mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-05 10:42:37 +01:00
91 lines
3.1 KiB
Python
91 lines
3.1 KiB
Python
import re
|
|
import urllib.parse
|
|
|
|
from .common import InfoExtractor
|
|
from ..compat import compat_urllib_parse_unquote
|
|
from ..utils import ExtractorError, clean_html
|
|
|
|
|
|
class PlayvidIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?playvid\.com/watch(\?v=|/)(?P<id>.+?)(?:#|$)'
|
|
_TESTS = [{
|
|
'url': 'http://www.playvid.com/watch/RnmBNgtrrJu',
|
|
'md5': 'ffa2f6b2119af359f544388d8c01eb6c',
|
|
'info_dict': {
|
|
'id': 'RnmBNgtrrJu',
|
|
'ext': 'mp4',
|
|
'title': 'md5:9256d01c6317e3f703848b5906880dc8',
|
|
'duration': 82,
|
|
'age_limit': 18,
|
|
},
|
|
'skip': 'Video removed due to ToS',
|
|
}, {
|
|
'url': 'http://www.playvid.com/watch/hwb0GpNkzgH',
|
|
'md5': '39d49df503ad7b8f23a4432cbf046477',
|
|
'info_dict': {
|
|
'id': 'hwb0GpNkzgH',
|
|
'ext': 'mp4',
|
|
'title': 'Ellen Euro Cutie Blond Takes a Sexy Survey Get Facial in The Park',
|
|
'age_limit': 18,
|
|
'thumbnail': r're:^https?://.*\.jpg$',
|
|
},
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
m_error = re.search(
|
|
r'<div class="block-error">\s*<div class="heading">\s*<div>(?P<msg>.+?)</div>\s*</div>', webpage)
|
|
if m_error:
|
|
raise ExtractorError(clean_html(m_error.group('msg')), expected=True)
|
|
|
|
video_title = None
|
|
duration = None
|
|
video_thumbnail = None
|
|
formats = []
|
|
|
|
# most of the information is stored in the flashvars
|
|
flashvars = self._html_search_regex(
|
|
r'flashvars="(.+?)"', webpage, 'flashvars')
|
|
|
|
infos = compat_urllib_parse_unquote(flashvars).split(r'&')
|
|
for info in infos:
|
|
videovars_match = re.match(r'^video_vars\[(.+?)\]=(.+?)$', info)
|
|
if videovars_match:
|
|
key = videovars_match.group(1)
|
|
val = videovars_match.group(2)
|
|
|
|
if key == 'title':
|
|
video_title = urllib.parse.unquote_plus(val)
|
|
if key == 'duration':
|
|
try:
|
|
duration = int(val)
|
|
except ValueError:
|
|
pass
|
|
if key == 'big_thumb':
|
|
video_thumbnail = val
|
|
|
|
videourl_match = re.match(
|
|
r'^video_urls\]\[(?P<resolution>[0-9]+)p', key)
|
|
if videourl_match:
|
|
height = int(videourl_match.group('resolution'))
|
|
formats.append({
|
|
'height': height,
|
|
'url': val,
|
|
})
|
|
|
|
# Extract title - should be in the flashvars; if not, look elsewhere
|
|
if video_title is None:
|
|
video_title = self._html_extract_title(webpage)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'formats': formats,
|
|
'title': video_title,
|
|
'thumbnail': video_thumbnail,
|
|
'duration': duration,
|
|
'description': None,
|
|
'age_limit': 18
|
|
}
|