mirror of
https://github.com/l1ving/youtube-dl
synced 2024-11-22 21:53:32 +08:00
133 lines
4.7 KiB
Python
133 lines
4.7 KiB
Python
# -*- coding: utf-8 -*-
|
|
from __future__ import unicode_literals
|
|
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import int_or_none
|
|
|
|
|
|
class TumblrIE(InfoExtractor):
|
|
_VALID_URL = r'http://(?P<blog_name>.*?)\.tumblr\.com/(?:post|video)/(?P<id>[0-9]+)(?:$|[/?#])'
|
|
_TESTS = [{
|
|
'url': 'http://tatianamaslanydaily.tumblr.com/post/54196191430/orphan-black-dvd-extra-behind-the-scenes',
|
|
'md5': '479bb068e5b16462f5176a6828829767',
|
|
'info_dict': {
|
|
'id': '54196191430',
|
|
'ext': 'mp4',
|
|
'title': 'tatiana maslany news, Orphan Black || DVD extra - behind the scenes ↳...',
|
|
'description': 'md5:37db8211e40b50c7c44e95da14f630b7',
|
|
'thumbnail': 're:http://.*\.jpg',
|
|
}
|
|
}, {
|
|
'url': 'http://5sostrum.tumblr.com/post/90208453769/yall-forgetting-the-greatest-keek-of-them-all',
|
|
'md5': 'bf348ef8c0ef84fbf1cbd6fa6e000359',
|
|
'info_dict': {
|
|
'id': '90208453769',
|
|
'ext': 'mp4',
|
|
'title': '5SOS STRUM ;]',
|
|
'description': 'md5:dba62ac8639482759c8eb10ce474586a',
|
|
'thumbnail': 're:http://.*\.jpg',
|
|
}
|
|
}, {
|
|
'url': 'http://hdvideotest.tumblr.com/post/130323439814/test-description-for-my-hd-video',
|
|
'md5': '7ae503065ad150122dc3089f8cf1546c',
|
|
'info_dict': {
|
|
'id': '130323439814',
|
|
'ext': 'mp4',
|
|
'title': 'HD Video Testing \u2014 Test description for my HD video',
|
|
'description': 'md5:97cc3ab5fcd27ee4af6356701541319c',
|
|
'thumbnail': 're:http://.*\.jpg',
|
|
},
|
|
'params': {
|
|
'format': 'hd',
|
|
},
|
|
}, {
|
|
'url': 'http://naked-yogi.tumblr.com/post/118312946248/naked-smoking-stretching',
|
|
'md5': 'de07e5211d60d4f3a2c3df757ea9f6ab',
|
|
'info_dict': {
|
|
'id': 'Wmur',
|
|
'ext': 'mp4',
|
|
'title': 'naked smoking & stretching',
|
|
'upload_date': '20150506',
|
|
'timestamp': 1430931613,
|
|
'age_limit': 18,
|
|
'uploader_id': '1638622',
|
|
'uploader': 'naked-yogi',
|
|
},
|
|
'add_ie': ['Vidme'],
|
|
}, {
|
|
'url': 'http://camdamage.tumblr.com/post/98846056295/',
|
|
'md5': 'a9e0c8371ea1ca306d6554e3fecf50b6',
|
|
'info_dict': {
|
|
'id': '105463834',
|
|
'ext': 'mp4',
|
|
'title': 'Cam Damage-HD 720p',
|
|
'uploader': 'John Moyer',
|
|
'uploader_id': 'user32021558',
|
|
},
|
|
'add_ie': ['Vimeo'],
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
m_url = re.match(self._VALID_URL, url)
|
|
video_id = m_url.group('id')
|
|
blog = m_url.group('blog_name')
|
|
|
|
url = 'http://%s.tumblr.com/post/%s/' % (blog, video_id)
|
|
webpage, urlh = self._download_webpage_handle(url, video_id)
|
|
|
|
iframe_url = self._search_regex(
|
|
r'src=\'(https?://www\.tumblr\.com/video/[^\']+)\'',
|
|
webpage, 'iframe url', default=None)
|
|
if iframe_url is None:
|
|
return self.url_result(urlh.geturl(), 'Generic')
|
|
|
|
iframe = self._download_webpage(iframe_url, video_id, 'Downloading iframe page')
|
|
|
|
duration = None
|
|
sources = []
|
|
|
|
sd_url = self._search_regex(
|
|
r'<source[^>]+src=(["\'])(?P<url>.+?)\1', iframe,
|
|
'sd video url', default=None, group='url')
|
|
if sd_url:
|
|
sources.append((sd_url, 'sd'))
|
|
|
|
options = self._parse_json(
|
|
self._search_regex(
|
|
r'data-crt-options=(["\'])(?P<options>.+?)\1', iframe,
|
|
'hd video url', default='', group='options'),
|
|
video_id, fatal=False)
|
|
if options:
|
|
duration = int_or_none(options.get('duration'))
|
|
hd_url = options.get('hdUrl')
|
|
if hd_url:
|
|
sources.append((hd_url, 'hd'))
|
|
|
|
formats = [{
|
|
'url': video_url,
|
|
'ext': 'mp4',
|
|
'format_id': format_id,
|
|
'height': int_or_none(self._search_regex(
|
|
r'/(\d{3,4})$', video_url, 'height', default=None)),
|
|
'quality': quality,
|
|
} for quality, (video_url, format_id) in enumerate(sources)]
|
|
|
|
self._sort_formats(formats)
|
|
|
|
# The only place where you can get a title, it's not complete,
|
|
# but searching in other places doesn't work for all videos
|
|
video_title = self._html_search_regex(
|
|
r'(?s)<title>(?P<title>.*?)(?: \| Tumblr)?</title>',
|
|
webpage, 'title')
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': video_title,
|
|
'description': self._og_search_description(webpage, default=None),
|
|
'thumbnail': self._og_search_thumbnail(webpage, default=None),
|
|
'duration': duration,
|
|
'formats': formats,
|
|
}
|