mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-23 19:35:19 +01:00
56 lines
1.9 KiB
Python
56 lines
1.9 KiB
Python
from __future__ import unicode_literals
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import unified_strdate
|
|
|
|
|
|
class ATTTechChannelIE(InfoExtractor):
|
|
_VALID_URL = r'https?://techchannel\.att\.com/play-video\.cfm/([^/]+/)*(?P<id>.+)'
|
|
_TEST = {
|
|
'url': 'http://techchannel.att.com/play-video.cfm/2014/1/27/ATT-Archives-The-UNIX-System-Making-Computers-Easier-to-Use',
|
|
'info_dict': {
|
|
'id': '11316',
|
|
'display_id': 'ATT-Archives-The-UNIX-System-Making-Computers-Easier-to-Use',
|
|
'ext': 'flv',
|
|
'title': 'AT&T Archives : The UNIX System: Making Computers Easier to Use',
|
|
'description': 'A 1982 film about UNIX is the foundation for software in use around Bell Labs and AT&T.',
|
|
'thumbnail': r're:^https?://.*\.jpg$',
|
|
'upload_date': '20140127',
|
|
},
|
|
'params': {
|
|
# rtmp download
|
|
'skip_download': True,
|
|
},
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
display_id = self._match_id(url)
|
|
|
|
webpage = self._download_webpage(url, display_id)
|
|
|
|
video_url = self._search_regex(
|
|
r"url\s*:\s*'(rtmp://[^']+)'",
|
|
webpage, 'video URL')
|
|
|
|
video_id = self._search_regex(
|
|
r'mediaid\s*=\s*(\d+)',
|
|
webpage, 'video id', fatal=False)
|
|
|
|
title = self._og_search_title(webpage)
|
|
description = self._og_search_description(webpage)
|
|
thumbnail = self._og_search_thumbnail(webpage)
|
|
upload_date = unified_strdate(self._search_regex(
|
|
r'[Rr]elease\s+date:\s*(\d{1,2}/\d{1,2}/\d{4})',
|
|
webpage, 'upload date', fatal=False), False)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'display_id': display_id,
|
|
'url': video_url,
|
|
'ext': 'flv',
|
|
'title': title,
|
|
'description': description,
|
|
'thumbnail': thumbnail,
|
|
'upload_date': upload_date,
|
|
}
|