X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/5ea61d104de9ca8a7d19d63d83173eb7391081d2..8b2307d66b10597b66d9491668d75d0e5ae5ab0c:/youtube_dl/extractor/patreon.py diff --git a/youtube_dl/extractor/patreon.py b/youtube_dl/extractor/patreon.py index 6cdc263..761a4b1 100644 --- a/youtube_dl/extractor/patreon.py +++ b/youtube_dl/extractor/patreon.py @@ -1,61 +1,75 @@ -# encoding: utf-8 +# coding: utf-8 from __future__ import unicode_literals from .common import InfoExtractor from ..utils import ( - js_to_json, + clean_html, + determine_ext, + int_or_none, + KNOWN_EXTENSIONS, + mimetype2ext, + parse_iso8601, + str_or_none, + try_get, ) class PatreonIE(InfoExtractor): - _VALID_URL = r'https?://(?:www\.)?patreon\.com/creation\?hid=(?P[^&#]+)' - _TESTS = [ - { - 'url': 'http://www.patreon.com/creation?hid=743933', - 'md5': 'e25505eec1053a6e6813b8ed369875cc', - 'info_dict': { - 'id': '743933', - 'ext': 'mp3', - 'title': 'Episode 166: David Smalley of Dogma Debate', - 'uploader': 'Cognitive Dissonance Podcast', - 'thumbnail': 're:^https?://.*$', - }, + _VALID_URL = r'https?://(?:www\.)?patreon\.com/(?:creation\?hid=|posts/(?:[\w-]+-)?)(?P\d+)' + _TESTS = [{ + 'url': 'http://www.patreon.com/creation?hid=743933', + 'md5': 'e25505eec1053a6e6813b8ed369875cc', + 'info_dict': { + 'id': '743933', + 'ext': 'mp3', + 'title': 'Episode 166: David Smalley of Dogma Debate', + 'description': 'md5:713b08b772cd6271b9f3906683cfacdf', + 'uploader': 'Cognitive Dissonance Podcast', + 'thumbnail': 're:^https?://.*$', + 'timestamp': 1406473987, + 'upload_date': '20140727', + 'uploader_id': '87145', + }, + }, { + 'url': 'http://www.patreon.com/creation?hid=754133', + 'md5': '3eb09345bf44bf60451b8b0b81759d0a', + 'info_dict': { + 'id': '754133', + 'ext': 'mp3', + 'title': 'CD 167 Extra', + 'uploader': 'Cognitive Dissonance Podcast', + 'thumbnail': 're:^https?://.*$', }, - { - 'url': 'http://www.patreon.com/creation?hid=754133', - 'md5': '3eb09345bf44bf60451b8b0b81759d0a', - 'info_dict': { - 'id': '754133', - 'ext': 'mp3', - 'title': 'CD 167 Extra', - 'uploader': 'Cognitive Dissonance Podcast', - 'thumbnail': 're:^https?://.*$', - }, + 'skip': 'Patron-only content', + }, { + 'url': 'https://www.patreon.com/creation?hid=1682498', + 'info_dict': { + 'id': 'SU4fj_aEMVw', + 'ext': 'mp4', + 'title': 'I\'m on Patreon!', + 'uploader': 'TraciJHines', + 'thumbnail': 're:^https?://.*$', + 'upload_date': '20150211', + 'description': 'md5:c5a706b1f687817a3de09db1eb93acd4', + 'uploader_id': 'TraciJHines', }, - { - 'url': 'https://www.patreon.com/creation?hid=1682498', - 'info_dict': { - 'id': 'SU4fj_aEMVw', - 'ext': 'mp4', - 'title': 'I\'m on Patreon!', - 'uploader': 'TraciJHines', - 'thumbnail': 're:^https?://.*$', - 'upload_date': '20150211', - 'description': 'md5:c5a706b1f687817a3de09db1eb93acd4', - 'uploader_id': 'TraciJHines', - }, - 'params': { - 'noplaylist': True, - 'skip_download': True, - } + 'params': { + 'noplaylist': True, + 'skip_download': True, } - ] + }, { + 'url': 'https://www.patreon.com/posts/episode-166-of-743933', + 'only_matching': True, + }, { + 'url': 'https://www.patreon.com/posts/743933', + 'only_matching': True, + }] # 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() + username, password = self._get_login_info() if username is None: return @@ -65,11 +79,11 @@ class PatreonIE(InfoExtractor): 'password': password, } - request = compat_urllib_request.Request( + request = sanitized_Request( 'https://www.patreon.com/processLogin', - compat_urllib_parse.urlencode(login_form).encode('utf-8') + compat_urllib_parse_urlencode(login_form).encode('utf-8') ) - login_page = self._download_webpage(request, None, note='Logging in as %s' % username) + login_page = self._download_webpage(request, None, note='Logging in') if re.search(r'onLoginFailed', login_page): raise ExtractorError('Unable to login, incorrect username and/or password', expected=True) @@ -80,38 +94,63 @@ class PatreonIE(InfoExtractor): def _real_extract(self, url): video_id = self._match_id(url) - webpage = self._download_webpage(url, video_id) - title = self._og_search_title(webpage).strip() - - attach_fn = self._html_search_regex( - r'
', - webpage, 'attachment URL', default=None) - embed = self._html_search_regex( - r']+id="watchCreation"[^>]*>\s*]+src="([^"]+)"', - webpage, 'embedded URL', default=None) - - if attach_fn is not None: - video_url = 'http://www.patreon.com' + attach_fn - thumbnail = self._og_search_thumbnail(webpage) - uploader = self._html_search_regex( - r'(.*?) is creating', webpage, 'uploader') - elif embed is not None: - return self.url_result(embed) - else: - playlist = self._parse_json(self._search_regex( - r'(?s)new\s+jPlayerPlaylist\(\s*\{\s*[^}]*},\s*(\[.*?,?\s*\])', - webpage, 'playlist JSON'), - video_id, transform_source=js_to_json) - data = playlist[0] - video_url = self._proto_relative_url(data['mp3']) - thumbnail = self._proto_relative_url(data.get('cover')) - uploader = data.get('artist') - - return { + post = self._download_json( + 'https://www.patreon.com/api/posts/' + video_id, video_id, query={ + 'fields[media]': 'download_url,mimetype,size_bytes', + 'fields[post]': 'comment_count,content,embed,image,like_count,post_file,published_at,title', + 'fields[user]': 'full_name,url', + 'json-api-use-default-includes': 'false', + 'include': 'media,user', + }) + attributes = post['data']['attributes'] + title = attributes['title'].strip() + image = attributes.get('image') or {} + info = { 'id': video_id, - 'url': video_url, - 'ext': 'mp3', 'title': title, - 'uploader': uploader, - 'thumbnail': thumbnail, + 'description': clean_html(attributes.get('content')), + 'thumbnail': image.get('large_url') or image.get('url'), + 'timestamp': parse_iso8601(attributes.get('published_at')), + 'like_count': int_or_none(attributes.get('like_count')), + 'comment_count': int_or_none(attributes.get('comment_count')), } + + for i in post.get('included', []): + i_type = i.get('type') + if i_type == 'media': + media_attributes = i.get('attributes') or {} + download_url = media_attributes.get('download_url') + ext = mimetype2ext(media_attributes.get('mimetype')) + if download_url and ext in KNOWN_EXTENSIONS: + info.update({ + 'ext': ext, + 'filesize': int_or_none(media_attributes.get('size_bytes')), + 'url': download_url, + }) + elif i_type == 'user': + user_attributes = i.get('attributes') + if user_attributes: + info.update({ + 'uploader': user_attributes.get('full_name'), + 'uploader_id': str_or_none(i.get('id')), + 'uploader_url': user_attributes.get('url'), + }) + + if not info.get('url'): + embed_url = try_get(attributes, lambda x: x['embed']['url']) + if embed_url: + info.update({ + '_type': 'url', + 'url': embed_url, + }) + + if not info.get('url'): + post_file = attributes['post_file'] + ext = determine_ext(post_file.get('name')) + if ext in KNOWN_EXTENSIONS: + info.update({ + 'ext': ext, + 'url': post_file['url'], + }) + + return info