Add Patreon extractor

pull/8/head
Erik Johnson 10 years ago
parent 65bc504db8
commit a00d73c8c8

@ -0,0 +1,127 @@
# encoding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..utils import (
ExtractorError,
compat_html_parser,
#compat_urllib_request,
#compat_urllib_parse,
)
class PatreonHTMLParser(compat_html_parser.HTMLParser):
_PREFIX = 'http://www.patreon.com'
_ATTACH_TAGS = 5 * ['div']
_ATTACH_CLASSES = [
'fancyboxhidden', 'box photo double', 'boxwrapper double',
'hiddendisplay shareinfo', 'attach'
]
_INFO_TAGS = 4 * ['div']
_INFO_CLASSES = [
'fancyboxhidden', 'box photo double', 'boxwrapper double',
'hiddendisplay shareinfo'
]
def get_creation_info(self, html_data):
self.tag_stack = []
self.attrs_stack = []
self.creation_info = {}
self.feed(html_data)
def handle_starttag(self, tag, attrs):
self.tag_stack.append(tag.lower())
self.attrs_stack.append(dict(attrs))
def handle_endtag(self, tag):
self.tag_stack.pop()
self.attrs_stack.pop()
def handle_data(self, data):
# Check first if this is a creation attachment
if self.tag_stack[-6:-1] == self._ATTACH_TAGS:
attrs_classes = [
x.get('class', '').lower() for x in self.attrs_stack[-6:-1]
]
if attrs_classes == self._ATTACH_CLASSES:
if self.tag_stack[-1] == 'a':
url = self._PREFIX + self.attrs_stack[-1].get('href')
self.creation_info['url'] = url
if '.' in data:
self.creation_info['ext'] = data.rsplit('.')[-1]
# Next, check if this is within the div containing the creation info
if self.tag_stack[-5:-1] == self._INFO_TAGS:
attrs_classes = [
x.get('class', '').lower() for x in self.attrs_stack[-5:-1]
]
if attrs_classes == self._INFO_CLASSES:
if self.attrs_stack[-1].get('class') == 'utitle':
self.creation_info['title'] = data.strip()
class PatreonIE(InfoExtractor):
IE_NAME = 'patreon'
_VALID_URL = r'https?://(?:www\.)?patreon\.com/creation\?hid=(.+)'
_TESTS = [
{
'url': 'http://www.patreon.com/creation?hid=743933',
'md5': 'e25505eec1053a6e6813b8ed369875cc',
'name': 'Patreon',
'info_dict': {
'id': '743933',
'ext': 'mp3',
'title': 'Episode 166: David Smalley of Dogma Debate',
'uploader': 'Cognitive Dissonance Podcast',
},
},
]
# Currently Patreon exposes download URL via hidden CSS, so login is not
# needed. Keeping this commented for when this inevitably changes.
'''
def _login(self):
(username, password) = self._get_login_info()
if username is None:
return
login_form = {
'redirectUrl': 'http://www.patreon.com/',
'email': username,
'password': password,
}
request = compat_urllib_request.Request(
'https://www.patreon.com/processLogin',
compat_urllib_parse.urlencode(login_form).encode('utf-8')
)
login_page = self._download_webpage(request, None, note='Logging in as %s' % username)
if re.search(r'onLoginFailed', login_page):
raise ExtractorError('Unable to login, incorrect username and/or password', expected=True)
def _real_initialize(self):
self._login()
'''
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
video_id = mobj.group(1)
info_page = self._download_webpage(url, video_id)
ret = {'id': video_id}
try:
ret['uploader'] = re.search(
r'<strong>(.+)</strong> is creating', info_page
).group(1)
except AttributeError:
pass
parser = PatreonHTMLParser()
parser.get_creation_info(info_page)
if not parser.creation_info.get('url'):
raise ExtractionError('Unable to retrieve creation URL')
ret.update(parser.creation_info)
return ret
Loading…
Cancel
Save