mirror of
				https://gitlab.com/ytdl-org/youtube-dl.git
				synced 2025-11-03 20:27:06 -05:00 
			
		
		
		
	
		
			
				
	
	
		
			157 lines
		
	
	
		
			5.9 KiB
		
	
	
	
		
			Python
		
	
	
	
	
	
			
		
		
	
	
			157 lines
		
	
	
		
			5.9 KiB
		
	
	
	
		
			Python
		
	
	
	
	
	
from __future__ import unicode_literals
 | 
						|
 | 
						|
import re
 | 
						|
import json
 | 
						|
 | 
						|
from .common import InfoExtractor
 | 
						|
from .youtube import YoutubeIE
 | 
						|
from ..utils import (
 | 
						|
    clean_html,
 | 
						|
    ExtractorError,
 | 
						|
    get_element_by_id,
 | 
						|
)
 | 
						|
 | 
						|
 | 
						|
class TechTVMITIE(InfoExtractor):
 | 
						|
    IE_NAME = 'techtv.mit.edu'
 | 
						|
    _VALID_URL = r'https?://techtv\.mit\.edu/(?:videos|embeds)/(?P<id>\d+)'
 | 
						|
 | 
						|
    _TEST = {
 | 
						|
        'url': 'http://techtv.mit.edu/videos/25418-mit-dna-learning-center-set',
 | 
						|
        'md5': '00a3a27ee20d44bcaa0933ccec4a2cf7',
 | 
						|
        'info_dict': {
 | 
						|
            'id': '25418',
 | 
						|
            'ext': 'mp4',
 | 
						|
            'title': 'MIT DNA and Protein Sets',
 | 
						|
            'description': 'md5:46f5c69ce434f0a97e7c628cc142802d',
 | 
						|
        },
 | 
						|
    }
 | 
						|
 | 
						|
    def _real_extract(self, url):
 | 
						|
        video_id = self._match_id(url)
 | 
						|
        raw_page = self._download_webpage(
 | 
						|
            'http://techtv.mit.edu/videos/%s' % video_id, video_id)
 | 
						|
        clean_page = re.compile(r'<!--.*?-->', re.S).sub('', raw_page)
 | 
						|
 | 
						|
        base_url = self._proto_relative_url(self._search_regex(
 | 
						|
            r'ipadUrl: \'(.+?cloudfront.net/)', raw_page, 'base url'), 'http:')
 | 
						|
        formats_json = self._search_regex(
 | 
						|
            r'bitrates: (\[.+?\])', raw_page, 'video formats')
 | 
						|
        formats_mit = json.loads(formats_json)
 | 
						|
        formats = [
 | 
						|
            {
 | 
						|
                'format_id': f['label'],
 | 
						|
                'url': base_url + f['url'].partition(':')[2],
 | 
						|
                'ext': f['url'].partition(':')[0],
 | 
						|
                'format': f['label'],
 | 
						|
                'width': f['width'],
 | 
						|
                'vbr': f['bitrate'],
 | 
						|
            }
 | 
						|
            for f in formats_mit
 | 
						|
        ]
 | 
						|
 | 
						|
        title = get_element_by_id('edit-title', clean_page)
 | 
						|
        description = clean_html(get_element_by_id('edit-description', clean_page))
 | 
						|
        thumbnail = self._search_regex(
 | 
						|
            r'playlist:.*?url: \'(.+?)\'',
 | 
						|
            raw_page, 'thumbnail', flags=re.DOTALL)
 | 
						|
 | 
						|
        return {
 | 
						|
            'id': video_id,
 | 
						|
            'title': title,
 | 
						|
            'formats': formats,
 | 
						|
            'description': description,
 | 
						|
            'thumbnail': thumbnail,
 | 
						|
        }
 | 
						|
 | 
						|
 | 
						|
class MITIE(TechTVMITIE):
 | 
						|
    IE_NAME = 'video.mit.edu'
 | 
						|
    _VALID_URL = r'https?://video\.mit\.edu/watch/(?P<title>[^/]+)'
 | 
						|
 | 
						|
    _TEST = {
 | 
						|
        'url': 'http://video.mit.edu/watch/the-government-is-profiling-you-13222/',
 | 
						|
        'md5': '7db01d5ccc1895fc5010e9c9e13648da',
 | 
						|
        'info_dict': {
 | 
						|
            'id': '21783',
 | 
						|
            'ext': 'mp4',
 | 
						|
            'title': 'The Government is Profiling You',
 | 
						|
            'description': 'md5:ad5795fe1e1623b73620dbfd47df9afd',
 | 
						|
        },
 | 
						|
    }
 | 
						|
 | 
						|
    def _real_extract(self, url):
 | 
						|
        mobj = re.match(self._VALID_URL, url)
 | 
						|
        page_title = mobj.group('title')
 | 
						|
        webpage = self._download_webpage(url, page_title)
 | 
						|
        embed_url = self._search_regex(
 | 
						|
            r'<iframe .*?src="(.+?)"', webpage, 'embed url')
 | 
						|
        return self.url_result(embed_url, ie='TechTVMIT')
 | 
						|
 | 
						|
 | 
						|
class OCWMITIE(InfoExtractor):
 | 
						|
    IE_NAME = 'ocw.mit.edu'
 | 
						|
    _VALID_URL = r'^http://ocw\.mit\.edu/courses/(?P<topic>[a-z0-9\-]+)'
 | 
						|
    _BASE_URL = 'http://ocw.mit.edu/'
 | 
						|
 | 
						|
    _TESTS = [
 | 
						|
        {
 | 
						|
            'url': 'http://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-041-probabilistic-systems-analysis-and-applied-probability-fall-2010/video-lectures/lecture-7-multiple-variables-expectations-independence/',
 | 
						|
            'info_dict': {
 | 
						|
                'id': 'EObHWIEKGjA',
 | 
						|
                'ext': 'mp4',
 | 
						|
                'title': 'Lecture 7: Multiple Discrete Random Variables: Expectations, Conditioning, Independence',
 | 
						|
                'description': 'In this lecture, the professor discussed multiple random variables, expectations, and binomial distribution.',
 | 
						|
                'upload_date': '20121109',
 | 
						|
                'uploader_id': 'MIT',
 | 
						|
                'uploader': 'MIT OpenCourseWare',
 | 
						|
            }
 | 
						|
        },
 | 
						|
        {
 | 
						|
            'url': 'http://ocw.mit.edu/courses/mathematics/18-01sc-single-variable-calculus-fall-2010/1.-differentiation/part-a-definition-and-basic-rules/session-1-introduction-to-derivatives/',
 | 
						|
            'info_dict': {
 | 
						|
                'id': '7K1sB05pE0A',
 | 
						|
                'ext': 'mp4',
 | 
						|
                'title': 'Session 1: Introduction to Derivatives',
 | 
						|
                'upload_date': '20090818',
 | 
						|
                'uploader_id': 'MIT',
 | 
						|
                'uploader': 'MIT OpenCourseWare',
 | 
						|
                'description': 'This section contains lecture video excerpts, lecture notes, an interactive mathlet with supporting documents, and problem solving videos.',
 | 
						|
            }
 | 
						|
        }
 | 
						|
    ]
 | 
						|
 | 
						|
    def _real_extract(self, url):
 | 
						|
        mobj = re.match(self._VALID_URL, url)
 | 
						|
        topic = mobj.group('topic')
 | 
						|
 | 
						|
        webpage = self._download_webpage(url, topic)
 | 
						|
        title = self._html_search_meta('WT.cg_s', webpage)
 | 
						|
        description = self._html_search_meta('Description', webpage)
 | 
						|
 | 
						|
        # search for call to ocw_embed_chapter_media(container_id, media_url, provider, page_url, image_url, start, stop, captions_file)
 | 
						|
        embed_chapter_media = re.search(r'ocw_embed_chapter_media\((.+?)\)', webpage)
 | 
						|
        if embed_chapter_media:
 | 
						|
            metadata = re.sub(r'[\'"]', '', embed_chapter_media.group(1))
 | 
						|
            metadata = re.split(r', ?', metadata)
 | 
						|
            yt = metadata[1]
 | 
						|
        else:
 | 
						|
            # search for call to ocw_embed_chapter_media(container_id, media_url, provider, page_url, image_url, captions_file)
 | 
						|
            embed_media = re.search(r'ocw_embed_media\((.+?)\)', webpage)
 | 
						|
            if embed_media:
 | 
						|
                metadata = re.sub(r'[\'"]', '', embed_media.group(1))
 | 
						|
                metadata = re.split(r', ?', metadata)
 | 
						|
                yt = metadata[1]
 | 
						|
            else:
 | 
						|
                raise ExtractorError('Unable to find embedded YouTube video.')
 | 
						|
        video_id = YoutubeIE.extract_id(yt)
 | 
						|
 | 
						|
        return {
 | 
						|
            '_type': 'url_transparent',
 | 
						|
            'id': video_id,
 | 
						|
            'title': title,
 | 
						|
            'description': description,
 | 
						|
            'url': yt,
 | 
						|
            'ie_key': 'Youtube',
 | 
						|
        }
 |