mirror of
https://codeberg.org/polarisfm/youtube-dl
synced 2024-11-05 01:44:33 +01:00
GoshGay Extractor
This commit is contained in:
parent
6e1e0e4b5b
commit
d6aa1967ad
@ -112,6 +112,7 @@ from .generic import GenericIE
|
|||||||
from .googleplus import GooglePlusIE
|
from .googleplus import GooglePlusIE
|
||||||
from .googlesearch import GoogleSearchIE
|
from .googlesearch import GoogleSearchIE
|
||||||
from .gorillavid import GorillaVidIE
|
from .gorillavid import GorillaVidIE
|
||||||
|
from .goshgay import GoshgayIE
|
||||||
from .hark import HarkIE
|
from .hark import HarkIE
|
||||||
from .helsinki import HelsinkiIE
|
from .helsinki import HelsinkiIE
|
||||||
from .hentaistigma import HentaiStigmaIE
|
from .hentaistigma import HentaiStigmaIE
|
||||||
|
72
youtube_dl/extractor/goshgay.py
Normal file
72
youtube_dl/extractor/goshgay.py
Normal file
@ -0,0 +1,72 @@
|
|||||||
|
# -*- coding: utf-8 -*-
|
||||||
|
from __future__ import unicode_literals
|
||||||
|
|
||||||
|
import re
|
||||||
|
|
||||||
|
from .common import InfoExtractor
|
||||||
|
from ..utils import (
|
||||||
|
compat_urlparse,
|
||||||
|
str_to_int,
|
||||||
|
ExtractorError,
|
||||||
|
)
|
||||||
|
import json
|
||||||
|
|
||||||
|
|
||||||
|
class GoshgayIE(InfoExtractor):
|
||||||
|
_VALID_URL = r'^(?:https?://)www.goshgay.com/video(?P<id>\d+?)($|/)'
|
||||||
|
_TEST = {
|
||||||
|
'url': 'http://www.goshgay.com/video4116282',
|
||||||
|
'md5': '268b9f3c3229105c57859e166dd72b03',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '4116282',
|
||||||
|
'ext': 'flv',
|
||||||
|
'title': 'md5:089833a4790b5e103285a07337f245bf',
|
||||||
|
'thumbnail': 're:http://.*\.jpg',
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
mobj = re.match(self._VALID_URL, url)
|
||||||
|
video_id = mobj.group('id')
|
||||||
|
|
||||||
|
webpage = self._download_webpage(url, video_id)
|
||||||
|
title = self._search_regex(r'class="video-title"><h1>(.+?)<', webpage, 'title')
|
||||||
|
|
||||||
|
player_config = self._search_regex(r'jwplayer\("player"\)\.setup\(({.+?})\)', webpage, 'config settings',
|
||||||
|
fatal=True, flags=re.S)
|
||||||
|
player_vars = json.loads(player_config.replace("'", '"'))
|
||||||
|
width = str_to_int(player_vars.get('width'))
|
||||||
|
height = str_to_int(player_vars.get('height'))
|
||||||
|
config_uri = player_vars.get('config')
|
||||||
|
|
||||||
|
if config_uri is None:
|
||||||
|
raise ExtractorError('Missing config URI')
|
||||||
|
node = self._download_xml(config_uri, video_id, 'Downloading player config XML',
|
||||||
|
errnote='Unable to download XML')
|
||||||
|
if node is None:
|
||||||
|
raise ExtractorError('Missing config XML')
|
||||||
|
if node.tag != 'config':
|
||||||
|
raise ExtractorError('Missing config attribute')
|
||||||
|
fns = node.findall('file')
|
||||||
|
imgs = node.findall('image')
|
||||||
|
if len(fns) != 1:
|
||||||
|
raise ExtractorError('Missing media URI')
|
||||||
|
video_url = fns[0].text
|
||||||
|
if len(imgs) < 1:
|
||||||
|
thumbnail = None
|
||||||
|
else:
|
||||||
|
thumbnail = imgs[0].text
|
||||||
|
|
||||||
|
url_comp = compat_urlparse.urlparse(url)
|
||||||
|
ref = "%s://%s%s" % (url_comp[0], url_comp[1], url_comp[2])
|
||||||
|
|
||||||
|
return {
|
||||||
|
'id': video_id,
|
||||||
|
'url': video_url,
|
||||||
|
'title': title,
|
||||||
|
'width': width,
|
||||||
|
'height': height,
|
||||||
|
'thumbnail': thumbnail,
|
||||||
|
'http_referer': ref,
|
||||||
|
'age_limit': 18,
|
||||||
|
}
|
Loading…
Reference in New Issue
Block a user