view_count = parse_count(extract_attributes(views).get('data-value'))
|
||||||
comment_count = str_to_int(self._search_regex(
|
comment_count = parse_count(self._search_regex(
|
||||||
r'>All [Cc]omments? \(([\d,.]+)\)',
|
r'>All [Cc]omments? \(([\d,.]+)\)',
|
||||||
webpage, 'comment count', default=None))
|
webpage, 'comment count', default=None))
|
||||||
|
|
||||||
@ -211,7 +235,7 @@ class YouPornIE(InfoExtractor):
|
|||||||
|
|
||||||
result = merge_dicts(data, {
|
result = merge_dicts(data, {
|
||||||
'id': video_id,
|
'id': video_id,
|
||||||
'display_id': display_id if display_id != video_id else None,
|
'display_id': display_id,
|
||||||
'title': title,
|
'title': title,
|
||||||
'description': description,
|
'description': description,
|
||||||
'thumbnail': thumbnail,
|
'thumbnail': thumbnail,
|
||||||
@ -225,4 +249,8 @@ class YouPornIE(InfoExtractor):
|
|||||||
'age_limit': age_limit,
|
'age_limit': age_limit,
|
||||||
'formats': formats,
|
'formats': formats,
|
||||||
})
|
})
|
||||||
|
# Remove promotional non-description
|
||||||
|
if result.get('description', '').startswith(
|
||||||
|
'Watch %s online' % (result['title'],)):
|
||||||
|
del result['description']
|
||||||
return result
|
return result
|
||||||
|
Loading…
Reference in New Issue
Block a user