|
|
- # coding: utf-8
- from __future__ import unicode_literals
-
- import re
-
- from .common import InfoExtractor
- from ..compat import (
- compat_str,
- compat_urllib_parse_unquote,
- )
- from ..utils import (
- ExtractorError,
- int_or_none,
- str_or_none,
- strip_or_none,
- try_get,
- urlencode_postdata,
- )
-
-
- class GaiaIE(InfoExtractor):
- _VALID_URL = r'https?://(?:www\.)?gaia\.com/video/(?P<id>[^/?]+).*?\bfullplayer=(?P<type>feature|preview)'
- _TESTS = [{
- 'url': 'https://www.gaia.com/video/connecting-universal-consciousness?fullplayer=feature',
- 'info_dict': {
- 'id': '89356',
- 'ext': 'mp4',
- 'title': 'Connecting with Universal Consciousness',
- 'description': 'md5:844e209ad31b7d31345f5ed689e3df6f',
- 'upload_date': '20151116',
- 'timestamp': 1447707266,
- 'duration': 936,
- },
- 'params': {
- # m3u8 download
- 'skip_download': True,
- },
- }, {
- 'url': 'https://www.gaia.com/video/connecting-universal-consciousness?fullplayer=preview',
- 'info_dict': {
- 'id': '89351',
- 'ext': 'mp4',
- 'title': 'Connecting with Universal Consciousness',
- 'description': 'md5:844e209ad31b7d31345f5ed689e3df6f',
- 'upload_date': '20151116',
- 'timestamp': 1447707266,
- 'duration': 53,
- },
- 'params': {
- # m3u8 download
- 'skip_download': True,
- },
- }]
- _NETRC_MACHINE = 'gaia'
- _jwt = None
-
- def _real_initialize(self):
- auth = self._get_cookies('https://www.gaia.com/').get('auth')
- if auth:
- auth = self._parse_json(
- compat_urllib_parse_unquote(auth.value),
- None, fatal=False)
- if not auth:
- username, password = self._get_login_info()
- if username is None:
- return
- auth = self._download_json(
- 'https://auth.gaia.com/v1/login',
- None, data=urlencode_postdata({
- 'username': username,
- 'password': password
- }))
- if auth.get('success') is False:
- raise ExtractorError(', '.join(auth['messages']), expected=True)
- if auth:
- self._jwt = auth.get('jwt')
-
- def _real_extract(self, url):
- display_id, vtype = re.search(self._VALID_URL, url).groups()
- node_id = self._download_json(
- 'https://brooklyn.gaia.com/pathinfo', display_id, query={
- 'path': 'video/' + display_id,
- })['id']
- node = self._download_json(
- 'https://brooklyn.gaia.com/node/%d' % node_id, node_id)
- vdata = node[vtype]
- media_id = compat_str(vdata['nid'])
- title = node['title']
-
- headers = None
- if self._jwt:
- headers = {'Authorization': 'Bearer ' + self._jwt}
- media = self._download_json(
- 'https://brooklyn.gaia.com/media/' + media_id,
- media_id, headers=headers)
- formats = self._extract_m3u8_formats(
- media['mediaUrls']['bcHLS'], media_id, 'mp4')
- self._sort_formats(formats)
-
- subtitles = {}
- text_tracks = media.get('textTracks', {})
- for key in ('captions', 'subtitles'):
- for lang, sub_url in text_tracks.get(key, {}).items():
- subtitles.setdefault(lang, []).append({
- 'url': sub_url,
- })
-
- fivestar = node.get('fivestar', {})
- fields = node.get('fields', {})
-
- def get_field_value(key, value_key='value'):
- return try_get(fields, lambda x: x[key][0][value_key])
-
- return {
- 'id': media_id,
- 'display_id': display_id,
- 'title': title,
- 'formats': formats,
- 'description': strip_or_none(get_field_value('body') or get_field_value('teaser')),
- 'timestamp': int_or_none(node.get('created')),
- 'subtitles': subtitles,
- 'duration': int_or_none(vdata.get('duration')),
- 'like_count': int_or_none(try_get(fivestar, lambda x: x['up_count']['value'])),
- 'dislike_count': int_or_none(try_get(fivestar, lambda x: x['down_count']['value'])),
- 'comment_count': int_or_none(node.get('comment_count')),
- 'series': try_get(node, lambda x: x['series']['title'], compat_str),
- 'season_number': int_or_none(get_field_value('season')),
- 'season_id': str_or_none(get_field_value('series_nid', 'nid')),
- 'episode_number': int_or_none(get_field_value('episode')),
- }
|