[curiositystream] Add new extractor
This commit is contained in:
		
							
								
								
									
										128
									
								
								youtube_dl/extractor/curiositystream.py
									
									
									
									
									
										Normal file
									
								
							
							
						
						
									
										128
									
								
								youtube_dl/extractor/curiositystream.py
									
									
									
									
									
										Normal file
									
								
							@@ -0,0 +1,128 @@
 | 
			
		||||
# coding: utf-8
 | 
			
		||||
from __future__ import unicode_literals
 | 
			
		||||
 | 
			
		||||
from .common import InfoExtractor
 | 
			
		||||
from ..utils import (
 | 
			
		||||
    int_or_none,
 | 
			
		||||
    urlencode_postdata,
 | 
			
		||||
    compat_str,
 | 
			
		||||
    ExtractorError,
 | 
			
		||||
)
 | 
			
		||||
 | 
			
		||||
 | 
			
		||||
class CuriosityStreamBaseIE(InfoExtractor):
 | 
			
		||||
    _NETRC_MACHINE = 'curiositystream'
 | 
			
		||||
    _auth_token = None
 | 
			
		||||
    _API_BASE_URL = 'https://api.curiositystream.com/v1/'
 | 
			
		||||
 | 
			
		||||
    def _handle_errors(self, result):
 | 
			
		||||
        error = result.get('error', {}).get('message')
 | 
			
		||||
        if error:
 | 
			
		||||
            if isinstance(error, dict):
 | 
			
		||||
                error = ', '.join(error.values())
 | 
			
		||||
            raise ExtractorError(
 | 
			
		||||
                '%s said: %s' % (self.IE_NAME, error), expected=True)
 | 
			
		||||
 | 
			
		||||
    def _call_api(self, path, video_id):
 | 
			
		||||
        headers = {}
 | 
			
		||||
        if self._auth_token:
 | 
			
		||||
            headers['X-Auth-Token'] = self._auth_token
 | 
			
		||||
        result = self._download_json(
 | 
			
		||||
            self._API_BASE_URL + path, video_id, headers=headers)
 | 
			
		||||
        self._handle_errors(result)
 | 
			
		||||
        return result['data']
 | 
			
		||||
 | 
			
		||||
    def _real_initialize(self):
 | 
			
		||||
        if not self._auth_token:
 | 
			
		||||
            user = self._downloader.cache.load('curiositystream', 'user') or {}
 | 
			
		||||
            self._auth_token = user.get('auth_token')
 | 
			
		||||
            if not self._auth_token:
 | 
			
		||||
                (email, password) = self._get_login_info()
 | 
			
		||||
                if email is None:
 | 
			
		||||
                    return
 | 
			
		||||
                result = self._download_json(
 | 
			
		||||
                    self._API_BASE_URL + 'login', None, data=urlencode_postdata({
 | 
			
		||||
                        'email': email,
 | 
			
		||||
                        'password': password,
 | 
			
		||||
                    }))
 | 
			
		||||
                self._handle_errors(result)
 | 
			
		||||
                self._auth_token = result['message']['auth_token']
 | 
			
		||||
                self._downloader.cache.store(
 | 
			
		||||
                    'curiositystream', 'user', {
 | 
			
		||||
                        'auth_token': self._auth_token,
 | 
			
		||||
                    })
 | 
			
		||||
 | 
			
		||||
    def _extract_media_info(self, media):
 | 
			
		||||
        video_id = compat_str(media['id'])
 | 
			
		||||
        limelight_media_id = media['limelight_media_id']
 | 
			
		||||
        title = media['title']
 | 
			
		||||
 | 
			
		||||
        subtitles = {}
 | 
			
		||||
        for closed_caption in media.get('closed_captions', []):
 | 
			
		||||
            sub_url = closed_caption.get('file')
 | 
			
		||||
            if not sub_url:
 | 
			
		||||
                continue
 | 
			
		||||
            lang = closed_caption.get('code') or closed_caption.get('language') or 'en'
 | 
			
		||||
            subtitles.setdefault(lang, []).append({
 | 
			
		||||
                'url': sub_url,
 | 
			
		||||
            })
 | 
			
		||||
 | 
			
		||||
        return {
 | 
			
		||||
            '_type': 'url_transparent',
 | 
			
		||||
            'id': video_id,
 | 
			
		||||
            'url': 'limelight:media:' + limelight_media_id,
 | 
			
		||||
            'title': title,
 | 
			
		||||
            'description': media.get('description'),
 | 
			
		||||
            'thumbnail': media.get('image_large') or media.get('image_medium') or media.get('image_small'),
 | 
			
		||||
            'duration': int_or_none(media.get('duration')),
 | 
			
		||||
            'tags': media.get('tags'),
 | 
			
		||||
            'subtitles': subtitles,
 | 
			
		||||
            'ie_key': 'LimelightMedia',
 | 
			
		||||
        }
 | 
			
		||||
 | 
			
		||||
 | 
			
		||||
class CuriosityStreamIE(CuriosityStreamBaseIE):
 | 
			
		||||
    IE_NAME = 'curiositystream'
 | 
			
		||||
    _VALID_URL = r'https?://app\.curiositystream\.com/video/(?P<id>\d+)'
 | 
			
		||||
    _TEST = {
 | 
			
		||||
        'url': 'https://app.curiositystream.com/video/2',
 | 
			
		||||
        'md5': 'a0074c190e6cddaf86900b28d3e9ee7a',
 | 
			
		||||
        'info_dict': {
 | 
			
		||||
            'id': '2',
 | 
			
		||||
            'ext': 'mp4',
 | 
			
		||||
            'title': 'How Did You Develop The Internet?',
 | 
			
		||||
            'description': 'Vint Cerf, Google\'s Chief Internet Evangelist, describes how he and Bob Kahn created the internet.',
 | 
			
		||||
            'timestamp': 1448388615,
 | 
			
		||||
            'upload_date': '20151124',
 | 
			
		||||
        }
 | 
			
		||||
    }
 | 
			
		||||
 | 
			
		||||
    def _real_extract(self, url):
 | 
			
		||||
        video_id = self._match_id(url)
 | 
			
		||||
        media = self._call_api('media/' + video_id, video_id)
 | 
			
		||||
        return self._extract_media_info(media)
 | 
			
		||||
 | 
			
		||||
 | 
			
		||||
class CuriosityStreamCollectionIE(CuriosityStreamBaseIE):
 | 
			
		||||
    IE_NAME = 'curiositystream:collection'
 | 
			
		||||
    _VALID_URL = r'https?://app\.curiositystream\.com/collection/(?P<id>\d+)'
 | 
			
		||||
    _TEST = {
 | 
			
		||||
        'url': 'https://app.curiositystream.com/collection/2',
 | 
			
		||||
        'info_dict': {
 | 
			
		||||
            'id': '2',
 | 
			
		||||
            'title': 'Curious Minds: The Internet',
 | 
			
		||||
            'description': 'How is the internet shaping our lives in the 21st Century?',
 | 
			
		||||
        },
 | 
			
		||||
        'playlist_mincount': 17,
 | 
			
		||||
    }
 | 
			
		||||
 | 
			
		||||
    def _real_extract(self, url):
 | 
			
		||||
        collection_id = self._match_id(url)
 | 
			
		||||
        collection = self._call_api(
 | 
			
		||||
            'collections/' + collection_id, collection_id)
 | 
			
		||||
        entries = []
 | 
			
		||||
        for media in collection.get('media', []):
 | 
			
		||||
            entries.append(self._extract_media_info(media))
 | 
			
		||||
        return self.playlist_result(
 | 
			
		||||
            entries, collection_id,
 | 
			
		||||
            collection.get('title'), collection.get('description'))
 | 
			
		||||
@@ -194,6 +194,10 @@ from .ctsnews import CtsNewsIE
 | 
			
		||||
from .ctv import CTVIE
 | 
			
		||||
from .ctvnews import CTVNewsIE
 | 
			
		||||
from .cultureunplugged import CultureUnpluggedIE
 | 
			
		||||
from .curiositystream import (
 | 
			
		||||
    CuriosityStreamIE,
 | 
			
		||||
    CuriosityStreamCollectionIE,
 | 
			
		||||
)
 | 
			
		||||
from .cwtv import CWTVIE
 | 
			
		||||
from .dailymail import DailyMailIE
 | 
			
		||||
from .dailymotion import (
 | 
			
		||||
 
 | 
			
		||||
		Reference in New Issue
	
	Block a user